Open questions — Alzheimer's disease¶
Last curated: 2026-08-31 (deepened)
Stable IDs (OQ-n) are retained when questions are refined. Tier 1 questions could change practice and are designable with existing tools. Tier 2 questions require enabling measurement, mechanistic or implementation work first. "Open" means the retrieved evidence leaves a decision unresolved, not that no study exists. This set replaces the six seed questions written when the condition was scoped; the seed questions are carried forward inside OQ-1, OQ-3, OQ-5, OQ-6, OQ-9 and OQ-16.
Tier 1 — practice-changing and designable now¶
OQ-1 — Is a group-mean CDR-SB difference of 0.45–0.67 points meaningful to patients and care partners?¶
Why open: Lecanemab slowed 18-month CDR-SB decline by 0.45 points (95% CI 0.23–0.67) and donanemab by 0.67 points (0.40–0.95) in low/medium-tau participants (van Dyck 2023, PMID 36449413; Sims 2023, PMID 37459141). Published minimal clinically important differences for CDR-SB are +1 in MCI and +2 in mild AD, and the review that assembled them concludes that anti-amyloid treatment effects are lower than those thresholds and that the thresholds themselves were derived without systematic patient and care-partner input (Muir 2024, PMID 38561021). Three regulators reached three different conclusions on the same data.
Design: A mixed-methods study deriving MCIDs de novo from people with early AD and their care partners, using anchor-based methods with patient-nominated anchors, then applying those thresholds to individual-participant data from CLARITY AD and TRAILBLAZER-ALZ 2 to report responder proportions and preference-weighted benefit rather than group means.
OQ-2 — Should anti-amyloid therapy be restricted to MCI and very mild dementia?¶
Why open: In 234 consecutively treated real-world patients, symptomatic ARIA occurred in 27% of those with mild dementia versus 1.8% of those with MCI or very mild dementia, while pooled ARIA incidence was 19% (95% CI 16–23) with an APOE4 gene-dose effect (RR 1.45 heterozygotes, 3.54 homozygotes) (Paczynski 2025, PMID 40354064; Qi 2026, PMID 41478817). Neither appropriate-use document stratifies eligibility by within-stage severity (Cummings 2023, PMID 37357276; Rabinovici 2025, PMID 40155270).
Design: Pooled individual-participant analysis of trial and registry cohorts modelling symptomatic ARIA and clinical benefit jointly against baseline CDR-SB, APOE genotype, microbleed count and CAA markers, producing a stratified net-benefit surface rather than a binary eligibility rule.
OQ-3 — What is the lifetime, competing-risk-adjusted probability that a biomarker-positive asymptomatic person develops dementia?¶
Why open: This is the number that would settle the AA-versus-IWG definitional dispute, and it does not exist. Available estimates are short-horizon: 5-year progression 11% at NIA-AA stage 1 and 56% at stage 3 (Vos 2013, PMID 24012374); 57% (95% CI 45–71) for A+T+ and 17% (13–22) for A+T− in cognitively unimpaired people (Moscoso 2025, PMID 40522652); pooled 20% (stage 1) to 73% (stage 3) (Parnetti 2019, PMID 30646955). The 2024 criteria diagnose on Core 1 alone; the IWG says most such people will not become symptomatic on a proximate timeline (Jack 2024, PMID 38934362; Dubois 2024, PMID 39483064).
Design: Pooled individual-participant multistate modelling across biomarker cohorts with death as a competing risk, stratified by age, sex, APOE genotype and A/T status, reporting lifetime and 10-year absolute risks with calibration against autopsy-confirmed outcomes.
OQ-4 — Can treatment be stopped once amyloid clearance is achieved, and does pathology re-accumulate?¶
Why open: The donanemab appropriate use recommendations already suggest considering discontinuation once amyloid clearance is demonstrated, typically 12–18 months after starting (Rabinovici 2025, PMID 40155270), but this is a pragmatic inference. TRAILBLAZER-ALZ 4 showed donanemab achieved clearance in 76.8% at 18 months versus 43.1% with aducanumab, with median time to clearance 359 versus 568 days (Salloway 2025, PMID 40390253) — it did not test what happens after stopping.
Design: Randomised discontinuation trial in participants who have achieved clearance, comparing stop versus continue, with amyloid PET, plasma p-tau217 and CDR-SB at 12, 24 and 36 months, powered for re-accumulation and for clinical divergence.
OQ-5 — Is anticoagulation an absolute or relative contraindication to anti-amyloid therapy?¶
Why open: The lecanemab appropriate use recommendations state that patients requiring anticoagulants should not receive lecanemab "until more data regarding this interaction are available," and FDA labelling advises caution because anticoagulant use was associated with an increased number of intracerebral haemorrhages (Cummings 2023, PMID 37357276). The donanemab document does not address it. The mechanism is plausible: ARIA regions show microinfarcts with haemosiderin arising from Aβ-laden vessels (Boon 2025, PMID 41109234). The data the recommendation waits for are not being generated by any retrieved trial.
Design: Prospective registry with prespecified anticoagulant strata and adjudicated intracerebral-haemorrhage outcomes, sized to detect a clinically relevant absolute risk increase; failing that, a target-trial emulation across national treatment registries.
OQ-6 — Should APOE4 homozygosity be operationalised as a distinct entity with its own pathway?¶
Why open: Almost all APOE4 homozygotes show AD pathology; by age 65 nearly all have abnormal CSF amyloid and 75% a positive amyloid scan; symptom onset is earlier (65.1 years) with a prediction interval as narrow as autosomal dominant disease (Fortea 2024, PMID 38710950). The same group carries the highest ARIA risk (RR 3.54) (Qi 2026, PMID 41478817), and the EMA excluded them from lecanemab authorisation. Guidance treats them as a risk stratum within a general population rather than as a genetic form.
Design: A dedicated APOE4/4 cohort with pre-specified prevention and treatment sub-studies, mirroring the DIAN and Down syndrome trial infrastructure; and formal comparison of ARIA-adjusted net benefit in homozygotes against exclusion.
OQ-7 — Which commercially available plasma assays actually meet the 90%/90% threshold, and who audits that?¶
Why open: The 2025 clinical practice guideline permits a blood test with ≥90% sensitivity and ≥90% specificity to substitute for amyloid PET or CSF in specialised care, while cautioning that many commercially available tests do not meet these thresholds, especially with a single cutoff (Palmqvist 2025, PMID 40729527). A head-to-head comparison found mass-spectrometry %p-tau217 at 0.93 accuracy versus 0.83–0.88 for immunoassays (Warmenhoven 2025, PMID 39468767), and ~90% of published accuracy studies used data-derived thresholds (Therriault 2025, PMID 40818474).
Design: A standing, independently funded assay-comparison programme testing every marketed assay on a common banked sample set with prespecified cutoffs and autopsy or PET reference, publishing per-assay performance on a fixed schedule.
OQ-8 — Does changing how a diagnosis is communicated change outcomes?¶
Why open: Dissatisfaction with diagnostic communication and inadequate post-diagnostic information recur across 35 qualitative papers and are implicated in disempowerment and self-stigma (Low 2018, PMID 28828999); the clinician's language determines whether people feel reassured or destabilised (O'Malley 2021, PMID 31647324); and suicide risk peaks in the first 90 days after diagnosis (SMR 3.40 at ages 65–74) (Schmutte 2022, PMID 34036738). A focused PubMed re-query on 2026-08-31 did not identify a randomised trial whose intervention was the diagnostic-disclosure encounter and whose outcomes covered the harms above; this is a dated design gap, not a claim that communication interventions do not exist.
Design: Cluster-randomised trial of a structured disclosure-and-support protocol versus usual disclosure, with 90-day and 12-month outcomes covering distress, self-stigma, service engagement, care-plan completion and safety events.
OQ-9 — Can a single-target therapy work in a population whose remaining pathology is unmeasured?¶
Why open: In 1,079 autopsies, 94% had at least one pathology, 78% two or more, and AD occurred in isolation in only 9%; more than 230 pathology combinations were observed and AD explained 22–100% of an individual's cognitive loss (Boyle 2018, PMID 29244218; Schneider 2007, PMID 17568013). Trials enrol biomarker-confirmed, low-microbleed participants; applying appropriate use criteria to a real clinic admitted about a quarter of amyloid-positive patients, with cerebrovascular findings the commonest disqualifier (Jeon 2025, PMID 41225766).
Design: Prespecified analysis of anti-amyloid trial data by in-vivo copathology burden (white-matter hyperintensity volume, microbleeds, medial temporal atrophy as a LATE proxy), reporting treatment effect as a function of estimated non-AD pathology load.
OQ-10 — Does lowering tau production produce clinical benefit where anti-tau antibodies did not?¶
Why open: A MAPT antisense oligonucleotide reduced CSF t-tau by 56% (95% CI 50–62) and lowered tau-PET signal (Edwards 2023, PMID 37902726), and a 416-participant phase 2 with a CDR-SB primary is under way (NCT05399888). Four anti-tau monoclonal antibodies across six RCTs and 2,193 patients produced no convincing clinical effect, with placebo ranking best on CDR-SB and ADCS-ADL (Cai 2024, PMID 39945003).
Design: The phase 2 already answers the first half; what is missing is a prespecified comparison of intracellular-lowering versus extracellular-clearance strategies on a common tau-PET and CSF MTBR-tau-243 readout, so that a negative clinical result can be attributed to compartment, timing or target.
OQ-11 — Does hearing intervention prevent cognitive decline in people at higher risk?¶
Why open: ACHIEVE was null overall (difference 0.002, 95% CI −0.077 to 0.081, P=0.96) with a prespecified interaction indicating a different effect in the higher-risk ARIC subcohort than in de-novo volunteers (p=0.010) (Lin 2023, PMID 37478886). Hearing loss carries a weighted population attributable fraction of 7.2% (95% CI 5.2–9.7) (Stephan 2024, PMID 38824956).
Design: A confirmatory trial enriched for the ARIC-like risk profile (older, more vascular risk factors, lower baseline cognition), powered for the interaction rather than for the main effect.
OQ-12 — Should blood-biomarker testing be permitted in primary care, and with what safeguards?¶
Why open: In prospective Swedish primary care, an APS2 score achieved AUC 0.96 and 91% diagnostic accuracy against physicians' 61% (Palmqvist 2024, PMID 39068545). The 2022 appropriate use recommendations advised against primary-care use pending more data (Hansson 2022, PMID 35908251); the 2025 guideline is scoped to specialised care (PMID 40729527). Diagnosis is delayed by 31–44 months on average, differentially by ethnicity (Lin 2021, PMID 34091580).
Design: Stepped-wedge implementation trial in primary care with a fixed assay and prespecified cutoffs, measuring time to diagnosis, referral appropriateness, downstream specialist workload, false-positive consequences and equity of access.
OQ-13 — Does treating psychosis or agitation early alter disease trajectory?¶
Why open: In 335 incident AD cases, psychosis carried HR 2.01 and agitation/aggression HR 2.95 for progression to severe dementia, and both predicted earlier death (Peters 2015, PMID 25585033). The treatment trials retrieved in a focused PubMed re-query on 2026-08-31 use short-term symptom endpoints rather than long-term progression; for example, the pivotal brexpiprazole trial ran 12 weeks with a CMAI endpoint (Lee 2023, PMID 37930669). They therefore do not establish whether symptom control alters disease trajectory.
Design: Randomised trial with a 24-month horizon and progression-to-severe-dementia as a co-primary endpoint alongside symptom control, with prespecified mortality surveillance given the antipsychotic signal (Schneider 2005, PMID 16234500; Ballard 2009, PMID 19138567).
OQ-14 — What is the long-term mortality risk of brexpiprazole in this population?¶
Why open: The mortality signal for atypical antipsychotics in dementia (OR 1.54, 95% CI 1.06–2.23) comes from 10–12-week trials, and DART-AD showed the hazard widens substantially with continued exposure — 24-month survival 46% versus 71% after withdrawal (Schneider 2005, PMID 16234500; Ballard 2009, PMID 19138567). Brexpiprazole's pivotal trial ran 12 weeks (Lee 2023, PMID 37930669).
Design: Active-comparator new-user cohort study in national registries with 24- and 36-month mortality, plus a randomised withdrawal trial in long-term users mirroring DART-AD.
OQ-15 — Should CAA burden be a formal eligibility variable rather than a monitoring variable?¶
Why open: ARIA regions show microinfarcts, complement activation and CD68-positive vessel walls arising from Aβ-laden leptomeningeal and penetrating vessels — iatrogenic destabilisation of vascular amyloid (Boon 2025, PMID 41109234). Boston criteria v2.0 can identify probable CAA in life with 74.5–92.5% sensitivity and 81.5–95.0% specificity depending on cohort (Charidimou 2022, PMID 35841910). The donanemab AUR excludes >4 microbleeds and cortical superficial siderosis but does not use formal CAA criteria (Rabinovici 2025, PMID 40155270).
Design: Apply Boston v2.0 retrospectively to trial and registry baseline MRIs and test whether CAA probability outperforms microbleed count in predicting symptomatic ARIA.
OQ-16 — How much dementia is actually preventable, and by what?¶
Why open: Weighted population attributable fractions give 32.0% (95% CI 26.6–37.5) for seven combined factors and are higher in low- and middle-income countries (Stephan 2024, PMID 38824956), while randomised multidomain trials have produced cognitive-composite differences of 0.022–0.029 units per year and no reduction in dementia incidence (Ngandu 2015, PMID 25771249; Baker 2025, PMID 40720610; Moll van Charante 2016, PMID 27474376). Midlife vascular risk predicted late-life amyloid while late-life risk did not (Gottesman 2017, PMID 28399252).
Design: A midlife-initiated prevention trial with plasma p-tau217 and amyloid PET as intermediate endpoints and dementia incidence as a long-term endpoint, in a cohort young enough that the exposure window is not already closed.
Tier 2 — requires enabling work¶
OQ-17 — Can LATE neuropathologic change be identified in life?¶
Why open: LATE-NC is present at cognitively relevant levels in about 25% of community autopsy brains, mimics amnestic AD, and has its own genetic risk profile (GRN, TMEM106B, ABCC9, KCNMB2, APOE); the consensus report names the absence of an antemortem biomarker as the principal obstacle to progress (Nelson 2019, PMID 31039256). Proposed 2025 clinical criteria now define probable LATE as progressive amnestic decline with substantial hippocampal atrophy and negative amyloid biomarkers, and possible LATE when amyloid is unavailable or present but hippocampal neurodegeneration is disproportionate; the authors explicitly present these as an initial framework requiring validation (Wolk 2025, PMID 39807681). Quantitative work found only p-tau burden and microinfarcts — not LATE-NC — independently predicting decline once tau was measured properly (Richardson 2026, PMID 42184025), which sharpens rather than resolves the biomarker question.
Enabling work: A TDP-43 PET ligand or a fluid marker with autopsy validation.
OQ-18 — Does Centiloid change measure parenchymal amyloid removal?¶
Why open: In five aducanumab-treated autopsies, Aβ clearance was localised to cortical layer I with no significant clearance in deeper layers, despite Centiloid reductions of −6% to −81% (Boon 2025, PMID 41109234). Centiloid change is the field's standard target-engagement readout (Klunk 2015, PMID 25443857) and the basis for stopping rules.
Enabling work: Larger treated-autopsy series across agents, with layer-resolved quantification paired to antemortem Centiloid values.
OQ-19 — Which clearance route fails first in sporadic AD?¶
Why open: Metabolic labelling established impaired Aβ40 and Aβ42 clearance with unchanged average production (Mawuenyega 2010, PMID 21148344), but that is a rate, not a route. Candidate routes — proteolysis, LRP1-mediated transport, perivascular/glymphatic drainage — are not separately measurable in humans; one night of sleep deprivation increased hippocampal and thalamic amyloid burden in 20 healthy adults (Shokri-Kojori 2018, PMID 29632177).
Enabling work: Route-specific in-vivo measures in humans.
OQ-20 — Is TREM2 agonism beneficial in humans?¶
Why open: The disease-associated microglia programme is TREM2-dependent in mouse but the human AD microglial signature differs and is attenuated in TREM2 risk-variant carriers (Zhou 2020, PMID 31932797). Higher CSF sTREM2 was associated with slower hippocampal atrophy and cognitive decline independently of p-tau181 (Stephenson 2026, PMID 41605308; Ewers 2019, PMID 31462511), and higher baseline TSPO binding predicted slower decline while rising binding predicted faster decline (Hamelin 2018, PMID 29608645). INVOKE-2 (AL002) then engaged TREM2 — CSF sTREM2 fell, osteopontin rose — without slowing CDR-SB at any of three doses (week-96 LS differences −0.31 to +0.13, all P>0.05) and with ARIA-like MRI as the leading TEAE (Mummery 2026, PMID 41787076; NCT04592874). The question is no longer whether this antibody, at this stage, works (it did not on the primary); it is whether an earlier, genotype-stratified, combination, or different-epitope TREM2 strategy would.
Enabling work: A human-validated pharmacodynamic marker distinguishing the protective from the detrimental activation profile; then a trial that is not a late-stage CDR-SB monotherapy of this antibody.
OQ-21 — Can tau-PET specificity be made reader-independent?¶
Why open: Against the same autopsies, five independent readers produced flortaucipir specificity ranging from 52.0% to 92.0% for Braak V–VI pathology, with sensitivity 92.3–100% (Fleisher 2020, PMID 32338734). Tau PET is rated "uncertain" in 6 of 17 clinical scenarios — the largest uncertainty block in any current appropriate-use document (Rabinovici 2025, PMID 39776249).
Enabling work: Quantitative reading standards or automated classification validated against autopsy, and cryo-EM-informed ligands selective for the AD paired-helical-filament fold (Fitzpatrick 2017, PMID 28678775).
OQ-22 — Do the four tau trajectories require different treatments and endpoints?¶
Why open: Data-driven modelling of tau-PET in 1,612 individuals resolved four spatiotemporal trajectories with prevalences of 18–33%, stable longitudinally, replicated with a second tracer, differing in demographics, cognition and outcome, and apparently spreading through distinct corticolimbic networks (Vogel 2021, PMID 33927414). Autopsy subtypes show the same structure (Murray 2011, PMID 21802369).
Enabling work: Prospective subtype assignment at enrolment in an existing trial, with subtype-specific outcome measures pre-registered.
OQ-23 — Does APOE4-associated blood–brain-barrier breakdown precede amyloid, and is sPDGFRβ a target?¶
Why open: BBB breakdown in hippocampus and medial temporal lobe distinguished APOE4 carriers from non-carriers even when cognitively unimpaired and was unrelated to CSF or PET amyloid and tau; baseline CSF sPDGFRβ predicted future cognitive decline in ε4 carriers only, after controlling for amyloid and tau (Montagne 2020, PMID 32376954; Nation 2019, PMID 30643288). Two independent CSF datasets broaden the association: sPDGFRβ correlated with albumin and tau markers in biomarker-defined AD (Miners 2019, PMID 31521199), and in 158 participants it peaked at CDR 0.5, related to albumin-ratio leakage and predicted MMSE change in that stage (Lv 2023, PMID 36915135). Neither identifies a causal ordering or validates sPDGFRβ as a treatment target.
Enabling work: Longitudinal ordering studies from midlife, and a tractable pharmacological handle on the cyclophilin A–MMP9 pathway.
OQ-24 — Can a polygenic risk score be made valid across ancestries?¶
Why open: Of 25 loci known in non-Hispanic White individuals, only APOE, ABCA7, TREM2, BIN1, CD2AP, FERMT2 and WWOX replicated at nominal significance in African American individuals, though pathways overlapped (Kunkle 2021, PMID 33074286). The ε2 protective effect is absent in East Asian populations and the ε4 effect attenuated in Hispanic populations without an ancestry explanation (Belloy 2023, PMID 37930705).
Enabling work: Adequately powered GWAS and sequencing in African, East Asian, South Asian, Hispanic/Latino and Middle Eastern populations.
OQ-25 — What are the insurance, employment and legal-capacity consequences of a biological diagnosis in an asymptomatic person?¶
Why open: The 2024 criteria make an asymptomatic Core 1-positive person diagnosable (Jack 2024, PMID 38934362); the IWG identifies disclosure to such people as "the most problematic implication" of a purely biological definition (Dubois 2024, PMID 39483064). Disclosure studies in research volunteers show mild, largely transient distress (Grill 2024, PMID 38585443; Clark 2024, PMID 39129396). Interviews with 17 dementia experts found three incompatible documentation practices—active AD, non-informative finding, or increased susceptibility—and those using the active-disease framing were more willing to disclose to employers and insurers (Vaishnav 2024, PMID 38393896). This closes the narrower question of whether clinician practice is uniform; it does not measure actual insurance, employment or capacity outcomes. A repeat live PubMed search on 2026-09-03 retrieved no longitudinal outcomes study.
Enabling work: Legal and actuarial mapping across jurisdictions, then prospective cohort follow-up of disclosed individuals.
OQ-26 — What do people with dementia in low- and middle-income countries prioritise?¶
Why open: 61% of people with dementia live in low- and middle-income countries but 74% of measured costs occur in high-income countries (Wimo 2023, PMID 36617519); the largest projected increases are in north Africa/Middle East (+367%) and eastern sub-Saharan Africa (+357%) (GBD 2019 Dementia Forecasting Collaborators 2022, PMID 34998485); only 7% of active trial sites are in LMICs (Cummings 2025, PMID 40555627); and in STRiDE community surveys fewer than 1% of people meeting dementia criteria had been diagnosed (Farina 2023, PMID 37278200). The audit search found relevant qualitative studies rather than an absence: patient–caregiver dyads in Colombia linked illness awareness to socioeconomic conditions and social support (Ramos 2026, PMID 41717228); caregiver studies in Iran and Uganda identified family, community, religious and formal-system supports, alongside financial and psychological burden (Sadeghi-Mahalli 2025, PMID 39097934; Abaasa 2023, PMID 37680685; Ainamani 2020, PMID 33043153); and Malaysian providers described fragmented pathways and late presentation (Goodson 2021, PMID 34368037). The remaining gap is that this evidence is geographically scattered, mostly caregiver- or provider-centred, and does not constitute a patient-led cross-country priority set.
Enabling work: Locally led qualitative and priority-setting research, and the organisational infrastructure to conduct it.
OQ-27 — What is the experience of anti-amyloid treatment itself?¶
Why open: Treatment involves fortnightly or monthly infusions, repeated surveillance MRI, and — in 19% of patients — an ARIA event (Qi 2026, PMID 41478817). This question was narrowed on 2026-09-03. Two qualitative studies now cover the decision either side of treatment: 22 people with confirmed AD who had discussed lecanemab with a clinician, reporting information sources, hope, and risk–benefit weighing shaped by family, insurance and trust (Parks 2025, PMID 40207637); and 22 Japanese care partners on donanemab and treat-to-clearance, reporting mental burden, relief at confirmed clearance and continuing worry after stopping (Katayama 2026, PMID 41307609). A repeat live PubMed search on the same date returned no qualitative study of the treatment period itself. The remaining gap is therefore specific: what infusion attendance, surveillance MRI and an ARIA event mean to patients and care partners while on treatment. Real-world cohorts quantify infusion reactions (37%) and withdrawal (9.8%) without reporting any of that (Paczynski 2025, PMID 40354064).
Enabling work: Prospective qualitative work embedded in treatment registries, with the ethics constraints set out in literature/patient-voice/README.md.
OQ-28 — Should staging schemes incorporate pathology density as well as topography?¶
Why open: Within Braak stage V alone, neocortical p-tau burden ranged from 0.2% to 53.7%, and only p-tau burden and microinfarcts independently affected cognitive decline; people with low burden (≤13%) had significantly better trajectories over their final 15 years than those with high burden (≥23.5%) (Richardson 2026, PMID 42184025). Braak staging is topographic by construction (Braak 1991, PMID 1759558) and supplies the "B" of the ABC score (Montine 2012, PMID 22101365).
Enabling work: Multi-centre replication with harmonised quantitative methods, then a revised consensus scoring system.
OQ-29 — Why do caregiver interventions improve self-efficacy but not strain?¶
Why open: D-CARE found no difference between health-system, community-based and usual dementia care on behavioural symptoms or caregiver strain across 2,176 dyads, while both active arms improved caregiver self-efficacy (Reuben 2025, PMID 39878968). Yet the 780-dyad Care Ecosystem reduced caregiver depression by 1.14 points and burden by 1.90 points while reducing emergency-department use (Possin 2019, PMID 31566651), and a component-optimisation trial found mindfulness reduced depression by 2.13 points while behavioural-problem management transiently increased anxiety (Kwok 2025, PMID 40094667). REACH II, twenty years earlier, reduced clinical depression from 22.7% to 12.6% (Belle 2006, PMID 17116917). “Care management” is therefore too coarse an intervention label to explain the contradiction.
Enabling work: Determine whether the difference lies in the intervention (structured skills training versus care navigation), in the outcome measures, or in improved usual care; then re-test the effective ingredient.
OQ-30 — Resolved 2026-08-31: semaglutide did not slow clinical progression¶
Resolution: The primary report was retrieved live during the audit. In evoke, week-104 CDR-SB changed 2.3 with semaglutide and 2.3 with placebo (difference −0.08, 95% CI −0.35 to 0.20; P=0.57); in evoke+, change was 2.2 versus 2.1 (difference +0.10, −0.17 to 0.38; P=0.46). Both trials were discontinued for negative clinical outcome. Treatment-emergent adverse events occurred in 91.2% of semaglutide recipients and 84.8% of placebo recipients; investigators considered five deaths treatment-related (one semaglutide, four placebo) (Cummings 2026, PMID 41865758; NCT04777396; NCT04777409).
Residual question: Why preclinical and observational signals failed to translate, and whether a prespecified metabolic subgroup behaves differently, remains open; that is a new mechanistic question rather than a reason to reinterpret two null primary endpoints.
OQ-31 — Can plasma-biomarker performance be made portable across assays and settings?¶
Why open: In a ten-assay head-to-head comparison, AUC for abnormal amyloid ranged from 0.642 to 0.947 and plasma–CSF correlations from 0.320 to 0.891 (Janelidze 2023, PMID 36087307). The systematic review behind the 2025 clinical guideline found sensitivity 49.3–91.4%, specificity 61.5–96.7% and GRADE certainty from moderate to very low across 31 tests; most studies had high risk of bias in at least one core diagnostic-accuracy domain (Pahlke 2025, PMID 41193403). The guideline’s ≥90%/≥90% substitution rule is test-specific, but clinical language still treats p-tau217 as a fungible analyte.
Enabling work: Prospective, blinded, multi-ancestry head-to-head studies using locked cut-points in primary and specialist care, with renal function, age, copathology and pre-test probability prespecified.
Dots not yet connected¶
Testable junctions between evidence bodies already represented in this knowledge base. "Missing junction" is a scoped synthesis gap, not a claim that no study exists anywhere.
| # | Dot A | Dot B | The missing junction | Powers |
|---|---|---|---|---|
| D1 | Plasma p-tau217 can identify AD pathology at ~90% accuracy in primary care | Anti-amyloid therapy requires infusion capacity and surveillance MRI that most systems lack | Model the case-finding-to-capacity ratio a national programme would generate, before deploying the test | OQ-12, OQ-2 |
| D2 | ARIA is iatrogenic destabilisation of cerebral amyloid angiopathy | Boston criteria v2.0 identify probable CAA in life with known sensitivity and specificity | Test CAA probability, not microbleed count, as the eligibility variable | OQ-15, OQ-5 |
| D3 | Midlife vascular risk predicts late-life amyloid; late-life risk does not | Every multidomain prevention trial has enrolled people over 60 | A midlife-initiated prevention trial with biomarker intermediate endpoints | OQ-16 |
| D4 | AD explains 22–100% of an individual's cognitive loss | Trials report a single average treatment effect | Report treatment effect as a function of in-vivo copathology burden | OQ-9, OQ-17 |
| D5 | Higher TREM2-dependent microglial activation predicts slower decline | INVOKE-2 engaged TREM2 without slowing CDR-SB and produced ARIA-like MRI; ADAPT was null | Separate the protective activation profile from the detrimental one before any further TREM2-agonist trial | OQ-20 |
| D6 | Suicide risk peaks in the first 90 days after diagnosis | Disclosure communication is the strongest modifiable theme in the qualitative literature | A disclosure-protocol trial with safety events as a prespecified outcome | OQ-8 |
| D7 | Tau-PET topography predicts where atrophy will occur at the individual level | Trials use whole-brain composite cognitive outcomes | Individualised, topography-matched cognitive endpoints in tau-directed trials | OQ-22, OQ-10 |
| D8 | APOE4 homozygotes have near-fully-penetrant AD biology and the earliest onset | APOE4 homozygotes have the highest ARIA risk and are excluded from EU authorisation | A dedicated homozygote pathway: earlier treatment, different agent, or prevention | OQ-6, OQ-2 |
| D9 | 25% of community autopsy brains have cognitively relevant LATE-NC | Amnestic presentations are treated as AD once a Core 1 biomarker is positive | Estimate how much biomarker-positive "AD" is clinically driven by LATE | OQ-17, OQ-3 |
| D10 | Weighted PAFs are higher in low- and middle-income countries | Only 7% of trial sites are in those countries, and the retrieved qualitative literature is small, geographically scattered and mostly caregiver/provider-centred | Prevention trials designed and led where the preventable fraction is largest | OQ-16, OQ-26 |
| D11 | Cognitive stimulation's largest and most certain effect is on communication and social interaction | Trials of dementia care models measure behavioural symptoms and caregiver strain | Use communication and social participation as primary outcomes in care-model trials | OQ-29 |
| D12 | Braak stage V contains a 250-fold range of neocortical tau burden | Tau-PET is being proposed as a prognostic gate for treatment | Calibrate tau-PET signal against quantitative post-mortem density, not stage alone | OQ-28, OQ-21 |
| D13 | APOE4 causes blood–brain-barrier breakdown independently of amyloid and tau | APOE4 is treated as an amyloid-clearance gene in therapeutic reasoning | Test whether APOE4-directed therapy should target the vasculature rather than amyloid | OQ-23, OQ-6 |
| D14 | Anti-amyloid trials exclude anticoagulated patients | Atrial fibrillation is common in the treatment-eligible age range | Quantify how much of the eligible population is excluded by anticoagulation alone, and at what cost | OQ-5, OQ-2 |
Maintenance rules¶
- Retire a question only when the evidence resolves the decision, not when one trial reports.
- Record neutral and harmful results as answers; OQ-30 was closed by two null primary results on 2026-08-31.
- Preserve the stable ID and add a dated resolution note when closing.
- Re-query ClinicalTrials.gov before describing any NCT status; every status on this page is as of 2026-08-31.
- Re-run the absence search before repeating any claim that no study exists — stale absences are the most common error in a file of this kind. As of the 2026-09-03 audit, the remaining dated gaps are OQ-25 (longitudinal insurance/employment/capacity outcomes, re-searched 2026-09-03 with no hit) and the narrowed OQ-27 (experience during anti-amyloid treatment — infusion, surveillance MRI, ARIA). OQ-26 was narrowed after LMIC studies were found, OQ-27 was narrowed on 2026-09-03 by two qualitative studies of treatment decisions, and OQ-30 was resolved by PMID 41865758.