Neurodegeneration Reversal

Established Every disease-modifying therapy approved for a neurodegenerative disease slows decline; none restores what has been lost, and the distinction is the whole subject. Two anti-amyloid antibodies now hold full approval in the United States on the strength of differences of roughly half a point to seven-tenths of a point on an eighteen-point clinical scale over eighteen months. Established A gene therapy for Huntington’s disease reported a large slowing against an external control in 2025. Cell-replacement grafts in Parkinson’s disease have survived and made dopamine in first-in-human trials. Established No published result in any of these diseases has returned a patient to a previous level of function and held them there.

Established The field’s central dispute is not whether the antibodies work but whether what they do is worth what it costs. The measured differences fall below most published estimates of the minimal clinically important difference on the same scale, while the treatments cause amyloid-related imaging abnormalities in a substantial minority of recipients. Established One national system authorised a product and then declined to fund it, which is the cleanest available statement that a real effect can be too small to buy.

Established Diagnosis has run decisively ahead of therapy, and that gap is now the field’s defining structural problem. A blood test measuring phosphorylated tau at residue 217 distinguishes Alzheimer’s pathology at accuracies around ninety per cent, against roughly sixty-one per cent for primary-care physicians and seventy-three per cent for specialists working without it, and a first blood-based diagnostic was cleared by a regulator in 2025. Frontier Revised diagnostic criteria define the disease biologically, which means that people with no symptoms can now be told they have it. What can then be offered them is a slowing therapy with a monitoring burden, or a prevention trial.

Established This is a synthesis and it does not repeat what its neighbours own. Neurogenetics holds the genetic-architecture record and the finding that every neurogenetic therapy reaching a patient has been monogenic; Longevity Therapies holds the ageing-intervention evidence these diseases sit downstream of; Brain Preservation holds the question of whether structure alone contains a person. Established The joint question none of them owns is the one here: what would it take to reverse rather than retard, and why has nothing done it?

Established A note on sourcing. This brief was commissioned in September 2026 from the Institute’s research base. Reading-list entries without links are cited from the bibliographic record rather than re-fetched, and claims are dated no later than early 2026 unless carried by a linked source.

Established The anti-amyloid numbers are small, real and precisely known, which is why the argument about them is so sharp. In an eighteen-month trial of 1,795 participants, lecanemab produced a difference of 0.45 points on the eighteen-point Clinical Dementia Rating Sum of Boxes against placebo, a relative slowing of about 27 per cent. Established In a seventy-six-week trial of 1,736 participants, donanemab produced a difference of 0.7 points on the same scale and about 2.9 points on a combined integrated scale. Established Published estimates of the minimal clinically important difference on that scale in mild cognitive impairment cluster around one to one-and-a-half points, above both results. The trials are positive and the effects are below the thresholds the field itself proposed.

Established The harms are also precisely known and are not rare. Amyloid-related imaging abnormalities with oedema occurred in about 12.6 per cent of lecanemab recipients against 1.7 per cent on placebo, and in roughly 24 per cent of donanemab recipients; haemorrhagic forms occurred in 17.3 and 31.4 per cent respectively. Established Risk concentrates in carriers of two copies of the APOE4 allele, and deaths have been reported in open-label extensions, several in people also receiving anticoagulation. The result is a therapy that requires genotyping, a scheduled MRI programme and infusion capacity.

Established Regulators reading the same dossiers reached different answers, which is the most informative single fact in the file. A US regulator granted traditional approval to lecanemab in 2023 and donanemab in July 2024; a European committee issued a negative opinion on lecanemab in July 2024, reversed it on re-examination later that year with use restricted by APOE4 genotype, and issued a negative opinion on donanemab in 2025. Established A UK regulator authorised lecanemab while the national cost-effectiveness body declined to fund it. Same data, three outcomes: the disagreement is about the value of the effect, not its existence.

Frontier A measured anomaly sits inside the positive trials and has not been resolved. Meta-analysis of amyloid-lowering trials finds accelerated loss of whole-brain volume and enlargement of the ventricles in treated arms, in the same trials that report slower clinical decline. Frontier Three explanations are in circulation — removal of amyloid mass and associated inflammatory tissue, treatment-related fluid shifts and ARIA, and genuine accelerated neurodegeneration — and no published analysis distinguishes them. Frontier A field that reports volumetric MRI as an outcome cannot indefinitely decline to explain it moving the wrong way.

Established Tau tracks symptoms better than amyloid does, and the therapies aimed at it have mostly failed. Tau pathology follows a stereotyped anatomical progression that correlates with cognitive stage far more closely than amyloid burden does, and four monoclonal antibodies against extracellular tau nonetheless failed to meet primary endpoints in Alzheimer’s disease or progressive supranuclear palsy between 2019 and 2023. Frontier The live approach is an antisense oligonucleotide that reduces production of the protein itself, which lowered cerebrospinal-fluid tau in early-phase work and is in a phase 2 trial whose readout is the most consequential pending result in the dementia field.

Established Parkinson’s disease has produced the only credible restoration attempts, and they replace cells rather than repair circuits. Two first-in-human cell-replacement trials reported in 2025 — one using embryonic-stem-cell-derived and one using induced-pluripotent-stem-cell-derived dopaminergic progenitors — showed graft survival, imaging evidence of dopamine production and exploratory motor improvement in small cohorts, with acceptable safety. Frontier It works in principle because the target is a diffusely projecting neuromodulatory population: the graft supplies a chemical, not a wiring diagram. Established It does not stop the underlying disease, and a phase 3 programme began in 2025.

Frontier Alpha-synuclein immunotherapy in Parkinson’s has produced missed primaries with persistent subgroup signals. One antibody failed outright. Prasinezumab missed its phase 2 primary endpoint; a subsequent analysis reported slower motor progression in a rapidly progressing subgroup, and a larger phase 2b then missed its own primary endpoint with a nominal trend in the same direction. Frontier This is the shape of an effect that is either small and real or absent and repeatedly rediscovered in post-hoc strata, and the design that would separate them — prospective enrichment for the fast-progressing phenotype — has not reported.

Established Huntington’s disease supplies the largest reported slowing and the weakest control condition. An AAV-delivered gene therapy reported in September 2025 a roughly 75 per cent slowing of progression on a composite clinical scale at thirty-six months, with neurofilament light below baseline; the comparator was a propensity-matched external control cohort rather than a concurrent randomised placebo arm. Established The same disease previously produced a sharp negative: an antisense therapy halted in 2021 when the higher-dose arm did worse than placebo, restarted at lower doses in a new trial.

Established Amyotrophic lateral sclerosis is where surrogate-endpoint approval has already been tested in both directions. Tofersen missed its primary clinical endpoint and was granted accelerated approval in 2023 on a reduction in neurofilament light; a different product approved in 2022 on a small phase 2 failed its confirmatory phase 3 in 2024 and was withdrawn by its sponsor. Frontier Both are evidence about the regulatory system rather than the biology.

Frontier Does removing amyloid before symptoms prevent the disease? This is the amyloid hypothesis’s remaining testable form, and the trials are enrolled and running: secondary-prevention studies in biomarker-positive people without cognitive symptoms, and a long-running study in families carrying autosomal-dominant mutations. Frontier Open-label extension data reported in 2025 from the familial study suggested that the longest-treated participants had a reduced risk of symptom onset; that is an uncontrolled comparison and is treated here as a hypothesis, not a result.

Frontier Can lowering tau production change the clinical course? Reducing the substrate rather than clearing the product is mechanistically distinct from everything that has failed, and the cerebrospinal-fluid effect is large. Frontier Whether a biochemical reduction translates into a cognitive difference is exactly the step at which amyloid-clearance therapies delivered less than expected.

Frontier Can cortical circuits be restored at all, or only replaced? Dopaminergic grafting works because the missing signal is diffuse. A cortical or hippocampal circuit encodes information in its specific connectivity, and no published work has regrown such a circuit in an adult mammal with its content intact. Speculative Mouse work has shown that memories apparently lost in a disease model can be retrieved by directly reactivating the cells that encoded them, which argues that some deficits are retrieval failures rather than erasures. Whether that generalises to human neurodegeneration is untested.

Frontier Does the adult human brain make new neurons, and does it matter here? Studies using different tissue-handling methods have reached opposite conclusions about adult hippocampal neurogenesis, and the practical stake is modest either way: even generous estimates are orders of magnitude below the neuronal losses in established dementia.

Established The endpoint is the first bottleneck, and it measures decline rather than recovery. Every approved disease-modifying agent is judged on a difference in rate of worsening over a fixed window, a design that cannot detect improvement because improvement is not what the scale is scored for. Frontier A restoration therapy tested on a slowing endpoint would be measured by the wrong instrument, and no regulator has accepted a recovery endpoint in these indications.

Established Delivery across the blood-brain barrier is the second and it shapes which modalities exist. Antibodies reach the brain at roughly a thousandth of their plasma concentration, which sets the infusion schedules. Established Oligonucleotides are delivered intrathecally and gene therapies by direct intraparenchymal infusion under stereotactic guidance, procedures that bound how many patients can be treated independently of how well the agent works.

Established Monitoring capacity is the third, and it is the reason approval has not meant access. An antibody with a double-digit rate of imaging abnormalities requires baseline and scheduled MRI, APOE genotyping, and a service able to act on incidental findings. Frontier Health systems have found this, rather than drug cost alone, to be the binding constraint on population-scale delivery.

Frontier Trial duration against disease duration is the fourth. These diseases run for one to two decades and trials run for eighteen months to two years, so every approved effect is an extrapolation from a short window on a long curve. Frontier Whether an eighteen-month slope difference persists, widens or converges over ten years is unknown for every agent in this brief, and open-label extensions cannot answer it because the control arm is gone.

Established Biomarker standardisation is the fifth and the most tractable. Blood assays for phosphorylated tau differ between platforms in cut-points and in performance in renal impairment or high body-mass index, so a result is not portable between health systems — which matters more as the tests move into primary care.

Established The field depends on biomarker-defined populations, which it now has. Every modern trial enrols on confirmed pathology rather than clinical syndrome, removing the misclassification that plagued earlier trial generations. Established It also means historical negative trials cannot be compared straightforwardly with current ones, because they were not testing the same patients.

Frontier It depends on gene-delivery platforms it shares with other fields. Neurogenetics records that the binding engineering constraint on neurogenetic therapy is a membrane, and that every therapy reaching a patient so far has been monogenic. Frontier Huntington’s disease and SOD1-associated ALS are the neurodegenerative diseases that fit that description, which is precisely why they have the strongest interventions.

Frontier It depends on stem-cell manufacturing at clinical grade. Dopaminergic cell products require differentiated, characterised, cryopreserved lots with reproducible potency, which is the same manufacturing problem faced by every cell therapy; Cellular Rejuvenation covers the reprogramming chemistry underneath it. Established Safety data from other central-nervous-system cell-therapy trials, including a long-follow-up phase 1 in spinal cord injury, is what makes ethics committees willing to approve these programmes at all.

Established It does not depend on symptomatic neuromodulation, which is a separate and more mature technology. Bioelectric Medicine records the adaptive deep-brain-stimulation results in Parkinson’s disease, including a blinded randomised feasibility trial and a cleared commercial system. Established Those devices manage symptoms well and modify no disease process; conflating the two is the commonest error in coverage of Parkinson’s therapeutics.

Frontier The decisive experiment is the secondary-prevention trial: removing amyloid from biomarker-positive people who have no symptoms, with clinical onset as the endpoint. It is already running in two large industry programmes and one long-standing familial-mutation study, and it is decisive in both directions. Frontier If pre-symptomatic clearance substantially delays onset, the modest effects in symptomatic disease are explained by timing and the field’s strategy is vindicated. Frontier If it does not, three decades of amyloid-directed development will have been tested at the stage most favourable to it and found wanting, and the field will have to relocate to tau, to inflammation, or to something not yet named.

Frontier The second experiment is the tau-lowering phase 2 readout. It is the only pending trial that tests a mechanistically distinct hypothesis in a population where the target correlates with symptoms. Frontier A positive result would make tau the field’s centre of gravity within a year; a negative one would leave prevention as the only live amyloid-era strategy.

Established The third is a randomised, concurrently controlled trial of the Huntington’s gene therapy. The 2025 result used a propensity-matched external control, and external controls in slowly progressive diseases have historically overstated effects because the matched cohort differs in unmeasured ways from people willing to undergo neurosurgery. Frontier The effect size reported is large enough that a concurrent control would be informative even at modest sample sizes, and it is the cleanest available test of whether a single-gene neurodegenerative disease can be substantially modified.

Frontier The fourth is the cell-replacement phase 3 in Parkinson’s disease, which is the only restoration-shaped trial in the field. Its endpoint is still a rating scale, but its mechanism is replacement rather than protection, so a clear positive would be the first evidence that lost function can be reinstated rather than preserved. Frontier A sham-surgical control is the design feature to watch, because the previous generation of intraputamenal trials produced strong imaging effects and no clinical separation.

Established Antibody manufacture is not the constraint; delivery infrastructure is. Monoclonal production at the required scale is routine biologics engineering; what is not routine is a health system that can genotype, image at baseline, infuse on schedule, re-image on protocol and triage imaging abnormalities, for a population counted in millions.

Frontier Intraparenchymal delivery is a neurosurgical engineering problem with real per-patient costs. Gene therapy for Huntington’s disease is infused stereotactically into deep structures over hours under MRI guidance, so throughput is set by operating-theatre time and trained surgical teams, which caps treatable population independently of vector supply.

Frontier Cell therapy adds a potency and identity problem that antibodies do not have. A dopaminergic product must be differentiated to a defined stage, free of residual pluripotent cells, viable after cryopreservation and delivered to a defined target volume. Established Graft survival has now been shown by imaging in humans, converting this from a biology question into a manufacturing and surgical one.

Established Blood-based diagnostics are the one component ready for scale. A plasma assay runs on existing immunoassay platforms in ordinary hospital laboratories at a cost far below amyloid PET or lumbar puncture, and a regulator cleared a first such test in 2025. Frontier The uncomfortable consequence: identification is the cheap part, and having anything to offer afterwards is the expensive one.

Established The genetics belong next door and they explain the pattern of successes. Neurogenetics records that every neurogenetic therapy to reach a patient is monogenic and that almost none of them edits anything. Established The neurodegenerative diseases with the strongest interventions — Huntington’s, SOD1-associated ALS — are the monogenic ones, and sporadic Alzheimer’s and Parkinson’s disease are not.

Frontier The ageing literature is upstream and is thinner than this field assumes. Longevity Therapies records that the mouse evidence for the leading geroprotective compounds is real, sex-split and smaller than advertised, and that the flagship human trial has been unfunded for a decade. Frontier Age is the dominant risk factor for every disease in this brief, so an intervention on ageing would be the highest-leverage prevention available — and the evidence base for one is weaker than the evidence base for the antibodies.

Speculative Preservation is the adjacency that states the limit case. Brain Preservation records that whole-brain structural preservation has been demonstrated and independently verified, that nobody has read one out, and that whether structure alone contains a person has never been tested in any organism. Speculative That is the same question restoration faces from the other side: if the connectivity encoding a memory is destroyed, no delivery improvement recovers it, and the problem stops being pharmacological.

Established The general pattern comes from outside neurology. Precision Medicine records a field that predicts harm well and benefit poorly. Established Neurodegeneration fits: APOE4 genotype predicts who will suffer imaging abnormalities with useful accuracy, and no marker predicts who will benefit from an anti-amyloid antibody.

Established Approval and access have decoupled, and the gap is now the field’s main institutional fact. A product can hold full marketing authorisation in one jurisdiction, a restricted authorisation in a second and a refusal in a third, while in a fourth it is authorised and not funded. Established All four positions were reached on essentially the same trial dossiers between 2023 and 2025, which means the variance is in how bodies value a sub-threshold effect against a double-digit adverse-event rate.

Frontier Surrogate endpoints are the live regulatory question and there is no settled standard. An ALS therapy was approved on a neurofilament reduction after missing its clinical endpoint; an amyloid-clearance therapy was approved in 2021 on a biomarker over the objection of an advisory committee and later discontinued by its sponsor. Frontier Until a body states what magnitude of biomarker change licenses an approval in these diseases, each decision is adjudicated separately and the field cannot plan development programmes against a known bar.

Established Diagnostic criteria have been revised ahead of the therapeutics, and the revision is contested. One set of criteria published in 2024 defines the disease by biomarker status alone, including in people without symptoms; a rival group argued in the same year that biomarker positivity without symptoms is a risk state rather than a disease. Frontier The disagreement has direct consequences: whether an asymptomatic biomarker-positive person is a patient determines insurance status, disclosure practice, driving and employment questions, and trial eligibility.

Frontier Capacity planning is where health systems will actually decide this. Delivering an infusion therapy with scheduled MRI to the eligible early-stage population of a mid-sized country implies imaging and infusion volumes that existing services cannot absorb without new capital. Frontier The decisions that follow — eligibility narrowing, triage by genotype, waiting lists — will shape measured population outcomes more than any further trial in the next five years.

Frontier Telling asymptomatic people they have a disease is a new practice with thin evidence behind it. Cheap blood testing plus biologically defined criteria makes pre-symptomatic diagnosis routine for the first time. Established Disclosure studies in genetic-risk settings found that most people cope well, but in self-selected, counselled cohorts; generalising to opportunistic primary-care testing is an assumption, not a finding.

Frontier Consent for an intervention with a double-digit imaging-abnormality rate and a sub-threshold benefit is genuinely difficult to obtain well. The honest disclosure is that the treatment is expected to slow decline by an amount smaller than most patients or families would notice directly, while carrying a measurable risk of brain oedema or microhaemorrhage. Frontier Studies of what patients understood after consent are scarce, which is itself a finding.

Established External-control designs raise a fairness question as well as a statistical one. Comparing treated patients against a matched historical cohort avoids randomising anyone to placebo, which is attractive in a fatal disease. Established It also systematically flatters the treatment when the treated group is selected for fitness to undergo a procedure, and patients consenting to such trials are rarely told how much of the reported effect the design itself could produce.

Frontier Unproven clinic offerings track this field closely and harm people. Clinics selling unregulated cell infusions for neurological conditions have operated at scale for a decade, documented in census work counting hundreds of US businesses. Established Every legitimate first-in-human cell-therapy result in Parkinson’s disease is recruited into their marketing, and the distinguishing feature — a characterised, regulator-reviewed product delivered surgically to a specific target — is what that marketing elides.

Frontier Equity is unusually stark here because the therapy is infrastructure-bound. Access requires genotyping, serial MRI and infusion capacity, which exist in wealthy urban systems and not elsewhere, so eligibility set by proximity to an MRI scanner produces outcome differences that look biological and are geographic.

Established The burden is large enough that even sub-threshold effects aggregate into something significant. Dementia affects tens of millions of people worldwide, and the care burden falls disproportionately on unpaid family members. Frontier A treatment delaying institutionalisation by months across a population is worth more in aggregate care-hours than its individual effect size suggests — the strongest argument for the antibodies and rarely the one made for them.

Frontier Pre-symptomatic identification without effective treatment creates a new social category. If millions of people can cheaply learn they are biomarker-positive decades before symptoms, the consequences run through insurance underwriting, employment, licensing and family planning. Frontier No jurisdiction has legislated for biomarker status the way several have legislated for genetic test results, and the gap is being filled by practice rather than by law.

Speculative If restoration ever worked, its effects would be different in kind from slowing. Slowing changes the shape of a decline; restoration would change who is in the workforce, who requires care, and what the last decades of life look like. Speculative The only demonstrated route to it replaces a diffusely projecting cell population, and there is no equivalent route for cortex, so the near-term civilizational consequence of this field is better prevention rather than recovery.

Handwave Claims that neurodegeneration will be cured by clearing a protein belong here, flagged. Three decades of amyloid-directed development have produced a slowing below the clinical threshold the field proposed for itself, with a non-trivial adverse-event profile. Handwave The hypothesis may still be right about causation and wrong about timing, which is exactly what the prevention trials are testing. Asserting the cure before those read out is the point at which this subject stops being about evidence.

These horizons track what would have to be demonstrated for a claim to change status, not predictions of arrival.

• 10 yr: Frontier The secondary-prevention readouts land on this horizon and are the field’s pivot point; so does the tau-lowering phase 2 and the Parkinson’s cell-therapy phase 3. Frontier A randomised concurrently controlled trial of the Huntington’s gene therapy is affordable now and its absence in ten years would say more about the field than any statement it makes. Frontier Blood-biomarker standardisation across platforms is a solved-in-principle problem that will or will not have been done, and it gates everything in primary care.
• 25 yr: Speculative If prevention works, this is the horizon on which health systems discover whether they can deliver it: population screening, monitoring capacity and a decades-long treatment commitment in asymptomatic people. Speculative If prevention fails, the field relocates on this horizon, and the honest expectation is a return to mechanism-finding with the biomarker toolkit it has acquired.
• 50 yr: Speculative Restoration of cortical function, as opposed to replacement of a neuromodulatory population, would arrive here if it arrives at all, and would require a demonstration nobody has produced in any adult mammal: regrowth of a circuit with its information content intact. Handwave Proposals to achieve this by scaling current cell therapies assume the hard part is cell supply rather than connectivity, which is the assumption the evidence least supports.
• 100 / 250+ yr: Handwave Claims about the abolition of dementia, or about recovering a person from a degenerated brain, belong here. Handwave Both require knowing whether the information constituting a person survives the pathology, which as a neighbouring brief records has never been tested in any organism, including those whose complete wiring diagrams have been available for decades.

• Depends on Three edges on this map are real. It waits on the delivery work recorded in Neurogenetics, whose binding constraint — getting a construct across a membrane into the right cells — is the same one that limits every gene-directed therapy here. It waits on the manufacturing and reprogramming chemistry in Cellular Rejuvenation, which supplies the cell products the only restoration-shaped trials use. And it sits downstream of Longevity Therapies: age is the dominant risk factor for every disease here, and an ageing intervention would outperform every agent in this brief on prevention if its evidence base were stronger than theirs, which that brief records it is not.
• Requires (not on this map) First, a clinical endpoint that can register improvement, since every instrument in current use scores the rate of worsening and therefore cannot detect the thing the field says it wants. Second, regrowth of a cortical or hippocampal circuit with its content intact in an adult mammal, which is the specific scientific result that separates replacement from restoration and which nobody is producing. Third, a stated regulatory standard for what magnitude of biomarker change licenses approval, without which each decision is adjudicated alone and development cannot be planned. Fourth, a legal status for asymptomatic biomarker-positive people covering insurance, employment and licensing, which several jurisdictions have written for genetic results and none has written for this. Fifth, imaging and infusion capacity sized for the eligible early-stage population rather than for specialist memory clinics, which is capital expenditure rather than science. Sixth, harmonisation of plasma phosphorylated-tau assays across platforms, so a result is portable between health systems.
• Enables A working prevention result would enable population biomarker screening to become a clinical programme rather than a research tool, and would make the cheap blood test the entry point to a care pathway rather than to bad news. A working restoration result would enable something this field has never had: a therapy judged on recovery rather than on the slope of decline. Neither is enabled by anything currently approved.
• Adjacent Bioelectric Medicine holds the symptomatic neuromodulation record that is routinely confused with disease modification in Parkinson’s disease. Precision Medicine supplies the pattern this field reproduces exactly: good at predicting harm, poor at predicting benefit. Brain Preservation is the adjacency that states the limit case, since both fields turn on whether structure alone contains a person.

Established “There is now a treatment that stops Alzheimer’s disease.” No approved agent stops it; two slow it by amounts below the field’s own clinical threshold. Established Sponsor analyses converting these differences into months of delay produce figures in the range of four to seven months, and independent analysts dispute the conversion method.

Established “The amyloid hypothesis has been disproven.” It has been partially confirmed and found insufficient, which is a different and more interesting state. Removing amyloid does change the clinical course, which earlier failures had not established, and the change is small. Frontier The remaining live version of the hypothesis is about timing rather than mechanism, and it is being tested in prevention trials that are enrolled and running.

Frontier “Regulators are just being slow.” They read the same data and reached opposite conclusions within months of each other. Four jurisdictions reached four positions on the same dossiers: a legitimate disagreement about whether a sub-threshold benefit justifies a double-digit adverse-event rate.

Frontier “Stem cells can replace lost brain cells.” In one disease, for one cell type, with important qualifications. Grafted dopaminergic progenitors have survived and produced dopamine in first-in-human Parkinson’s trials, because the missing signal is diffusely projected and does not encode information in its wiring. Established Nothing comparable has been shown for cortical or hippocampal neurons, and the grafts do not halt the underlying disease in the patient’s own cells.

Established “A 75 per cent slowing has been achieved in Huntington’s disease.” That is the reported figure and the control was external. The comparison was against a propensity-matched historical cohort rather than a concurrent randomised placebo arm. Frontier External controls in slowly progressive disease have a documented tendency to overstate effects, and the appropriate response is not to dismiss the result but to run the randomised trial the effect size makes affordable.

Frontier “GLP-1 drugs prevent dementia.” Two large randomised trials tested that and their sponsor reported in late 2025 that both missed their primary endpoint. The hypothesis came from post-hoc analyses of trials run for other indications, which is the evidence class that most reliably generates enthusiasm and most reliably fails confirmation. Established The negative result is the most important dementia-trial news of that year and received a fraction of the coverage the hypothesis did.