abstract for alzheimer's research paper

• Published: 09 July 2009

Alzheimer's disease therapeutic research: the path forward

• Paul S Aisen 1  

Alzheimers Res Ther volume  1 , Article number:  2 ( 2009 ) Cite this article

13k Accesses

40 Citations

Metrics details

The field of Alzheimer's disease therapeutic research seems poised to bring to clinic the next generation of treatments, moving beyond symptomatic benefits to modification of the underlying neurobiology of the disease. But a series of recent trials has had disappointingly negative results that raise questions about our drug development strategies. Consideration of ongoing programs demonstrates difficult pitfalls. But a clear path forward is emerging. Successful strategies will utilize newly available tools to reconsider issues of diagnosis, assessment and analysis, facilitating the study of new treatments at early stages in the disease process at which they are most likely to yield major clinical benefits.

Alzheimer's disease (AD) was described just over 100 years ago as an uncommon devastating dementia affecting people in middle age. In the 1970s, Dr Robert Katzman demonstrated that AD is in fact an epidemic of enormous proportions, affecting a substantial segment of the aging population [ 1 ]. This spurred basic and clinical therapeutic research activity, leading to the development of modestly effective symptomatic treatments. While efforts to improve cognitive and behavioral symptoms continue, the major focus of AD therapeutic research is now disease modification – that is, slowing the progression of the underlying neurobiology of AD [ 2 ]. Alois Alzheimer described neuronal loss with formation of plaques and tangles. Today's leading programs target the biochemical pathways leading to amyloid accumulation and neurofibrillary tangle formation, and aim to protect neuronal cells and synapses against dysfunction and destruction.

Clear targets have been identified. Two enzymes, beta secretase and the gamma secretase complex, appear to be essential for cleavage of the amyloidogenic Aβ fragment from its transmembrane amyloid precursor protein (APP); inhibition of one or both is expected to reduce amyloid accumulation [ 3 ]. Genetic evidence provides strong support for these approaches: all known genetic causes of AD either increase the expression of APP or increase the generation of amyloidogenic fragments. There is also hope that inhibiting receptors that mediate Aβ trafficking [ 4 , 5 ] and toxicity [ 5 , 6 ] may modify AD neurodegeneration. Tangle-related targets, including kinase inhibitors aiming to reduce the hyperphosphorylation that characterizes the abnormal tau protein in tangles [ 7 ], have seen more limited efforts. Neurotrophic programs include direct neurosurgical delivery of nerve growth factor to the nucleus basalis [ 8 ] using a viral vector.

But despite the proliferation of clinical development programs, early results have been quite disappointing. The first two anti-amyloid drugs to reach the pivotal stage of development, tramiprosate and tarenflurbil, failed in phase III. What are the implications of these failures? Are the targets wrong? Can the field afford to invest the huge efforts and funds necessary to continue to test potential disease-modifying treatments? Is there any realistic likelihood of success?

Tramiprosate

Tramiprosate (also referred to as homotaurine and 3-amino-1-propanesulfonic acid, or 3APS) is an Aβ-binding compound that was developed using in vitro and in vivo model systems [ 9 ] that left some uncertainty regarding the brain concentration necessary for a pharmacodynamic effect in human AD. While a phase II study did suggest a reduction in cerebrospinal fluid Aβ in AD subjects treated with tramiprosate [ 10 ], it was unknown whether the degree of reduction would be sufficient to translate into clinical benefit. The small and brief phase II program was not designed to demonstrate clinically a disease-slowing effect; as expected, subjects in the 12 week treatment trial treated with placebo showed no decline, and, therefore, there was no possibility of showing reduced decline with treatment. The development of tramiprosate as a pharmaceutical treatment for AD was halted when the first phase III trial failed to demonstrate significant beneficial effects on the primary analysis of cognitive and clinical outcomes.

Tarenflurbil

Similar problems were faced in the tarenflurbil development program. In vitro and in vivo , the drug clearly modulates gamma secretase activity, reducing generation of Aβ [ 11 , 12 ]. In early human studies, however, there was no strong biomarker evidence of amyloid reduction in cerebrospinal fluid (CSF), and concerns about inadequate brain concentrations in humans were voiced. The phase II trial, though relatively large, was underpowered to show a disease-modifying effect, and the primary analysis of the impact of treatment on cognitive and functional outcomes was negative [ 13 ]. However, post hoc analyses appeared to be consistent with a beneficial drug effect, leading to the launch of large phase III trials. The program was terminated when the first phase III trial showed no evidence of beneficial effect.

Bapineuzumab

A somewhat similar situation has arisen in the development program of bapineuzumab, a monoclonal amino terminus-specific anti-amyloid antibody [ 14 ]. Perhaps misled by encouraging cognitive data from a small phase I trial hinting at a symptomatic effect, the sponsors sought evidence of efficacy in the modestly sized phase II program. Though there was evidence of benefit in a number of secondary analyses, particularly in the apolipoprotein E ε4 negative subgroup, the primary cognitive efficacy analysis was negative. The sponsors, Elan and Wyeth, have nonetheless proceeded with a very large phase III program.

Why have so many programs yielded discouraging efficacy data in phase II and III clinical trials? In phase II, the problem may be primarily one of statistical power. Most programs seek to be able to demonstrate a 25% to 33% slowing of progression of mild or mild to moderate AD. But in view of substantial inter-subject and inter-site variance issues, a large and long trial is necessary. Estimates using preliminary data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) [ 15 ] suggest that to demonstrate a 33% reduction in progression rate (as measured by change in Alzheimer's Disease Assessment Scale-cognitive subscale (ADAS-cog) or Clinical Dementia Rating 'sum of boxes' (CDR-SB)) in an 18 month trial in mild AD, approximately 300 subjects per group are required (M Donohue et al ., unpublished). No phase II program has approached this size. It should be noted that the ADNI experience probably underestimates the sample size required, in that it may be expected that variance will be greater in commercial trials, particularly those that are international, than among the academic North American sites participating in ADNI.

Other recent trials: Rember and dimebon

As with the anti-amyloid agents tramiprosate, tarenflurbil and bapineuzumab, the phase II trial of the anti-tangle compound Rember (methylene blue) did not meet its primary efficacy objectives [ 16 ]. In view of the modest group sizes and short duration, this too is not surprising, regardless of whether the drug ultimately proves effective in pivotal trials. But as with the anti-amyloid programs, caution must be exercised in the interpretation of post hoc analyses of the Rember trial data.

The development of dimebon represents the one recent AD program with strikingly positive results. At the primary 6 month analysis, strongly significant beneficial effects were seen on all outcome measures [ 17 ]. With continuation of the blind through 12 months of treatment, the effects on outcome measures increased, consistent with (though not definitive evidence of) a disease-modifying effect [ 17 ]. The success of this modestly sized trial is indicative of the immediate symptomatic benefit associated with the treatment. If a putative disease modifying drug yields short-term benefits, short (6 month) trials may be sufficient for regulatory approval. Symptomatic effects are plausible with neuroprotective and anti-amyloid drugs. But in the absence of such effects, a modest, phase II-type trial will be insufficient; little or no cognitive and clinical decline can be observed in 6 months, so no slowing of progression can be demonstrated. A consensus has arisen that 18 months or longer is an appropriate duration of treatment for studies aiming to show slowing of decline in AD.

But what about the two negative phase III trials of plausible anti-amyloid agents? There was certainly a 'phase II problem' – that is, phase III proceeded without evidence of efficacy in phase II. As expected, the small phase II tramiprosate study did not show any efficacy signal, but the modest reduction in CSF Aβ42 was considered encouraging. But it is unknown what the size of this biomarker signal must be to predict clinical efficacy with prolonged treatment. Further, there were questions about the magnitude and consistency of central nervous system drug penetration and concentration. But in addition to these uncertainties, the power of the phase III North American tramiprosate trial was lower than expected. The placebo group decline was smaller, and the standard deviation of the change score was higher, than expected; the power to demonstrate the target effect size of 25% slowing with the group sizes of 350 was limited. The tarenflurbil phase III program followed a phase II study that (not surprisingly) failed to achieve its primary efficacy objectives, so the risk of a negative phase III program had to be considered substantial; only the post hoc phase II analyses were encouraging. In addition, there was no convincing evidence of pharmacodynamic effect; specifically, no reduction in CSF Aβ in humans had been demonstrated. The negative trial results may reflect inadequate brain penetration in humans to yield a sufficient reduction in the generation of Aβ.

On the basis of these plausible explanations, the negative results of the phase II and III anti-amyloid trials cannot be considered to be strong evidence against the amyloid cascade hypothesis. The scientific basis for the hypothesis remains quite compelling. Aβ42 is highly toxic to neuronal cells and synaptic function, particularly in its oligomeric states. Each of the known genetic causes of AD is closely linked to Aβ generation: Down syndrome to APP over-expression, and familial autosomal dominant AD to mutations of APP and presinilins 1 and 2 that increase amyloidogenic cleavage of APP. Occam's Razor points strongly to amyloid as the pivotal molecule. Recent reports of a small number of AD patients with progressive dementia despite apparent amyloid plaque clearance resulting from active vaccination [ 18 ] does not disprove the hypothesis; clinical data on these individuals are limited, plaques are probably not the most important form of Aβ, the course of disease had these patients not been treated is unknown, and perhaps earlier treatment is necessary for a profound effect on outcome.

The need for early intervention

This last point may be key. There is strong evidence that the pathiobiology of AD precedes dementia by many years. In Down syndrome, amyloid deposition in brain precedes dementia by years or decades [ 19 , 20 ]. There is a high prevalence of brain amyloid in non-demented elderly individuals at autopsy [ 21 ], perhaps an indication of a long pre-symptomatic stage. Similarly, neuroimaging evidence of brain amyloid deposition is common [ 22 ]. Subtle memory symptoms and cognitive decline have been documented more than a decade before dementia onset [ 23 ]. If, as suggested by the recent active vaccine study autopsy report, dementia can progress despite elimination of amyloid plaques in AD patients [ 18 ], perhaps it is necessary to intervene earlier in the disease process.

At present, the diagnosis of AD requires the presence of dementia. What is the relationship of pre-dementia cognitive dysfunction to AD? What is the significance of amyloid brain deposition in the absence of cognitive impairment? With two plausible (if not yet proven) methods for identifying brain amyloid deposition, positron emission tomography (PET) scanning and CSF measurement of Aβ42, identification of such individuals is quite feasible. If either subtle cognitive impairment or amyloid deposition in brain consistently predicts AD dementia, it should be considered an early stage of AD. That is, we should revise the standard NINCDS-ADRDA criteria for AD [ 24 ].

Dubois and colleagues [ 25 ] have proposed one possible revision. They suggest that a 'research diagnosis' of AD be based on the presence of gradually progressive episodic memory impairment with evidence of AD neurobiology documented by the presence of one or more among several characteristic biomarker signals. The biomarker signals include medial temporal lobe atrophy by volumetric magnetic resonance imaging (MRI), temporal parietal hypoperfusion by [18F]fluorodeoxyglucose (FDG)-PET, amyloid deposition by PET, or CSF findings (elevated tau or phospho-tau, and/or low Aβ42) characteristic of AD. The proposed criteria can be applied in the pre-dementia or dementia stages.

It is plausible that effective disease-modifying interventions might be only minimally effective or even futile at the dementia stage; neuroprotection or favorable effects on the inciting amyloid dysregulation might be overwhelmed by extensive neuronal/synaptic degeneration and plaque pathology. Extending the diagnosis of AD to include individuals with mild cognitive impairment and even normal cognition when there is biomarker evidence of AD-type pathophysiology might facilitate the development of disease-modifying drugs for the treatment of individuals most likely to respond. The earlier the disease-modifying intervention, the greater the expected impact on the disease course. This idea is supported by a number of therapeutic studies in transgenic animal models [ 26 , 27 ].

The design of early intervention trials

Altering the definition of AD may not be necessary for the development of early interventions. A possible development strategy for early intervention using the current diagnostic criteria would be to conduct studies aiming to demonstrate that an intervention increases the time to dementia diagnosis. Several completed studies have enrolled subjects with amnestic mild cognitive impairment (MCI), with the primary analysis of survival to consensus diagnosis of AD [ 28 ]. This design has the advantage of clear clinical validity, a desirable feature in consideration of the uncertain regulatory status of the MCI designation. At least one set of MCI criteria seems to predict a high likelihood of AD diagnosis (approximately 15% per year), so that such a trial can have a reasonable size with adequate power to demonstrate a treatment effect [ 29 ]. But progression from MCI to AD is not a discrete event; the loss of function necessary to meet criteria for dementia occurs gradually, and it is challenging to assign a specific date to dementia onset. This subjectivity may be aggravated in large international trials. The progression of cognitive and functional impairment caused by AD pathobiology is insidious; defining a discrete disease onset seems arbitrary.

If the diagnostic criteria for AD are modified to encompass individuals prior to the onset of dementia, it would be straightforward to design trials with standard AD co-primary outcome measures. The ADNI longitudinal data demonstrate acceptable decline rate and variance for the ADAS-cog and CDR-SB in amnestic MCI; the size of trials adequately powered to demonstrate slowing of cognitive/clinical progression would be large but perhaps manageable. Adding selection criteria, and perhaps covariates to adjust for disease state, will reduce sample sizes substantially. In particular, for the development of anti-amyloid programs such as secretase inhibitor, anti-aggregation agents and anti-amyloid immunotherapy, trials can select MCI patients with biomarker evidence of brain amyloid deposition. Two options are feasible, though each presents challenges. The advent of F18 amyloid binding radiotracers has established the feasibility of amyloid brain imaging at most sites with PET scanners, but this is an expensive undertaking that has not yet been fully validated. CSF Aβ42 is strongly associated with neuroimaging evidence of amyloid deposition [ 30 ] and is essentially universally available, though requiring lumbar puncture during study screening may not be welcomed by investigators and especially subjects. The addition of covariates such as MRI volumetric measures to analysis plans will reduce unexplained variance and further increase statistical power to demonstrate slowing of progression. If the community of AD investigators, clinicians and regulators were to adopt early AD diagnostic criteria, feasible early AD trials could be launched immediately (Table 1 ).

While disease-modifying treatment of MCI is expected to yield more dramatic benefits than treatment of mild AD, perhaps the most appropriate population for intervention is the earlier, pre-symptomatic (or very mildly symptomatic) subjects with biomarker evidence suggestive of AD. The ultimate goal of disease-modification programs, prevention of AD dementia, is conceivable if treatment is started before appreciable neuronal damage and synaptic dysfunction have occurred. Preliminary evidence from some studies suggest that markers of amyloid accumulation predict dementia even in asymptomatic individuals [ 31 ].

Validated surrogate markers in AD

But in the absence of symptoms, it will not be possible to fulfill the conventional US Food and Drug Administration requirement for AD drug development: demonstration of efficacy on co-primary measures, specifically a cognitive performance test and a functional/global measure. It would require huge and lengthy studies to show slowing of cognitive and clinical progression or delay to diagnosis of dementia in subjects not yet showing any symptoms. To study interventions in this population, we will require validated surrogate markers.

A biomarker is any objectively measured characteristic that reflects normal or pathological processes, or responses to therapeutic intervention. As discussed above, biomarkers can be valuable in selecting subjects for clinical trials and for therapeutic interventions, for reducing unexplained variance and thus improving statistical power, and for establishing proof of concept in early phase drug development.

In rare cases, a biomarker can take the place of a clinical endpoint for establishing efficacy in a phase III clinical trial; that is, a biomarker can be validated as a surrogate endpoint. Examples of such surrogate markers include blood glucose and hgA1c in diabetes, blood pressure and cholesterol in cardiovascular disease, intraocular pressure in glaucoma, and lymphocyte subset ratios and viral load in HIV disease. To validate a biomarker as a surrogate endpoint, several issues must be addressed. There must be a well-accepted scientific framework connecting the biomarker to disease mechanisms and the prediction of clinical outcomes. Further, drug effects on the biomarker must be related to drug effects on clinical outcome; ideally, the biomarker should fully capture treatment effects, as confirmed by clinical trials of multiple interventions.

It is unlikely that an ideal surrogate for disease-modifying intervention in AD will become available in the foreseeable future. However, in consideration of the enormous clinical need, and the likelihood that the development of highly effective disease-modifying treatments will require the use of surrogate endpoints, it is reasonable to assume that regulatory agencies will consider acceptance of surrogates that are less than ideal.

A validated surrogate marker is essential for the study of AD interventions in asymptomatic or very mildly symptomatic individuals. It may be feasible to gain acceptance of a surrogate AD biomarker with a small number of trials demonstrating concordant treatment effects on the biomarkers and clinical symptoms. Even if the benefits of disease-modifying treatments in mild AD dementia are limited, they may well be sufficient to establish this concordance. Indeed, ongoing anti-amyloid trials that have incorporated biomarkers could provide this evidence. Consensus among clinical experts, based on robust data, that candidate biomarkers track disease progression at various stages of disease will strengthen the case for validation. A leading candidate surrogate marker is brain atrophy rate as measured by volumetric MRI; a huge body of evidence supports a link between regional brain atrophy and progression of AD pathobiology [ 32 – 34 ].

Paving the path forward

Building the consensus necessary to shift regulatory guidelines, clinical trial design and clinical practice will require large-scale cooperation among pharmaceutical and biotech companies, academic leaders, advocacy groups, funders and regulators. It is fortuitous that such cooperative efforts have been steadily gaining traction in recent years. Regular meetings involving all of the stakeholder groups have been productive; the semiannual Alzheimer's Association Research Roundtable, the annual Leon Thal Symposium sponsored by the Lou Ruvo Brain Institute, and the meetings of the Task Force on Use of Biomarkers in Alzheimer's Trials are leading examples that have advanced the field. They demonstrate the eagerness of many companies to share experience and ideas in pursuit of solutions to problems in AD therapeutic research.

Perhaps the best example of a cooperative effort among many to overcome the hurdles in drug development is ADNI. Led by Michael Weiner at the University of California San Francisco, ADNI is jointly funded by the National Institute on Aging, the Alzheimer's Association and other foundations, and contributions from pharmaceutical companies. It is a long-term effort to collect longitudinal cognitive, clinical, CSF and neuroimaging data on cohorts of individuals with mild AD, mild cognitive impairment and normal cognitive aging to allow optimal use of biomarkers in trial design. ADNI brings together leaders from academia, industry, government agencies and advocacy groups on at least a biweekly basis to jointly assess the study progress, and to discuss roadblocks and paths forward. To maximize scientific advance, all ADNI data are publicly posted in real-time; a huge number of presentations and publications from ADNI as well as outside investigators bears evidence of its success. ADNI has also spawned or supported similar collaborative efforts in Europe, Japan, Australia and China.

Data and ideas arising from ADNI and the various collaborative meetings have provided the ideas and data behind the discussion in this article. With continuation of these efforts, the common goal of optimal trial design is readily achievable. The challenges of determining populations for study, cognitive and clinical outcome measures, validation of biomarkers and analytic plans can be met within a few years. Consensus will lead to practical regulatory pathways, and the successful introduction of disease-modifying interventions that will blunt the AD epidemic that is growing with the aging world populations.

Abbreviations

Alzheimer's disease

Alzheimer's Disease Assessment Scale-cognitive subscale

Alzheimer's Disease Neuroimaging Initiative

amyloid precursor protein

Clinical Dementia Rating 'sum of boxes'

cerebrospinal fluid

mild cognitive impairment

magnetic resonance imaging

positron emission tomography.

Katzman R: Editorial: The prevalence and malignancy of Alzheimer disease. A major killer. Arch Neurol. 1976, 33: 217-218.

Article   CAS   PubMed   Google Scholar  

Vellas B, Coley N, Andrieu S: Disease modifying trials in Alzheimer's disease: perspectives for the future. J Alzheimers Dis. 2008, 15: 289-301.

PubMed   Google Scholar  

Hardy J, Selkoe DJ: The amyloid hypothesis of Alzheimer's disease: progress and problems on the road to therapeutics. Science. 2002, 297: 353-356. 10.1126/science.1072994.

Schmidt AM, Yan SD, Yan SF, Stern DM: The biology of the receptor for advanced glycation end products and its ligands. Biochim Biophys Acta. 2000, 1498: 99-111. 10.1016/S0167-4889(00)00087-2.

Yan SD, Chen X, Fu J, Chen M, Zhu HJ, Roher A, Slattery T, Zhao L, Nagashima M, Morser J, Migheli A, Nawroth P, Stern D, Schmidt AM: RAGE and amyloid-Beta peptide neurotoxicity in Alzheimer's disease. Nature. 1996, 382: 685-691. 10.1038/382685a0.

Lauren J, Gimbel DA, Nygaard HB, Gilbert JW, Strittmatter SM: Cellular prion protein mediates impairment of synaptic plasticity by amyloid-beta oligomers. Nature. 2009, 457: 1128-1132. 10.1038/nature07761.

Article   PubMed Central   CAS   PubMed   Google Scholar  

Avila J, Hernandez F: GSK-3 inhibitors for Alzheimer's disease. Expert Rev Neurother. 2007, 7: 1527-1533. 10.1586/14737175.7.11.1527.

Tuszynski MH, Thal L, Pay M, Salmon DP, U HS, Bakay R, Patel P, Blesch A, Vahlsing HL, Ho G, Tong G, Potkin SG, Fallon J, Hansen L, Mufson EJ, Kordower JH, Gall C, Conner J: A phase 1 clinical trial of nerve growth factor gene therapy for Alzheimer disease. Nat Med. 2005, 11: 551-555. 10.1038/nm1239.

Gervais F, Paquette J, Morissette C, Krzywkowski P, Yu M, Azzi M, Lacombe D, Kong X, Aman A, Laurin J, Szarek WA, Tremblay P: Targeting soluble Abeta peptide with Tramiprosate for the treatment of brain amyloidosis. Neurobiol Aging. 2007, 28: 537-547. 10.1016/j.neurobiolaging.2006.02.015.

Aisen PS, Saumier D, Briand R, Laurin J, Gervais F, Tremblay P, Garceau D: A phase II study targeting amyloid-beta with 3APS in mild-to-moderate Alzheimer disease. Neurology. 2006, 67: 1757-1763. 10.1212/01.wnl.0000244346.08950.64.

Weggen S, Eriksen JL, Das P, Sagi SA, Wang R, Pietrzik CU, Findlay KA, Smith TE, Murphy MP, Bulter T, Kang DE, Marquez-Sterling N, Golde TE, Koo EH: A subset of NSAIDs lower amy-loidogenic Abeta42 independently of cyclooxygenase activity. Nature. 2001, 414: 212-216. 10.1038/35102591.

Kukar T, Prescott S, Eriksen JL, Holloway V, Murphy MP, Koo EH, Golde TE, Nicolle MM: Chronic administration of R-flurbiprofen attenuates learning impairments in transgenic amyloid precursor protein mice. BMC Neurosci. 2007, 8: 54-10.1186/1471-2202-8-54.

Article   PubMed Central   PubMed   Google Scholar  

Wilcock GK, Black SE, Hendrix SB, Zavitz KH, Swabb EA, Laughlin MA: Efficacy and safety of tarenflurbil in mild to moderate Alzheimer's disease: a randomised phase II trial. Lancet Neurol. 2008, 7: 483-493. 10.1016/S1474-4422(08)70090-5.

Relkin NR: Current state of immunotherapy for Alzheimer's disease. CNS Spectr. 2008, 13: 39-41.

Alzheimer's Disease Neuroimaging Initiative. [ http://adni-info.org ]

Gura T: Hope in Alzheimer's fight emerges from unexpected places. Nat Med. 2008, 14: 894-10.1038/nm0908-894.

Doody RS, Gavrilova SI, Sano M, Thomas RG, Aisen PS, Bachurin SO, Seely L, Hung D: Effect of dimebon on cognition, activities of daily living, behaviour, and global function in patients with mild-to-moderate Alzheimer's disease: a randomised, double-blind, placebo-controlled study. Lancet. 2008, 372: 207-215. 10.1016/S0140-6736(08)61074-0.

Holmes C, Boche D, Wilkinson D, Yadegarfar G, Hopkins V, Bayer A, Jones RW, Bullock R, Love S, Neal JW, Zotova E, Nicoll JA: Long-term effects of Abeta42 immunisation in Alzheimer's disease: follow-up of a randomised, placebo-controlled phase I trial. Lancet. 2008, 372: 216-223. 10.1016/S0140-6736(08)61075-2.

Lemere CA, Blusztajn JK, Yamaguchi H, Wisniewski T, Saido TC, Selkoe DJ: Sequence of deposition of heterogeneous amyloid beta-peptides and APO E in Down syndrome: implications for initial events in amyloid plaque formation. Neurobiol Dis. 1996, 3: 16-32. 10.1006/nbdi.1996.0003.

Leverenz JB, Raskind MA: Early amyloid deposition in the medial temporal lobe of young Down syndrome patients: a regional quantitative analysis. Exp Neurol. 1998, 150: 296-304. 10.1006/exnr.1997.6777.

Davis DG, Schmitt FA, Wekstein DR, Markesbery WR: Alzheimer neuropathologic alterations in aged cognitively normal subjects. J Neuropathol Exp Neurol. 1999, 58: 376-388. 10.1097/00005072-199904000-00008.

Aizenstein HJ, Nebes RD, Saxton JA, Price JC, Mathis CA, Tsopelas ND, Ziolko SK, James JA, Snitz BE, Houck PR, Bi W, Cohen AD, Lopresti BJ, DeKosky ST, Halligan EM, Klunk WE: Frequent amyloid deposition without significant cognitive impairment among the elderly. Arch Neurol. 2008, 65: 1509-1517. 10.1001/archneur.65.11.1509.

Amieva H, Le Goff M, Millet X, Orgogozo JM, Peres K, Barberger-Gateau P, Jacqmin-Gadda H, Dartigues JF: Prodromal Alzheimer's disease: successive emergence of the clinical symptoms. Ann Neurol. 2008, 64: 492-498. 10.1002/ana.21509.

Article   PubMed   Google Scholar  

McKhann G, Drachman D, Folstein M, Katzman R, Price D, Stadlan EM: Clinical diagnosis of Alzheimer's disease: Report of the NINCDS-ADRDA Work Group. Neurology. 1984, 34: 939-944.

Dubois B, Feldman HH, Jacova C, Dekosky ST, Barberger-Gateau P, Cummings J, Delacourte A, Galasko D, Gauthier S, Jicha G, Meguro K, O'Brien J, Pasquier F, Robert P, Rossor M, Salloway S, Stern Y, Visser PJ, Scheltens P: Research criteria for the diagnosis of Alzheimer's disease: revising the NINCDS-ADRDA criteria. Lancet Neurol. 2007, 6: 734-746. 10.1016/S1474-4422(07)70178-3.

Schenk D, Barbour R, Dunn W, Gordon G, Grajeda H, Guido T, Hu K, Huang J, Johnson-Wood K, Khan K, Kholodenko D, Lee M, Liao Z, Lieberburg I, Motter R, Mutter L, Soriano F, Shopp G, Vasquez N, Vandevert C, Walker S, Wogulis M, Yednock T, Games D, Seubert P: Immunization with amyloid-beta attenuates Alzheimer-disease-like pathology in the PDAPP mouse. Nature. 1999, 400: 173-177. 10.1038/22124.

McLaurin J, Kierstead ME, Brown ME, Hawkes CA, Lambermon MH, Phinney AL, Darabie AA, Cousins JE, French JE, Lan MF, Chen F, Wong SS, Mount HT, Fraser PE, Westaway D, St George-Hyslop P: Cyclohexanehexol inhibitors of Abeta aggregation prevent and reverse Alzheimer phenotype in a mouse model. Nat Med. 2006, 12: 801-808. 10.1038/nm1423.

Jelic V, Kivipelto M, Winblad B: Clinical trials in mild cognitive impairment: lessons for the future. J Neurol Neurosurg Psychiatry. 2006, 77: 429-438. 10.1136/jnnp.2005.072926.

Petersen RC, Thomas RG, Grundman M, Bennett D, Doody R, Ferris S, Galasko D, Jin S, Kaye J, Levey A, Pfeiffer E, Sano M, van Dyck CH, Thal LJ: Vitamin E and Donepezil for the Treatment of Mild Cognitive Impairment. N Engl J Med. 2005, 352: 2379-2388. 10.1056/NEJMoa050151.

Fagan AM, Mintun MA, Mach RH, Lee SY, Dence CS, Shah AR, LaRossa GN, Spinner ML, Klunk WE, Mathis CA, DeKosky ST, Morris JC, Holtzman DM: Inverse relation between in vivo amyloid imaging load and cerebrospinal fluid Abeta42 in humans. Ann Neurol. 2006, 59: 512-519. 10.1002/ana.20730.

Fagan AM, Roe CM, Xiong C, Mintun MA, Morris JC, Holtzman DM: Cerebrospinal fluid tau/beta-amyloid(42) ratio as a prediction of cognitive decline in nondemented older adults. Arch Neurol. 2007, 64: 343-349. 10.1001/archneur.64.3.noc60123.

Smith AD, Jobst KA: Use of structural imaging to study the progression of Alzheimer's disease. Br Med Bull. 1996, 52: 575-586.

Scahill RI, Fox NC: Longitudinal imaging in dementia. Br J Radiol. 2007, 80 (Spec No 2): S92-98. 10.1259/bjr/78981552.

Hampel H, Burger K, Teipel SJ, Bokde AL, Zetterberg H, Blennow K: Core candidate neurochemical and imaging biomarkers of Alzheimer's disease. Alzheimers Dement. 2008, 4: 38-48. 10.1016/j.jalz.2007.08.006.

Download references

Acknowledgements

This work was supported in part by grants (U01-AG010483, U01-AG024904) from the National Institute on Aging of the National Institutes of Health.

Author information

Authors and affiliations.

Department of Neurosciences, University of California San Diego, Gilman Drive, La Jolla, CA, 92093, USA

Paul S Aisen

You can also search for this author in PubMed   Google Scholar

Corresponding author

Correspondence to Paul S Aisen .

Additional information

Competing interests.

PSA is the recipient of grant awards from Pfizer, Baxter, Neuro-Hitech, Abbott and Martek, and is a consultant to Elan, Wyeth, Eisai, Neurochem, Schering-Plough, Bristol Myers Squibb, Lilly, Neurophage, Merck, Roche, Amgen, Genentech, Abbott, Pfizer, Novartis and Medivation. He holds stock options in Medivation.

Rights and permissions

Reprints and permissions

About this article

Cite this article.

Aisen, P.S. Alzheimer's disease therapeutic research: the path forward. Alz Res Therapy 1 , 2 (2009). https://doi.org/10.1186/alzrt2

Download citation

Published : 09 July 2009

DOI : https://doi.org/10.1186/alzrt2

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

• Mild Cognitive Impairment
• Down Syndrome
• Brain Amyloid Deposition
• Primary Efficacy Objective
• Magnetic Resonance Imaging Volumetric Measure

Alzheimer's Research & Therapy

ISSN: 1758-9193

• General enquiries: [email protected]

Alzheimer’s Disease Detection using Machine Learning: A Review

Ieee account.

• Change Username/Password
• Update Address

Purchase Details

• Payment Options
• Order History
• View Purchased Documents

Profile Information

• Communications Preferences
• Profession and Education
• Technical Interests
• US & Canada: +1 800 678 4333
• Worldwide: +1 732 981 0060
• Contact & Support
• About IEEE Xplore
• Accessibility
• Terms of Use
• Nondiscrimination Policy
• Privacy & Opting Out of Cookies

A not-for-profit organization, IEEE is the world's largest technical professional organization dedicated to advancing technology for the benefit of humanity. © Copyright 2024 IEEE - All rights reserved. Use of this web site signifies your agreement to the terms and conditions.

• Open access
• Published: 02 October 2023

Clinical trials of new drugs for Alzheimer disease: a 2020–2023 update

• Li-Kai Huang 1 , 2 , 3   na1 ,
• Yi-Chun Kuan 2 , 3 , 4 , 5   na1 ,
• Ho-Wei Lin 6 &
• Chaur-Jong Hu   ORCID: orcid.org/0000-0002-4900-5967 1 , 2 , 3 , 4  

Journal of Biomedical Science volume  30 , Article number:  83 ( 2023 ) Cite this article

18k Accesses

16 Citations

24 Altmetric

Metrics details

Alzheimer's disease (AD) is the leading cause of dementia, presenting a significant unmet medical need worldwide. The pathogenesis of AD involves various pathophysiological events, including the accumulation of amyloid and tau, neuro-inflammation, and neuronal injury. Clinical trials focusing on new drugs for AD were documented in 2020, but subsequent developments have emerged since then. Notably, the US-FDA has approved Aducanumab and Lecanemab, both antibodies targeting amyloid, marking the end of a nearly two-decade period without new AD drugs. In this comprehensive report, we review all trials listed in clinicaltrials.gov, elucidating their underlying mechanisms and study designs. Ongoing clinical trials are investigating numerous promising new drugs for AD. The main trends in these trials involve pathophysiology-based, disease-modifying therapies and the recruitment of participants in earlier stages of the disease. These trends underscore the significance of conducting fundamental research on pathophysiology, prevention, and intervention prior to the occurrence of brain damage caused by AD.

Alzheimer disease (AD) represents a major global medical, social, and economic burden. The World Alzheimer Report 2022 revealed that more than 55 million people have AD or related conditions worldwide, and this number is projected to reach 82 million by 2030 and 138 million by 2050 [ 1 ]. Typically, AD first manifests as progressive memory decline accompanied or followed by other cognitive dysfunctions, such as visuospatial abnormalities, navigation difficulties, executive problems, and language disturbances. These cognitive impairments affect the performance of activities of daily living. During the course of AD, many behavioral and psychological symptoms of dementia (BPSD) occur [ 2 , 3 , 4 ].

Although the exact causes of AD remain unclear, the disease has two pathological hallmarks: plaques composed of amyloid-beta (Aβ) fibrils and neurofibrillary tangles (NFTs) consisting of hyperphosphorylated tau protein [ 5 , 6 , 7 ]. The key event in AD pathogenesis is believed to be Aβ accumulation. Cerebral Aβ fibril deposition may occur decades before the onset of clinical symptoms [ 8 ]. Brain atrophy, particularly in the hippocampus, is major indicator of early Aβ accumulation, particularly in the presubiculum [ 9 , 10 ]. Aβ accumulation was discovered to be crucial by three independent research groups in 1991 [ 11 , 12 , 13 ]. In familial AD, mutant autosomal-dominant genes, including the genes for amyloid precursor protein ( APP ), presenilin-1 ( PSEN1 ), and presenilin-2 ( PSEN2 ), encode the major proteins involved in amyloid metabolism [ 13 , 14 , 15 ]. Individuals with trisomy 21 (Down syndrome) have an extra copy of the APP gene, which may result in increased amyloid production and AD risk in middle age [ 16 ]. At present, the predominant theory regarding the cause of AD is the amyloid hypothesis; crucial advancements in AD therapy have been made on the basis of the proposed role of amyloid accumulation in the AD development. The United States Food and Drug Administration (US FDA) granted traditional approval for Leqembi (lecanemab-irmb) on July 6, 2023, for the treatment of AD [ 17 ]. The approval of this treatment not only affirms the pathophysiological significance of amyloid in AD but also marks a notable advance in clinical approaches to AD treatment, remedying the scarcity of new drugs in the market for nearly two decades.

Tau is a microtubule-associated protein that aids in microtubule assembly and stabilization. In AD, tau becomes hyperphosphorylated and aggregates to form paired helical filaments, a major component of NFTs within the neuronal cytoplasm. As the disease progresses, the gradual spread of tau pathology throughout brain regions has been suggested to be caused by the transfer of abnormal types of tau protein from one neuron to another [ 18 ]. The accumulation of NFTs might be initiated between the accumulation of Aβ and the development of clinical symptoms of AD [ 19 ]. NFTs and quantitative neuronal loss may be more closely correlated with disease severity and dementia progression than the amyloid plaque burden [ 20 , 21 , 22 ]. Positron emission tomography (PET) investigations have revealed a strong correlation between the binding characteristics of tau tracers and the severity of clinical manifestations in individuals with AD [ 23 ]. Molecular imaging modalities (PET) and cerebrospinal fluid (CSF) and blood–based biomarkers have extended the diagnostic scope of AD pathology to both clinical and even preclinical settings. The analysis of a combination of biomarkers such as amyloid, tau, and neurodegeneration (collectively, ATN classification) has been proposed by research on AD [ 24 , 25 ]. Furthermore, the exceptional diagnostic accuracy of plasma-based biomarkers has facilitated the clinical transition of fluid biomarkers from research settings to clinical practice. A recent presentation at the Alzheimer’s Association International Conference in 2023 highlighted the clinical and research applications of two fundamental AD biomarker categories, labeled as A and T. The A category pertains to biomarkers associated with the Aβ proteinopathy pathway, and the T category pertains to biomarkers linked to tau proteinopathy [ 26 ].

Aβ serves as a proinflammatory agent and triggers the nuclear factor κB (NF-κB) pathway in astrocytes, increasing complement C3 release. Subsequently, by binding to C3a receptors, C3 causes neuronal dysfunction and microglial activation [ 27 ]. In the early stage of AD, activated microglia may play a protective, anti-neuroinflammatory role by clearing amyloids and releasing nerve growth factors. However, activated microglia induce neurotoxic A1 astrocyte reactivity through the release of IL-1α, C1q, and TNF-α, resulting in a feedback loop of dysregulated inflammation in AD [ 28 ]. The excessive accumulation of Aβ or other toxic compounds activates proinflammatory phenotypes, resulting in neuronal damage [ 29 ]. Sustained inflammation has been observed in the brains of patients with AD [ 30 , 31 ]. The inadequate clearance of Aβ along with the aggregation of tau disrupts microglial defense mechanisms, resulting in sustained and harmful microglial activation [ 32 ]. The sequential occurrence of amyloid plaque formation, microglial activation, and the pathological phosphorylation and aggregation of tau proteins to form NFTs is the fundamental notion of the amyloid cascade–inflammation hypothesis. In the Multi-Ethnic Study of Atherosclerosis (multiple covariates were controlled for), vascular risk factor profiles and Aβ deposition significantly predicted cognitive decline [ 33 ]. Vascular risk factors can also lead to inflammation in the brain, which damages neuronal cells and further increases the likelihood of AD dementia [ 34 ].

The role of autophagy impairment is proposed in a novel hypothesis concerning plaque formation in AD. Among neurons that are compromised but still maintain some integrity, autophagic vacuoles (AVs) containing abundant Aβ are notably present. These AVs cluster within expansive membrane blebs, exhibiting a distinctive flower-like arrangement termed PANTHOS. These formations constitute the primary source of the majority of amyloid plaques found in mouse models of AD [ 35 ]. Neuroprotective therapies, including free radical scavengers, regeneration enhancers, and the suppression of excitable amino acid signaling pathways, have been proposed for preventing neuronal death or brain atrophy caused by amyloid, tau, and neuroinflammation [ 36 ]. Pathological evidence indicates that AD is also associated with degeneration in cholinergic neuron-rich regions, such as the nucleus basalis of Meynert, frontal cortex, and anterior and posterior cingulate cortex, which can lead to the symptoms of memory impairment and agitation. Acetylcholine (ACh) plays a vital role in memory function, including memory encoding, consolidation, and retrieval processes, and increasing Ach levels by using cholinesterase inhibitors (AChEIs) has become a standard therapy for the symptoms of AD [ 37 ].

Clinical trials of early or preventive interventions based on amyloid/tau theory and those targeting other pathophysiologies are ongoing or have been initiated. Many ongoing clinical trials on AD are focused on disease-modifying therapies (DMTs) that target the causes and can change the course of AD. The other trials involve symptomatic treatments—for example, enhancing cognitive function and relieving BPSD (Fig.  1 ). In this review, we summarize the new drugs being examined in ongoing trials (listed on ClinicalTrials.gov) and discuss the trends in and obstacles in AD clinical trials.

According to the amyloid hypothesis, the pathophysiology and clinical course of Alzheimer's disease progress as follows: amyloid accumulation, neuroinflammation, tau accumulation, brain metabolism dysfunction, brain atrophy, cognitive decline (from mild cognitive impairment to dementia), and the development of dementia symptoms. Novel drugs should target at least one of these events. AD Alzheimer's disease, aMCI amnestic mild cognitive impairment, BPSD behavioral psychological symptoms of dementia

Anti-amyloid therapy

Table 1 summarizes the US FDA approval status of anti-amyloid agents. Tables 2 and 3 summarize the ongoing phase 3 and phase 2 trials of anti-amyloid therapy respectively.

Aducanumab (brand name: Aduhelm) is a high-affinity, fully human immunoglobulin gamma 1 (IgG1) monoclonal antibody that binds to the N-terminus of Aβ fibrils and blocks amyloid aggregation [ 38 ]. In August 2015, two phase 3 clinical trials, namely ENGAGE and EMERGE studies, were initiated. These trials compared monthly intravenous infusions of aducanumab at one of three doses with infusions of placebo over 18 months, and the primary outcomes were cognitive and functional decline, which were assessed using the Clinical Dementia Rating (CDR) scale Sum of Boxes (CDR-SB). The secondary outcomes were other cognitive and functional measures. The trials were conducted in 150 centers across North America, Europe, Australia, and Asia. However, the findings of the EMERGE trial reached statistical significance, whereas the primary endpoint was not reached in the ENGAGE trial. An exploratory analysis revealed that a subgroup of the participants in the ENGAGE trial who received a high dose of aducanumab exhibited slow decline, which was similar to that observed among the participants in the EMERGE trial. The US FDA approved aducanumab in June 2021 on the basis of the data of the EMERGE and ENGAGE trials. Both trials presented evidence of an intermediate effect of the drug on biomarkers, indicating amyloid removal, which is likely linked to the clinical benefit of aducanumab. Further trials must be conducted to confirm the potential benefit of aducanumab [ 39 ]. The phase 3b/4 ENVISION trial (NCT05310071), which began in 2022, will enroll 1,512 patients with early AD who will receive either monthly doses of aducanumab of up to 10 mg/kg or placebo for 18 months. The aim of the trial is to determine the efficacy of aducanumab in delaying cognitive and functional decline in comparison with placebo, which would be determined on the basis of CDR-SB scores. The secondary endpoints of the trial include scores on the Alzheimer’s Disease Assessment Scale–Cognitive Subscale (ADAS-Cog) 13, Alzheimer’s Disease Cooperative Study–Activities of Daily Living Inventory (ADCS-ADL)–Mild Cognitive Impairment Version, Integrated Alzheimer’s Disease Rating Scale (iADRS), Mini-Mental State Examination, and Neuropsychiatric Inventory. The trial intends to recruit 18% of its participants from Black and Latinx populations in the United States and will have a long-term follow-up of up to 4 years, with results expected by 2026. The EMBARK trial (NCT04241068) is a phase 3b open-label study including 1,696 participants from previous aducanumab trials (from trials 221AD103, 221AD301, 221AD302, and 221AD205) that will assess aducanumab safety and tolerability over 100 weeks after a wash-out period. Participants will receive an intravenous infusion of aducanumab at 10 mg/kg monthly for 2 years, and eligible participants will continue to receive the infusion for another 52 weeks during the long-term extended treatment period. The primary outcomes are safety and tolerability, and the efficacy endpoints are the same as those in the EMERGE and ENGAGE trials, and Caregiver Global Impression of Change evaluations will be conducted every 6 months. All participants will undergo volumetric magnetic resonance imaging (MRI) scans, and a subset of the study population will undergo biomarker testing, including amyloid PET, tau PET, and CSF testing.

Lecanemab (brand name: Leqembi), a humanized IgG1 antibody derived from mAb158, selectively binds to soluble Aβ protofibrils [ 40 ]. The US FDA approved it on January 6, 2023, through an accelerated approval pathway on the basis of evidence of amyloid removal in a phase 2 trial (NCT01767311) and because it had a likelihood of having clinical benefits [ 41 ] A double-blind, placebo-controlled phase 2 trial recruited 856 patients with AD with mild cognitive impairment (MCI) or mild dementia and verified amyloid pathology through amyloid PET or CSF Aβ1-42 [ 42 ]. The results revealed a significant and dose-dependent reduction of amyloid plaques in the lecanemab group (10 mg/kg, intravenous infusion every 2 weeks) from baseline to week 79 compared with the placebo group. At the time of writing this paper, three phase 3 clinical trials on lecanemab are underway. The first trial, Clarity AD (NCT03887455), was initiated in March 2019 and was conducted at 250 sites around the world. It reported favorable outcomes for all primary and secondary measures, including ADAS-Cog14, AD Composite Score (ADCOMS), and ADCS-MCI-ADL scores [ 43 ]. The second trial is AHEAD 3–45 (NCT04468659), which was initiated in July 2020 as a 4-year trial comprising two substudies, one of which is A3, and the other one is A45. A3 is enrolling 400 people whose amyloid levels are below the brain-wide threshold for positivity; participants will receive 5 mg/kg lecanemab titrated to 10 mg/kg or placebo every month for 216 weeks. A45 is enrolling 1,000 cognitively healthy participants with positive amyloid PET scans, and they will receive lecanemab titrated to 10 mg/kg every 2 weeks for 96 weeks, followed by 10 mg/kg every month through week 216. The trial is expected to run until October 2027. The third phase 3 clinical trial is the Dominantly Inherited Alzheimer Network Trials Unit (DIAN-TU) Next Generation trial (DIAN-TU-001 (E2814), NCT05269394), in which a combination of lecanemab and the anti-tau antibody E2814 (phase 2) will be administered to 168 people with familial AD mutations. On July 6, 2023, Leqembi (lecanemab-irmb) received traditional approval from the US FDA for the treatment of AD based on Phase 3 data from the Clarity AD clinical trial [ 17 ].

The appropriate use recommendations (AURs) for lecanemab and aducanumab highlight the importance of patient selection, surveillance for adverse events, and clinician preparedness [ 44 , 45 ]. The AURs for both drugs have several similarities with respect to age criteria, biomarker requirements (positive amyloid PET or CSF findings indicative of AD), diagnosis (MCI due to AD or mild AD dementia), and MRI exclusion criteria (e.g., microhemorrhages and cortical infarction). The AURs also emphasize the importance of monitoring for amyloid-related imaging abnormalities (ARIAs), which can occur in patients receiving these drugs. APOE genotyping is recommended for informing risk discussions with candidate participants because APOE4 allele carriers, especially APOE4 homozygotes, are at a high risk of ARIAs. Patients receiving treatment must have care partners or family members who can provide necessary support and who clearly understand the nature and requirements of the therapy. Discontinuation of treatment is recommended in the following situations: when a patient is taking drugs with associated risks, such as anticoagulation agents for conditions like atrial fibrillation, deep vein thrombosis, or pulmonary embolism; or when any of the following conditions occur: a hypercoagulable state, or the development of any of the following: cerebral macrohemorrhage, multiple areas of superficial siderosis, more than 10 instances of microhemorrhages since treatment initiation, severe symptoms of ARIAs, or two or more episodes of ARIAs.

Donanemab is a humanized monoclonal antibody developed from mouse mE8-IgG2a. It recognizes Aβ (3–42), an aggregated form of Aβ found in amyloid plaques [ 46 ]. It was discovered to be bound to approximately one-third of amyloid plaques in postmortem brain samples of patients with AD or Down syndrome, and it strongly reacted with the plaque core [ 47 ]. In the phase 2 TRAILBLAZER-ALZ study, the safety, tolerability, and efficacy of donanemab alone and in combination with the Beta-Secretase 1 (BACE1) inhibitor LY3202626 (developed by Eli Lilly and Company) were evaluated over 18 months. The trial met its primary endpoint of delaying decline—which was determined on the basis of iADRS scores—by 32% compared with placebo. Amyloid burden reduction was correlated with improvement in iADRS scores only in ApoE4 carriers [ 48 ]. Donanemab reduced the tau burden in the temporal, parietal, and frontal lobes and significantly decreased plasma pTau217 by 24% in the treatment group, whereas the placebo group exhibited a 6% increase in plasma pTau217 at the end of the trial [ 49 ]. At the time of writing this paper, five phase 3 trials of donanemab are underway: TRAILBLAZER-ALZ 2, TRAILBLAZER-ALZ 3, TRAILBLAZER-ALZ 4, TRAILBLAZER-ALZ 5 and TRAILBLAZER-ALZ 6. The TRAILBLAZER-ALZ 2 (NCT04437511) trial was initially started in June 2020 as a phase 2 safety and efficacy trial, and 500 patients with early AD were recruited. Inclusion criteria of TRAILBLAZER-ALZ 2 are similar to those of TRAILBLAZER-ALZ: a ≥ 6-month history of worsening memory and positive amyloid (flortaucipir) PET. The trial was subsequently extended to a phase 3 trial with 1,800 participants. The primary outcome is iADRS, and the effectiveness of treatment is being measured using a disease-progression model rather than solely on the basis of changes at the final time point. Trial results for 1,736 participants were published to report donanemab’s impact on early symptomatic AD. Using PET imaging to categorize individuals into groups with low/medium or high tau pathology load, the study spanned 18 months and assessed cognitive and functional scales. Donanemab achieved significant cognitive improvement in the low/medium tau group (iADRS change: − 6.02 vs. − 9.27 placebo) and combined population (change: − 10.2 vs. − 13.1 placebo). The drug notably reduced decline by 60% in patients with early-stage AD, supporting the efficacy of short-term dosing. Twenty-four outcomes were evaluated, with significant findings for 23 outcomes. Adverse effects included amyloid-related imaging problems (24% donanemab vs. 2.1% placebo) and infusion-related reactions (8.7% donanemab vs. 0.5% placebo). The study findings indicated the potential of donanemab to slow AD progression, particularly in the early stage [ 50 ]. In the TRAILBLAZER-ALZ study, donanemab slowed disease progression by 32% at 18 months ( p  = 0.04 vs. placebo), thus demonstrating clinical efficacy [ 51 ]. TRAILBLAZER-ALZ 3 (NCT05026866) is a placebo-controlled phase 3 prevention trial that was started in August 2021. The trial plans to enroll 3,300 cognitively healthy people aged 50–55 years who are at high risk of AD, as determined by elevated plasma pTau217 levels and Telephone Interview for Cognitive Status-modified scores. The primary outcome is the time to clinical progression, which is measured using global CDR scores. Participants are to be monitored every 6 months until cognitive impairment is noted (i.e., a score above 0 on the CDR for two consecutive evaluations) in 434 participants. The trial has a decentralized design and is being conducted at more than 200 sites in the United States, Japan, and Puerto Rico until November 2027. TRAILBLAZER-ALZ 4 (NCT05108922) is a phase 3, open-label, head-to-head comparison of amyloid clearance by either donanemab or aducanumab that began in November 2021 after the US FDA approval of aducanumab. The trial enrolled 200 people with early symptomatic AD, as indicated by a global CDR score of 0.5 or 1, at 31 sites in the United States. The primary outcome is the percentage of participants who achieve complete amyloid plaque clearance after 6 months for each treatment group, with clearance determined using amyloid (florbetapir) PET. The trial has 17 secondary outcomes, which are all related to amyloid PET measurements at up to 18 months. The preliminary results were presented at the 2022 Clinical Trial of AD (CTAD) conference: 38% of the patients on donanemab exhibited amyloid levels below the amyloid positivity threshold after 6 months, whereas only 2% of the patients on aducanumab has such findings. Plasma pTau217 levels decreased by 25% for the participants receiving donanemab, but not at all for those receiving aducanumab. The side effect of ARIA-edema occurred in 22% of the participants in both groups. TRAILBLAZER-ALZ 5 (NCT05508789) is being conducted to assess the safety and efficacy of donanemab in individuals with early symptomatic AD. The trial started in October 2022; 1,500 participants will be recruited by using the same criteria as those of TRAILBLAZER-ALZ 2 from 148 sites across China, Korea, Taiwan, and Europe; and the trial is expected to run until mid-2025. Participants will be administered monthly infusions of either donanemab or placebo, and the primary outcome will be measured on the basis of iADRS score changes after 18 months. TRAILBLAZER-ALZ 6 (NCT05738486) is a phase 3b study that will assess the impact of various dosing regimens of donanemab on the occurrence and severity of ARIA-E (ARIA with edema or effusion) in 800 adults with early symptomatic AD. The study also seeks to identify participant characteristics that predict the risk of ARIA-E. The trial is divided into four arms, each with a distinct donanemab dose.

Remternetug is a monoclonal antibody that recognizes a pyroglutamated form of Aβ that aggregates into amyloid plaques. In August 2022, Eli Lilly and Company initiated a phase 3 trial called TRAILRUNNER-ALZ1 (NCT05463731) that will randomize 600 patients with early symptomatic AD across 75 sites in the United States and 2 sites in Japan into groups receiving the antibody or placebo through intravenous infusion or subcutaneous injection for 1 year. The primary outcome is the percentage of patients whose amyloid plaques are cleared by the end of the treatment period. The secondary outcomes include the measurement of amyloid clearance, pharmacokinetics, and treatment-emergent anti-drug antibodies. The study also plans to conduct a year-long, blinded crossover extension. An additional safety cohort of 640 patients will receive open-label remternetug for 1 year.

Solanezumab is a humanized monoclonal antibody that targets the mid-domain of the Aβ peptide for increasing Aβ clearance [ 52 ]. Phase 3 trials of solanezumab, including EXPEDITION-1 and EXPEDITION-2, which enrolled 2,052 patients with mild-to-moderate AD, did not reveal improvements in ADAS-Cog11 and ADCS-ADL scores, which were the primary outcome measures. Similarly, the phase 3 trial EXPEDITION-3 demonstrated that 400 mg solanezumab administered every 4 weeks did not have significant effects on cognitive decline in patients with mild AD [ 52 ]. A4 (NCT02008357) is a phase 3 prevention trial focused on slowing memory and cognitive decline in elderly individuals without cognitive impairment or dementia. A4 is using a sensitive cognitive battery—the Alzheimer Disease Cooperative Study Preclinical Alzheimer Cognitive Composite—and was initiated in February 28, 2014. On March 8, 2023, Eli Lilly and Company reported that solanezumab did not slow cognitive decline or clear amyloid plaques in individuals with preclinical AD in the A4 study. DIAN-TU-001 (NCT01760005) is another ongoing phase 3 clinical trial that is testing the combination of solanezumab and gantenerumab in 210 asymptomatic and mildly symptomatic carriers of autosomal-dominant mutations in AD genes. However, on February 10, 2020, the study investigators announced that the primary endpoint was not achieved in the trial, namely treatment-related changes on the DIAN-Multivariate Cognitive Endpoint. The results indicated that the solanezumab-treated group had greater cognitive decline on some measures relative to placebo, and that solanezumab treatment did not exert any beneficial effects on downstream biomarkers, whereas gantenerumab significantly reduced amyloid plaques, CSF total tau, and phospho-tau181 and attenuated increases in neurofilament light chain [ 53 ]. The participants were offered an open-label extension with high-dose gantenerumab because of its positive effects on imaging and other biomarkers, such as normalized CSF Aβ42, and because it reduced CSF total tau and pTau181 levels.

ALZ-801 is a prodrug of tramiprosate, a small molecule of anti-Aβ oligomers and an aggregation inhibitor [ 54 ]. The phase 3 trial APOLLOE4 (NCT04770220) is evaluating the safety and efficacy of ALZ-801 for patients with early AD and carrying the homozygous ε4 allele on the apolipoprotein E gene ( APOE4/4 ). The recruited patients are receiving 265 mg ALZ-801 or placebo twice daily for 18 months. The trial started in May 2021. The primary endpoint is ADAS-Cog scores, and the secondary endpoints are scores of the Disability Assessment for Dementia, CDR-SB, and Amsterdam-iADL. The biomarkers of interest include the hippocampal volume, as determined through MRI and based on CSF and plasma pTau181 levels. Another phase 2 trial (NCT04693520) is investigating the effects of oral ALZ-801 administered to participants with early AD who have the APOE4/4 or APOE3/4 genotype with biomarkers of core AD pathology. The study is also assessing the efficacy, safety, and tolerability of ALZ-801.

Simufilam (PTI-125) is a drug that binds to filamin, a scaffolding protein that stabilizes the interaction between soluble Aβ42 and the α7 nicotinic acetylcholine receptor [ 55 ]. Two phase 3 trials, namely RETHINK-ALZ (NCT04994483) and REFOCUS-ALZ (NCT05026177), were commenced in November 2021. Both are safety and efficacy studies of simufilam and have enrolled participants with mild-to-moderate AD. RETHINK-ALZ will randomize 750 participants with AD and CDR scores of 0.5, 1, or 2 into groups receiving either placebo or 100 mg of simufilam twice a day for 1 year (52 weeks). The coprimary outcomes of this trial are ADAS-Cog12 and ADCS-ADL scores, and the trial is set to run through October 2023. REFOCUS-ALZ will randomize 1,083 participants into groups receiving placebo or 50 or 100 mg of simufilam (1:1:1) for 76 weeks. The primary outcome measures are similar to those of the RETHINK-ALZ trial. A phase 3 trial of simufilam (NCT05575076) was started in November 2022 to assess the long-term safety and tolerability of simufilam in participants with mild-to-moderate AD. That open-label extension study is intended to assess the long-term safety and tolerability of simufilam 100 mg twice daily in patients who have completed the RETHINK-ALZ or REFOCUS-ALZ Phase 3 clinical trials. The primary outcome measure is adverse event monitoring from baseline to week 52.

Varoglutamstat (PQ912) is a glutaminyl cyclase inhibitor that reduces pGlu-Aβ generation [ 56 ]. Glutaminyl cyclase catalyzes the cyclization of an exposed glutamate at the N-terminus of Aβ, resulting in the formation of toxic pGlu-Aβ, a major component of amyloid plaques. Two ongoing phase 2 clinical trials, namely VIVA-MIND and VIVIAD, are evaluating the safety, tolerability, and efficacy of varoglutamstat in participants with MCI and mild dementia due to AD. VIVA-MIND (NCT03919162) is a phase 2A multicenter, randomized, double-blind, placebo-controlled, parallel-group study of varoglutamstat, with a stage gate to phase 2B. Phase 2A involves an adaptive dosing evaluation of three doses of varoglutamstat or placebo for ≥ 24 weeks. VIVIAD (NCT04498650) is a phase 2B, multicenter, randomized, double-blind, placebo-controlled, parallel-group, dose-finding study being conducted to evaluate the safety, tolerability, and efficacy of varoglutamstat in 259 participants with MCI and mild dementia due to AD.

ABBV-916 is a monoclonal antibody to Aβ. It recognizes N-terminal truncated Aβ modified with pyroglutamate at position 3 (N3), a form of Aβ that is aggregated into amyloid plaques. A two-stage phase 2 trial of ABBV-916 is ongoing (NCT05291234). Stage A is a multiple ascending dose study, and participants have a 25% chance of receiving placebo. Stage B is a proof-of-concept study, and participants have a 20% chance of receiving placebo. The first 6 months of the study are a double-blinded period, which is to be followed by a 2-year extension period in which all participants receive ABBV-916. Approximately 195 participants aged 50–90 years are to be enrolled at approximately 90 sites across the world. The participants are to receive intravenous doses of ABBV-916 or placebo once every 4 weeks for 24 weeks and are to be followed up for an additional 16 weeks.

CT1812 is a ligand that targets the component 1 subunit of the sigma2/progesterone membrane receptor. It functions as a negative allosteric regulator, reducing the affinity of oligomeric Aβ and interfering with Aβ-induced synaptic toxicity [ 57 ]. START(COG0203) study (NCT05531656) is a phase 2, multicenter, randomized, double-blind, placebo-controlled trial that was initiated in September 2022 for evaluating the efficacy and safety of CT1812. START is comparing the effects of CT1812 (100 or 300 mg) with those of placebo over 18 months in 540 people with MCI or mild dementia due to AD. The SHINE (COG0201) study (NCT03507790) is a multicenter, randomized, double-blind, placebo-controlled, parallel-group, 36-week phase 2 study of two doses of CT1812 in adults with mild-to-moderate AD. The study is evaluating the safety, tolerability, pharmacokinetics, and efficacy of CT1812.

Anti-tau therapy

Table 4 summarizes the ongoing phase 2 trials of anti-tau therapy.

Bepranemab (UCB0107) is a monoclonal IgG4 antibody that targets a central tau epitope. An ongoing phase 2 trial (NCT04867616) enrolling 421 participants with prodromal or mild AD is investigating the safety, tolerability, and efficacy of bepranemab. After an 80-week double-blinded treatment period, the participants are eligible to enter a 48-week open-label extension period, in which they are to receive bepranemab treatment for 44 weeks. Subsequently, they are to participate in a safety evaluation visit 20 weeks after the last infusion. The primary outcome measure is the CDR-SB score.

JNJ-63733657 is a humanized IgG1 monoclonal antibody that targets the microtubule-binding region of tau and prevents the cell-to-cell propagation of pathogenic tau aggregates. The AUTONOMY trial (NCT04619420) is an ongoing phase 2, randomized, double-blind, placebo-controlled, parallel-group multicenter study. Participants with early AD symptoms and a positive tau PET scan are randomized to groups receiving JNJ-63733657 or placebo. This trial is enrolling 420 participants and is expected to be completed by November 2025. The primary outcome measure is clinical decline, as determined using the iADRS.

ACI-35 is a liposome-based vaccine that targets pathological conformations of phosphorylated tau. A phase 1b/2a multicenter, double-blind, randomized, placebo-controlled trial (NCT04445831) was conducted to evaluate the safety, tolerability, and immunogenicity of various doses, regimens, and combinations of tau-targeting vaccines in individuals with early AD. The vaccines tested were JACI-35.054 and ACI-35.030 at various dose levels. The findings were presented at the 2022 CTAD conference. The results indicated that participants who received ACI-35.030 exhibited a strong and sustained immune response against pathological tau proteins (pTau) and nonphosphorylated tau (ePHF), particularly in the mid- and low-dose groups. Recipients of JACI-35.054 also displayed a robust immune response against ePHF and pTau, but without a clear dose–effect relationship. The trial has been conducted across nine centers in Finland, Sweden, the Netherlands, and the United Kingdom and is expected to be completed by October 2023.

E2814 is a monoclonal IgG1 antibody that targets an HVPGG epitope in the microtubule-binding domain of tau, prevents cell-to-cell propagation, and mediates the clearance of pathogenic tau proteins. The DIAN-TU-001 (E2814) trial (NCT05269394) is a phase 2/3 multicenter, randomized, double-blind, placebo-controlled platform trial of potential disease-modifying therapies with biomarker, cognitive, and clinical endpoints. The trial is enrolling patients with dominantly inherited AD. The study design involves the use of the anti-amyloid antibody lecanemab. Some participants are receiving a matching placebo plus lecanemab, whereas others are receiving concurrent therapy with E2814 plus lecanemab.

LY3372689 is a small-molecule inhibitor of O-GlcNAcase, which promotes tau glycosylation and prevents tau aggregation [ 58 ]. A phase 2 trial (NCT05063539) was initiated in September 2021 for assessing the safety, tolerability, and efficacy of LY3372689 in 330 patients with early symptomatic AD with progressive memory changes for ≥ 6 months and who met the criterion of having a positive flortaucipir-PET scan.

BIIB080 is a tau DNA/RNA-based antisense oligonucleotide that inhibits the translation of tau mRNA into protein, thus suppressing tau expression. CELIA (NCT05399888) is an ongoing phase 2 trial that is aiming to determine whether BIIB080 can delay AD progression in comparison with placebo and to identify the most effective dose of BIIB080. In March 2019, Biogen/Ionis performed a 4-year open-label extension trial of quarterly injections for individuals who completed the randomized portion of the trial. The initial data of this trial were reported at the Alzheimer’s Association International Conference (2021), revealing no serious adverse events from the intrathecal injection of BIIB080 at either of three doses every month for 3 months or two high-dose injections 3 months apart. BIIB080 led to a dose-dependent reductions of 30%–50% in total tau and pTau181 levels in CSF.

Neuroprotectors and cognitive enhancers

Table 5 summarizes the ongoing phase 3 trials for therapies other than anti-amyloid/tau treatment.

The active metabolite of fosgonimeton (ATH-1017) is a positive modulator of hepatocyte growth factor (HGF)/MET signaling [ 59 ]. A phase 3 trial of fosgonimeton (NCT04488419) was initiated in September 2020 and is expected to be completed in February 2024. This study is evaluating the safety and efficacy of fosgonimeton in participants with mild-to-moderate AD, with double-blind, parallel-arm treatment implemented for 26 weeks. The primary outcome measure is the overall treatment effect of fosgonimeton, as measured using the Global Statistical Test, which combines cognition (ADAS-Cog) and function (ADCS-ADL) scores.

AR-1001 selectively inhibits phosphodiesterase 5 and suppresses cGMP hydrolysis, resulting in the activation of protein kinase G and the increased phosphorylation of the cAMP-responsive element-binding protein at Ser133. It can rescue long-term potentiation impairment and cognitive dysfunction in animal models of AD [ 60 ]. A phase 3 trial of AR-1001 (NCT05531526) was started in December 2022 and is estimated to be completed in December 2027. The study aims to evaluate the efficacy and safety of AR1001 in participants with early AD. The primary outcome measure is the change in the CDR-SB from baseline to week 52.

BPDO-1603 is a potential cognitive-enhancing drug for AD, but its mechanism of action remains unknown [ 61 ]. A phase 3 trial of BPDO-1603 (NCT04229927) was started in February 2020 and is estimated to be completed in March 2023. The study has been undertaken to evaluate the efficacy and safety of BPDO-1603 in patients with moderate-to-severe AD. The primary outcome measures are the change in Severe Impairment Battery total scores from baseline to week 24, and CIBIC-plus total scores at week 24.

Buntanetap is a novel translational inhibitor of multiple neurotoxic proteins, including APP, tau, and α-synuclein, by enhancing the binding of the atypical iron response element in the 5′UTR regions of the mRNA of the neurotoxic proteins to iron regulatory protein 1 [ 62 ]. In February 2023, phase 2 and 3 trials (NCT05686044) were initiated to measure the efficacy and safety of three doses of buntanetap in comparison with placebo in participants with mild-to-moderate AD. The primary outcome measures are ADAS-Cog and ADCS Clinical Global Impression of Change (ADCS-CGIC) scores.

Caffeine is an adenosine receptor antagonist that has been reported to be associated with slower cognitive decline and lower cerebral amyloid accumulation [ 63 ]. A phase 3 trial of caffeine (NCT04570085) was started in March 2021 to evaluate the efficacy of 30 weeks of caffeine intake in comparison with placebo on cognitive decline in patients with mild-to-moderate AD dementia (Mini-Mental State Examination scores: 16–24). The primary outcome measure is changes in neuropsychological test battery scores between the randomized value and the value after 30 weeks of treatment.

Hydralazine may have anti-neurodegenerative effects because it activates the Nrf2 pathway, which involves more than 200 antioxidant proteins; improves mitochondrial function; and increases respiration capacity and the production of adenosine triphosphate; hydralazine also activates autophagy, which aids in the clearance of intracellular aggregates [ 64 , 65 , 66 ]. A phase 3 trial of hydralazine (NCT04842552) was started in August 2021 and is anticipated to be completed in December 2023. The study is comparing the effects of 75 mg hydralazine versus placebo in patients with mild-to-moderate AD. Various cognitive and function tests, including olfactory tests, biochemical analyses, and adverse effect monitoring, are being conducted regularly during follow-up.

KarXT (xanomeline-trospium), comprised of muscarinic agonist xanomeline and muscarinic antagonist trospium, is designed to preferentially activate muscarinic receptor in the CNS and ameliorate the peripheral muscarinic side effects. It is reported that KarXT improves cognition in patients with AD and schizophrenia [ 67 ]. A 38-week phase 3 trial comparing the effects of KarXT (NCT05511363) and placebo in participants with psychosis associated with AD dementia was started in August 2022. The trial is analyzing the time from randomization to relapse (primary outcome) as well as the time from randomization to discontinuation for any reason and the safety and tolerability of KarXT (secondary outcomes).

Metformin, a commonly prescribed antidiabetic medication, has been reported to improve cognition or mood in many neurological disorders [ 68 , 69 ]. A phase 3 trial of metformin (NCT04098666) was started in March 2021 and is anticipated to be completed in April 2026. The primary outcome measure is the total recall of the Free and Cued Selective Reminding Test at 24 months.

Nilotinib is a tyrosine kinase inhibitor that preferentially targets discoidin domain receptors and can effectively reduce the occurrence of misfolded proteins in animal models of neurodegeneration by crossing the blood–brain barrier and promoting Aβ and tau degradation [ 70 ]. A phase 3 trial (NCT05143528) was initiated in February 2022 to investigate the safety and efficacy of nilotinib BE (bioequivalent) in individuals with early AD. The primary outcome measure is changes in CDR-SB scores between baseline and week 72.

Piromelatine is a melatonin MT1/2/3 and serotonin 5-HT-1A/1D receptor agonist and was developed as a treatment for mild AD [ 71 ]. In May 2022, a randomized trial (NCT05267535) was initiated in 225 noncarriers of a specific polymorphism, and these participants with mild dementia due to AD are allocated at a ratio of 1:1 to receive piromelatine or placebo for 26 weeks. A 12-month extension involves treating the placebo group with piromelatine to assess the drug’s disease-modifying effects. The primary analysis will be conducted after the initial 26 weeks. If efficacy is not confirmed, the study is to end without the extension phase.

Semaglutide is a peptidic GLP-1 receptor agonist that may regulate the aggregation of Aβ in AD. GLP-1 receptors are involved in cognition, synaptic transmission in hippocampal neurons, and cell apoptosis; thus, they may serve as targets for exploring candidate drugs with neuroprotective and cognition-enhancing effects [ 72 ]. A phase 3 trial of semaglutide (NCT04777396) was started in May 2021 to investigate the efficacy of semaglutide in individuals with early AD. The primary outcome measure is changes in the CDR-SB score from baseline to week 104.

Tricaprilin, a semisynthetic medium-chain triglyceride, is hydrolyzed to octanoic acid after administration and is further metabolized to ketones, which serve as an alternative energy substrate for the brain [ 73 ]. Therefore, tricaprilin can be used as a ketogenic source for the management of mild-to-moderate AD. A phase 3 trial (NCT04187547) was started in June 2022 to evaluate the efficacy and safety of tricaprilin in participants with mild-to-moderate AD. The primary outcome measure is changes in ADAS-Cog scores from baseline to week 20.

Anti-neuroinflammation therapy

Masitinib, an oral tyrosine kinase inhibitor, exerts effects by inhibiting mast cell and microglia/macrophage activity, with significant CNS penetration [ 74 ]. It is currently undergoing a phase 3 trial (NCT05564169) with 600 participants, employing a randomized, double-blind, placebo-controlled, parallel-group design over 24 weeks, followed by a 24-week extension phase. Quadruple masking ensures blinding. The study aims to evaluate Masitinib as an adjunct therapy for mild to moderate AD. Estimated to conclude on December 15, 2025, the trial assesses primary outcomes through changes from baseline in ADAS-Cog-11 and ADCS-ADL scores, measuring cognitive and functional abilities, respectively.

NE3107 is an anti-inflammatory insulin sensitizer that can cross the blood–brain barrier and bind to ERK. NE3107 can selectively inhibit inflammation-driven ERK- and NF-κB-stimulated inflammatory mediators, including TNF-α, without disturbing their homeostatic functions [ 75 ]. A multicenter phase 3 trial (NCT04669028) was started in August 2021 to investigate the safety and efficacy of NE3107 at 20 mg that was orally administered twice daily versus placebo in adult participants with mild-to-moderate AD. The primary outcome measures are changes in ADAS-Cog12 and ADCS-CGIC scores from baseline to week 30 [ 76 ].

BPSD-relieving therapy

Masupirdine, a selective 5‐HT6 receptor antagonist with favorable physicochemical properties and absorption, distribution, metabolism, and excretion properties, may have beneficial effects on agitation, aggression, and psychosis in patients with moderate AD [ 77 ]. A phase 3 trial (NCT05397639) was started in November 2022 to evaluate the efficacy, safety, tolerability, and pharmacokinetics of masupirdine in comparison with placebo for treating agitation in participants with AD dementia. The primary outcome measure is the change in the score of the Cohen–Mansfield Agitation Inventory from baseline to week 12.

Nabilone is a partial agonist of cannabinoid receptor 1 (CB1) and CB2 in the brain and in peripheral tissues, and it has been reported to provide effective treatment for agitation in patients with AD [ 78 ]. A phase 3 trial (NCT04516057) was started in February 2021 to investigate whether nabilone is an effective treatment for agitation in AD patients. The primary outcome measure is agitation (Cohen–Mansfield Agitation Inventory) between baseline and week 8.

Phase 4 and repurposing trials

Table 6 summarizes ongoing phase 4 trials.

Escitalopram, a selective-serotonin reuptake inhibitor, is a commonly used antidepressant. It ameliorates cognitive impairment and could selectively attenuate phosphorylated tau accumulation in stressed rats by regulating hypothalamic–pituitary–adrenal axis activity and the insulin receptor substrate/glycogen synthase kinase-3β signaling pathway [ 79 ]. A phase 4 trial (NCT05004987) was started in February 2022 to investigate whether a reduction in depressive symptoms owing to the administration of escitalopram oxalate is associated with the normalization of AD biomarkers in CSF and inflammatory markers in the peripheral blood. The primary outcome measures are changes in CSF Aβ40 and Aβ42 levels, vascular dysfunction biomarker levels, and scores of the Montgomery–Asberg Depression Ratio Scale at week 8.

Sodium oligomannate (GV-971), a marine-derived oligosaccharide, can reconstitute the gut microbiota, reduce bacterial metabolite–driven peripheral infiltration of immune cells into the brain, inhibit amyloid-β fibril formation, and inhibit neuroinflammation in the brain, as demonstrated in animal studies [ 80 , 81 ]. A phase 4 trial (NCT05181475) was initiated in December 2021 to examine the long-term efficacy and safety of GV-971 as well as changes in blood and gut microbiota biomarkers and thereby validate its mechanism of action and establish guidance for the more rational use of drugs in clinical practice. The primary outcome measure is changes in ADAS-Cog11 scores from baseline to week 48. Another phase 4 trial was started in July 2022 and is comparing the efficacy and safety of memantine and GV-971 monotherapy and combination therapy in patients with moderate-to-severe AD. The primary outcome measure is changes in cognitive function at weeks 12, 24, 36, and 48.

Spironolactone, an aldosterone mineralocorticoid receptor antagonist, has been commonly used to treat cardiovascular diseases, including hypertension. It has anti-inflammatory effects on the peripheral tissues and central nervous system and therefore may have beneficial effects on neurological disorders [ 82 ]. A phase 4 trial (NCT04522739) was started in September 2022 to investigate whether spironolactone can be tolerated by older Black American adults with MCI and to determine its effect on memory and thinking abilities, as measured by participant performance on cognitive tests. The primary outcome measures are the number of adverse events and the attrition rate.

Published results

Among the clinical trials newly registered in the last 4 years, four articles pertaining to two trials have been published in peer-reviewed scientific journals. The characteristics of the published randomized controlled trials are summarized in Table 7 [ 43 , 53 , 83 , 84 ]. Two articles reported the results of NCT03887455 [ 43 , 84 ], and the other two reported the results of NCT01760005 [ 53 , 83 ]. The articles were published between 2018 and 2023. The results of both NCT03887455 (Clarity AD) and NCT01760005 have been discussed in the anti-amyloid section. The methodological quality of these studies is summarized in Table 8 . Both trials (NCT03887455 and NCT01760005) had a overall low risk of bias [ 43 , 53 , 83 , 84 ].

Our understanding of AD originated from clinical research, and how pathological findings are associated with clinical presentation of AD has continued to intrigue the neuroscience research community over the past century. DMTs have become the core of new drug development, and the accumulation of knowledge is leading to the evolution of diagnostic criteria and clinical outcome measurements. The view of clinical outcomes has shifted from considering them as solely determinative to considering them to be just one of the determinants. In accordance with the 2018 NIA-AA Research Framework criteria [ 25 ] or the new 2023 NIA-AA revised criteria for AD [ 26 ], the incorporation of biomarkers is necessary in clinical practice.

This review documented that in terms of the number of AD drug trials and the number of recruited participants, the majority of trials continue to focus on mechanisms involving amyloid and tau. Our 2020 report highlighted that due to the failure of early anti-amyloid trials to achieve their intended outcomes, particularly studies involving BACE inhibitors and monoclonal antibodies, some have questioned whether amyloid remains clinically relevant in AD. This shift in perspective has led to a change in the focus of research toward populations in the prodromal or preclinical stage with positive results for diagnostic biomarkers. Additionally, the validity of the amyloid hypothesis has been contested, resulting in a significant reduction in the number of anti-amyloid phase 3 trials since 2019. However, the targets of both phase 1 and phase 2 trials are diverse, with a noticeable increase in the number of phase 1 trials focusing on neuroprotection and phase 2 trials focusing on anti-neuroinflammation [ 85 ]. Since the positive outcomes in terms of slow decline in cognitive abilities in the lecanemab Clarity AD trial [ 43 ] and the donanemab trial TRAILBLAZER-ALZ [ 86 ], the impact of amyloid and consequent pathological alterations is likely to become the main focus of clinical trials. The incorporation of amyloid-related therapy either as an add-on or as a link to specific aspects of AD pathophysiology might become an important trend in clinical trials of new drugs in the future. However, despite this expansion of research areas, the scope of indications for novel anti-amyloid monoclonal antibody therapy remains limited. The mode of treatment administration and the high monitoring costs along with the need for specialized facilities and imaging scans remain challenges. Other unmet needs, such as addressing BPSD and enhancing cognitive function, necessitate pharmaceutical research. Examining drugs with diverse mechanisms necessitates thorough evaluation that extends beyond mere clinical measurements to encompass their intermediate impact on biomarkers. It is essential to investigate the potential synergy between a new drug and existing medications approved by the US FDA. This approach could even be extended to situations where adjuvant treatment, such as tau-related treatments, is provided after amyloid clearance has been achieved. Clinical trials related to AD have also exhibited a shift in focus toward the earlier stages of AD, such as MCI, or even cognitively healthy participants for developing prevention interventions.

Successful phase 3 trials such as Clarity AD (lecanemab) and EMERGE (aducanumab) have evaluated anti-amyloid treatment in mild AD (Fig.  2 ). Trials that do not target specific pathophysiologies are becoming fewer in all phases (Figs.  2 and 3 ). However, an increasing number of early-phase trials of therapies for symptoms, including cognitive enhancers and agents for relieving BPSD, are being conducted. This reflects the unmet clinical need for such therapies (Figs.  2 and 3 ). Similarly, an increasing number of phase 1 trials involving DMTs, particularly those targeting both anti-amyloid and anti-tau mechanisms, has been noted, indicating the importance of basic research (Fig.  3 ). Outcome measurement tools have also become more diverse, which has enabled meaningful improvements in AD and the efficacy of treatments to be clearly determined in clinical trials. Overall, the field of AD clinical trials is evolving, and additional promising treatments for AD are likely to be developed in the near future.

Trends in Phase 3 trials, 2020–2023, categorized according to event-related themes in ClinicalTrials.gov. Left: Number of Phase 3 trials. Right: Percentage of Phase 3 trials. A anti-amyloid therapy, B anti-tau therapy, C neuroprotection, D anti-neuroinflammation, E cognitive enhancer, F relief of behavioral psychological symptoms of dementia, G others, U undisclosed

Trends in Phase 1 and 2 trials, 2020–2023, categorized according to event-related themes in ClinicalTrials.gov. Left: Number of Phase 2 trials. Right: Number of Phase 1 trials; A anti-amyloid therapy, B anti-tau therapy, C neuroprotection, D anti-neuroinflammation, E cognitive enhancer, F relief of behavioral psychological symptoms of dementia, G others, U undisclosed

Availability of data and materials

Not applicable.

Abbreviations

Amyloid-beta

Acetylcholine

Cholinesterase inhibitors

• Alzheimer disease

Alzheimer’s Disease Assessment Scale–Cognitive Subscale

Alzheimer’s Disease Cooperative Study–Activities of Daily Living Inventory–Mild Cognitive Impairment Version

Apolipoprotein gene

Amyloid precursor protein

Amyloid-related imaging abnormalities

Amyloid, tau, and neurodegeneration biomarkers

Appropriate use recommendations

Autophagic vacuoles

Beta-secretase 1

Behavioral psychological symptoms of dementia

Clinical Dementia Rating scale

Clinical Dementia Rating scale Sum of Box

Caregiver Global Impression of Change

Cerebrospinal fluid

Clinical Trial of AD

Disease-modifyung therapies

Integrated Alzheimer’s Disease Rating Scale

Immunoglobulin gamma 1

Mild cognitive impairment

Magnetic resonance imaging

Nuclear factor κB

Neurofibrillary tangles

Neuropsychiatric Inventory

Positron emission tomography

Presenilin-1

Presenilin-2

Gauthier S WC, Servaes S, Morais JA, Rosa-Neto P. World Alzheimer Report 2022: Life after diagnosis: Navigating treatment, care and support; 2022.

Apostolova LG. Alzheimer disease. Continuum (Minneap Minn). 2016;22(2 Dementia):419–34.

Mega MS, Cummings JL, Fiorello T, Gornbein J. The spectrum of behavioral changes in Alzheimer’s disease. Neurology. 1996;46(1):130–5.

Article   CAS   PubMed   Google Scholar  

McKhann GM, Knopman DS, Chertkow H, Hyman BT, Jack CR Jr, Kawas CH, et al. The diagnosis of dementia due to Alzheimer’s disease: recommendations from the National Institute on Aging-Alzheimer’s Association workgroups on diagnostic guidelines for Alzheimer’s disease. Alzheimers Dement. 2011;7(3):263–9.

Article   PubMed   PubMed Central   Google Scholar  

Glenner GG, Wong CW. Alzheimer’s disease: initial report of the purification and characterization of a novel cerebrovascular amyloid protein. Biochem Biophys Res Commun. 1984;120(3):885–90.

Masters CL, Simms G, Weinman NA, Multhaup G, McDonald BL, Beyreuther K. Amyloid plaque core protein in Alzheimer disease and Down syndrome. Proc Natl Acad Sci U S A. 1985;82(12):4245–9.

Article   CAS   PubMed   PubMed Central   Google Scholar  

Goedert M, Spillantini M, Crowther R. Tau proteins and neurofibrillary degeneration. Brain Pathol. 1991;1(4):279–86.

Sperling R, Mormino E, Johnson K. The evolution of preclinical Alzheimer’s disease: implications for prevention trials. Neuron. 2014;84(3):608–22.

De Leon MJ, George AE, Golomb J, Tarshish C, Convit A, Kluger A, et al. Frequency of hippocampal formation atrophy in normal aging and Alzheimer’s disease. Neurobiol Aging. 1997;18(1):1–11.

Article   PubMed   Google Scholar  

Parker TD, Cash DM, Lane CAS, Lu K, Malone IB, Nicholas JM, et al. Hippocampal subfield volumes and pre-clinical Alzheimer’s disease in 408 cognitively normal adults born in 1946. PLoS ONE. 2019;14(10): e0224030.

Beyreuther K, Masters CL. Amyloid precursor protein (APP) and beta A4 amyloid in the etiology of Alzheimer’s disease: precursor-product relationships in the derangement of neuronal function. Brain Pathol. 1991;1(4):241–51.

Hardy J, Allsop D. Amyloid deposition as the central event in the aetiology of Alzheimer’s disease. Trends Pharmacol Sci. 1991;12(10):383–8.

Selkoe DJ. The molecular pathology of Alzheimer’s disease. Neuron. 1991;6(4):487–98.

Kang J, Lemaire H-G, Unterbeck A, Salbaum JM, Masters CL, Grzeschik K-H, et al. The precursor of Alzheimer’s disease amyloid A4 protein resembles a cell-surface receptor. Nature. 1987;325(6106):733–6.

Gunawardena S, Yang G, Goldstein LS. Presenilin controls kinesin-1 and dynein function during APP-vesicle transport in vivo. Hum Mol Genet. 2013;22(19):3828–43.

Kolata G. Down Syndrome—Alzheimer’s Linked: Down syndrome adults get Alzheimer-like changes in their brains and many become demented, leading researchers to ask about the connection. Science. 1985;230(4730):1152–3.

FDA Converts Novel Alzheimer’s Disease Treatment to Traditional Approval-Action Follows Confirmatory Trial to Verify Clinical Benefit. 2023.

Medina M, Avila J. The role of extracellular Tau in the spreading of neurofibrillary pathology. Front Cell Neurosci. 2014;8:113.

Jack CR, Knopman DS, Jagust WJ, Shaw LM, Aisen PS, Weiner MW, et al. Hypothetical model of dynamic biomarkers of the Alzheimer’s pathological cascade. Lancet Neurol. 2010;9(1):119–28.

Gómez-Isla T, Hollister R, West H, Mui S, Growdon JH, Petersen RC, et al. Neuronal loss correlates with but exceeds neurofibrillary tangles in Alzheimer’s disease. Ann Neurol. 1997;41(1):17–24.

Arriagada PV, Growdon JH, Hedley-Whyte ET, Hyman BT. Neurofibrillary tangles but not senile plaques parallel duration and severity of Alzheimer’s disease. Neurology. 1992;42(3 Pt 1):631–9.

Bierer LM, Hof PR, Purohit DP, Carlin L, Schmeidler J, Davis KL, et al. Neocortical neurofibrillary tangles correlate with dementia severity in Alzheimer’s disease. Arch Neurol. 1995;52(1):81–8.

Okamura N, Yanai K. Applications of tau PET imaging. Nat Rev Neurol. 2017;13(4):197–8.

Jack CR Jr, Bennett DA, Blennow K, Carrillo MC, Feldman HH, Frisoni GB, et al. A/T/N: an unbiased descriptive classification scheme for Alzheimer disease biomarkers. Neurology. 2016;87:539–47.

Jack CR Jr, Bennett DA, Blennow K, Carrillo MC, Dunn B, Haeberlein SB, et al. NIA-AA Research Framework: Toward a biological definition of Alzheimer’s disease. Alzheimers Dement. 2018;14(4):535–62.

NIA-AA Revised Clinical Criteria for Alzheimer's Disease. 2023.

Lian H, Yang L, Cole A, Sun L, Chiang AC, Fowler SW, et al. NFκB-activated astroglial release of complement C3 compromises neuronal morphology and function associated with Alzheimer’s disease. Neuron. 2015;85(1):101–15.

Liddelow SA, Guttenplan KA, Clarke LE, Bennett FC, Bohlen CJ, Schirmer L, et al. Neurotoxic reactive astrocytes are induced by activated microglia. Nature. 2017;541(7638):481–7.

Hamelin L, Lagarde J, Dorothée G, Leroy C, Labit M, Comley RA, et al. Early and protective microglial activation in Alzheimer’s disease: a prospective study using 18 F-DPA-714 PET imaging. Brain. 2016;139(4):1252–64.

Gomez-Nicola D, Boche D. Post-mortem analysis of neuroinflammatory changes in human Alzheimer’s disease. Alzheimers Res Ther. 2015;7(1):42.

Zimmer ER, Leuzy A, Benedet AL, Breitner J, Gauthier S, Rosa-Neto P. Tracking neuroinflammation in Alzheimer’s disease: the role of positron emission tomography imaging. J Neuroinflammation. 2014;11(1):1–12.

Article   Google Scholar  

McGeer PL, McGeer EG. The amyloid cascade-inflammatory hypothesis of Alzheimer disease: implications for therapy. Acta Neuropathol. 2013;126(4):479–97.

Lockhart SN, Schaich CL, Craft S, Sachs BC, Rapp SR, Jung Y, et al. Associations among vascular risk factors, neuroimaging biomarkers, and cognition: Preliminary analyses from the Multi-Ethnic Study of Atherosclerosis (MESA). Alzheimers Dement. 2022;18(4):551–60.

Wanleenuwat P, Iwanowski P, Kozubski W. Alzheimer’s dementia: pathogenesis and impact of cardiovascular risk factors on cognitive decline. Postgrad Med. 2019;131(7):415–22.

Lee J-H, Yang D-S, Goulbourne CN, Im E, Stavrides P, Pensalfini A, et al. Faulty autolysosome acidification in Alzheimer’s disease mouse models induces autophagic build-up of Aβ in neurons, yielding senile plaques. Nat Neurosci. 2022;25(6):688–701.

Chi H, Chang H-Y, Sang T-K. Neuronal cell death mechanisms in major neurodegenerative diseases. Int J Mol Sci. 2018;19(10):3082.

Raina P, Santaguida P, Ismaila A, Patterson C, Cowan D, Levine M, et al. Effectiveness of cholinesterase inhibitors and memantine for treating dementia: evidence review for a clinical practice guideline. Ann Intern Med. 2008;148(5):379–97.

Arndt JW, Qian F, Smith BA, Quan C, Kilambi KP, Bush MW, et al. Structural and kinetic basis for the selectivity of aducanumab for aggregated forms of amyloid-β. Sci Rep. 2018;8(1):6412.

Dunn B, Stein P, Cavazzoni P. Approval of aducanumab for Alzheimer disease—the FDA’s perspective. JAMA Intern Med. 2021;181(10):1276–8.

Tucker S, Möller C, Tegerstedt K, Lord A, Laudon H, Sjödahl J, et al. The murine version of BAN2401 (mAb158) selectively reduces amyloid-β protofibrils in brain and cerebrospinal fluid of tg-ArcSwe mice. J Alzheimers Dis. 2015;43(2):575–88.

Canady VA. FDA approves new treatment for Alzheimer’s disease. Ment Heal Wkly. 2023;33(3):6–7.

Swanson CJ, Zhang Y, Dhadda S, Wang J, Kaplow J, Lai RY, et al. A randomized, double-blind, phase 2b proof-of-concept clinical trial in early Alzheimer’s disease with lecanemab, an anti-Aβ protofibril antibody. Alzheimers Res Ther. 2021;13(1):80.

Van Dyck CH, Swanson CJ, Aisen P, Bateman RJ, Chen C, Gee M, et al. Lecanemab in early Alzheimer’s disease. N Engl J Med. 2023;388(1):9–21.

Cummings J, Rabinovici G, Atri A, Aisen P, Apostolova L, Hendrix S, et al. Aducanumab: appropriate use recommendations update. J Prev Alzheimers Di s. 2022;9(2):221–30.

CAS   Google Scholar  

Cummings J, Apostolova L, Rabinovici G, Atri A, Aisen P, Greenberg S, et al. Lecanemab: appropriate use recommendations. J Prev Alzheimers Dis. 2023;10(3):362–77.

CAS   PubMed   Google Scholar  

Irizarry MC, Sims JR, Lowe SL, Nakano M, Hawdon A, Willis BA, et al. O4–08-06: Safety, pharmacokinetics (PK), and florbetapir F-18 positron emission tomography (PET) after multiple dose administration of LY3002813, a β-amyloid plaque-specific antibody, in Alzheimer’s disease. Alzheimers Dement. 2016;12:P352–3.

Bouter Y, Liekefeld H, Pichlo S, Westhoff AC, Fenn L, Bakrania P, et al. Donanemab detects a minor fraction of amyloid-β plaques in post-mortem brain tissue of patients with Alzheimer’s disease and Down syndrome. Acta Neuropathol. 2022;143(5):601–3.

Shcherbinin S, Evans CD, Lu M, Andersen SW, Pontecorvo MJ, Willis BA, et al. Association of amyloid reduction after donanemab treatment with tau pathology and clinical outcomes: the TRAILBLAZER-ALZ randomized clinical trial. JAMA Neurol. 2022;79(10):1015–24.

Pontecorvo MJ, Lu M, Burnham SC, Schade AE, Dage JL, Shcherbinin S, et al. Association of donanemab treatment with exploratory plasma biomarkers in early symptomatic Alzheimer disease: a secondary analysis of the TRAILBLAZER-ALZ randomized clinical trial. JAMA Neurol. 2022;79(12):1250–9.

Sims JR, Zimmer JA, Evans CD, Lu M, Ardayfio P, Sparks J, et al. Donanemab in early symptomatic Alzheimer disease: the TRAILBLAZER-ALZ 2 randomized clinical trial. JAMA. 2023;330(6):512–27.

Wessels AM, Dennehy EB, Dowsett SA, Dickson SP, Hendrix SB. Meaningful clinical changes in Alzheimer disease measured with the iADRS and illustrated using the donanemab TRAILBLAZER-ALZ study findings. Neurol Clin Pract. 2023;13(2): e200127.

Honig LS, Vellas B, Woodward M, Boada M, Bullock R, Borrie M, et al. Trial of solanezumab for mild dementia due to Alzheimer’s disease. N Engl J Med. 2018;378(4):321–30.

Salloway S, Farlow M, McDade E, Clifford DB, Wang G, Llibre-Guerra JJ, et al. A trial of gantenerumab or solanezumab in dominantly inherited Alzheimer’s disease. Nat Med. 2021;27(7):1187–96.

Hey JA, Yu JY, Versavel M, Abushakra S, Kocis P, Power A, et al. Clinical pharmacokinetics and safety of ALZ-801, a novel prodrug of tramiprosate in development for the treatment of Alzheimer’s disease. Clin Pharmacokinet. 2018;57(3):315–33.

Wang HY, Bakshi K, Frankfurt M, Stucky A, Goberdhan M, Shah SM, et al. Reducing amyloid-related Alzheimer’s disease pathogenesis by a small molecule targeting filamin A. J Neurosci. 2012;32(29):9773–84.

Lues I, Weber F, Meyer A, Bühring U, Hoffmann T, Kühn-Wache K, et al. A phase 1 study to evaluate the safety and pharmacokinetics of PQ912, a glutaminyl cyclase inhibitor, in healthy subjects. Alzheimers Dement (N Y). 2015;1(3):182–95.

Grundman M, Morgan R, Lickliter JD, Schneider LS, DeKosky S, Izzo NJ, et al. A phase 1 clinical trial of the sigma-2 receptor complex allosteric antagonist CT1812, a novel therapeutic candidate for Alzheimer’s disease. Alzheimers Dement (N Y). 2019;5:20–6.

Paul S, Haskali MB, Liow JS, Zoghbi SS, Barth VN, Kolodrubetz MC, et al. Evaluation of a PET radioligand to image O-GlcNAcase in brain and periphery of rhesus monkey and knock-out mouse. J Nucl Med. 2019;60(1):129–34.

Johnston JL, Reda SM, Setti SE, Taylor RW, Berthiaume A-A, Walker WE, et al. Fosgonimeton, a novel positive modulator of the HGF/MET system, promotes neurotrophic and procognitive effects in models of dementia. Neurotherapeutics. 2023;20(2):431–51.

Kang BW, Kim F, Cho J-Y, Kim S, Rhee J, Choung JJ. Phosphodiesterase 5 inhibitor mirodenafil ameliorates Alzheimer-like pathology and symptoms by multimodal actions. Alzheimers Res Ther. 2022;14(1):92.

Cummings J, Lee G, Nahed P, Kambar MEZN, Zhong K, Fonseca J, et al. Alzheimer’s disease drug development pipeline: 2022. Alzheimers Dement (N Y). 2022;8(1): e12295.

Fang C, Hernandez P, Liow K, Damiano E, Zetterberg H, Blennow K, et al. Buntanetap, a novel translational inhibitor of multiple neurotoxic proteins, proves to be safe and promising in both Alzheimer’s and Parkinson’s patients. J Prev Alzheimers Dis. 2023;10(1):25–33.

Gardener SL, Rainey-Smith SR, Villemagne VL, Fripp J, Doré V, Bourgeat P, et al. Higher coffee consumption is associated with slower cognitive decline and less cerebral Aβ-Amyloid accumulation over 126 months: data from the Australian imaging, biomarkers, and lifestyle study. Front Aging Neurosci. 2021;13: 744872.

Chhunchha B, Kubo E, Krueger RR, Singh D. Hydralazine revives cellular and ocular lens health-span by ameliorating the aging and oxidative-dependent loss of the Nrf2-activated cellular stress response. Antioxidants (Basel). 2023;12(1):140.

Chang T-T, Chen J-W. Potential Impacts of Hydralazine as a Novel Antioxidant on Cardiovascular and Renal Disease—Beyond Vasodilation and Blood Pressure Lowering. Antioxidants (Basel). 2022;11(11):2224.

Dehghan E, Goodarzi M, Saremi B, Lin R, Mirzaei H. Hydralazine targets cAMP-dependent protein kinase leading to sirtuin1/5 activation and lifespan extension in C. elegans . Nat Commun. 2019;10(1):4905.

Sauder C, Allen LA, Baker E, Miller AC, Paul SM, Brannan SK. Effectiveness of KarXT (xanomeline-trospium) for cognitive impairment in schizophrenia: post hoc analyses from a randomised, double-blind, placebo-controlled phase 2 study. Transl Psychiatry. 2022;12(1):491.

Li N, Zhou T, Fei E. Actions of metformin in the brain: a new perspective of metformin treatments in related neurological disorders. Int J Mol Sci. 2022;23(15):8281.

Khaleghi-Mehr M, Delshad A-A, Shafie-Damavandi S, Roghani M. Metformin mitigates amyloid β1-40-induced cognitive decline via attenuation of oxidative/nitrosative stress and neuroinflammation. Metab Brain Dis. 2023;38(4):1127–42.

Turner RS, Hebron ML, Lawler A, Mundel EE, Yusuf N, Starr JN, et al. Nilotinib effects on safety, tolerability, and biomarkers in Alzheimer’s disease. Ann Neurol. 2020;88(1):183–94.

Schneider L, Laudon M, Nir T, Caceres J, Ianniciello G, Capulli M, et al. A polymorphism cluster at the 2q12 locus may predict response to piromelatine in patients with mild Alzheimer’s disease. J Prev Alzheimers Dis. 2022;9(2):247–54.

Mahapatra MK, Karuppasamy M, Sahoo BM. Therapeutic potential of semaglutide, a newer GLP-1 receptor agonist, in abating obesity, non-alcoholic steatohepatitis and neurodegenerative diseases: a narrative review. Pharm Res. 2022;39(6):1233–48.

Li Z, et al. Modeling digestion, absorption, and ketogenesis after administration of tricaprilin formulations to humans. Eur J Pharm Biopharm. 2023;182:41–52.

Li T, Martin E, Abada YS, et al. Effects of chronic masitinib treatment in APPswe/PSEN1dE9 transgenic mice modeling Alzheimer’s disease. J Alzheimers Dis. 2020;76(4):1339–45.

Reading CL, Ahlem CN, Murphy MF. NM101 Phase III study of NE3107 in Alzheimer’s disease: rationale, design and therapeutic modulation of neuroinflammation and insulin resistance. Neurodegener Dis Manag. 2021;11(4):289–98.

Nirogi R, Jayarajan P, Benade V, Shinde A, Goyal VK, Jetta S, et al. Potential beneficial effects of masupirdine (SUVN-502) on agitation/aggression and psychosis in patients with moderate Alzheimer’s disease: exploratory post hoc analyses. Int J Geriatr Psychiatry. 2022. https://doi.org/10.1002/gps.5813 .

Herrmann N, Ruthirakuhan M, Gallagher D, Verhoeff N, Kiss A, Black SE, et al. Randomized Placebo-Controlled Trial of Nabilone for Agitation in Alzheimer’s Disease. Am J Geriatr Psychiatry. 2019;27(11):1161–73.

Wu C, Gong W-G, Wang Y-J, Sun J-J, Zhou H, Zhang Z-J, et al. Escitalopram alleviates stress-induced Alzheimer’s disease-like tau pathologies and cognitive deficits by reducing hypothalamic-pituitary-adrenal axis reactivity and insulin/GSK-3β signal pathway activity. Neurobiol Aging. 2018;67:137–47.

Xiao S, Chan P, Wang T, Hong Z, Wang S, Kuang W, et al. A 36-week multicenter, randomized, double-blind, placebo-controlled, parallel-group, phase 3 clinical trial of sodium oligomannate for mild-to-moderate Alzheimer’s dementia. Alzheimers Res Ther. 2021;13(1):62.

Lu J, Pan Q, Zhou J, Weng Y, Chen K, Shi L, et al. Pharmacokinetics, distribution, and excretion of sodium oligomannate, a recently approved anti-Alzheimer’s disease drug in China. J Pharm Anal. 2022;12(1):145–55.

Mehdipour M, Emamghoreishi M, Farrokhi MR, Amirinezhadfard E, Keshavarz M. The effect of spironolactone on β-amyloid-induced memory impairment in male rats: the role of microglial inhibition. Adv Pharm Bull. 2022;12(3):623–31.

Bateman RJ, Benzinger TL, Berry S, Clifford DB, Duggan C, Fagan AM, et al. The DIAN-TU Next Generation Alzheimer’s prevention trial: Adaptive design and disease progression model. Alzheimers Dement. 2017;13(1):8–19.

Tahami Monfared AA, Ye W, Sardesai A, Folse H, Chavan A, Aruffo E, et al. A path to improved Alzheimer’s care: simulating long-term health outcomes of lecanemab in early Alzheimer’s disease from the CLARITY AD trial. Neurol Ther. 2023;12(3):863–81.

Huang L-K, Chao S-P, Hu C-J. Clinical trials of new drugs for Alzheimer disease. J Biomed Sci. 2020;27(1):18.

Mintun MA, Lo AC, Duggan Evans C, Wessels AM, Ardayfio PA, Andersen SW, et al. Donanemab in early Alzheimer’s disease. N Engl J Med. 2021;384(18):1691–704.

Download references

Acknowledgements

This manuscript was edited by Wallace Academic Editing.

The authors have not received any funding.

Author information

Li-Kai Huang and Yi-Chun Kuan contributed equally to this work.

Authors and Affiliations

PhD Program in Medical Neuroscience, College of Medical Science and Technology, Taipei Medical University, No. 291, Zhong Zheng Road, Zhonghe District, New Taipei City, Taiwan

Li-Kai Huang & Chaur-Jong Hu

Taipei Neuroscience Institute, Taipei Medical University, New Taipei City, Taiwan

Li-Kai Huang, Yi-Chun Kuan & Chaur-Jong Hu

Dementia Center and Department of Neurology, Shuang-Ho Hospital, Taipei Medical University, New Taipei City, Taiwan

Department of Neurology, School of Medicine, College of Medicine, Taipei Medical University, Taipei, Taiwan

Yi-Chun Kuan & Chaur-Jong Hu

Department of Biomedical Engineering, National Taiwan University, Taipei, Taiwan

Yi-Chun Kuan

School of Medicine, College of Medicine, Taipei Medical University, Taipei, Taiwan

You can also search for this author in PubMed   Google Scholar

Contributions

LKH and YCK: Conducted literature search, developed the study concept and design, extracted information from trials and studies, and contributed to manuscript drafting and revision. HWL: Extracted information from trials and studies and contributed to manuscript drafting and revision. CJH: Contributed to the study concept and design, interpreted the data and information, finalized and revised the manuscript, and provided overall supervision of the entire project.

Corresponding author

Correspondence to Chaur-Jong Hu .

Ethics declarations

Ethics approval and consent to participate, consent for publication, competing interests.

The authors declare no competing interests.

Additional information

Publisher's note.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ . The Creative Commons Public Domain Dedication waiver ( http://creativecommons.org/publicdomain/zero/1.0/ ) applies to the data made available in this article, unless otherwise stated in a credit line to the data.

Reprints and permissions

About this article

Cite this article.

Huang, LK., Kuan, YC., Lin, HW. et al. Clinical trials of new drugs for Alzheimer disease: a 2020–2023 update. J Biomed Sci 30 , 83 (2023). https://doi.org/10.1186/s12929-023-00976-6

Download citation

Received : 31 May 2023

Accepted : 26 September 2023

Published : 02 October 2023

DOI : https://doi.org/10.1186/s12929-023-00976-6

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

• Clinical trials
• Anti-amyloid
• Neuroprotection
• Cognitive enhancement
• Neuroinflammation

Journal of Biomedical Science

ISSN: 1423-0127

• Submission enquiries: Access here and click Contact Us
• General enquiries: [email protected]

• Reference Manager
• Simple TEXT file

People also looked at

Original research article, alzheimer’s disease and epilepsy: the top 100 cited papers.

• 1 Department of General Practice and International Medicine, Second Affiliated Hospital, School of Medicine, Zhejiang University, Hangzhou, China
• 2 Department of Neurology, Epilepsy Center, Second Affiliated Hospital, School of Medicine, Zhejiang University, Hangzhou, China

Background: Alzheimer’s disease (AD) is one of the common neurodegenerative diseases, which often coexists with epilepsy. It is very significant to study the treatment options and the relationship between AD and epilepsy.

Aims: The purpose of this study was to analyze the top 100 cited papers about AD and epilepsy using bibliometrics, and to describe the current situation and predict research hot spots.

Methods: Top 100 papers were obtained from the Web of Science Core Collection (WoSCC). The WoSCC was used to analyze the author, institution, country, title, keywords, abstract, citation, subject category, publication year, impact factor (IF), and other functions. SPSS25 software was used for statistical analysis and CiteSpace V.5.7.R2 was used to visualize the information through collaborative networks.

Results: The number of publications gradually increased from 2000 to 2021. The total citation count for the top 100 papers ranged from 15 to 433(mean = 67.43). The largest number of papers were published in 2016 ( n = 11). Meanwhile, USA (centrality: 0.93) and Columbia University (centrality: 0.06) were the most influential research country and institutions, respectively. The top contributing journals was Journal of Alzheimer’s Disease (8%). The IF for journals ranged from 1.819 to 53.44. A network analysis of the author’s keywords showed that “beta” (centrality: 0.39), “amyloid beta” (centrality: 0.29), “hyperexcitability” (centrality: 0.29) and “disease” (centrality: 0.29) had a high degree of centrality.

Conclusion: AD and epilepsy have been intensively studied in the past few years. The relationships, mechanisms and treatment of AD and epilepsy will be subjects of active research hotpots in future. These findings provide valuable information for clinicians and scientists to identify new perspectives with potential collaborators and cooperative countries.

Introduction

Alzheimer’s disease (AD) and epilepsy are common neurological diseases. The number of patients with dementia and epilepsy in the global population is increasing, representing a growing problem for global health. The overall lifetime prevalence of epilepsy is 7.60 per 1,000 population [95% confidence interval (CI) 6.17–9.38] and the prevalence of epilepsy tends to peak in the elderly ( Beghi, 2020 ). The prevalence of AD in Europe was estimated at 5.05% (95% CI, 4.73–5.39), and similar to epilepsy, AD prevalence increases with age ( Niu et al., 2017 ). In fact, these two diseases are related. Epilepsy occurs more frequently in patients with AD than in those with non-Alzheimer’s disease ( Samson et al., 1996 ). Recent findings showed that seizures could accelerate the decline of cognitive ability in patients with AD and that there might be an important bidirectional relationship between epilepsy and AD ( Sen et al., 2018 ). AD and Epilepsy also share many pathological similarities ( Lehmann et al., 2021 ), e.g., temporal lobe atrophy, neuronal death, gliosis, neuritic alterations, and neuroinflammation ( Struble et al., 2010 ). In addition, AD plus epilepsy would lead to more serious clinical consequences, such as cognitive decline, weakness, anxiety, depression, social withdrawal, psychological and behavioral comorbidity, and poor treatment compliance ( Cretin, 2021 ). Overtime, a large amount of literature has been published comprising a wide range of relevant research and clinical themes. However, the precise mechanisms leading to the development of seizures in the setting of AD are still under investigation and require further study. A meta-analysis showed that the quality of evidence on the treatment outcome of epilepsy in patients with AD was very low ( Liu and Wang, 2021 ).

As a discipline emerging since its formal foundation, a bibliometric review of the literature was warranted to aid the synthesis and implementation of the evidence base. Despite citation analysis across a broad range of neurosciences ( Yeung et al., 2017 ), there is limited information in the field of AD and epilepsy, with few published studies. Citation counting is an important metric to understand the significance of the contribution of research to a research field ( Fox et al., 2021 ). Previous reviews only relied on individuals to study the research through literature summary and extraction, and thus cannot fully reflect the temporal and spatial distribution of researchers, institutions, and journals. Moreover, it is difficult to visualize the internal structure of the knowledge base and research focus, and systematic, comprehensive, and visual research are rarely found. Therefore, the present study aims to comprehensively analyze the current status, research hotpots, and development trends through a bibliometric analysis of the top 100 papers on AD and epilepsy published from 2000 to 2021. The findings may help follow-up researchers study the association between AD and epilepsy, identify journal publications and collaborators, and analyze keywords and research trends, which might promote research aiming to determine the cause, mechanism, and treatment of the disease.

Materials and methods

Data source.

The retrieval data for measurement and statistical analysis were screened from the Web of Science Core Collection (WoSCC), which provided the citation search, giving access to multiple databases that reference cross-disciplinary research and allowing an in-depth exploration of specialized subfields ( Wu et al., 2021 ). We conducted a literature search from the WoSCC on June 5th, 2022. In this study, the search criteria in the WoSCC database were as follows: ((TI = (epilep* OR seizure* OR convuls*)) OR KP = (epilep* OR seizure* OR convuls*)) AND ((TI = (dement* OR Alzheimer* OR “cognit* impair*” OR AD)) OR KP = (dement* OR Alzheimer* OR “cognit* impair*” OR AD)). Timespan: 2000-01-01 to 2021-12-31 (Publication Date). Document type: articles and reviews; Language: English. A total of 2601 records were retrieved. Then, two independent investigators reviewed the titles and abstracts and deleted studies that were not associated with AD and epilepsy, which excluded 1237 papers according to the criteria ( Guo et al., 2021 ). And 1264 papers were excluded after the 100th rank from the selected literature. Finally, the top 100 studies were determined. A Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flow diagram for the WoSCC results is provided in Figure 1 .

Figure 1. Flow chart of literature collection.

Data analysis

In the present study, WoSCC was used to analyze the author, institution, country, title, keywords, abstract, citation, subject category, publication year, impact factor (IF), and other functions. The selected documents were imported into Excel (Microsoft Corporation, Redmond, WA, United States) and CiteSpace 5.8.R3 (64-bit) ( Palop and Mucke, 2009 ). SPSS 25.0 statistical software (IBM Corp., Armonk, NY, United States) was used for statistical analysis. Continuous variables were expressed as the mean ± SD. Categorical variables were expressed as a percentage. CiteSpace, a bibliometric analysis tool, was created by Dr. Chaomei Chen (School of Information Science and Technology, Drexel University, Philadelphia, PA, United States) and his team. It visualizes countries/regions, institutions, authors and their cooperative relationships, co-cited references, and co-occurrence words through collaborative networks, which has been widely used in biomedical research fields.

Three folders of the research were created, including input folder placing the data downloaded from WoSCC, a data folder containing the data after deleting duplicate documents, and a project folder containing the data processed by cite. We did not find duplicate documents that needed to be deleted. The overall selected time span was from January 2000 to December 2021. Then, the slice length was set as 2 years. The node type was selected according to the type of analysis performed. The link lines between the nodes indicated the collaborative relationships. The size of the circles represented the number of papers published by the country/region, institute, or author. Purple rings indicated that these countries/regions, institutes, or authors had greater centrality.

The 100 most-cited publications

We retrieved the 100 most frequently cited papers related to AD and epilepsy. The results were ranked according to citation counts to represent the 100 most-cited publications (76 articles and 24 reviews). A comprehensive list of the 100 publications and a citation details are presented in the Table 1 .

Table 1. The 100 most-cited publications.

As shown in Table 1 , The 100 most-cited articles received a total of 6743 citations (according to Web of science, WOS). The median number citations was 39, with a range of 15–433. For annual citations, the mean value was 7.41 with a range of 0.8–35.70. Seventeen papers were cited more than 100 times, and 36 were cited more than 50 times. The review entitled “Epilepsy and Cognitive Impairments in Alzheimer Disease” from Palop and Mucke (2009) was the most-cited publication ( n = 433).

Publication years

As shown in Figure 2 , the top 100 most-cited papers were published from 2000 to 2020. Overall, publications showed a fluctuating upward trend. The largest number of studies was published in 2016 ( n = 11), including eight articles and three reviews. The number of papers published in 2011 was higher than the number of papers published between 2000 and 2010.

Figure 2. The number of publications among different types of papers according to publication year.

Contributions of countries

Overall, 22 countries contributed to the included studies, with nine countries publishing only one study. The United States was the largest contributor of studies (38%), followed by Italy (11%) and the United Kingdom (9%). The contributing countries are shown in Figure 3 .

Figure 3. Countries among the top 100 most-cited papers.

Generating a country map using CiteSpace resulted in 22 nodes and 44 links ( Figure 4 ). The 100 papers were published by research groups in 22 countries. The top five countries were the United States, Italy, the United Kingdom, France, and Finland. The top three countries in terms of centrality were the United States (0.93), the United Kingdom (0.44), and France (0.21). An analysis in terms of publication and centrality indicated that the United States, the United Kingdom, Italy, and France were the main research powers in this research. United States has established cooperation with 12 countries, and the strongest collaborations were identified between United States, Canada, Israel, Greece, India, and Switzerland.

Figure 4. Contributions of countries. Purple rings indicated that these countries/regions had greater centrality (no less than 0.1). The size of the circle represents the number of papers published by the country. The shorter the distance between two circles, the greater the cooperation between the two countries. United States was the most influential research country.

Contributions of institutions

A total of 178 institutions published at least one top-cited paper, the distribution of institutions was very scattered, with 21 (11.80%) institutions publishing two papers and 148 (83.15%) institutions publishing only one paper. A small number of institutions accounted for a high proportion of the highest cited papers. Table 2 shows that the top nine institutions collectively published at least three papers. Baylor College of Medicine from United States topped the list with eight papers.

Table 2. Top nine institutions with at least three papers in the top 100 most-cited papers.

Figure 5 shows that generating an institution map resulted in 149 nodes and 308 links. The top 100 publications were distributed among 149 research institutions. The top five institutions were Baylor College Medicine, University of Eastern Finland, Columbia University, and Johns Hopkins University and Kuopio University Hospital. The top four institutions in terms of centrality were Columbia University (0.06), Baylor College of Medicine (0.04), Johns Hopkins University (0.04) and Indiana University (0.03). Analysis in terms of centrality indicated that Columbia University was the most influential research institution.

Figure 5. Contributions of institutions. Nodes represent institutes, and the size of each node corresponds to the co-occurrence frequency of the institutes. The size of the circle represents the number of papers published by the institute. The shorter the distance between two circles, the greater the cooperation between the two institutes. Columbia University was the most influential research institutions.

Distribution of journals

The 100 most-cited articles were published in 59 journals; 18 journals published more than one study, which were distributed in the four partitions of Journal Citation Reports (JCR) ( McVeigh, 2008 ). The JCR data provide both detailed journal information and flexible context, and category information to allow people to understand the way each journal functions in the literature. The major contributing journals are presented in Table 3 . The top five journals that published the 100 most-cited AD and epilepsy studies included Journal of Alzheimer’s Disease ( n = 8), Epilepsia ( n = 6), Epilepsy and Behavior ( n = 4), Neurobiology of Aging ( n = 4) and Epilepsy Research ( n = 4). With regard to the average citation number per paper, Journal of Neuroscience ranked first with a mean of 20.98 citations per paper, followed by Nature Medicine, with a mean of 17.84.

Table 3. Journals contributed ≥ 2 papers in the top 100 most-cited papers.

The IF for journals in the top 100 most-cited papers ranged from 1.819 to 53.44, among which 37 journals had an IF between 3 and 5, 34 journals had an IF between 5 and 10, and 3 journals had an IF above 20 ( Figure 6 ).

Figure 6. The number of articles corresponding to different IF in the top 100 most-cited papers.

In Table 4 , the top 100 papers were classified into different study fields on the basis of WOS categories. The leading WOS category was “Clinical Neurology” ( n = 54), following by “Neurosciences” ( n = 49) and “Behavioral Sciences” ( n = 10).

Table 4. The number of study fields on the basis of WOS categories.

Major contributing authors

Overall, a total of 606 authors contributed to the 100 studies. There was wide, disparate authorship of first authors, with 90 different first authors represented in the 100 included publications. A total of four contributors published the most articles, namely Tanila Heikki, Noebels Jeffrey, Pitkänen, Asla and Scharfman, Helen E, who all published five publications, the total number of citations for the papers were 629, 592, 578, and 234 respectively. Only three authors have published three studies as a first author. Keith A. Vossel from the University of California, as a first author and corresponding author, had the largest number of total citations in 2021 ( n = 185). Table 5 presents results for authors who contributed three or more of the 100 most-cited papers.

Table 5. Major contributing authors in the top 100 most-cited list.

Analysis of co-occurring keywords

CiteSpace was used to extract the keywords from the top 100 papers on AD and epilepsy. A network analysis of the author’s keywords or subject words was carried out during the publication time of the articles, namely, 2000–2021. Table 6 shows that the top five keywords are epilepsy ( n = 41), dementia ( n = 21), mouse model ( n = 21), mild cognitive impairment ( n = 13), and Alzheimer’s disease ( n = 11). The greater the centrality value, the more cooperation between the node and other nodes. Figure 7 shows that “beta” (centrality: 0.39), “amyloid beta” (centrality: 0.29), “hyperexcitability” (centrality: 0.29) and “disease” (centrality: 0.29) had a high degree of centrality during this period. A comprehensive analysis of centrality showed that “beta” ( n = 7, centrality: 0.39), “disease” ( n = 3, centrality: 0.29), “dementia” ( n = 21, centrality: 0.25) and “brain” ( n = 4, centrality: 0.25) are the most influential keywords in this field.

Table 6. Frequency of co-occurring keywords.

Figure 7. Analysis of co-occurrence Keywords. Nodes represent keywords, and the size of each node corresponds to the co-occurring frequency of the keywords. The color of the lines that appear together between keywords indicates chronological order: grey represents the oldest, and orange the newest. “beta” had a highest degree of centrality.

The aim of this paper was to perform a bibliometric study of the top 100 most-cited publications on AD and epilepsy over the last 22 years. To the best of our knowledge, this was the first study to provide an overview the current main status of development, hot spots of study, and the future trends in AD and epilepsy. Our study is of great significance for students, researchers, and clinicians working in the field. Here, we summarize several characteristics of these papers to help understand the history and professional prospects comprehensively.

The average number of citations for the 100 most-cited clinical trials in AD and epilepsy was 67.43 (range: 15–433), well below the number of citations observed in general neuroscience articles (3,087) ( Yeung et al., 2017 ). However, citation counts are not a direct measure of scientific quality and importance. The number of citations can be regarded as a relatively reasonable index to evaluate the quality of papers, which varies in different sub disciplines and depends on the size of the scientific community. Citation count analysis might be related to journal IF, publication frequency, and publication year. In general, an article with 100 or more citations is considered a “classic” in the research field and might even be a seminal paper ( Xiong et al., 2021 ). However, our study showed that only 17 papers were cited more than 100 times, this can possibly come from the fact that comorbid epilepsy in AD may not be an interesting topic for many clinicians. In fact, cognitivists dealing with AD patients are not highly aware of epilepsy and, inversely, epileptologists may be poorly informed of dementing diseases. Therefore, the AD-epilepsy topic may have interest only for a small research community, which can result in a relatively low citation index.

The most-cited publication was “Epilepsy and Cognitive Impairments in Alzheimer’s Disease” ( Palop and Mucke, 2009 ), wrote by Palop and Mucke (2009) , with 433 total citations, 31 annual citations, and 39 citations in 2021. It may be relevant to current research and have far-reaching implications. However, in general, consensus and position papers, guides, and systematic reviews receive more citations than original articles, which is a bias that must be considered in citation analysis. The article that ranked second, which was based on a “Amyloid beta-Induced Neuronal Hyperexcitability Triggers Progressive Epilepsy” ( Minkeviciene et al., 2009 ) had a total of 405 citations, 28.93 annual citations and 54 citations in 2021; it was published in Journal of Neuroscience in 2009 and written by Minkeviciene et al. (2009) . The team performed video-EEG recordings in mice carrying mutant human APPswe and PS1dE9 genes (APdE9 mice) and their wild-type littermates, identifying fibrillar Abeta may be the cause of epileptiform activity.

The number of papers published has increased in recent years. Although the most cited publications tend to be from the first few years (2000–2010), the number of highly cited articles published from 2011–2020 is nearly three times that published from 2000–2010. In 2016, the number of publications was the largest, reaching 11, which showed that a recently published article is likely to gradually improve the quality of research in recent years and have a potential academic importance in the future.

The majority of research originated from the United States (38%). A bibliometric analysis of the most cited articles in neurocritical care research showed that United States was the country with most articles (60, 35 primary research) and citations (6115) among the top 100 ( Ramos et al., 2019 ). Baylor College of Medicine from United States topped the list with eight papers. The distribution of country and institutions was very scattered. The United States and its institutions have played a leading role in AD and epilepsy research. The United States was also ranked first in other fields of neurology ( Xiong et al., 2021 ), and it is also a research leader in terms of quality and quantity. However, compared with other studies, our study showed that the cooperation between countries and institutions was not close ( Yin et al., 2019 ). In fact, just like the study of Parkinson’s disease ( Li et al., 2008 ), more cooperation between different countries and institutions might be an effective way to promote the development of AD and epilepsy research worldwide. Contrastingly, distribution analysis showed that research was widespread all over the world and the diseases had research value.

Baylor College of Medicine and Columbia University played important roles in in the field of AD and epilepsy research. Baylor College of Medicine, as one of the top colleges and universities in Texas, United States, focused on cooperative research programs, discovered basic insights into human health and diseases through extensive interdisciplinary and interdisciplinary cooperation, and applied their findings to develop new diagnostic tools and treatments. Columbia University is a world-class private research university located in Manhattan, NY, United States. It carried out extensive research in the field of neuroscience.

The 100 included articles were published in 59 journals, among which 18 journals published more than one study. The papers were distributed in four partitions of JCR. The IF of the top five journals was less than six. The top journal among the list of the 100 most-cited studies was Journal of Alzheimer’s Disease (IF = 4.472). The leading WOS categories focused on “Clinical Neurology” and “Neurosciences,” with few interdisciplinary studies. Resulting from its interdisciplinary nature, Psycho-Oncology has been subject to extensive interdisciplinary research, which has provided great help for the development of the discipline ( Fox et al., 2021 ). Therefore, we look forward to more interdisciplinary research.

Author analysis revealed a network of core author collaborations in the field of AD and epilepsy research. This information might be relevant to clinical researchers and research institutions who are searching for a network of research leaders in the field to explore potential collaborations. Our authorship analysis does not show as many authors as other areas in research. The most contributing author was Tanila, Heikki from Kuopio University Hospital from Finland. Related research focuses on the pathogenesis of epilepsy in mouse models of AD, which is a research hotspot in this field.

Keyword analysis showed that “beta,” “amyloid beta,” “hyperexcitability,” “disease,” “dementia,” and “brain” had high influential. indicating that these research directions are very significant. This result showed that researchers have begun to extend their research to the pathogenesis. These results also show that AD and epilepsy is still a disease that requires to be solved urgently. We must explore the deficiencies and innovations in this field, such as treatment, pathological mechanism, and disease management on improve the quality of life of patients.

Limitations

The present study has some limitations. First, there are several intrinsic limitations of using citation analysis to evaluate the academic importance of a specific article, author, or journal. There is a certain bias in citation analysis, such as the fact that papers written in English, papers that can be accessed through open access, and papers published in journals with high IFs are more likely to be cited. In addition, through a “snowball effect,” people tended to cite publications that are already highly cited ( Yeung et al., 2017 ). We selected the top 100 papers, but citation searches are “time-dependent,” older articles are likely to be cited more often, and the newest list of highly cited articles may be dominated by some older articles. Furthermore, citation analysis might severely underestimate the impact of clinical research as compared to basic research ( van Eck et al., 2013 ). Second, the search was limited to the WOS database. It did not record citations by textbooks and other databases. Our study only selected papers written in English, which might have yielded an incomplete search.

We identified the 100 most cited papers in the field of AD and epilepsy. By reviewing these top cited papers, researchers can immediately understand the hot topics and research collaborations on AD and epilepsy, and improve their work. This study shows that the relationship, mechanism, and treatment of AD and epilepsy have been widely studied, and in recent years, this field has shown new vitality; however, there is a general lack of cooperation between countries, and the mechanism of epilepsy and AD is unclear, which deserves further study.

Data availability statement

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found in the article/supplementary material.

Author contributions

G-FZ and YG designed the study. G-FZ drafted and edited the manuscript. G-FZ and W-XG analyzed the data. G-FZ, Z-Y-RX, YG, and W-XG revised the manuscript. All authors contributed to the article and approved the submitted version.

This study was supported by the National Natural Science Foundation of China (Grant No. 81871010).

Conflict of interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Beghi, E. (2020). The Epidemiology of Epilepsy. Neuroepidemiology 54, 185–191. doi: 10.1159/000503831

PubMed Abstract | CrossRef Full Text | Google Scholar

Cretin, B. (2021). Treatment of Seizures in Older Patients with Dementia. Drugs Aging 38, 181–192. doi: 10.1007/s40266-020-00826-2

Fox, S., Lynch, J., D’Alton, P., and Carr, A. (2021). Psycho-Oncology: a Bibliometric Review of the 100 Most-Cited Articles. Healthcare 9:1008. doi: 10.3390/healthcare9081008

Guo, Y., Xu, Z. Y. R., Cai, M. T., Gong, W. X., and Shen, C. H. (2021). Epilepsy With Suicide: a Bibliometrics Study and Visualization Analysis via CiteSpace. Front. Neurol. 12:823474. doi: 10.3389/fneur.2021.823474

Lehmann, L., Lo, A., Knox, K. M., and Barker-Haliski, M. (2021). Alzheimer’s Disease and Epilepsy: a Perspective on the Opportunities for Overlapping Therapeutic Innovation. Neurochem. Res. 46, 1895–1912. doi: 10.1007/s11064-021-03332-y

Li, T., Ho, Y. S., and Li, C. Y. (2008). Bibliometric analysis on global Parkinson’s disease research trends during 1991-2006. Neurosci. Lett. 441, 248–252. doi: 10.1016/j.neulet.2008.06.044

Liu, J., and Wang, L. N. (2021). Treatment of epilepsy for people with Alzheimer’s disease. Cochrane Database Syst. Rev. 12:CD011922. doi: 10.1002/14651858.CD011922.pub4

McVeigh, M. E. (2008). Neurology in the journal citation reports. Neurology 71, 1848–1849. doi: 10.1212/01.wnl.0000338902.81462.88

Minkeviciene, R., Rheims, S., Dobszay, M. B., Zilberter, M., Hartikainen, J., and Fülöp, L. (2009). Amyloid beta-induced neuronal hyperexcitability triggers progressive epilepsy. J. Neurosci. 29, 3453–3462. doi: 10.1523/JNEUROSCI.5215-08.2009

Niu, H., Álvarez-Álvarez, I., Guillén-Grima, F., and Aguinaga-Ontoso, I. (2017). Prevalence and incidence of Alzheimer’s disease in Europe: a meta-analysis. Neurologia 32, 523–532. doi: 10.1016/j.nrl.2016.02.016

Palop, J. J., and Mucke, L. (2009). Epilepsy and Cognitive Impairments in Alzheimer Disease. Arch. Neurol. 66, 435–440. doi: 10.1001/archneurol.2009.15

Ramos, M. B., Koterba, E., Rosi Júnior, J., Teixeira, M. J., and Figueiredo, E. G. A. (2019). Bibliometric Analysis of the Most Cited Articles in Neurocritical Care Research. Neurocrit. Care 31, 365–372. doi: 10.1007/s12028-019-00731-6

Samson, W. N., van Duijn, C. M., Hop, W. C., and Hofman, A. (1996). Clinical features and mortality in patients with early-onset Alzheimer’s disease. Eur. Neurol. 36, 103–106. doi: 10.1159/000117218

Sen, A., Capelli, V., and Husain, M. (2018). Cognition and dementia in older patients with epilepsy. Brain 141, 1592–1608. doi: 10.1093/brain/awy022

Struble, R. G., Ala, T., Patrylo, P. R., Brewer, G. J., and Yan, X. X. (2010). Is brain amyloid production a cause or a result of dementia of the Alzheimer’s type? J. Alzheimers Dis. 22, 393–399. doi: 10.3233/JAD-2010-100846

van Eck, N. J., Waltman, L., van Raan, A. F., Klautz, R. J., and Peul, W. C. (2013). Citation analysis may severely underestimate the impact of clinical research as compared to basic research. PLoS One. 8:e62395. doi: 10.1371/journal.pone.0062395

Wu, M. Q., Wu, D. Q., Hu, C. P., and Iao, L. S. (2021). Studies on Children With Developmental Coordination Disorder in the Past 20 Years: a Bibliometric Analysis via CiteSpace. Front Psychiatry 12:776883. doi: 10.3389/fpsyt.2021.776883

Xiong, H. Y., Liu, H., and Wang, X. Q. (2021). Top 100 Most-Cited Papers in Neuropathic Pain From 2000 to 2020: a Bibliometric Study. Front. Neurol. 12:765193. doi: 10.3389/fneur.2021.765193

Yeung, A. W. K., Goto, T. K., and Leung, W. K. (2017). At the Leading Front of Neuroscience: a Bibliometric Study of the 100 Most-Cited Articles. Front. Hum. Neurosci. 11:363. doi: 10.3389/fnhum.2017.00363

Yin, X., Cheng, F., Wang, X., Mu, J., Ma, C., Zhai, C., et al. (2019). Top 100 cited articles on rheumatoid arthritis: a bibliometric analysis. Medicine 98:e14523. doi: 10.1097/md.0000000000014523

Keywords : Alzheimer’s disease, epilepsy, bibliometric study, top-cited, citation

Citation: Zhang G-F, Gong W-X, Xu Z-Y-R and Guo Y (2022) Alzheimer’s disease and epilepsy: The top 100 cited papers. Front. Aging Neurosci. 14:926982. doi: 10.3389/fnagi.2022.926982

Received: 23 April 2022; Accepted: 30 June 2022; Published: 22 July 2022.

Reviewed by:

Copyright © 2022 Zhang, Gong, Xu and Guo. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY) . The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

*Correspondence: Yi Guo, [email protected]

This article is part of the Research Topic

Insights in Alzheimer’s Disease and Related Dementias

Logging off

The Application of Artificial Intelligence in Alzheimer’s Research

Alzheimer’s disease (AD) is an irreversible and neurodegenerative disease that slowly impairs memory and neurocognitive function, but the etiology of AD is still unclear. With the explosive growth of electronic health data, the application of artificial intelligence (AI) in the healthcare setting provides excellent potential for exploring etiology and personalized treatment approaches, and improving the disease’s diagnostic and prognostic outcome. This paper first briefly introduces AI technologies and applications in medicine, and then presents a comprehensive review of AI in AD. In simple, it includes etiology discovery based on genetic data, computer-aided diagnosis (CAD), computer-aided prognosis (CAP) of AD using multi-modality data (genetic, neuroimaging and linguistic data), and pharmacological or non-pharmacological approaches for treating AD. Later, some popular publicly available AD datasets are introduced, which are important for advancing AI technologies in AD analysis. Finally, core research challenges and future research directions are discussed.

FullText for HTML: https://doi.org/.....

References( 149 )

Publication history, acknowledgements.

This work was partially supported by the National Key R&D Program of China (No. 2020YFB2104402) and Beijing Postdoctoral Research Foundation (No. 2022ZZ075).

Rights and permissions

The articles published in this open access journal are distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/).

京ICP备 10035462号-42

Researchers call for a major rethink of how Alzheimer's treatments are evaluated

• PMID: 38480971
• DOI: 10.1038/d41586-024-00756-8

Keywords: Alzheimer's disease; Diseases; Health care; Medical research.

Publication types

• Alzheimer Disease* / drug therapy
• Disease Progression
• Research Personnel*
• Time Factors

Advertisement

Abstract 5562: Altered tumor microenvironment in animal model of concomitant GBM and Alzheimer's pathology

• Split-Screen
• Article contents
• Figures & tables
• Supplementary Data
• Peer Review
• Get Permissions
• Cite Icon Cite
• Search Site
• Version of Record March 22 2024

David Nascari , Ryan Eghlimi , Angad Beniwal , Drake Alton , John Fryer , Nhan L. Tran; Abstract 5562: Altered tumor microenvironment in animal model of concomitant GBM and Alzheimer's pathology. Cancer Res 15 March 2024; 84 (6_Supplement): 5562. https://doi.org/10.1158/1538-7445.AM2024-5562

Download citation file:

• Ris (Zotero)
• Reference Manager

Glioblastoma (GBM) and Alzheimer’s disease (AD) are two devastating central nervous system diagnoses with no known cure. Both diseases are associated with advanced age. The majority of GBM tumors form in the frontal and temporal lobes, two of the brain regions most impacted by amyloid and tau pathology in AD. Females have a higher risk of AD, while males have a higher risk of GBM. Several large case series have established epidemiological evidence that the two diseases are inversely correlated. Nevertheless, it remains unknown whether presence of subclinical amyloid or tau pathology antagonizes the establishment or progression of GBM. The present study sought to characterize GBM growth, progression, and immune response in the setting of concomitant AD pathology using the syngeneic murine GL261 tumor model. To further understand the tumor cell changes that occur in the presence or absence of AD pathology, we engineered GL261 with a dual-tagged vector consisting of RiboTag (to allow for full-length mRNA profiling) and TurboID (to allow for proteomic profiling). GL261 cells were injected intracranially into APP NL-G-F/NL-G-F - MAPT human/human , APP NL-G-F/+ - MAPT human/+ , and wildtype C57BL/6J mice. Tumors in both wildtype and AD conditions contained ~50% intratumoral myeloid cells (Iba+, P2RY12-). Differential activation of intratumoral myeloid cells will be assessed with immunohistochemistry. Microgliosis (IBA1, P2RY12) and astrogliosis (GFAP) around the tumor in the setting of Alzheimer’s pathology versus normal conditions will be compared. Invasiveness, proliferation, DNA damage (pH2AX), and apoptosis (cleaved PARP, cleaved Caspase 3) will be compared. Tumor cell-specific changes unique to tumors in the AD microenvironment will be assessed transcriptionally with RiboTag profiling as well as proteomically with TurboID purifications. Our preliminary data suggests that the presence of concomitant AD pathology affects GBM growth and progression as well as the immune response to the tumor, and warrants further investigation.

Citation Format: David Nascari, Ryan Eghlimi, Angad Beniwal, Drake Alton, John Fryer, Nhan L. Tran. Altered tumor microenvironment in animal model of concomitant GBM and Alzheimer's pathology [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 5562.

Citing articles via

Email alerts.

• Online First
• Collections
• Online ISSN 1538-7445
• Print ISSN 0008-5472

AACR Journals

• Blood Cancer Discovery
• Cancer Discovery
• Cancer Epidemiology, Biomarkers & Prevention
• Cancer Immunology Research
• Cancer Prevention Research
• Cancer Research
• Cancer Research Communications
• Clinical Cancer Research
• Molecular Cancer Research
• Molecular Cancer Therapeutics
• Info for Advertisers
• Information for Institutions/Librarians

• Privacy Policy
• Copyright © 2023 by the American Association for Cancer Research.

This Feature Is Available To Subscribers Only

Sign In or Create an Account

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

• View all journals
• My Account Login
• Explore content
• About the journal
• Publish with us
• Sign up for alerts
• Open access
• Published: 19 March 2024

TacticAI: an AI assistant for football tactics

• Zhe Wang   ORCID: orcid.org/0000-0002-0748-5376 1   na1 ,
• Petar Veličković   ORCID: orcid.org/0000-0002-2820-4692 1   na1 ,
• Daniel Hennes   ORCID: orcid.org/0000-0002-3646-5286 1   na1 ,
• Nenad Tomašev   ORCID: orcid.org/0000-0003-1624-0220 1 ,
• Laurel Prince 1 ,
• Michael Kaisers 1 ,
• Yoram Bachrach 1 ,
• Romuald Elie 1 ,
• Li Kevin Wenliang 1 ,
• Federico Piccinini 1 ,
• William Spearman 2 ,
• Ian Graham 3 ,
• Jerome Connor 1 ,
• Yi Yang 1 ,
• Adrià Recasens 1 ,
• Mina Khan 1 ,
• Nathalie Beauguerlange 1 ,
• Pablo Sprechmann 1 ,
• Pol Moreno 1 ,
• Nicolas Heess   ORCID: orcid.org/0000-0001-7876-9256 1 ,
• Michael Bowling   ORCID: orcid.org/0000-0003-2960-8418 4 ,
• Demis Hassabis 1 &
• Karl Tuyls   ORCID: orcid.org/0000-0001-7929-1944 5  

Nature Communications volume  15 , Article number:  1906 ( 2024 ) Cite this article

15k Accesses

237 Altmetric

Metrics details

• Computational science
• Information technology

Identifying key patterns of tactics implemented by rival teams, and developing effective responses, lies at the heart of modern football. However, doing so algorithmically remains an open research challenge. To address this unmet need, we propose TacticAI, an AI football tactics assistant developed and evaluated in close collaboration with domain experts from Liverpool FC. We focus on analysing corner kicks, as they offer coaches the most direct opportunities for interventions and improvements. TacticAI incorporates both a predictive and a generative component, allowing the coaches to effectively sample and explore alternative player setups for each corner kick routine and to select those with the highest predicted likelihood of success. We validate TacticAI on a number of relevant benchmark tasks: predicting receivers and shot attempts and recommending player position adjustments. The utility of TacticAI is validated by a qualitative study conducted with football domain experts at Liverpool FC. We show that TacticAI’s model suggestions are not only indistinguishable from real tactics, but also favoured over existing tactics 90% of the time, and that TacticAI offers an effective corner kick retrieval system. TacticAI achieves these results despite the limited availability of gold-standard data, achieving data efficiency through geometric deep learning.

Similar content being viewed by others

Automated recognition of the cricket batting backlift technique in video footage using deep learning architectures

Tevin Moodley, Dustin van der Haar & Habib Noorbhai

Synchronization of passes in event and spatiotemporal soccer data

Henrik Biermann, Rumena Komitova, … Daniel Memmert

A Random Forest approach to identify metrics that best predict match outcome and player ranking in the esport Rocket League

Tim D. Smithies, Mark J. Campbell, … Adam J. Toth

Introduction

Association football, or simply football or soccer, is a widely popular and highly professionalised sport, in which two teams compete to score goals against each other. As each football team comprises up to 11 active players at all times and takes place on a very large pitch (also known as a soccer field), scoring goals tends to require a significant degree of strategic team-play. Under the rules codified in the Laws of the Game 1 , this competition has nurtured an evolution of nuanced strategies and tactics, culminating in modern professional football leagues. In today’s play, data-driven insights are a key driver in determining the optimal player setups for each game and developing counter-tactics to maximise the chances of success 2 .

When competing at the highest level the margins are incredibly tight, and it is increasingly important to be able to capitalise on any opportunity for creating an advantage on the pitch. To that end, top-tier clubs employ diverse teams of coaches, analysts and experts, tasked with studying and devising (counter-)tactics before each game. Several recent methods attempt to improve tactical coaching and player decision-making through artificial intelligence (AI) tools, using a wide variety of data types from videos to tracking sensors and applying diverse algorithms ranging from simple logistic regression to elaborate neural network architectures. Such methods have been employed to help predict shot events from videos 3 , forecast off-screen movement from spatio-temporal data 4 , determine whether a match is in-play or interrupted 5 , or identify player actions 6 .

The execution of agreed-upon plans by players on the pitch is highly dynamic and imperfect, depending on numerous factors including player fitness and fatigue, variations in player movement and positioning, weather, the state of the pitch, and the reaction of the opposing team. In contrast, set pieces provide an opportunity to exert more control on the outcome, as the brief interruption in play allows the players to reposition according to one of the practiced and pre-agreed patterns, and make a deliberate attempt towards the goal. Examples of such set pieces include free kicks, corner kicks, goal kicks, throw-ins, and penalties 2 .

Among set pieces, corner kicks are of particular importance, as an improvement in corner kick execution may substantially modify game outcomes, and they lend themselves to principled, tactical and detailed analysis. This is because corner kicks tend to occur frequently in football matches (with ~10 corners on average taking place in each match 7 ), they are taken from a fixed, rigid position, and they offer an immediate opportunity for scoring a goal—no other set piece simultaneously satisfies all of the above. In practice, corner kick routines are determined well ahead of each match, taking into account the strengths and weaknesses of the opposing team and their typical tactical deployment. It is for this reason that we focus on corner kick analysis in particular, and propose TacticAI, an AI football assistant for supporting the human expert with set piece analysis, and the development and improvement of corner kick routines.

TacticAI is rooted in learning efficient representations of corner kick tactics from raw, spatio-temporal player tracking data. It makes efficient use of this data by representing each corner kick situation as a graph—a natural representation for modelling relationships between players (Fig.  1 A, Table  2 ), and these player relationships may be of higher importance than the absolute distances between them on the pitch 8 . Such a graph input is a natural candidate for graph machine learning models 9 , which we employ within TacticAI to obtain high-dimensional latent player representations. In the Supplementary Discussion section, we carefully contrast TacticAI against prior art in the area.

A How corner kick situations are converted to a graph representation. Each player is treated as a node in a graph, with node, edge and graph features extracted as detailed in the main text. Then, a graph neural network operates over this graph by performing message passing; each node’s representation is updated using the messages sent to it from its neighbouring nodes. B How TacticAI processes a given corner kick. To ensure that TacticAI’s answers are robust in the face of horizontal or vertical reflections, all possible combinations of reflections are applied to the input corner, and these four views are then fed to the core TacticAI model, where they are able to interact with each other to compute the final player representations—each internal blue arrow corresponds to a single message passing layer from ( A ). Once player representations are computed, they can be used to predict the corner’s receiver, whether a shot has been taken, as well as assistive adjustments to player positions and velocities, which increase or decrease the probability of a shot being taken.

Uniquely, TacticAI takes advantage of geometric deep learning 10 to explicitly produce player representations that respect several symmetries of the football pitch (Fig.  1 B). As an illustrative example, we can usually safely assume that under a horizontal or vertical reflection of the pitch state, the game situation is equivalent. Geometric deep learning ensures that TacticAI’s player representations will be identically computed under such reflections, such that this symmetry does not have to be learnt from data. This proves to be a valuable addition, as high-quality tracking data is often limited—with only a few hundred matches played each year in every league. We provide an in-depth overview of how we employ geometric deep learning in TacticAI in the “Methods” section.

From these representations, TacticAI is then able to answer various predictive questions about the outcomes of a corner—for example, which player is most likely to make first contact with the ball, or whether a shot will take place. TacticAI can also be used as a retrieval system—for mining similar corner kick situations based on the similarity of player representations—and a generative recommendation system, suggesting adjustments to player positions and velocities to maximise or minimise the estimated shot probability. Through several experiments within a case study with domain expert coaches and analysts from Liverpool FC, the results of which we present in the next section, we obtain clear statistical evidence that TacticAI readily provides useful, realistic and accurate tactical suggestions.

To demonstrate the diverse qualities of our approach, we design TacticAI with three distinct predictive and generative components: receiver prediction, shot prediction, and tactic recommendation through guided generation, which also correspond to the benchmark tasks for quantitatively evaluating TacticAI. In addition to providing accurate quantitative insights for corner kick analysis with its predictive components, the interplay between TacticAI’s predictive and generative components allows coaches to sample alternative player setups for each routine of interest, and directly evaluate the possible outcomes of such alternatives.

We will first describe our quantitative analysis, which demonstrates that TacticAI’s predictive components are accurate at predicting corner kick receivers and shot situations on held-out test corners and that the proposed player adjustments do not strongly deviate from ground-truth situations. However, such an analysis only gives an indirect insight into how useful TacticAI would be once deployed. We tackle this question of utility head-on and conduct a comprehensive case study in collaboration with our partners at Liverpool FC—where we directly ask human expert raters to judge the utility of TacticAI’s predictions and player adjustments. The following sections expand on the specific results and analysis we have performed.

In what follows, we will describe TacticAI’s components at a minimal level necessary to understand our evaluation. We defer detailed descriptions of TacticAI’s components to the “Methods” section. Note that, all our error bars reported in this research are standard deviations.

Benchmarking TacticAI

We evaluate the three components of TacticAI on a relevant benchmark dataset of corner kicks. Our dataset consists of 7176 corner kicks from the 2020 to 2021 Premier League seasons, which we randomly shuffle and split into a training (80%) and a test set (20%). As previously mentioned, TacticAI operates on graphs. Accordingly, we represent each corner kick situation as a graph, where each node corresponds to a player. The features associated with each node encode the movements (velocities and positions) and simple profiles (heights and weights) of on-pitch players at the timestamp when the corresponding corner kick was being taken by the attacking kicker (see the “Methods” section), and no information of ball movement was encoded. The graphs are fully connected; that is, for every pair of players, we will include the edge connecting them in the graph. Each of these edges encodes a binary feature, indicating whether the two players are on opposing teams or not. For each task, we generated the relevant dataset of node/edge/graph features and corresponding labels (Tables  1 and 2 , see the “Methods” section). The components were then trained separately with their corresponding corner kick graphs. In particular, we only employ a minimal set of features to construct the corner kick graphs, without encoding the movements of the ball nor explicitly encoding the distances between players into the graphs. We used a consistent training-test split for all benchmark tasks, as this made it possible to benchmark not only the individual components but also their interactions.

Accurate receiver and shot prediction through geometric deep learning

One of TacticAI’s key predictive models forecasts the receiver out of the 22 on-pitch players. The receiver is defined as the first player touching the ball after the corner is taken. In our evaluation, all methods used the same set of features (see the “Receiver prediction” entry in Table  1 and the “Methods” section). We leveraged the receiver prediction task to benchmark several different TacticAI base models. Our best-performing model—achieving 0.782 ± 0.039 in top-3 test accuracy after 50,000 training steps—was a deep graph attention network 11 , 12 , leveraging geometric deep learning 10 through the use of D 2 group convolutions 13 . We supplement this result with a detailed ablation study, verifying that both our choice of base architecture and group convolution yielded significant improvements in the receiver prediction task (Supplementary Table  2 , see the subsection “Ablation study” in the “Methods” section). Considering that corner kick receiver prediction is a highly challenging task with many factors that are unseen by our model—including fatigue and fitness levels, and actual ball trajectory—we consider TacticAI’s top-3 accuracy to reflect a high level of predictive power, and keep the base TacticAI architecture fixed for subsequent studies. In addition to this quantitative evaluation with the evaluation dataset, we also evaluate the performance of TacticAI’s receiver prediction component in a case study with human raters. Please see the “Case study with expert raters” section for more details.

For shot prediction, we observe that reusing the base TacticAI architecture to directly predict shot events—i.e., directly modelling the probability \({\mathbb{P}}(\,{{\mbox{shot}}}| {{\mbox{corner}}}\,)\) —proved challenging, only yielding a test F 1 score of 0.52 ± 0.03, for a GATv2 base model. Note that here we use the F 1 score—the harmonic mean of precision and recall—as it is commonly used in binary classification problems over imbalanced datasets, such as shot prediction. However, given that we already have a potent receiver predictor, we decided to use its output to give us additional insight into whether or not a shot had been taken. Hence, we opted to decompose the probability of taking a shot as

where \({\mathbb{P}}(\,{{\mbox{receiver}}}| {{\mbox{corner}}}\,)\) are the probabilities computed by TacticAI’s receiver prediction system, and \({\mathbb{P}}(\,{{\mbox{shot}}}| {{\mbox{receiver}}},{{\mbox{corner}}}\,)\) models the conditional shot probability after a specific player makes first contact with the ball. This was implemented through providing an additional global feature to indicate the receiver in the corresponding corner kick (Table  1 ) while the architecture otherwise remained the same as that of receiver prediction (Supplementary Fig.  2 , see the “Methods” section). At training time, we feed the ground-truth receiver as input to the model—at inference time, we attempt every possible receiver, weighing their contributions using the probabilities given by TacticAI’s receiver predictor, as per Eq. ( 1 ). This two-phased approach yielded a final test F 1 score of 0.68 ± 0.04 for shot prediction, which encodes significantly more signal than the unconditional shot predictor, especially considering the many unobservables associated with predicting shot events. Just as for receiver prediction, this performance can be further improved using geometric deep learning; a conditional GATv2 shot predictor with D 2 group convolutions achieves an F 1 score of 0.71 ± 0.01.

Moreover, we also observe that, even just through predicting the receivers, without explicitly classifying any other salient features of corners, TacticAI learned generalisable representations of the data. Specifically, team setups with similar tactical patterns tend to cluster together in TacticAI’s latent space (Fig.  2 ). However, no clear clusters are observed in the raw input space (Supplementary Fig.  1 ). This indicates that TacticAI can be leveraged as a useful corner kick retrieval system, and we will present our evaluation of this hypothesis in the “Case study with expert raters” section.

We visualise the latent representations of attacking and defending teams in 1024 corner kicks using t -SNE. A latent team embedding in one corner kick sample is the mean of the latent player representations on the same attacking ( A – C ) or defending ( D ) team. Given the reference corner kick sample ( A ), we retrieve another corner kick sample ( B ) with respect to the closest distance of their representations in the latent space. We observe that ( A ) and ( B ) are both out-swing corner kicks and share similar patterns of their attacking tactics, which are highlighted with rectangles having the same colours, although they bear differences with respect to the absolute positions and velocities of the players. All the while, the latent representation of an in-swing attack ( C ) is distant from both ( A ) and ( B ) in the latent space. The red arrows are only used to demonstrate the difference between in- and out-swing corner kicks, not the actual ball trajectories.

Lastly, it is worth emphasising that the utility of the shot predictor likely does not come from forecasting whether a shot event will occur—a challenging problem with many imponderables—but from analysing the difference in predicted shot probability across multiple corners. Indeed, in the following section, we will show how TacticAI’s generative tactic refinements can directly influence the predicted shot probabilities, which will then corresponds to highly favourable evaluation by our expert raters in the “Case study with expert raters” section.

Controlled tactic refinement using class-conditional generative models

Equipped with components that are able to potently relate corner kicks with their various outcomes (e.g. receivers and shot events), we can explore the use of TacticAI to suggest adjustments of tactics, in order to amplify or reduce the likelihood of certain outcomes.

Specifically, we aim to produce adjustments to the movements of players on one of the two teams, including their positions and velocities, which would maximise or minimise the probability of a shot event, conditioned on the initial corner setup, consisting of the movements of players on both teams and their heights and weights. In particular, although in real-world scenarios both teams may react simultaneously to the movements of each other, in our study, we focus on moderate adjustments to player movements, which help to detect players that are not responding to a tactic properly. Due to this reason, we simplify the process of tactic refinement through generating the adjustments for only one team while keeping the other fixed. The way we train a model for this task is through an auto-encoding objective: we feed the ground-truth shot outcome (a binary indicator) as an additional graph-level feature to TacticAI’s model (Table  1 ), and then have it learn to reconstruct a probability distribution of the input player coordinates (Fig.  1 B, also see the “Methods” section). As a consequence, our tactic adjustment system does not depend on the previously discussed shot predictor—although we can use the shot predictor to evaluate whether the adjustments make a measurable difference in shot probability.

This autoencoder-based generative model is an individual component that separates from TacticAI’s predictive systems. All three systems share the encoder architecture (without sharing parameters), but use different decoders (see the “Methods” section). At inference time, we can instead feed in a desired shot outcome for the given corner setup, and then sample new positions and velocities for players on one team using this probability distribution. This setup, in principle, allows for flexible downstream use, as human coaches can optimise corner kick setups through generating adjustments conditioned on the specific outcomes of their interest—e.g., increasing shot probability for the attacking team, decreasing it for the defending team (Fig.  3 ) or amplifying the chance that a particular striker receives the ball.

TacticAI makes it possible for human coaches to redesign corner kick tactics in ways that help maximise the probability of a positive outcome for either the attacking or the defending team by identifying key players, as well as by providing temporally coordinated tactic recommendations that take all players into consideration. As demonstrated in the present example ( A ), for a corner kick in which there was a shot attempt in reality ( B ), TacticAI can generate a tactically-adjusted setting in which the shot probability has been reduced, by adjusting the positioning of the defenders ( D ). The suggested defender positions result in reduced receiver probability for attacking players 2–5 (see bottom row), while the receiver probability of Attacker 1, who is distant from the goalpost, has been increased ( C ). The model is capable of generating multiple such scenarios. Coaches can inspect the different options visually and additionally consult TacticAI’s quantitative analysis of the presented tactics.

We first evaluate the generated adjustments quantitatively, by verifying that they are indistinguishable from the original corner kick distribution using a classifier. To do this, we synthesised a dataset consisting of 200 corner kick samples and their corresponding conditionally generated adjustments. Specifically, for corners without a shot event, we generated adjustments for the attacking team by setting the shot event feature to 1, and vice-versa for the defending team when a shot event did happen. We found that the real and generated samples were not distinguishable by an MLP classifier, with an F 1 score of 0.53 ± 0.05, indicating random chance level accuracy. This result indicates that the adjustments produced by TacticAI are likely similar enough to real corner kicks that the MLP is unable to tell them apart. Note that, in spite of this similarity, TacticAI recommends player-level adjustments that are not negligible—in the following section we will illustrate several salient examples of this. To more realistically validate the practical indistinguishability of TacticAI’s adjustments from realistic corners, we also evaluated the realism of the adjustments in a case study with human experts, which we will present in the following section.

In addition, we leveraged our TacticAI shot predictor to estimate whether the proposed adjustments were effective. We did this by analysing 100 corner kick samples in which threatening shots occurred, and then, for each sample, generated one defensive refinement through setting the shot event feature to 0. We observed that the average shot probability significantly decreased, from 0.75 ± 0.14 for ground-truth corners to 0.69 ± 0.16 for adjustments ( z  = 2.62,  p  < 0.001). This observation was consistent when testing for attacking team refinements (shot probability increased from 0.18 ± 0.16 to 0.31 ± 0.26 ( z  = −4.46,  p  < 0.001)). Moving beyond this result, we also asked human raters to assess the utility of TacticAI’s proposed adjustments within our case study, which we detail next.

Case study with expert raters

Although quantitative evaluation with well-defined benchmark datasets was critical for the technical development of TacticAI, the ultimate test of TacticAI as a football tactic assistant is its practical downstream utility being recognised by professionals in the industry. To this end, we evaluated TacticAI through a case study with our partners at Liverpool FC (LFC). Specifically, we invited a group of five football experts: three data scientists, one video analyst, and one coaching assistant. Each of them completed four tasks in the case study, which evaluated the utility of TacticAI’s components from several perspectives; these include (1) the realism of TacticAI’s generated adjustments, (2) the plausibility of TacticAI’s receiver predictions, (3) effectiveness of TacticAI’s embeddings for retrieving similar corners, and (4) usefulness of TacticAI’s recommended adjustments. We provide an overview of our study’s results here and refer the interested reader to Supplementary Figs.  3 – 5 and the  Supplementary Methods for additional details.

We first simultaneously evaluated the realism of the adjusted corner kicks generated by TacticAI, and the plausibility of its receiver predictions. Going through a collection of 50 corner kick samples, we first asked the raters to classify whether a given sample was real or generated by TacticAI, and then they were asked to identify the most likely receivers in the corner kick sample (Supplementary Fig.  3 ).

On the task of classifying real and generated samples, first, we found that the raters’ average F 1 score of classifying the real vs. generated samples was only 0.60 ± 0.04, with individual F 1 scores ( \({F}_{1}^{A}=0.54,{F}_{1}^{B}=0.64,{F}_{1}^{C}=0.65,{F}_{1}^{D}=0.62,{F}_{1}^{E}=0.56\) ), indicating that the raters were, in many situations, unable to distinguish TacticAI’s adjustments from real corners.

The previous evaluation focused on analysing realism detection performance across raters. We also conduct a study that analyses realism detection across samples. Specifically, we assigned ratings for each sample—assigning +1 to a sample if it was identified as real by a human rater, and 0 otherwise—and computed the average rating for each sample across the five raters. Importantly, by studying the distribution of ratings, we found that there was no significant difference between the average ratings assigned to real and generated corners ( z  = −0.34,  p  > 0.05) (Fig.  4 A). Hence, the real and generated samples were assigned statistically indistinguishable average ratings by human raters.

In task 1, we tested the statistical difference between the real corner kick samples and the synthetic ones generated by TacticAI from two aspects: ( A.1 ) the distributions of their assigned ratings, and ( A.2 ) the corresponding histograms of the rating values. Analogously, in task 2 (receiver prediction), ( B.1 ) we track the distributions of the top-3 accuracy of receiver prediction using those samples, and ( B.2 ) the corresponding histogram of the mean rating per sample. No statistical difference in the mean was observed in either cases (( A.1 ) ( z  = −0.34,  p  > 0.05), and ( B.1 ) ( z  = 0.97,  p  > 0.05)). Additionally, we observed a statistically significant difference between the ratings of different raters on receiver prediction, with three clear clusters emerging ( C ). Specifically, Raters A and E had similar ratings ( z  = 0.66,  p  > 0.05), and Raters B and D also rated in similar ways ( z  = −1.84,  p  > 0.05), while Rater C responded differently from all other raters. This suggests a good level of variety of the human raters with respect to their perceptions of corner kicks. In task 3—identifying similar corners retrieved in terms of salient strategic setups—there were no significant differences among the distributions of the ratings by different raters ( D ), suggesting a high level of agreement on the usefulness of TacticAI’s capability of retrieving similar corners ( F 1,4  = 1.01,  p  > 0.1). Finally, in task 4, we compared the ratings of TacticAI’s strategic refinements across the human raters ( E ) and found that the raters also agreed on the general effectiveness of the refinements recommended by TacticAI ( F 1,4  = 0.45,  p  > 0.05). Note that the violin plots used in B.1 and C – E model a continuous probability distribution and hence assign nonzero probabilities to values outside of the allowed ranges. We only label y -axis ticks for the possible set of ratings.

For the task of identifying receivers, we rated TacticAI’s predictions with respect to a rater as +1 if at least one of the receivers identified by the rater appeared in TacticAI’s top-3 predictions, and 0 otherwise. The average top-3 accuracy among the human raters was 0.79 ± 0.18; specifically, 0.81 ± 0.17 for the real samples, and 0.77 ± 0.21 for the generated ones. These scores closely line up with the accuracy of TacticAI in predicting receivers for held-out test corners, validating our quantitative study. Further, after averaging the ratings for receiver prediction sample-wise, we found no statistically significant difference between the average ratings of predicting receivers over the real and generated samples ( z  = 0.97,  p  > 0.05) (Fig.  4 B). This indicates that TacticAI was equally performant in predicting the receivers of real corners and TacticAI-generated adjustments, and hence may be leveraged for this purpose even in simulated scenarios.

There is a notably high variance in the average receiver prediction rating of TacticAI. We hypothesise that this is due to the fact that different raters may choose to focus on different salient features when evaluating the likely receivers (or even the amount of likely receivers). We set out to validate this hypothesis by testing the pair-wise similarity of the predictions by the human raters through running a one-away analysis of variance (ANOVA), followed by a Tukey test. We found that the distributions of the five raters’ predictions were significantly different ( F 1,4  = 14.46,  p  < 0.001) forming three clusters (Fig.  4 C). This result indicates that different human raters—as suggested by their various titles at LFC—may often use very different leads when suggesting plausible receivers. The fact that TacticAI manages to retain a high top-3 accuracy in such a setting suggests that it was able to capture the salient patterns of corner kick strategies, which broadly align with human raters’ preferences. We will further test this hypothesis in the third task—identifying similar corners.

For the third task, we asked the human raters to judge 50 pairs of corners for their similarity. Each pair consisted of a reference corner and a retrieved corner, where the retrieved corner was chosen either as the nearest-neighbour of the reference in terms of their TacticAI latent space representations, or—as a feature-level heuristic—the cosine similarities of their raw features (Supplementary Fig.  4 ) in our corner kick dataset. We score the raters’ judgement of a pair as +1 if they considered the corners presented in the case to be usefully similar, otherwise, the pair is scored with 0. We first computed, for each rater, the recall with which they have judged a baseline- or TacticAI-retrieved pair as usefully similar—see description of Task 3 in the  Supplementary Methods . For TacticAI retrievals, the average recall across all raters was 0.59 ± 0.09, and for the baseline system, the recall was 0.36 ± 0.10. Secondly, we assess the statistical difference between the results of the two methods by averaging the ratings for each reference–retrieval pair, finding that the average rating of TacticAI retrievals is significantly higher than the average rating of baseline method retrievals ( z  = 2.34,  p  < 0.05). These two results suggest that TacticAI significantly outperforms the feature-space baseline as a method for mining similar corners. This indicates that TacticAI is able to extract salient features from corners that are not trivial to extract from the input data alone, reinforcing it as a potent tool for discovering opposing team tactics from available data. Finally, we observed that this task exhibited a high level of inter-rater agreement for TacticAI-retrieved pairs ( F 1,4  = 1.01,  p  > 0.1) (Fig.  4 D), suggesting that human raters were largely in agreement with respect to their assessment of TacticAI’s performance.

Finally, we evaluated TacticAI’s player adjustment recommendations for their practical utility. Specifically, each rater was given 50 tactical refinements together with the corresponding real corner kick setups—see Supplementary Fig.  5 , and the “Case study design” section in the  Supplementary Methods . The raters were then asked to rate each refinement as saliently improving the tactics (+1), saliently making them worse (−1), or offering no salient differences (0). We calculated the average rating assigned by each of the raters (giving us a value in the range [− 1, 1] for each rater). The average of these values across all five raters was 0.7 ± 0.1. Further, for 45 of the 50 situations (90%), the human raters found TacticAI’s suggestion to be favourable on average (by majority voting). Both of these results indicate that TacticAI’s recommendations are salient and useful to a downstream football club practitioner, and we set out to validate this with statistical tests.

We performed statistical significance testing of the observed positive ratings. First, for each of the 50 situations, we averaged its ratings across all five raters and then ran a t -test to assess whether the mean rating was significantly larger than zero. Indeed, the statistical test indicated that the tactical adjustments recommended by TacticAI were constructive overall ( \({t}_{49}^{{{{{{{{\rm{avg}}}}}}}}}=9.20,\, p \, < \, 0.001\) ). Secondly, we verified that each of the five raters individually found TacticAI’s recommendations to be constructive, running a t -test on each of their ratings individually. For all of the five raters, their average ratings were found to be above zero with statistical significance ( \({t}_{49}^{A}=5.84,\, {p}^{A} \, < \, 0.001;{t}_{49}^{B}=7.88,\; {p}^{B} \, < \, 0.001;{t}_{49}^{C}=7.00,\; {p}^{C} \, < \, 0.001;{t}_{49}^{D}=6.04,\; {p}^{D} \, < \, 0.001;{t}_{49}^{E}=7.30,\, {p}^{E} \, < \, 0.001\) ). In addition, their ratings also shared a high level of inter-agreement ( F 1,4  = 0.45,  p  > 0.05) (Fig.  4 E), suggesting a level of practical usefulness that is generally recognised by human experts, even though they represent different backgrounds.

Taking all of these results together, we find TacticAI to possess strong components for prediction, retrieval, and tactical adjustments on corner kicks. To illustrate the kinds of salient recommendations by TacticAI, in Fig.  5 we present four examples with a high degree of inter-rater agreement.

These examples are selected from our case study with human experts, to illustrate the breadth of tactical adjustments that TacticAI suggests to teams defending a corner. The density of the yellow circles coincides with the number of times that the corresponding change is recognised as constructive by human experts. Instead of optimising the movement of one specific player, TacticAI can recommend improvements for multiple players in one generation step through suggesting better positions to block the opposing players, or better orientations to track them more efficiently. Some specific comments from expert raters follow. In A , according to raters, TacticAI suggests more favourable positions for several defenders, and improved tracking runs for several others—further, the goalkeeper is positioned more deeply, which is also beneficial. In B , TacticAI suggests that the defenders furthest away from the corner make improved covering runs, which was unanimously deemed useful, with several other defenders also positioned more favourably. In C , TacticAI recommends improved covering runs for a central group of defenders in the penalty box, which was unanimously considered salient by our raters. And in D , TacticAI suggests substantially better tracking runs for two central defenders, along with a better positioning for two other defenders in the goal area.

We have demonstrated an AI assistant for football tactics and provided statistical evidence of its efficacy through a comprehensive case study with expert human raters from Liverpool FC. First, TacticAI is able to accurately predict the first receiver after a corner kick is taken as well as the probability of a shot as the direct result of the corner. Second, TacticAI has been shown to produce plausible tactical variations that improve outcomes in a salient way, while being indistinguishable from real scenarios by domain experts. And finally, the system’s latent player representations are a powerful means to retrieve similar set-piece tactics, allowing coaches to analyse relevant tactics and counter-tactics that have been successful in the past.

The broader scope of strategy modelling in football has previously been addressed from various individual angles, such as pass prediction 14 , 15 , 16 , shot prediction 3 or corner kick tactical classification 7 . However, to the best of our knowledge, our work stands out by combining and evaluating predictive and generative modelling of corner kicks for tactic development. It also stands out in its method of applying geometric deep learning, allowing for efficiently incorporating various symmetries of the football pitch for improved data efficiency. Our method incorporates minimal domain knowledge and does not rely on intricate feature engineering—though its factorised design naturally allows for more intricate feature engineering approaches when such features are available.

Our methodology requires the position and velocity estimates of all players at the time of execution of the corner and subsequent events. Here, we derive these from high-quality tracking and event data, with data availability from tracking providers limited to top leagues. Player tracking based on broadcast video would increase the reach and training data substantially, but would also likely result in noisier model inputs. While the attention mechanism of GATs would allow us to perform introspection of the most salient factors contributing to the model outcome, our method does not explicitly model exogenous (aleatoric) uncertainty, which would be valuable context for the football analyst.

While the empirical study of our method’s efficacy has been focused on corner kicks in association football, it readily generalises to other set pieces (such as throw-ins, which similarly benefit from similarity retrieval, pass and/or shot prediction) and other team sports with suspended play situations. The learned representations and overall framing of TacticAI also lay the ground for future research to integrate a natural language interface that enables domain-grounded conversations with the assistant, with the aim to retrieve particular situations of interest, make predictions for a given tactical variant, compare and contrast, and guide through an interactive process to derive tactical suggestions. It is thus our belief that TacticAI lays the groundwork for the next-generation AI assistant for football.

We devised TacticAI as a geometric deep learning pipeline, further expanded in this section. We process labelled spatio-temporal football data into graph representations, and train and evaluate on benchmarking tasks cast as classification or regression. These steps are presented in sequence, followed by details on the employed computational architecture.

Raw corner kick data

The raw dataset consisted of 9693 corner kicks collected from the 2020–21, 2021–22, and 2022–23 (up to January 2023) Premier League seasons. The dataset was provided by Liverpool FC and comprises four separate data sources, described below.

Our primary data source is spatio-temporal trajectory frames (tracking data), which tracked all on-pitch players and the ball, for each match, at 25 frames per second. In addition to player positions, their velocities are derived from position data through filtering. For each corner kick, we only used the frame in which the kick is being taken as input information.

Secondly, we also leverage event stream data, which annotated the events or actions (e.g., passes, shots and goals) that have occurred in the corresponding tracking frames.

Thirdly, the line-up data for the corresponding games, which recorded the players’ profiles, including their heights, weights and roles, is also used.

Lastly, we have access to miscellaneous game data, which contains the game days, stadium information, and pitch length and width in meters.

Graph representation and construction

We assumed that we were provided with an input graph \({{{{{{{\mathcal{G}}}}}}}}=({{{{{{{\mathcal{V}}}}}}}},\,{{{{{{{\mathcal{E}}}}}}}})\) with a set of nodes \({{{{{{{\mathcal{V}}}}}}}}\) and edges \({{{{{{{\mathcal{E}}}}}}}}\subseteq {{{{{{{\mathcal{V}}}}}}}}\times {{{{{{{\mathcal{V}}}}}}}}\) . Within the context of football games, we took \({{{{{{{\mathcal{V}}}}}}}}\) to be the set of 22 players currently on the pitch for both teams, and we set \({{{{{{{\mathcal{E}}}}}}}}={{{{{{{\mathcal{V}}}}}}}}\times {{{{{{{\mathcal{V}}}}}}}}\) ; that is, we assumed all pairs of players have the potential to interact. Further analyses, leveraging more specific choices of \({{{{{{{\mathcal{E}}}}}}}}\) , would be an interesting avenue for future work.

Additionally, we assume that the graph is appropriately featurised. Specifically, we provide a node feature matrix, \({{{{{{{\bf{X}}}}}}}}\in {{\mathbb{R}}}^{| {{{{{{{\mathcal{V}}}}}}}}| \times k}\) , an edge feature tensor, \({{{{{{{\bf{E}}}}}}}}\in {{\mathbb{R}}}^{| {{{{{{{\mathcal{V}}}}}}}}| \times | {{{{{{{\mathcal{V}}}}}}}}| \times l}\) , and a graph feature vector, \({{{{{{{\bf{g}}}}}}}}\in {{\mathbb{R}}}^{m}\) . The appropriate entries of these objects provide us with the input features for each node, edge, and graph. For example, \({{{{{{{{\bf{x}}}}}}}}}_{u}\in {{\mathbb{R}}}^{k}\) would provide attributes of an individual player \(u\in {{{{{{{\mathcal{V}}}}}}}}\) , such as position, height and weight, and \({{{{{{{{\bf{e}}}}}}}}}_{uv}\in {{\mathbb{R}}}^{l}\) would provide the attributes of a particular pair of players \((u,\, v)\in {{{{{{{\mathcal{E}}}}}}}}\) , such as their distance, and whether they belong to the same team. The graph feature vector, g , can be used to store global attributes of interest to the corner kick, such as the game time, current score, or ball position. For a simplified visualisation of how a graph neural network would process such an input, refer to Fig.  1 A.

To construct the input graphs, we first aligned the four data sources with respect to their game IDs and timestamps and filtered out 2517 invalid corner kicks, for which the alignment failed due to missing data, e.g., missing tracking frames or event labels. This filtering yielded 7176 valid corner kicks for training and evaluation. We summarised the exact information that was used to construct the input graphs in Table  2 . In particular, other than player heights (measured in centimeters (cm)) and weights (measured in kilograms (kg)), the players were anonymous in the model. For the cases in which the player profiles were missing, we set their heights and weights to 180 cm and 75 kg, respectively, as defaults. In total, we had 385 such occurrences out of a total of 213,246( = 22 × 9693) during data preprocessing. We downscaled the heights and weights by a factor of 100. Moreover, for each corner kick, we zero-centred the positions of on-pitch players and normalised them onto a 10 m × 10 m pitch, and their velocities were re-scaled accordingly. For the cases in which the pitch dimensions were missing, we used a standard pitch dimension of 110 m × 63 m as default.

We summarised the grouping of the features in Table  1 . The actual features used in different benchmark tasks may differ, and we will describe this in more detail in the next section. To focus on modelling the high-level tactics played by the attacking and defending teams, other than a binary indicator for ball possession—which is 1 for the corner kick taker and 0 for all other players—no information of ball movement, neither positions nor velocities, was used to construct the input graphs. Additionally, we do not have access to the player’s vertical movement, therefore only information on the two-dimensional movements of each player is provided in the data. We do however acknowledge that such information, when available, would be interesting to consider in a corner kick outcome predictor, considering the prevalence of aerial battles in corners.

Benchmark tasks construction

TacticAI consists of three predictive and generative models, which also correspond to three benchmark tasks implemented in this study. Specifically, (1) Receiver prediction, (2) Threatening shot prediction, and (3) Guided generation of team positions and velocities (Table  1 ). The graphs of all the benchmark tasks used the same feature space of nodes and edges, differing only in the global features.

For all three tasks, our models first transform the node features to a latent node feature matrix, \({{{{{{{\bf{H}}}}}}}}={f}_{{{{{{{{\mathcal{G}}}}}}}}}({{{{{{{\bf{X}}}}}}}},\, {{{{{{{\bf{E}}}}}}}},\, {{{{{{{\bf{g}}}}}}}})\) , from which we could answer queries: either about individual players—in which case we learned a relevant classifier or regressor over the h u vectors (the rows of H )—or about the occurrence of a global event (e.g. shot taken)—in which case we classified or regressed over the aggregated player vectors, ∑ u h u . In both cases, the classifiers were trained using stochastic gradient descent over an appropriately chosen loss function, such as categorical cross-entropy for classifiers, and mean squared error for regressors.

For different tasks, we extracted the corresponding ground-truth labels from either the event stream data or the tracking data. Specifically, (1) We modelled receiver prediction as a node classification task and labelled the first player to touch the ball after the corner was taken as the target node. This player could be either an attacking or defensive player. (2) Shot prediction was modelled as graph classification. In particular, we considered a next-ball-touch action by the attacking team as a shot if it was a direct corner, a goal, an aerial, hit on the goalposts, a shot attempt saved by the goalkeeper, or missing target. This yielded 1736 corners labelled as a shot being taken, and 5440 corners labelled as a shot not being taken. (3) For guided generation of player position and velocities, no additional label was needed, as this model relied on a self-supervised reconstruction objective.

The entire dataset was split into training and evaluation sets with an 80:20 ratio through random sampling, and the same splits were used for all tasks.

Graph neural networks

The central model of TacticAI is the graph neural network (GNN) 9 , which computes latent representations on a graph by repeatedly combining them within each node’s neighbourhood. Here we define a node’s neighbourhood, \({{{{{{{{\mathcal{N}}}}}}}}}_{u}\) , as the set of all first-order neighbours of node u , that is, \({{{{{{{{\mathcal{N}}}}}}}}}_{u}=\{v\,| \,(v,\, u)\in {{{{{{{\mathcal{E}}}}}}}}\}\) . A single GNN layer then transforms the node features by passing messages between neighbouring nodes 17 , following the notation of related work 10 , and the implementation of the CLRS-30 benchmark baselines 18 :

where \(\psi :{{\mathbb{R}}}^{k}\times {{\mathbb{R}}}^{k}\times {{\mathbb{R}}}^{l}\times {{\mathbb{R}}}^{m}\to {{\mathbb{R}}}^{{k}^{{\prime} }}\) and \(\phi :{{\mathbb{R}}}^{k}\times {{\mathbb{R}}}^{{k}^{{\prime} }}\to {{\mathbb{R}}}^{{k}^{{\prime} }}\) are two learnable functions (e.g. multilayer perceptrons), \({{{{{{{{\bf{h}}}}}}}}}_{u}^{(t)}\) are the features of node u after t GNN layers, and ⨁ is any permutation-invariant aggregator, such as sum, max, or average. By definition, we set \({{{{{{{{\bf{h}}}}}}}}}_{u}^{(0)}={{{{{{{{\bf{x}}}}}}}}}_{u}\) , and iterate Eq. ( 2 ) for T steps, where T is a hyperparameter. Then, we let \({{{{{{{\bf{H}}}}}}}}={f}_{{{{{{{{\mathcal{G}}}}}}}}}({{{{{{{\bf{X}}}}}}}},\, {{{{{{{\bf{E}}}}}}}},\, {{{{{{{\bf{g}}}}}}}})={{{{{{{{\bf{H}}}}}}}}}^{(T)}\) be the final node embeddings coming out of the GNN.

It is well known that Eq. ( 2 ) is remarkably general; it can be used to express popular models such as Transformers 19 as a special case, and it has been argued that all discrete deep learning models can be expressed in this form 20 , 21 . This makes GNNs a perfect framework for benchmarking various approaches to modelling player–player interactions in the context of football.

Different choices of ψ , ϕ and ⨁ yield different architectures. In our case, we utilise a message function that factorises into an attentional mechanism, \(a:{{\mathbb{R}}}^{k}\times {{\mathbb{R}}}^{k}\times {{\mathbb{R}}}^{l}\times {{\mathbb{R}}}^{m}\to {\mathbb{R}}\) :

yielding the graph attention network (GAT) architecture 12 . In our work, specifically, we use a two-layer multilayer perceptron for the attentional mechanism, as proposed by GATv2 11 :

where \({{{{{{{{\bf{W}}}}}}}}}_{1},\, {{{{{{{{\bf{W}}}}}}}}}_{2}\in {{\mathbb{R}}}^{k\times h}\) , \({{{{{{{{\bf{W}}}}}}}}}_{e}\in {{\mathbb{R}}}^{l\times h}\) , \({{{{{{{{\bf{W}}}}}}}}}_{g}\in {{\mathbb{R}}}^{m\times h}\) and \({{{{{{{\bf{a}}}}}}}}\in {{\mathbb{R}}}^{h}\) are the learnable parameters of the attentional mechanism, and LeakyReLU is the leaky rectified linear activation function. This mechanism computes coefficients of interaction (a single scalar value) for each pair of connected nodes ( u ,  v ), which are then normalised across all neighbours of u using the \({{{{{{{\rm{softmax}}}}}}}}\) function.

Through early-stage experimentation, we have ascertained that GATs are capable of matching the performance of more generic choices of ψ (such as the MPNN 17 ) while being more scalable. Hence, we focus our study on the GAT model in this work. More details can be found in the subsection “Ablation study” section.

Geometric deep learning

In spite of the power of Eq. ( 2 ), using it in its full generality is often prone to overfitting, given the large number of parameters contained in ψ and ϕ . This problem is exacerbated in the football analytics domain, where gold-standard data is generally very scarce—for example, in the English Premier League, only a few hundred games are played every season.

In order to tackle this issue, we can exploit the immense regularity of data arising from football games. Strategically equivalent game states are also called transpositions, and symmetries such as arriving at the same chess position through different move sequences have been exploited computationally since the 1960s 22 . Similarly, game rotations and reflections may yield equivalent strategic situations 23 . Using the blueprint of geometric deep learning (GDL) 10 , we can design specialised GNN architectures that exploit this regularity.

That is, geometric deep learning is a generic methodology for deriving mathematical constraints on neural networks, such that they will behave predictably when inputs are transformed in certain ways. In several important cases, these constraints can be directly resolved, directly informing neural network architecture design. For a comprehensive example of point clouds under 3D rotational symmetry, see Fuchs et al. 24 .

To elucidate several aspects of the GDL framework on a high level, let us assume that there exists a group of input data transformations (symmetries), \({\mathfrak{G}}\) under which the ground-truth label remains unchanged. Specifically, if we let y ( X ,  E ,  g ) be the label given to the graph featurised with X ,  E ,  g , then for every transformation \({\mathfrak{g}}\in {\mathfrak{G}}\) , the following property holds:

This condition is also referred to as \({\mathfrak{G}}\) -invariance. Here, by \({\mathfrak{g}}({{{{{{{\bf{X}}}}}}}})\) we denote the result of transforming X by \({\mathfrak{g}}\) —a concept also known as a group action. More generally, it is a function of the form \({\mathfrak{G}}\times {{{{{{{\mathcal{S}}}}}}}}\to {{{{{{{\mathcal{S}}}}}}}}\) for some state set \({{{{{{{\mathcal{S}}}}}}}}\) . Note that a single group element, \({\mathfrak{g}}\in {\mathfrak{G}}\) can easily produce different actions on different \({{{{{{{\mathcal{S}}}}}}}}\) —in this case, \({{{{{{{\mathcal{S}}}}}}}}\) could be \({{\mathbb{R}}}^{| {{{{{{{\mathcal{V}}}}}}}}| \times k}\) ( X ), \({{\mathbb{R}}}^{| {{{{{{{\mathcal{V}}}}}}}}| \times | {{{{{{{\mathcal{V}}}}}}}}| \times l}\) ( E ) and \({{\mathbb{R}}}^{m}\) ( g ).

It is worth noting that GNNs may also be derived using a GDL perspective if we set the symmetry group \({\mathfrak{G}}\) to \({S}_{| {{{{{{{\mathcal{V}}}}}}}}}|\) , the permutation group of \(| {{{{{{{\mathcal{V}}}}}}}}|\) objects. Owing to the design of Eq. ( 2 ), its outputs will not be dependent on the exact permutation of nodes in the input graph.

Frame averaging

A simple mechanism to enforce \({\mathfrak{G}}\) -invariance, given any predictor \({f}_{{{{{{{{\mathcal{G}}}}}}}}}({{{{{{{\bf{X}}}}}}}},\, {{{{{{{\bf{E}}}}}}}},\, {{{{{{{\bf{g}}}}}}}})\) , performs frame averaging across all \({\mathfrak{G}}\) -transformed inputs:

This ensures that all \({\mathfrak{G}}\) -transformed versions of a particular input (also known as that input’s orbit) will have exactly the same output, satisfying Eq. ( 5 ). A variant of this approach has also been applied in the AlphaGo architecture 25 to encode symmetries of a Go board.

In our specific implementation, we set \({\mathfrak{G}}={D}_{2}=\{{{{{{{{\rm{id}}}}}}}},\leftrightarrow,\updownarrow,\leftrightarrow \updownarrow \}\) , the dihedral group. Exploiting D 2 -invariance allows us to encode quadrant symmetries. Each element of the D 2 group encodes the presence of vertical or horizontal reflections of the input football pitch. Under these transformations, the pitch is assumed completely symmetric, and hence many predictions, such as which player receives the corner kick, or takes a shot from it, can be safely assumed unchanged. As an example of how to compute transformed features in Eq. ( 6 ), ↔( X ) horizontally reflects all positional features of players in X (e.g. the coordinates of the player), and negates the x -axis component of their velocity.

Group convolutions

While the frame averaging approach of Eq. ( 6 ) is a powerful way to restrict GNNs to respect input symmetries, it arguably misses an opportunity for the different \({\mathfrak{G}}\) -transformed views to interact while their computations are being performed. For small groups such as D 2 , a more fine-grained approach can be assumed, operating over a single GNN layer in Eq. ( 2 ), which we will write shortly as \({{{{{{{{\bf{H}}}}}}}}}^{(t)}={g}_{{{{{{{{\mathcal{G}}}}}}}}}({{{{{{{{\bf{H}}}}}}}}}^{(t-1)},\, {{{{{{{\bf{E}}}}}}}},\, {{{{{{{\bf{g}}}}}}}})\) . The condition that we need a symmetry-respecting GNN layer to satisfy is as follows, for all transformations \({\mathfrak{g}}\in {\mathfrak{G}}\) :

that is, it does not matter if we apply \({\mathfrak{g}}\) it to the input or the output of the function \({g}_{{{{{{{{\mathcal{G}}}}}}}}}\) —the final answer is the same. This condition is also referred to as \({\mathfrak{G}}\) -equivariance, and it has recently proved to be a potent paradigm for developing powerful GNNs over biochemical data 24 , 26 .

To satisfy D 2 -equivariance, we apply the group convolution approach 13 . Therein, views of the input are allowed to directly interact with their \({\mathfrak{G}}\) -transformed variants, in a manner very similar to grid convolutions (which is, indeed, a special case of group convolutions, setting \({\mathfrak{G}}\) to be the translation group). We use \({{{{{{{{\bf{H}}}}}}}}}_{{\mathfrak{g}}}^{(t)}\) to denote the \({\mathfrak{g}}\) -transformed view of the latent node features at layer t . Omitting E and g inputs for brevity, and using our previously designed layer \({g}_{{{{{{{{\mathcal{G}}}}}}}}}\) as a building block, we can perform a group convolution as follows:

Here, ∥ is the concatenation operation, joining the two node feature matrices column-wise; \({{\mathfrak{g}}}^{-1}\) is the inverse transformation to \({\mathfrak{g}}\) (which must exist as \({\mathfrak{G}}\) is a group); and \({{\mathfrak{g}}}^{-1}{\mathfrak{h}}\) is the composition of the two transformations.

Effectively, Eq. ( 8 ) implies our D 2 -equivariant GNN needs to maintain a node feature matrix \({{{{{{{{\bf{H}}}}}}}}}_{{\mathfrak{g}}}^{(t)}\) for every \({\mathfrak{G}}\) -transformation of the current input, and these views are recombined by invoking \({g}_{{{{{{{{\mathcal{G}}}}}}}}}\) on all pairs related together by applying a transformation \({\mathfrak{h}}\) . Note that all reflections are self-inverses, hence, in D 2 , \({\mathfrak{g}}={{\mathfrak{g}}}^{-1}\) .

It is worth noting that both the frame averaging in Eq. ( 6 ) and group convolution in Eq. ( 8 ) are similar in spirit to data augmentation. However, whereas standard data augmentation would only show one view at a time to the model, a frame averaging/group convolution architecture exhaustively generates all views and feeds them to the model all at once. Further, group convolutions allow these views to explicitly interact in a way that does not break symmetries. Here lies the key difference between the two approaches: frame averaging and group convolutions rigorously enforce the symmetries in \({\mathfrak{G}}\) , whereas data augmentation only provides implicit hints to the model about satisfying them. As a consequence of the exhaustive generation, Eqs. ( 6 ) and ( 8 ) are only feasible for small groups like D 2 . For larger groups, approaches like Steerable CNNs 27 may be employed.

Network architectures

While the three benchmark tasks we are performing have minor differences in the global features available to the model, the neural network models designed for them all have the same encoder–decoder architecture. The encoder has the same structure in all tasks, while the decoder model is tailored to produce appropriately shaped outputs for each benchmark task.

Given an input graph, TacticAI’s model first generates all relevant D 2 -transformed versions of it, by appropriately reflecting the player coordinates and velocities. We refer to the original input graph as the identity view, and the remaining three D 2 -transformed graphs as reflected views.

Once the views are prepared, we apply four group convolutional layers (Eq. ( 8 )) with a GATv2 base model (Eqs. ( 3 ) and ( 4 )) as the \({g}_{{{{{{{{\mathcal{G}}}}}}}}}\) function. Specifically, this means that, in Eqs. ( 3 ) and ( 4 ), every instance of \({{{{{{{{\bf{h}}}}}}}}}_{u}^{(t-1)}\) is replaced by the concatenation of \({({{{{{{{{\bf{h}}}}}}}}}_{{\mathfrak{h}}}^{(t-1)})}_{u}\parallel {({{{{{{{{\bf{h}}}}}}}}}_{{{\mathfrak{g}}}^{-1}{\mathfrak{h}}}^{(t-1)})}_{u}\) . Each GATv2 layer has eight attention heads and computes four latent features overall per player. Accordingly, once the four group convolutions are performed, we have a representation of \({{{{{{{\bf{H}}}}}}}}\in {{\mathbb{R}}}^{4\times 22\times 4}\) , where the first dimension corresponds to the four views ( \({{{{{{{{\bf{H}}}}}}}}}_{{{{{{{{\rm{id}}}}}}}}},\, {{{{{{{{\bf{H}}}}}}}}}_{\leftrightarrow },\, {{{{{{{{\bf{H}}}}}}}}}_{\updownarrow },\, {{{{{{{{\bf{H}}}}}}}}}_{\leftrightarrow \updownarrow }\in {{\mathbb{R}}}^{22\times 4}\) ), the second dimension corresponds to the players (eleven on each team), and the third corresponds to the 4-dimensional latent vector for each player node in this particular view. How this representation is used by the decoder depends on the specific downstream task, as we detail below.

For receiver prediction, which is a fully invariant function (i.e. reflections do not change the receiver), we perform simple frame averaging across all views, arriving at

and then learn a node-wise classifier over the rows of \({{{{{{{{\bf{H}}}}}}}}}^{{{{{{{{\rm{node}}}}}}}}}\in {{\mathbb{R}}}^{22\times 4}\) . We further decode H node into a logit vector \({{{{{{{\bf{O}}}}}}}}\in {{\mathbb{R}}}^{22}\) with a linear layer before computing the corresponding softmax cross entropy loss.

For shot prediction, which is once again fully invariant (i.e. reflections do not change the probability of a shot), we can further average the frame-averaged features across all players to get a global graph representation:

and then learn a binary classifier over \({{{{{{{{\bf{h}}}}}}}}}^{{{{{{{{\rm{graph}}}}}}}}}\in {{\mathbb{R}}}^{4}\) . Specifically, we decode the hidden vector into a single logit with a linear layer and compute the sigmoid binary cross-entropy loss with the corresponding label.

For guided generation (position/velocity adjustments), we generate the player positions and velocities with respect to a particular outcome of interest for the human coaches, predicted over the rows of the hidden feature matrix. For example, the model may adjust the defensive setup to decrease the shot probability by the attacking team. The model output is now equivariant rather than invariant—reflecting the pitch appropriately reflects the predicted positions and velocity vectors. As such, we cannot perform frame averaging, and take only the identity view’s features, \({{{{{{{{\bf{H}}}}}}}}}_{{{{{{{{\rm{id}}}}}}}}}\in {{\mathbb{R}}}^{22\times 4}\) . From this latent feature matrix, we can then learn a conditional distribution from each row, which models the positions or velocities of the corresponding player. To do this, we extend the backbone encoder with conditional variational autoencoder (CVAE 28 , 29 ). Specifically, for the u -th row of H id , h u , we first map its latent embedding to the parameters of a two-dimensional Gaussian distribution \({{{{{{{\mathcal{N}}}}}}}}({\mu }_{u}| {\sigma }_{u})\) , and then sample the coordinates and velocities from this distribution. At training time, we can efficiently propagate gradients through this sampling operation using the reparameterisation trick 28 : sample a random value \({\epsilon }_{u} \sim {{{{{{{\mathcal{N}}}}}}}}(0,1)\) for each player from the unit Gaussian distribution, and then treat μ u  +  σ u ϵ u as the sample for this player. In what follows, we omit edge features for brevity. For each corner kick sample X with the corresponding outcome o (e.g. a binary value indicating a shot event), we extend the standard VAE loss 28 , 29 to our case of outcome-conditional guided generation as

where h u is the player embedding corresponding to the u th row of H id , and \({\mathbb{KL}}\) is Kullback–Leibler (KL) divergence. Specifically, the first term is the generation loss between the real player input x u and the reconstructed sample decoded from h u with the decoder p ϕ . Using the KL term, the distribution of the latent embedding h u is regularised towards p ( h u ∣ o ), which is a multivariate Gaussian in our case.

A complete high-level summary of the generic encoder–decoder equivariant architecture employed by TacticAI can be summarised in Supplementary Fig.  2 . In the following section, we will provide empirical evidence for justifying these architectural decisions. This will be done through targeted ablation studies on our predictive benchmarks (receiver prediction and shot prediction).

Ablation study

We leveraged the receiver prediction task as a way to evaluate various base model architectures, and directly quantitatively assess the contributions of geometric deep learning in this context. We already see that the raw corner kick data can be better represented through geometric deep learning, yielding separable clusters in the latent space that could correspond to different attacking or defending tactics (Fig.  2 ). In addition, we hypothesise that these representations can also yield better performance on the task of receiver prediction. Accordingly, we ablate several design choices using deep learning on this task, as illustrated by the following four questions:

Does a factorised graph representation help? To assess this, we compare it against a convolutional neural network (CNN 30 ) baseline, which does not leverage a graph representation.

Does a graph structure help? To assess this, we compare against a Deep Sets 31 baseline, which only models each node in isolation without considering adjacency information—equivalently, setting each neighbourhood \({{{{{{{{\mathcal{N}}}}}}}}}_{u}\) to a singleton set { u }.

Are attentional GNNs a good strategy? To assess this, we compare against a message passing neural network 32 , MPNN baseline, which uses the fully potent GNN layer from Eq. ( 2 ) instead of the GATv2.

Does accounting for symmetries help? To assess this, we compare our geometric GATv2 baseline against one which does not utilise D 2 group convolutions but utilises D 2 frame averaging, and one which does not explicitly utilise any aspect of D 2 symmetries at all.

Each of these models has been trained for a fixed budget of 50,000 training steps. The test top- k receiver prediction accuracies of the trained models are provided in Supplementary Table  2 . As already discussed in the section “Results”, there is a clear advantage to using a full graph structure, as well as directly accounting for reflection symmetry. Further, the usage of the MPNN layer leads to slight overfitting compared to the GATv2, illustrating how attentional GNNs strike a good balance of expressivity and data efficiency for this task. Our analysis highlights the quantitative benefits of both graph representation learning and geometric deep learning for football analytics from tracking data. We also provide a brief ablation study for the shot prediction task in Supplementary Table  3 .

Training details

We train each of TacticAI’s models in isolation, using NVIDIA Tesla P100 GPUs. To minimise overfitting, each model’s learning objective is regularised with an L 2 norm penalty with respect to the network parameters. During training, we use the Adam stochastic gradient descent optimiser 33 over the regularised loss.

All models, including baselines, have been given an equal hyperparameter tuning budget, spanning the number of message passing steps ({1, 2, 4}), initial learning rate ({0.0001, 0.00005}), batch size ({128, 256}) and L 2 regularisation coefficient ({0.01, 0.005, 0.001, 0.0001, 0}). We summarise the chosen hyperparameters of each TacticAI model in Supplementary Table  1 .

Data availability

The data collected in the human experiments in this study have been deposited in the Zenodo database under accession code https://zenodo.org/records/10557063 , and the processed data which is used in the statistical analysis and to generate the relevant figures in the main text are available under the same accession code. The input and output data generated and/or analysed during the current study are protected and are not available due to data privacy laws and licensing restrictions. However, contact details of the input data providers are available from the corresponding authors on reasonable request.

Code availability

All the core models described in this research were built with the Graph Neural Network processors provided by the CLRS Algorithmic Reasoning Benchmark 18 , and their source code is available at https://github.com/google-deepmind/clrs . We are unable to release our code for this work as it was developed in a proprietary context; however, the corresponding authors are open to answer specific questions concerning re-implementations on request. For general data analysis, we used the following freely available packages: numpy v1.25.2 , pandas v1.5.3 , matplotlib v3.6.1 , seaborn v0.12.2 and scipy v1.9.3 . Specifically, the code of the statistical analysis conducted in this study is available at https://zenodo.org/records/10557063 .

The International Football Association Board (IFAB). Laws of the Game (The International Football Association Board, 2023).

Tuyls, K. et al. Game plan: what AI can do for football, and what football can do for AI. J. Artif. Intell. Res. 71 , 41–88 (2021).

Article   Google Scholar  

Goka, R., Moroto, Y., Maeda, K., Ogawa, T. & Haseyama, M. Prediction of shooting events in soccer videos using complete bipartite graphs and players’ spatial–temporal relations. Sensors 23 , 4506 (2023).

Article   ADS   PubMed   PubMed Central   Google Scholar  

Omidshafiei, S. et al. Multiagent off-screen behavior prediction in football. Sci. Rep. 12 , 8638 (2022).

Article   ADS   CAS   PubMed   PubMed Central   Google Scholar  

Lang, S., Wild, R., Isenko, A. & Link, D. Predicting the in-game status in soccer with machine learning using spatiotemporal player tracking data. Sci. Rep. 12 , 16291 (2022).

Baccouche, M., Mamalet, F., Wolf, C., Garcia, C. & Baskurt, A. Action classification in soccer videos with long short-term memory recurrent neural networks. In International Conference on Artificial Neural Networks (eds Diamantaras, K., Duch, W. & Iliadis, L. S.) pages 154–159 (Springer, 2010).

Shaw, L. & Gopaladesikan, S. Routine inspection: a playbook for corner kicks. In Machine Learning and Data Mining for Sports Analytics: 7th International Workshop, MLSA 2020, Co-located with ECML/PKDD 2020 , Proceedings, Ghent, Belgium, September 14–18, 2020, Vol . 7 , 3–16 (Springer, 2020).

Araújo, D. & Davids, K. Team synergies in sport: theory and measures. Front. Psychol. 7 , 1449 (2016).

Article   PubMed   PubMed Central   Google Scholar  

Veličković, P. Everything is connected: graph neural networks. Curr. Opin. Struct. Biol. 79 , 102538 (2023).

Article   PubMed   Google Scholar  

Bronstein, M. M., Bruna, J., Cohen, T. & Veličković, P. Geometric deep learning: grids, groups, graphs, geodesics, and gauges. arXiv preprint arXiv:2104.13478 (2021).

Brody, S., Alon, U. & Yahav, E. How attentive are graph attention networks? In International Conference on Learning Representations (ICLR, 2022). https://openreview.net/forum?id=F72ximsx7C1 .

Veličković, P. et al. Graph attention networks. In International Conference on Learning Representations (ICLR, 2018). https://openreview.net/forum?id=rJXMpikCZ .

Cohen, T. & Welling, M. Group equivariant convolutional networks. In International Conference on Machine Learning (eds Balcan, M. F. & Weinberger, K. Q.) 2990–2999 (PMLR, 2016).

Honda, Y., Kawakami, R., Yoshihashi, R., Kato, K. & Naemura, T. Pass receiver prediction in soccer using video and players’ trajectories. In 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) 3502–3511 (2022). https://ieeexplore.ieee.org/document/9857310 .

Hubáček, O., Sourek, G. & Železný, F. Deep learning from spatial relations for soccer pass prediction. In MLSA@PKDD/ECML (eds Brefeld, U., Davis, J., Van Haaren, J. & Zimmermann, A.) Vol. 11330, (Lecture Notes in Computer Science, Springer, Cham, 2018).

Sanyal, S. Who will receive the ball? Predicting pass recipient in soccer videos. J Visual Commun. Image Represent. 78 , 103190 (2021).

Gilmer, J., Schoenholz, S. S., Riley, P. F., Vinyals, O. & Dahl, G. E. Neural message passing for quantum chemistry. In International Conference on Machine Learning (Precup, D. & Teh, Y. W.) 1263–1272 (PMLR, 2017).

Veličković, P. et al. The CLRS algorithmic reasoning benchmark. In International Conference on Machine Learning (eds Chaudhuri, K. et al.) 22084–22102 (PMLR, 2022).

Vaswani, A. et al. Attention is all you need. In Advances in Neural Information Processing Systems (eds Guyon, I. et al.) Vol. 30 (Curran Associates, Inc., 2017).

Veličković, P. Message passing all the way up. In ICLR 2022 Workshop on Geometrical and Topological Representation Learning (GTRL, 2022). https://openreview.net/forum?id=Bc8GiEZkTe5 .

Baranwal, A., Kimon, F. & Aukosh, J. Optimality of message-passing architectures for sparse graphs. In Thirty-seventh Conference on Neural Information Processing Systems (2023). https://papers.nips.cc/paper_files/paper/2023/hash/7e991aa4cd2fdf0014fba2f000f542d0-Abstract-Conference.html .

Greenblatt, R. D., Eastlake III, D. E. & Crocker, S. D. The Greenblatt chess program. In Proc. Fall Joint Computer Conference , 14–16 , 801–810 (Association for Computing Machinery, 1967). https://dl.acm.org/doi/10.1145/1465611.1465715 .

Schijf, M., Allis, L. V. & Uiterwijk, J. W. Proof-number search and transpositions. ICGA J. 17 , 63–74 (1994).

Fuchs, F., Worrall, D., Fischer, V. & Welling, M. SE(3)-transformers: 3D roto-translation equivariant attention networks. Adv. Neural Inf. Process. Syst. 33 , 1970–1981 (2020).

Google Scholar  

Silver, D. et al. Mastering the game of Go with deep neural networks and tree search. Nature 529 , 484–489 (2016).

Article   ADS   CAS   PubMed   Google Scholar  

Satorras, V. G., Hoogeboom, E. & Welling, M. E ( n ) equivariant graph neural networks. In International Conference on Machine Learning (eds Meila, M. & Zhang, T.) 9323–9332 (PMLR, 2021).

Cohen, T. S. & Welling, M. Steerable CNNs. In International Conference on Learning Representations (ICLR, 2017). https://openreview.net/forum?id=rJQKYt5ll .

Kingma, D. P. & Welling, M. Auto-encoding variational Bayes. In 2nd International Conference on Learning Representations, ICLR 2014, Banff, AB, Canada, April 14–16, 2014, Conference Track Proceedings (ICLR, 2014). https://openreview.net/forum?id=33X9fd2-9FyZd .

Sohn, K., Lee, H. & Yan, X. Learning structured output representation using deep conditional generative models. In Advances in Neural Information Processing Systems (eds Cortes, C, Lawrence, N., Lee, D., Sugiyama, M. & Garnett, R.) Vol. 28 (Curran Associates, Inc., 2015).

Fernández, J. & Bornn, L. Soccermap: a deep learning architecture for visually-interpretable analysis in soccer. In Machine Learning and Knowledge Discovery in Databases. Applied Data Science and Demo Track: European Conference, ECML PKDD 2020, Ghent, Belgium, September 14–18, 2020, Proceedings, Part V (eds Dong, Y., Ifrim, G., Mladenić, D., Saunders, C. & Van Hoecke, S.) 491–506 (Springer, 2021).

Zaheer, M. et al. Deep sets. In Advances in Neural Information Processing Systems Vol. 30 (eds Guyon, I., et al.) (Curran Associates, Inc., 2017).

Gilmer, J., Schoenholz, S. S., Riley, P. F., Vinyals, O. & Dahl, G. E. Neural message passing for quantum chemistry. In Proc. 34th International Conference on Machine Learning , Vol. 70 of Proceedings of Machine Learning Research, 6–11 Aug 2017 (eds Precup, D. & Whye Teh, Y) 1263–1272 (PMLR, 2017).

Kingma, G. E. & Ba, J. Adam: a method for stochastic optimization. In ICLR (Poster) , (eds Bengio, Y. & LeCun, Y.) (International Conference of Learning Representations (ICLR), 2015). https://openreview.net/forum?id=8gmWwjFyLj .

Download references

Acknowledgements

We gratefully acknowledge the support of James French, Timothy Waskett, Hans Leitert and Benjamin Hervey for their extensive efforts in analysing TacticAI’s outputs. Further, we are thankful to Kevin McKee, Sherjil Ozair and Beatrice Bevilacqua for useful technical discussions, and Marc Lanctôt and Satinder Singh for reviewing the paper prior to submission.

Author information

These authors contributed equally: Zhe Wang, Petar Veličković, Daniel Hennes.

Authors and Affiliations

Google DeepMind, 6-8 Handyside Street, London, N1C 4UZ, UK

Zhe Wang, Petar Veličković, Daniel Hennes, Nenad Tomašev, Laurel Prince, Michael Kaisers, Yoram Bachrach, Romuald Elie, Li Kevin Wenliang, Federico Piccinini, Jerome Connor, Yi Yang, Adrià Recasens, Mina Khan, Nathalie Beauguerlange, Pablo Sprechmann, Pol Moreno, Nicolas Heess & Demis Hassabis

Liverpool FC, AXA Training Centre, Simonswood Lane, Kirkby, Liverpool, L33 5XB, UK

William Spearman

Liverpool FC, Kirkby, UK

University of Alberta, Amii, Edmonton, AB, T6G 2E8, Canada

Michael Bowling

Google DeepMind, London, UK

You can also search for this author in PubMed   Google Scholar

Contributions

Z.W., D. Hennes, L.P. and K.T. coordinated and organised the research effort leading to this paper. P.V. and Z.W. developed the core TacticAI models. Z.W., W.S. and I.G. prepared the Premier League corner kick dataset used for training and evaluating these models. P.V., Z.W., D. Hennes and N.T. designed the case study with human experts and Z.W. and P.V. performed the qualitative evaluation and statistical analysis of its outcomes. Z.W., P.V., D. Hennes, N.T., L.P., M. Kaisers, Y.B., R.E., L.K.W., F.P., W.S., I.G., N.H., M.B., D. Hassabis and K.T. contributed to writing the paper and providing feedback on the final manuscript. J.C., Y.Y., A.R., M. Khan, N.B., P.S. and P.M. contributed valuable technical and implementation discussions throughout the work’s development.

Corresponding authors

Correspondence to Zhe Wang , Petar Veličković or Karl Tuyls .

Ethics declarations

Competing interests.

The authors declare no competing interests but the following competing interests: TacticAI was developed during the course of the Authors’ employment at Google DeepMind and Liverpool Football Club, as applicable to each Author.

Peer review

Peer review information.

Nature Communications thanks Rui Luo and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary information

Supplementary information, peer review file, rights and permissions.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ .

Reprints and permissions

About this article

Cite this article.

Wang, Z., Veličković, P., Hennes, D. et al. TacticAI: an AI assistant for football tactics. Nat Commun 15 , 1906 (2024). https://doi.org/10.1038/s41467-024-45965-x

Download citation

Received : 13 October 2023

Accepted : 08 February 2024

Published : 19 March 2024

DOI : https://doi.org/10.1038/s41467-024-45965-x

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

By submitting a comment you agree to abide by our Terms and Community Guidelines . If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.

Quick links

• Explore articles by subject
• Guide to authors
• Editorial policies

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

An official website of the United States government

The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

• Publications
• Account settings

Preview improvements coming to the PMC website in October 2024. Learn More or Try it out now .

• Advanced Search
• Journal List
• Front Aging Neurosci

Alzheimer’s disease and epilepsy: The top 100 cited papers

Gui-fen zhang.

1 Department of General Practice and International Medicine, Second Affiliated Hospital, School of Medicine, Zhejiang University, Hangzhou, China

Wen-Xin Gong

Zheng-yan-ran xu.

2 Department of Neurology, Epilepsy Center, Second Affiliated Hospital, School of Medicine, Zhejiang University, Hangzhou, China

Associated Data

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found in the article/supplementary material.

Alzheimer’s disease (AD) is one of the common neurodegenerative diseases, which often coexists with epilepsy. It is very significant to study the treatment options and the relationship between AD and epilepsy.

The purpose of this study was to analyze the top 100 cited papers about AD and epilepsy using bibliometrics, and to describe the current situation and predict research hot spots.

Top 100 papers were obtained from the Web of Science Core Collection (WoSCC). The WoSCC was used to analyze the author, institution, country, title, keywords, abstract, citation, subject category, publication year, impact factor (IF), and other functions. SPSS25 software was used for statistical analysis and CiteSpace V.5.7.R2 was used to visualize the information through collaborative networks.

The number of publications gradually increased from 2000 to 2021. The total citation count for the top 100 papers ranged from 15 to 433(mean = 67.43). The largest number of papers were published in 2016 ( n = 11). Meanwhile, USA (centrality: 0.93) and Columbia University (centrality: 0.06) were the most influential research country and institutions, respectively. The top contributing journals was Journal of Alzheimer’s Disease (8%). The IF for journals ranged from 1.819 to 53.44. A network analysis of the author’s keywords showed that “beta” (centrality: 0.39), “amyloid beta” (centrality: 0.29), “hyperexcitability” (centrality: 0.29) and “disease” (centrality: 0.29) had a high degree of centrality.

AD and epilepsy have been intensively studied in the past few years. The relationships, mechanisms and treatment of AD and epilepsy will be subjects of active research hotpots in future. These findings provide valuable information for clinicians and scientists to identify new perspectives with potential collaborators and cooperative countries.

Introduction

Alzheimer’s disease (AD) and epilepsy are common neurological diseases. The number of patients with dementia and epilepsy in the global population is increasing, representing a growing problem for global health. The overall lifetime prevalence of epilepsy is 7.60 per 1,000 population [95% confidence interval (CI) 6.17–9.38] and the prevalence of epilepsy tends to peak in the elderly ( Beghi, 2020 ). The prevalence of AD in Europe was estimated at 5.05% (95% CI, 4.73–5.39), and similar to epilepsy, AD prevalence increases with age ( Niu et al., 2017 ). In fact, these two diseases are related. Epilepsy occurs more frequently in patients with AD than in those with non-Alzheimer’s disease ( Samson et al., 1996 ). Recent findings showed that seizures could accelerate the decline of cognitive ability in patients with AD and that there might be an important bidirectional relationship between epilepsy and AD ( Sen et al., 2018 ). AD and Epilepsy also share many pathological similarities ( Lehmann et al., 2021 ), e.g., temporal lobe atrophy, neuronal death, gliosis, neuritic alterations, and neuroinflammation ( Struble et al., 2010 ). In addition, AD plus epilepsy would lead to more serious clinical consequences, such as cognitive decline, weakness, anxiety, depression, social withdrawal, psychological and behavioral comorbidity, and poor treatment compliance ( Cretin, 2021 ). Overtime, a large amount of literature has been published comprising a wide range of relevant research and clinical themes. However, the precise mechanisms leading to the development of seizures in the setting of AD are still under investigation and require further study. A meta-analysis showed that the quality of evidence on the treatment outcome of epilepsy in patients with AD was very low ( Liu and Wang, 2021 ).

As a discipline emerging since its formal foundation, a bibliometric review of the literature was warranted to aid the synthesis and implementation of the evidence base. Despite citation analysis across a broad range of neurosciences ( Yeung et al., 2017 ), there is limited information in the field of AD and epilepsy, with few published studies. Citation counting is an important metric to understand the significance of the contribution of research to a research field ( Fox et al., 2021 ). Previous reviews only relied on individuals to study the research through literature summary and extraction, and thus cannot fully reflect the temporal and spatial distribution of researchers, institutions, and journals. Moreover, it is difficult to visualize the internal structure of the knowledge base and research focus, and systematic, comprehensive, and visual research are rarely found. Therefore, the present study aims to comprehensively analyze the current status, research hotpots, and development trends through a bibliometric analysis of the top 100 papers on AD and epilepsy published from 2000 to 2021. The findings may help follow-up researchers study the association between AD and epilepsy, identify journal publications and collaborators, and analyze keywords and research trends, which might promote research aiming to determine the cause, mechanism, and treatment of the disease.

Materials and methods

Data source.

The retrieval data for measurement and statistical analysis were screened from the Web of Science Core Collection (WoSCC), which provided the citation search, giving access to multiple databases that reference cross-disciplinary research and allowing an in-depth exploration of specialized subfields ( Wu et al., 2021 ). We conducted a literature search from the WoSCC on June 5th, 2022. In this study, the search criteria in the WoSCC database were as follows: ((TI = (epilep* OR seizure* OR convuls*)) OR KP = (epilep* OR seizure* OR convuls*)) AND ((TI = (dement* OR Alzheimer* OR “cognit* impair*” OR AD)) OR KP = (dement* OR Alzheimer* OR “cognit* impair*” OR AD)). Timespan: 2000-01-01 to 2021-12-31 (Publication Date). Document type: articles and reviews; Language: English. A total of 2601 records were retrieved. Then, two independent investigators reviewed the titles and abstracts and deleted studies that were not associated with AD and epilepsy, which excluded 1237 papers according to the criteria ( Guo et al., 2021 ). And 1264 papers were excluded after the 100th rank from the selected literature. Finally, the top 100 studies were determined. A Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flow diagram for the WoSCC results is provided in Figure 1 .

Flow chart of literature collection.

Data analysis

In the present study, WoSCC was used to analyze the author, institution, country, title, keywords, abstract, citation, subject category, publication year, impact factor (IF), and other functions. The selected documents were imported into Excel (Microsoft Corporation, Redmond, WA, United States) and CiteSpace 5.8.R3 (64-bit) ( Palop and Mucke, 2009 ). SPSS 25.0 statistical software (IBM Corp., Armonk, NY, United States) was used for statistical analysis. Continuous variables were expressed as the mean ± SD. Categorical variables were expressed as a percentage. CiteSpace, a bibliometric analysis tool, was created by Dr. Chaomei Chen (School of Information Science and Technology, Drexel University, Philadelphia, PA, United States) and his team. It visualizes countries/regions, institutions, authors and their cooperative relationships, co-cited references, and co-occurrence words through collaborative networks, which has been widely used in biomedical research fields.

Three folders of the research were created, including input folder placing the data downloaded from WoSCC, a data folder containing the data after deleting duplicate documents, and a project folder containing the data processed by cite. We did not find duplicate documents that needed to be deleted. The overall selected time span was from January 2000 to December 2021. Then, the slice length was set as 2 years. The node type was selected according to the type of analysis performed. The link lines between the nodes indicated the collaborative relationships. The size of the circles represented the number of papers published by the country/region, institute, or author. Purple rings indicated that these countries/regions, institutes, or authors had greater centrality.

The 100 most-cited publications

We retrieved the 100 most frequently cited papers related to AD and epilepsy. The results were ranked according to citation counts to represent the 100 most-cited publications (76 articles and 24 reviews). A comprehensive list of the 100 publications and a citation details are presented in the Table 1 .

The 100 most-cited publications.

As shown in Table 1 , The 100 most-cited articles received a total of 6743 citations (according to Web of science, WOS). The median number citations was 39, with a range of 15–433. For annual citations, the mean value was 7.41 with a range of 0.8–35.70. Seventeen papers were cited more than 100 times, and 36 were cited more than 50 times. The review entitled “Epilepsy and Cognitive Impairments in Alzheimer Disease” from Palop and Mucke (2009) was the most-cited publication ( n = 433).

Publication years

As shown in Figure 2 , the top 100 most-cited papers were published from 2000 to 2020. Overall, publications showed a fluctuating upward trend. The largest number of studies was published in 2016 ( n = 11), including eight articles and three reviews. The number of papers published in 2011 was higher than the number of papers published between 2000 and 2010.

The number of publications among different types of papers according to publication year.

Contributions of countries

Overall, 22 countries contributed to the included studies, with nine countries publishing only one study. The United States was the largest contributor of studies (38%), followed by Italy (11%) and the United Kingdom (9%). The contributing countries are shown in Figure 3 .

Countries among the top 100 most-cited papers.

Generating a country map using CiteSpace resulted in 22 nodes and 44 links ( Figure 4 ). The 100 papers were published by research groups in 22 countries. The top five countries were the United States, Italy, the United Kingdom, France, and Finland. The top three countries in terms of centrality were the United States (0.93), the United Kingdom (0.44), and France (0.21). An analysis in terms of publication and centrality indicated that the United States, the United Kingdom, Italy, and France were the main research powers in this research. United States has established cooperation with 12 countries, and the strongest collaborations were identified between United States, Canada, Israel, Greece, India, and Switzerland.

Contributions of countries. Purple rings indicated that these countries/regions had greater centrality (no less than 0.1). The size of the circle represents the number of papers published by the country. The shorter the distance between two circles, the greater the cooperation between the two countries. United States was the most influential research country.

Contributions of institutions

A total of 178 institutions published at least one top-cited paper, the distribution of institutions was very scattered, with 21 (11.80%) institutions publishing two papers and 148 (83.15%) institutions publishing only one paper. A small number of institutions accounted for a high proportion of the highest cited papers. Table 2 shows that the top nine institutions collectively published at least three papers. Baylor College of Medicine from United States topped the list with eight papers.

Top nine institutions with at least three papers in the top 100 most-cited papers.

Figure 5 shows that generating an institution map resulted in 149 nodes and 308 links. The top 100 publications were distributed among 149 research institutions. The top five institutions were Baylor College Medicine, University of Eastern Finland, Columbia University, and Johns Hopkins University and Kuopio University Hospital. The top four institutions in terms of centrality were Columbia University (0.06), Baylor College of Medicine (0.04), Johns Hopkins University (0.04) and Indiana University (0.03). Analysis in terms of centrality indicated that Columbia University was the most influential research institution.

Contributions of institutions. Nodes represent institutes, and the size of each node corresponds to the co-occurrence frequency of the institutes. The size of the circle represents the number of papers published by the institute. The shorter the distance between two circles, the greater the cooperation between the two institutes. Columbia University was the most influential research institutions.

Distribution of journals

The 100 most-cited articles were published in 59 journals; 18 journals published more than one study, which were distributed in the four partitions of Journal Citation Reports (JCR) ( McVeigh, 2008 ). The JCR data provide both detailed journal information and flexible context, and category information to allow people to understand the way each journal functions in the literature. The major contributing journals are presented in Table 3 . The top five journals that published the 100 most-cited AD and epilepsy studies included Journal of Alzheimer’s Disease ( n = 8), Epilepsia ( n = 6), Epilepsy and Behavior ( n = 4), Neurobiology of Aging ( n = 4) and Epilepsy Research ( n = 4). With regard to the average citation number per paper, Journal of Neuroscience ranked first with a mean of 20.98 citations per paper, followed by Nature Medicine, with a mean of 17.84.

Journals contributed ≥ 2 papers in the top 100 most-cited papers.

The IF for journals in the top 100 most-cited papers ranged from 1.819 to 53.44, among which 37 journals had an IF between 3 and 5, 34 journals had an IF between 5 and 10, and 3 journals had an IF above 20 ( Figure 6 ).

The number of articles corresponding to different IF in the top 100 most-cited papers.

In Table 4 , the top 100 papers were classified into different study fields on the basis of WOS categories. The leading WOS category was “Clinical Neurology” ( n = 54), following by “Neurosciences” ( n = 49) and “Behavioral Sciences” ( n = 10).

The number of study fields on the basis of WOS categories.

Major contributing authors

Overall, a total of 606 authors contributed to the 100 studies. There was wide, disparate authorship of first authors, with 90 different first authors represented in the 100 included publications. A total of four contributors published the most articles, namely Tanila Heikki, Noebels Jeffrey, Pitkänen, Asla and Scharfman, Helen E, who all published five publications, the total number of citations for the papers were 629, 592, 578, and 234 respectively. Only three authors have published three studies as a first author. Keith A. Vossel from the University of California, as a first author and corresponding author, had the largest number of total citations in 2021 ( n = 185). Table 5 presents results for authors who contributed three or more of the 100 most-cited papers.

Major contributing authors in the top 100 most-cited list.

Analysis of co-occurring keywords

CiteSpace was used to extract the keywords from the top 100 papers on AD and epilepsy. A network analysis of the author’s keywords or subject words was carried out during the publication time of the articles, namely, 2000–2021. Table 6 shows that the top five keywords are epilepsy ( n = 41), dementia ( n = 21), mouse model ( n = 21), mild cognitive impairment ( n = 13), and Alzheimer’s disease ( n = 11). The greater the centrality value, the more cooperation between the node and other nodes. Figure 7 shows that “beta” (centrality: 0.39), “amyloid beta” (centrality: 0.29), “hyperexcitability” (centrality: 0.29) and “disease” (centrality: 0.29) had a high degree of centrality during this period. A comprehensive analysis of centrality showed that “beta” ( n = 7, centrality: 0.39), “disease” ( n = 3, centrality: 0.29), “dementia” ( n = 21, centrality: 0.25) and “brain” ( n = 4, centrality: 0.25) are the most influential keywords in this field.

Frequency of co-occurring keywords.

Analysis of co-occurrence Keywords. Nodes represent keywords, and the size of each node corresponds to the co-occurring frequency of the keywords. The color of the lines that appear together between keywords indicates chronological order: grey represents the oldest, and orange the newest. “beta” had a highest degree of centrality.

The aim of this paper was to perform a bibliometric study of the top 100 most-cited publications on AD and epilepsy over the last 22 years. To the best of our knowledge, this was the first study to provide an overview the current main status of development, hot spots of study, and the future trends in AD and epilepsy. Our study is of great significance for students, researchers, and clinicians working in the field. Here, we summarize several characteristics of these papers to help understand the history and professional prospects comprehensively.

The average number of citations for the 100 most-cited clinical trials in AD and epilepsy was 67.43 (range: 15–433), well below the number of citations observed in general neuroscience articles (3,087) ( Yeung et al., 2017 ). However, citation counts are not a direct measure of scientific quality and importance. The number of citations can be regarded as a relatively reasonable index to evaluate the quality of papers, which varies in different sub disciplines and depends on the size of the scientific community. Citation count analysis might be related to journal IF, publication frequency, and publication year. In general, an article with 100 or more citations is considered a “classic” in the research field and might even be a seminal paper ( Xiong et al., 2021 ). However, our study showed that only 17 papers were cited more than 100 times, this can possibly come from the fact that comorbid epilepsy in AD may not be an interesting topic for many clinicians. In fact, cognitivists dealing with AD patients are not highly aware of epilepsy and, inversely, epileptologists may be poorly informed of dementing diseases. Therefore, the AD-epilepsy topic may have interest only for a small research community, which can result in a relatively low citation index.

The most-cited publication was “Epilepsy and Cognitive Impairments in Alzheimer’s Disease” ( Palop and Mucke, 2009 ), wrote by Palop and Mucke (2009) , with 433 total citations, 31 annual citations, and 39 citations in 2021. It may be relevant to current research and have far-reaching implications. However, in general, consensus and position papers, guides, and systematic reviews receive more citations than original articles, which is a bias that must be considered in citation analysis. The article that ranked second, which was based on a “Amyloid beta-Induced Neuronal Hyperexcitability Triggers Progressive Epilepsy” ( Minkeviciene et al., 2009 ) had a total of 405 citations, 28.93 annual citations and 54 citations in 2021; it was published in Journal of Neuroscience in 2009 and written by Minkeviciene et al. (2009) . The team performed video-EEG recordings in mice carrying mutant human APPswe and PS1dE9 genes (APdE9 mice) and their wild-type littermates, identifying fibrillar Abeta may be the cause of epileptiform activity.

The number of papers published has increased in recent years. Although the most cited publications tend to be from the first few years (2000–2010), the number of highly cited articles published from 2011–2020 is nearly three times that published from 2000–2010. In 2016, the number of publications was the largest, reaching 11, which showed that a recently published article is likely to gradually improve the quality of research in recent years and have a potential academic importance in the future.

The majority of research originated from the United States (38%). A bibliometric analysis of the most cited articles in neurocritical care research showed that United States was the country with most articles (60, 35 primary research) and citations (6115) among the top 100 ( Ramos et al., 2019 ). Baylor College of Medicine from United States topped the list with eight papers. The distribution of country and institutions was very scattered. The United States and its institutions have played a leading role in AD and epilepsy research. The United States was also ranked first in other fields of neurology ( Xiong et al., 2021 ), and it is also a research leader in terms of quality and quantity. However, compared with other studies, our study showed that the cooperation between countries and institutions was not close ( Yin et al., 2019 ). In fact, just like the study of Parkinson’s disease ( Li et al., 2008 ), more cooperation between different countries and institutions might be an effective way to promote the development of AD and epilepsy research worldwide. Contrastingly, distribution analysis showed that research was widespread all over the world and the diseases had research value.

Baylor College of Medicine and Columbia University played important roles in in the field of AD and epilepsy research. Baylor College of Medicine, as one of the top colleges and universities in Texas, United States, focused on cooperative research programs, discovered basic insights into human health and diseases through extensive interdisciplinary and interdisciplinary cooperation, and applied their findings to develop new diagnostic tools and treatments. Columbia University is a world-class private research university located in Manhattan, NY, United States. It carried out extensive research in the field of neuroscience.

The 100 included articles were published in 59 journals, among which 18 journals published more than one study. The papers were distributed in four partitions of JCR. The IF of the top five journals was less than six. The top journal among the list of the 100 most-cited studies was Journal of Alzheimer’s Disease (IF = 4.472). The leading WOS categories focused on “Clinical Neurology” and “Neurosciences,” with few interdisciplinary studies. Resulting from its interdisciplinary nature, Psycho-Oncology has been subject to extensive interdisciplinary research, which has provided great help for the development of the discipline ( Fox et al., 2021 ). Therefore, we look forward to more interdisciplinary research.

Author analysis revealed a network of core author collaborations in the field of AD and epilepsy research. This information might be relevant to clinical researchers and research institutions who are searching for a network of research leaders in the field to explore potential collaborations. Our authorship analysis does not show as many authors as other areas in research. The most contributing author was Tanila, Heikki from Kuopio University Hospital from Finland. Related research focuses on the pathogenesis of epilepsy in mouse models of AD, which is a research hotspot in this field.

Keyword analysis showed that “beta,” “amyloid beta,” “hyperexcitability,” “disease,” “dementia,” and “brain” had high influential. indicating that these research directions are very significant. This result showed that researchers have begun to extend their research to the pathogenesis. These results also show that AD and epilepsy is still a disease that requires to be solved urgently. We must explore the deficiencies and innovations in this field, such as treatment, pathological mechanism, and disease management on improve the quality of life of patients.

Limitations

The present study has some limitations. First, there are several intrinsic limitations of using citation analysis to evaluate the academic importance of a specific article, author, or journal. There is a certain bias in citation analysis, such as the fact that papers written in English, papers that can be accessed through open access, and papers published in journals with high IFs are more likely to be cited. In addition, through a “snowball effect,” people tended to cite publications that are already highly cited ( Yeung et al., 2017 ). We selected the top 100 papers, but citation searches are “time-dependent,” older articles are likely to be cited more often, and the newest list of highly cited articles may be dominated by some older articles. Furthermore, citation analysis might severely underestimate the impact of clinical research as compared to basic research ( van Eck et al., 2013 ). Second, the search was limited to the WOS database. It did not record citations by textbooks and other databases. Our study only selected papers written in English, which might have yielded an incomplete search.

We identified the 100 most cited papers in the field of AD and epilepsy. By reviewing these top cited papers, researchers can immediately understand the hot topics and research collaborations on AD and epilepsy, and improve their work. This study shows that the relationship, mechanism, and treatment of AD and epilepsy have been widely studied, and in recent years, this field has shown new vitality; however, there is a general lack of cooperation between countries, and the mechanism of epilepsy and AD is unclear, which deserves further study.

Data availability statement

Author contributions.

G-FZ and YG designed the study. G-FZ drafted and edited the manuscript. G-FZ and W-XG analyzed the data. G-FZ, Z-Y-RX, YG, and W-XG revised the manuscript. All authors contributed to the article and approved the submitted version.

Conflict of interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

This study was supported by the National Natural Science Foundation of China (Grant No. 81871010).

• Beghi E. (2020). The Epidemiology of Epilepsy. Neuroepidemiology 54 185–191. 10.1159/000503831 [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Cretin B. (2021). Treatment of Seizures in Older Patients with Dementia. Drugs Aging 38 181–192. 10.1007/s40266-020-00826-2 [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Fox S., Lynch J., D’Alton P., Carr A. (2021). Psycho-Oncology: a Bibliometric Review of the 100 Most-Cited Articles. Healthcare 9 :1008. 10.3390/healthcare9081008 [ PMC free article ] [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Guo Y., Xu Z. Y. R., Cai M. T., Gong W. X., Shen C. H. (2021). Epilepsy With Suicide: a Bibliometrics Study and Visualization Analysis via CiteSpace. Front. Neurol. 12 :823474. 10.3389/fneur.2021.823474 [ PMC free article ] [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Lehmann L., Lo A., Knox K. M., Barker-Haliski M. (2021). Alzheimer’s Disease and Epilepsy: a Perspective on the Opportunities for Overlapping Therapeutic Innovation. Neurochem. Res. 46 1895–1912. 10.1007/s11064-021-03332-y [ PMC free article ] [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Li T., Ho Y. S., Li C. Y. (2008). Bibliometric analysis on global Parkinson’s disease research trends during 1991-2006. Neurosci. Lett. 441 248–252. 10.1016/j.neulet.2008.06.044 [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Liu J., Wang L. N. (2021). Treatment of epilepsy for people with Alzheimer’s disease. Cochrane Database Syst. Rev. 12 :CD011922. 10.1002/14651858.CD011922.pub4 [ PMC free article ] [ PubMed ] [ CrossRef ] [ Google Scholar ]
• McVeigh M. E. (2008). Neurology in the journal citation reports. Neurology 71 1848–1849. 10.1212/01.wnl.0000338902.81462.88 [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Minkeviciene R., Rheims S., Dobszay M. B., Zilberter M., Hartikainen J., Fülöp L. (2009). Amyloid beta-induced neuronal hyperexcitability triggers progressive epilepsy. J. Neurosci. 29 3453–3462. 10.1523/JNEUROSCI.5215-08.2009 [ PMC free article ] [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Niu H., Álvarez-Álvarez I., Guillén-Grima F., Aguinaga-Ontoso I. (2017). Prevalence and incidence of Alzheimer’s disease in Europe: a meta-analysis. Neurologia 32 523–532. 10.1016/j.nrl.2016.02.016 [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Palop J. J., Mucke L. (2009). Epilepsy and Cognitive Impairments in Alzheimer Disease. Arch. Neurol. 66 435–440. 10.1001/archneurol.2009.15 [ PMC free article ] [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Ramos M. B., Koterba E., Rosi Júnior J., Teixeira M. J., Figueiredo E. G. A. (2019). Bibliometric Analysis of the Most Cited Articles in Neurocritical Care Research. Neurocrit. Care 31 365–372. 10.1007/s12028-019-00731-6 [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Samson W. N., van Duijn C. M., Hop W. C., Hofman A. (1996). Clinical features and mortality in patients with early-onset Alzheimer’s disease. Eur. Neurol. 36 103–106. 10.1159/000117218 [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Sen A., Capelli V., Husain M. (2018). Cognition and dementia in older patients with epilepsy. Brain 141 1592–1608. 10.1093/brain/awy022 [ PMC free article ] [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Struble R. G., Ala T., Patrylo P. R., Brewer G. J., Yan X. X. (2010). Is brain amyloid production a cause or a result of dementia of the Alzheimer’s type? J. Alzheimers Dis. 22 393–399. 10.3233/JAD-2010-100846 [ PMC free article ] [ PubMed ] [ CrossRef ] [ Google Scholar ]
• van Eck N. J., Waltman L., van Raan A. F., Klautz R. J., Peul W. C. (2013). Citation analysis may severely underestimate the impact of clinical research as compared to basic research. PLoS One. 8 :e62395. 10.1371/journal.pone.0062395 [ PMC free article ] [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Wu M. Q., Wu D. Q., Hu C. P., Iao L. S. (2021). Studies on Children With Developmental Coordination Disorder in the Past 20 Years: a Bibliometric Analysis via CiteSpace. Front Psychiatry 12 :776883. 10.3389/fpsyt.2021.776883 [ PMC free article ] [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Xiong H. Y., Liu H., Wang X. Q. (2021). Top 100 Most-Cited Papers in Neuropathic Pain From 2000 to 2020: a Bibliometric Study. Front. Neurol. 12 :765193. 10.3389/fneur.2021.765193 [ PMC free article ] [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Yeung A. W. K., Goto T. K., Leung W. K. (2017). At the Leading Front of Neuroscience: a Bibliometric Study of the 100 Most-Cited Articles. Front. Hum. Neurosci. 11 :363. 10.3389/fnhum.2017.00363 [ PMC free article ] [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Yin X., Cheng F., Wang X., Mu J., Ma C., Zhai C., et al. (2019). Top 100 cited articles on rheumatoid arthritis: a bibliometric analysis. Medicine 98 :e14523. 10.1097/md.0000000000014523 [ PMC free article ] [ PubMed ] [ CrossRef ] [ Google Scholar ]