Executive Summary
The most important lesson from failed Alzheimer's programs is not that a single therapeutic hypothesis was disproved. The failures show that success depends on alignment across mechanism, disease stage, exposure, biomarker response, clinical outcome measurement, safety, and the proposed approval pathway.
The approved treatment landscape prevents an overly broad interpretation of the negative record. Regulatory success has occurred through distinct routes, including treatment of cognitive symptoms across dementia stages, treatment of agitation associated with Alzheimer's dementia, and amyloid-directed treatment initiated in biomarker-confirmed mild cognitive impairment or mild dementia.
For regulatory strategy, negative trials are most useful when separated into distinct failure modes. Some programs engaged a target without changing downstream disease. Some changed a biomarker without producing clinical benefit. Some moved into earlier disease but still lacked sufficient biological effect. Others revealed that the exposure needed to test the hypothesis could not be delivered with an acceptable benefit-risk profile.
A 2026-ready development program should therefore be built as a connected evidence system rather than a sequence of conventional phases. Each transition should answer a specific question: Is the enrolled population biologically defined? Is the drug reaching and altering the intended target? Is the magnitude and duration of that effect sufficient? Is there concordant movement in downstream biology and clinical trajectory? Can the effect be reproduced in a design that supports the intended regulatory pathway?
The FDA-Approved Treatment Context
The current U.S. treatment landscape is broader than either symptomatic therapy or anti-amyloid therapy alone. It includes cholinesterase inhibitors for dementia of the Alzheimer's type, memantine for moderate-to-severe dementia, a memantine and donepezil combination, two amyloid beta-directed antibodies for treatment initiated in mild cognitive impairment or mild dementia, and brexpiprazole for agitation associated with dementia due to Alzheimer's disease. 12345678910
| FDA-approved product or active ingredient | Therapeutic role | Labeled disease stage or use | Regulatory context |
|---|---|---|---|
| Donepezil, including Aricept | Acetylcholinesterase inhibition | Dementia of the Alzheimer's type; efficacy demonstrated in mild, moderate, and severe disease | Symptomatic cognitive treatment across multiple dementia stages |
| Rivastigmine, including Exelon Patch | Cholinesterase inhibition | Dementia of the Alzheimer's type; patch efficacy demonstrated in mild, moderate, and severe disease | Symptomatic treatment with oral and transdermal development history |
| Galantamine, including Razadyne | Cholinesterase inhibition | Mild-to-moderate dementia of the Alzheimer's type | Symptomatic cognitive treatment |
| Benzgalantamine, marketed as Zunveyl | Cholinesterase inhibition through a galantamine prodrug | Mild-to-moderate dementia of the Alzheimer's type in adults | Newer oral product within the established symptomatic treatment category |
| Memantine, including Namenda XR | NMDA receptor antagonism | Moderate-to-severe dementia of the Alzheimer's type | Symptomatic treatment for later-stage dementia |
| Memantine and donepezil extended-release combination | NMDA receptor antagonism plus acetylcholinesterase inhibition | Moderate-to-severe dementia of the Alzheimer's type in patients stabilized on donepezil | Fixed-combination symptomatic treatment |
| Lecanemab, marketed as Leqembi, with Leqembi Iqlik maintenance dosing | Amyloid beta-directed antibody | Treatment of Alzheimer's disease initiated in mild cognitive impairment or mild dementia after confirmation of amyloid beta pathology | Converted from accelerated to traditional approval after verification of clinical benefit; subcutaneous maintenance presentation subsequently approved |
| Donanemab, marketed as Kisunla | Amyloid beta-directed antibody | Treatment of Alzheimer's disease initiated in mild cognitive impairment or mild dementia after confirmation of amyloid beta pathology | Approved early-stage disease treatment with product-specific dosing, imaging, and safety provisions |
| Brexpiprazole, marketed as Rexulti | Atypical antipsychotic | Agitation associated with dementia due to Alzheimer's disease | Behavior-specific indication; not approved for as-needed treatment |
The approved products occupy different regulatory categories. Donepezil, rivastigmine, galantamine, benzgalantamine, memantine, and the memantine-donepezil combination are approved for treatment of dementia at specified severity stages. Brexpiprazole has a behavior-specific indication for agitation associated with dementia due to Alzheimer's disease and is not indicated for as-needed use. Lecanemab and donanemab are approved for treatment of Alzheimer's disease with initiation in patients with mild cognitive impairment or mild dementia, following confirmation of amyloid beta pathology. 1234567910
Lecanemab is especially important to the regulatory history because FDA converted it from accelerated approval to traditional approval after determining that a confirmatory trial verified clinical benefit. Its subsequent development also produced an approved subcutaneous maintenance presentation, Leqembi Iqlik, for use after the initial treatment period specified in labeling. Donanemab provides a second approved amyloid beta-directed product in the same early clinical stages, but with its own dosing, stopping, imaging, and safety framework. 11789
These approvals change how the negative amyloid record should be interpreted. The existence of successful lecanemab and donanemab programs means that failed amyloid trials do not support a class-wide conclusion that amyloid lowering is clinically or regulatorily futile. Taken together, the record instead supports a product-specific analysis of target species, epitope, central exposure, magnitude and timing of plaque reduction, enrolled disease stage, endpoint architecture, and management of amyloid-related imaging abnormalities. 117912131415
Aducanumab should be treated as historical regulatory context rather than part of the current approved treatment landscape. FDA's accelerated approval status report identifies Aduhelm as having received accelerated approval in 2021, without conversion to traditional approval, and records withdrawal of the application effective November 1, 2024. 16
The Failure Archive Is Mechanistically Diverse
The Alzheimer's failure record spans secretase inhibition, antibodies directed at different amyloid species, plaque-removing antibodies, and antibodies directed at extracellular tau. These programs failed in different populations and for different reasons, so they should not be compressed into a single conclusion about whether amyloid or tau is valid. The more defensible interpretation is that each program tested a specific product, exposure profile, target species, disease stage, endpoint architecture, and treatment duration. 171218131920211415222324
| Program | Mechanistic approach | Population tested | Principal negative finding | Regulatory lesson |
|---|---|---|---|---|
| Semagacestat | Gamma-secretase inhibition | Mild-to-moderate Alzheimer's disease | No cognitive improvement; the higher dose worsened function and adverse events were more frequent | Target biology cannot be separated from mechanism-wide safety and normal physiological function |
| Bapineuzumab | Anti-amyloid monoclonal antibody | Mild-to-moderate Alzheimer's disease, analyzed by APOE epsilon 4 status | No improvement in clinical outcomes despite biomarker differences in carriers | Biomarker movement and genotype stratification do not substitute for replicated clinical benefit |
| Solanezumab | Antibody targeting soluble monomeric amyloid beta | Mild dementia and later preclinical amyloid-positive disease | No significant slowing in mild dementia; no slowing of cognitive and functional decline in the preclinical A4 trial | Moving earlier does not rescue a mechanism without adequate biological effect |
| Verubecestat | BACE1 inhibition | Mild-to-moderate and prodromal Alzheimer's disease | No slowing of cognitive or functional decline; adverse effects complicated the benefit-risk profile | Large upstream pathway suppression can fail clinically and create mechanism-related liabilities |
| Lanabecestat | BACE1 inhibition | Early Alzheimer's disease and mild dementia | Two large trials were stopped for futility and showed no slowing despite target engagement | Pharmacodynamic activity must be linked prospectively to downstream and clinical effects |
| Gantenerumab | Antibody targeting aggregated amyloid beta | Early symptomatic Alzheimer's disease | Amyloid burden was reduced, but clinical decline was not slowed | The magnitude, timing, and clinical translation of plaque removal are distinct questions |
| Crenezumab | Anti-amyloid monoclonal antibody | Cognitively unimpaired PSEN1 mutation carriers | Five to eight years of therapy did not significantly improve primary clinical outcomes and did not significantly remove plaques | Early intervention is not enough when the product does not generate the required biological perturbation |
| Gosuranemab, semorinemab, and tilavonemab | Extracellular tau-directed antibodies | Prodromal or early Alzheimer's disease | Target engagement or acceptable tolerability was observed, but clinical progression was not slowed | Epitope selection, compartment, tau species, and downstream biomarker response must be validated before pivotal escalation |
The regulatory value of the archive is therefore taxonomic. A program that failed because target engagement was insufficient carries a different implication from one that achieved large biomarker effects without clinical benefit, and both differ from a program in which safety prevented adequate testing of the mechanism. 2514152217
Biological Diagnosis and Disease Stage
FDA's 2024 revised draft guidance recommends that eligibility for early Alzheimer's trials be grounded in the biological presence of disease rather than syndromic definitions alone. It also separates early disease into conceptual stages based on pathophysiology, subtle cognitive abnormality, and functional impairment, and asks sponsors to identify both the stage at enrollment and the stage expected at the primary outcome assessment. 25
That framework addresses a historical weakness, but biological confirmation and earlier enrollment are not sufficient by themselves. Solanezumab failed to slow decline in amyloid-positive, cognitively unimpaired participants in A4. Crenezumab failed in cognitively unimpaired carriers of a highly penetrant PSEN1 mutation after five to eight years of treatment. Gantenerumab failed to slow clinical decline in early symptomatic disease despite reducing amyloid burden. 131514
The approved lecanemab and donanemab labels show that early-stage selection can support a successful program when it is paired with confirmed amyloid pathology, product-specific biological activity, clinically interpretable outcomes, and an operational safety framework. The lesson is therefore not simply to move earlier, but to establish why the selected mechanism remains modifiable and clinically relevant at the selected stage. 79131514
A sponsor should be able to explain why the targeted species is causal or modifiable at the selected stage, how much change is required, how quickly downstream biology should respond, and which clinical trajectory remains alterable. A generic claim that treatment is being moved earlier is not a substitute for this quantitative disease model. 251315
FDA also notes that copathology is common and may support exclusion criteria or preplanned subgroup analyses. For development teams, this elevates tau burden, vascular disease, neurodegeneration, and other relevant pathology from exploratory characterization to potential sources of treatment-effect dilution, safety heterogeneity, or indication boundaries. 25
Target Engagement and Biomarker Interpretation
Several failed programs demonstrated that a proximal pharmacodynamic effect can be real while the therapeutic hypothesis remains unproven. Gantenerumab reduced amyloid plaque burden without slowing clinical decline. Lanabecestat produced substantial dose-related reductions in cerebrospinal fluid amyloid beta but did not slow cognitive or functional decline. Gosuranemab reduced unbound N-terminal tau in cerebrospinal fluid without favorable effects on cognitive or functional outcomes. 142122
The approved anti-amyloid products do not weaken this lesson. They make it more precise. A biomarker effect becomes regulatorily consequential when its magnitude, timing, product-specific mechanism, disease-stage context, and relationship to clinical benefit form a coherent evidentiary package. FDA's conversion of lecanemab to traditional approval followed verification of clinical benefit rather than reliance on amyloid reduction alone. 11714
This distinction is directly relevant to accelerated approval. FDA states that a surrogate endpoint considered reasonably likely to predict clinical benefit may support accelerated approval, but its acceptability depends on the disease stage, population, mechanism, and available treatments. The guidance explicitly cautions that a surrogate accepted for one product or trial population should not be assumed appropriate for another. 25
A credible biomarker strategy therefore needs at least three layers: evidence that the drug reaches and alters the intended target, evidence that the magnitude of change is biologically sufficient, and evidence that downstream disease processes or clinical trajectory move in the predicted direction. Failure at the second or third layer should prevent a program from treating target engagement as validation of the full mechanism. 25141522
Dose Selection and Benefit-Risk
Semagacestat is the clearest warning against interpreting pathway suppression as a clean therapeutic intervention. The phase 3 trial did not show cognitive benefit, the higher dose significantly worsened functional ability, and treatment was associated with more adverse events, including skin cancers and infections. 17
The BACE inhibitor experience reinforces the same point from a different mechanism. Verubecestat did not slow cognitive or functional decline in mild-to-moderate disease and was associated with treatment-related adverse events; moving the program into prodromal disease did not establish benefit. Lanabecestat also failed in early and mild disease despite evidence of target engagement. 192021
The lecanemab and donanemab labels also demonstrate that successful dose strategy includes more than selecting a biologically active regimen. Both require amyloid confirmation and product-specific provisions for imaging, dosing, and management of amyloid-related imaging abnormalities. Leqembi's subsequent maintenance options and Kisunla's updated titration further show that dosing strategy can continue to evolve after initial approval. 789
For regulatory planning, the relevant dose is not merely the maximum tolerated dose or a dose that changes a proximal biomarker. It is the exposure that produces a prespecified, biologically sufficient effect in the relevant compartment while preserving an acceptable safety margin for a chronic, often presymptomatic population. When the safety ceiling sits below the exposure needed to test the mechanism, the program may be scientifically interesting but not regulatorily viable. 17192025
This argues for earlier exposure-response modeling across pharmacokinetics, target engagement, downstream biomarkers, imaging, and safety. Pivotal dose selection should be framed as a quantitative test of the causal chain, not as a conventional choice among tolerated regimens. 142122
Earlier Intervention and Stage Selection
The field progressively moved from mild-to-moderate dementia into mild disease, prodromal disease, biomarker-positive preclinical disease, and genetically determined presymptomatic disease. Yet solanezumab, verubecestat, gantenerumab, and crenezumab all produced negative results in one or more earlier populations. 1813201415
These failures do not show that early intervention is unimportant. The approved indications for lecanemab and donanemab establish early symptomatic disease as a viable treatment context, but only within a biomarker-confirmed and product-specific development framework. The failures show that timing cannot compensate for an ineffective target species, inadequate brain exposure, insufficient magnitude of pathology reduction, an endpoint that does not capture the expected effect, or a mechanism that does not alter the dominant driver of progression at that stage. 79141513
A sponsor moving into Stage 1 or Stage 2 should therefore predefine what success looks like before clinical decline is readily measurable. FDA notes that these trials may require longer duration, sensitive cognitive measures, time-to-event approaches, or biomarker-based strategies, with detailed discussion of whether the evidence could support accelerated or traditional approval. 25
The practical implication is that prevention programs need stronger, not weaker, mechanistic validation. Long duration and slow progression increase the cost of being wrong about the target, dose, biomarker, or expected treatment window. 251315
Endpoint Strategy and Intended Claim
FDA continues to distinguish endpoint expectations by disease stage. For stages with detectable cognitive and functional impairment, the cognitive and functional or global co-primary approach remains generally acceptable. In the earliest stages, FDA may consider cognition alone with strong justification, a time-to-event endpoint tied to meaningful progression, or a surrogate endpoint under the appropriate approval framework. 25
The approved landscape demonstrates that endpoint architecture also depends on the intended therapeutic claim. Symptomatic dementia products, a treatment for agitation associated with Alzheimer's dementia, and amyloid-directed products for early-stage disease address different clinical constructs and therefore rely on different outcome frameworks. An endpoint suitable for cognition in established dementia cannot automatically support a behavioral indication, disease-modification claim, or prevention strategy. 12510119
| Development context | Primary evidentiary problem | Endpoint implication | Regulatory question to resolve early |
|---|---|---|---|
| Symptomatic dementia treatment | Demonstrating improvement or reduced worsening in cognition, function, or global status at the indicated severity stage | Outcomes must measure the symptomatic construct addressed by the proposed claim | Is the effect clinically interpretable within the proposed dementia stage and treatment duration? |
| Behavior-specific treatment | Demonstrating benefit on a defined neuropsychiatric manifestation such as agitation | The endpoint must directly measure the behavior and its severity rather than general dementia progression | Is the indication sufficiently specific, and is the treatment effect distinguishable from sedation or nonspecific behavioral suppression? |
| Stage 3 or later symptomatic disease | Demonstrating clinically meaningful slowing when cognition and function are both impaired | Cognitive plus functional or global assessment remains a generally acceptable model | Does the endpoint support the intended claim and remain interpretable across the enrolled severity range? |
| Stage 2 disease | Subtle cognitive abnormality without functional impairment | Sensitive cognition, sufficiently long follow-up, or time to meaningful progression may be needed | Can cognition alone be clinically persuasive, and what corroborating biological evidence is required? |
| Stage 1 disease | No measurable clinical impairment at baseline and variable latency to symptoms | Biomarker patterns, stage transition, or long-duration clinical outcomes may be considered | Is the proposed surrogate reasonably likely to predict benefit for this mechanism and population? |
| Accelerated approval strategy | Surrogate effect may precede verified clinical benefit | Clinical outcome assessments should still be embedded and confirmatory work must verify benefit | Is the surrogate product-specific, quantitatively persuasive, and linked to a feasible confirmatory plan? |
Failed programs show why endpoint selection cannot be treated as a statistical afterthought. EXPEDITION3 did not confirm a significant cognitive benefit in mild dementia. A4 used a long preclinical design but did not show slowing of cognitive and functional decline. The crenezumab prevention trial used two clinical primary outcomes over five to eight years and still did not show significant benefit. 181315
The lesson is not simply to use longer trials or more sensitive scales. The endpoint must sit at the point in the causal chain where the proposed effect should become observable during the study, and the expected placebo trajectory must be sufficiently reliable to support the planned effect size. For Stage 1 and Stage 2 programs, progression-event definitions, repeated-measures properties, practice effects, missingness, and transitions between stages become part of the core regulatory argument. 251315
Subgroups, Replication, and Claim Control
The solanezumab program illustrates the danger of promoting an exploratory subgroup into a pivotal thesis without sufficient independent validation. Earlier phase 3 studies did not meet their primary endpoints, although analyses in participants with mild disease generated a hypothesis that was tested prospectively in EXPEDITION3. That confirmatory trial did not show a significant effect on cognitive decline. 2618
Bapineuzumab similarly incorporated APOE epsilon 4 stratification into parallel phase 3 trials, but clinical outcomes were not improved in either genetic stratum, even though biomarker differences were observed in carriers. 12
Genetic and biomarker stratification remain important in approved anti-amyloid treatment, particularly for diagnosis and safety planning, but that does not convert every subgroup analysis into evidence of differential efficacy. For regulatory teams, subgroup evidence should be governed as a claim-risk issue. The subgroup must have a biologically coherent interaction hypothesis, reliable classification, adequate power, controlled multiplicity, and prospective replication. 79261812
Without those elements, a subgroup may support mechanistic learning or a subsequent trial, but it should not carry the evidentiary weight of a positive primary analysis. 261812
Diagnostic and Operational Feasibility
Biological enrichment improves disease specificity but creates major operational consequences. In AMARANTH, 6,871 people were screened and 2,218 were randomized; in DAYBREAK-ALZ, 5,706 were screened and 1,722 were randomized. Both trials confirmed amyloid pathology before enrollment and were ultimately stopped after futility analyses. 21
FDA states that early Alzheimer's trials are expected to use biomarker evidence to establish diagnosis. When a needed FDA-approved or cleared diagnostic is unavailable, sponsors should engage the relevant review division early regarding possible companion diagnostic codevelopment. 25
The approved lecanemab and donanemab labels convert this diagnostic principle into treatment infrastructure. Both require confirmation of amyloid beta pathology before treatment, while their safety frameworks require access to magnetic resonance imaging and the ability to identify and manage amyloid-related imaging abnormalities. 79
This means the clinical protocol, diagnostic strategy, assay validation, site capacity, screening algorithm, imaging or fluid-biomarker logistics, and eventual indicated population cannot be developed independently. A highly selective trial may strengthen internal validity while creating recruitment, generalizability, labeling, or implementation constraints that must be addressed before phase 3. 252179
Operational performance also affects evidentiary quality in long studies. Prevention trials require durable retention, stable assessment procedures, control of missing data, and plans for participants who transition between stages. The crenezumab trial achieved 94 percent completion over a five-to-eight-year common-close design, but the primary clinical outcomes remained negative, showing that strong execution cannot rescue an inadequate biological effect. 15
A 2026 Evidence Architecture
The accumulated failures and the approved product landscape support a gated model in which each development transition requires evidence that the preceding link in the causal chain is intact. The purpose is not to eliminate uncertainty. It is to prevent phase 3 from becoming the first credible test of target relevance, biological sufficiency, dose adequacy, endpoint sensitivity, diagnostic feasibility, or operational deliverability. 251191721141522
The approved therapies also show that the evidence gates differ by intended claim. A symptomatic cognitive treatment, a behavior-specific therapy, an early-stage disease-modifying therapy, and a prevention therapy do not require identical biomarkers, endpoints, duration, diagnostics, or benefit-risk arguments. Regulatory strategy should begin with the proposed claim and work backward to the evidence architecture required to support it. 15107925
| Evidence gate | Minimum question before escalation | Typical evidence package | Failure pattern it is designed to prevent |
|---|---|---|---|
| Intended claim | Is the program seeking symptomatic, behavioral, disease-modifying, or prevention labeling? | Target product profile, stage-specific indication, endpoint and duration rationale | Applying an evidentiary model from one treatment category to a different proposed claim |
| Biological population | Do enrolled participants have the target disease biology at the intended stage? | Validated diagnostic biomarkers, stage definition, copathology plan | Dilution by non-Alzheimer syndromes or biologically heterogeneous populations |
| Relevant target engagement | Does the drug reach and alter the intended target species in the relevant compartment? | Pharmacokinetics, cerebrospinal fluid or imaging pharmacodynamics, dose-response | Advancing on peripheral or mechanistically ambiguous activity |
| Biological sufficiency | Is the magnitude and duration of target change large enough to alter downstream disease? | Quantitative threshold, downstream biomarker movement, temporal concordance | Treating any statistically significant biomarker change as mechanism validation |
| Clinical translation | Does the predicted clinical trajectory move in the selected stage? | Stage-matched cognitive, functional, global, behavioral, or progression-event outcomes | Discovering in phase 3 that the endpoint is insensitive or disconnected from the proposed claim |
| Benefit-risk at required exposure | Can the biologically sufficient exposure be maintained chronically? | Exposure-safety modeling, mechanism-specific monitoring, discontinuation rules | Selecting a tolerated dose that cannot test the hypothesis, or an active dose that is not acceptable |
| Diagnostic and delivery feasibility | Can eligible patients be identified and treatment safely implemented? | Assay strategy, imaging capacity, site readiness, monitoring and treatment workflow | Securing an approval that cannot be operationalized in the intended population |
| Regulatory pathway | Does the evidence support traditional approval, accelerated approval, or further learning? | Surrogate justification, clinical outcome data, confirmatory plan, diagnostic plan | Assuming precedent for another product automatically applies |
| Replication and claim control | Are subgroup and secondary findings reproducible and prospectively controlled? | Prespecified hypotheses, multiplicity control, independent confirmation | Converting exploratory signal into an unsupported pivotal claim |
For non-amyloid mechanisms, the standard should not be weaker merely because the target is novel. The tau antibody trials show that plausible pathology, target engagement, and tolerability can coexist with no measurable effect on clinical progression. Novel mechanisms need mechanism-specific biomarkers and a defined downstream signature before large efficacy studies. 222324
For programs considering accelerated approval, the confirmatory strategy should be designed while the pivotal surrogate trial is being designed. FDA recommends including clinical outcome assessments even when a surrogate is the primary measure, and postapproval trials are required to verify and describe clinical benefit under accelerated approval. The lecanemab conversion and the subsequent withdrawal of the Aduhelm application illustrate two different possible outcomes following accelerated approval. 251116
Conclusion
The approved treatment landscape prevents the Alzheimer's development record from being reduced to a story of universal failure. Symptomatic cognitive therapies, a behavior-specific treatment, and amyloid-directed products for biomarker-confirmed early disease demonstrate that distinct regulatory strategies can succeed.
The failures nevertheless narrow the space for weak development logic. It is no longer sufficient to identify a plausible target, show a proximal biomarker change, move into an earlier population, and rely on a large pivotal trial to resolve the remaining uncertainty.
The stronger regulatory strategy is to make every link explicit: the intended claim, biological diagnosis, stage-specific causal rationale, sufficient central exposure, quantitative target modification, downstream biological response, clinically interpretable outcome, acceptable chronic safety, operational feasibility, and a pathway-specific confirmatory plan.
The failure archive should therefore be treated as a design asset. Its value lies not in proving that an entire target class is futile, but in showing exactly which assumptions separated successful products from programs that could not establish a favorable benefit-risk profile or a persuasive clinical effect.