Open questions — vascular dementia¶
Last curated: 2026-09-02 (second deepening pass) · status: draft research agenda
Questions OQ-1 to OQ-20 and junctions D1–D10 are the seeded agenda, revised below where later passes changed their status. OQ-21 to OQ-30 were added by the 2026-08-31 deepening pass, OQ-31 to OQ-36 and junctions D18–D27 by the 2026-09-02 pass.
Tier 1 — practice-changing and designable now¶
OQ-1 — Which contemporary criteria best identify a reproducible, prognostically coherent VCI population?¶
Why open: VASCOG, VICCCS, and VasCog-2-WSO converge on a spectrum model but prospective head-to-head reliability, prognosis, and treatment-response data remain limited (Sachdev 2014, PMID 24632990; Skrobot 2018, PMID 29055812; VasCog-2-WSO Criteria Consortium 2025, PMID 40955506).
Sharpened 2026-08-31. The bar the older criteria set is now quantified and it is low. Against autopsy, ADDTC probable VaD reaches sensitivity 0.25 and NINDS–AIREN probable 0.20, and Gold and colleagues found no statistically significant relationship between neuropathological diagnosis and ICD-10, ADDTC probable, or NINDS–AIREN probable criteria (Gold 2002, PMID 11772694). Interrater reliability across four criteria sets ranged κ 0.30–0.61, with the oldest instrument the most reproducible (Chui 2000, PMID 10681076). In the oldest old, 42% of pathologically pure vascular dementias had no stroke history at all, so stroke-anchored criteria are structurally blind to them (Bacchetta 2007, PMID 16580095). No modern criteria set — VASCOG, VICCCS, or VasCog-2-WSO — has been subjected to an equivalent autopsy validation. That, not further consensus refinement, is the missing study.
Design: Multinational prospective cohort, blinded parallel criteria assignment, common MRI/AD biomarkers, 3-year cognition/function/stroke outcomes.
OQ-2 — What BP target maximizes cognitive net benefit in MRI-defined SVD?¶
Why open: Pooled trials show small benefit, while SPRINT MIND reduced MCI but not probable dementia and trial populations were not enriched by SVD (Williamson 2019, PMID 30688979; Hughes 2020, PMID 32427305).
Sharpened 2026-09-02, and the parent question is now partly closed. Whether blood-pressure lowering prevents dementia at all has an affirmative randomized answer: a cluster-randomized trial in 33,995 rural Chinese adults with uncontrolled hypertension achieved a 22.0 mm Hg systolic contrast over 48 months and reduced all-cause dementia (RR 0.85, 95% CI 0.76–0.95, P=0.0035) with fewer serious adverse events (He 2025, PMID 40258956). What remains open is exactly this question's original form — the target, in SVD-enriched populations, with subtype adjudication. The new trial reported all-cause dementia only, was open-label, and recruited a previously untreated population, so it does not establish a target for people already on treatment, nor a vascular-dementia-specific effect. Two further constraints have been quantified: the population attributable fraction of hypertension, diabetes and smoking for dementia occurring after age 80 is only 2–8%, against 21.8–44.0% for dementia by 80 depending on the age at which the risk factor was measured (Smith 2025, PMID 40455489); and randomized deprescribing in older adults changed nothing, including falls (RR 1.00, 0.89–1.13) (Alsubaiei 2026, PMID 41550131). The trial that remains undone is an SVD-enriched, subtype-adjudicated target comparison in people under 75.
Sharpened 2026-08-31. Two findings now bracket this question from opposite sides. Intensive BP control produced a large, dose-dependent reduction in a validated global SVD imaging factor (Cohen's d −0.40, 95% CI −0.62 to −0.17; 21–39% less progression across attained-reduction strata) (Charisis 2026, PMID 42614618), while the Cochrane pooling of placebo-controlled trials in people without prior cerebrovascular disease found OR 0.89 (0.72–1.09) for incident dementia at very low certainty (Cunningham 2021, PMID 34028812). Meanwhile a 2×2 factorial trial that randomized systolic <130 mm Hg plus atorvastatin, with or without exercise, in 513 at-risk older adults was flatly null on cognition at 24 months (Zhang 2026, PMID 41870419). The question is no longer only "what target" but "does later-life vascular risk modification move cognition at all, and if the imaging moves while cognition does not, which endpoint is wrong."
Design: Pragmatic target trial stratified by SVD burden, orthostasis, and frailty; outcomes cognition, stroke, falls, syncope, kidney injury.
OQ-3 — Does optimized recurrent-stroke prevention reduce delayed post-stroke dementia?¶
Why open: Recurrent stroke carries high dementia burden and several risk factors are actionable, but cognitive prevention has rarely been a powered endpoint (Pendlebury 2009, PMID 19782001; Filler 2024, PMID 38101426).
Design: Cluster implementation trial with prespecified accessible cognitive and functional outcomes at 3 years.
OQ-4 — Which post-stroke screening schedule improves outcomes rather than detection alone?¶
Why open: Any post-stroke NCD may affect 53.4%, but no universal timing or instrument links screening to improved daily function (Barbay 2018, PMID 30504699; Swartz 2025, PMID 39822128).
Design: Randomize systematic repeated screening plus action pathway versus usual care.
OQ-5 — Can Alzheimer biomarkers identify the subgroup with worthwhile cholinesterase response?¶
Why open: Mean benefits are small and older trials did not adequately separate mixed Alzheimer disease (Battle 2021, PMID 33704781; Schneider 2007, PMID 17568013).
Sharpened 2026-08-31. The stratification tool now exists. Plasma NfL tracks small-vessel-disease burden while p-tau217 tracks Alzheimer-related neurodegeneration, and GFAP tracks both, in memory-clinic patients validated against amyloid/tau PET (Dewenter 2026, PMID 42252508; Lockhart 2026, PMID 41630612). A biomarker-stratified discontinuation trial is therefore designable today with assays already in clinical use. The counterweight: only 16 placebo-controlled VaD trials exist and all but one were published in 2012 or earlier (Cheng 2026, PMID 42635820), so the stratified question would have to be answered by a new trial, not by re-analysis.
Design: Biomarker-stratified randomized discontinuation trial with patient-defined functional targets.
OQ-6 — What is the net benefit of anticoagulation across CAA/microbleed burden?¶
Why open: Microbleeds raise ICH risk, while AF anticoagulation prevents embolic stroke; existing estimates are largely observational (Charidimou 2017, PMID 29117953; Zhao 2024, PMID 38560733).
Partly answered 2026-08-31, and re-sharpened. Randomized individual-participant data from 412 people across four trials show that after intracranial haemorrhage, anticoagulation reduces ischaemic major adverse cardiovascular events from 19% to 4% (HR 0.27, 95% CI 0.13–0.56) with a smaller, statistically uncertain increase in haemorrhagic events (HR 1.80, 95% CI 0.77–4.21) and unchanged death-or-dependence (COCROACH; Al-Shahi Salman 2023, PMID 37839434). What remains open is the subgroup question: within a randomized trial's imaging subanalysis, the category "probable CAA" did not predict recurrence while cortical superficial siderosis did (HR 7.7, 95% CI 1.4–42.2) (Fandler-Höfler 2025, PMID 41197104). The question should now be posed at marker level, not category level. Note also that HAS-BLED discriminates worse than chance in this population (C 0.41, 95% CI 0.29–0.53) (Wilson 2018, PMID 29778365).
Design: Prospective registry-embedded trial/causal cohort with standardized susceptibility MRI.
OQ-7 — Which MRI measure is a valid SVD treatment-response endpoint?¶
Why open: STRIVE-2 catalogs quantitative candidates, but change in a marker is not yet proven to mediate cognitive benefit (Duering 2023, PMID 37236211; Markus 2022, PMID 35969390).
Sharpened 2026-08-31. The candidate field has narrowed usefully. PSMD explains cognition beyond WMH volume across four independent cohorts (Luckey 2024, PMID 39569745); MRI free water is the only MarkVCID kit that has predicted accelerated decline prospectively (Maillard 2022, PMID 36523847); cerebrovascular reactivity replicated a prespecified cognitive association at three sites (Liu 2024, PMID 38951718). Against WMH volume specifically there are now three separate strikes: growth correlates with memory decline but regression correlates with nothing (Bahrani 2023, PMID 37840499); genetically determined WMH is causal in two-sample MR and null in individual-level longitudinal analysis of the same construct (Sargurupremraj 2024, PMID 38776079); and vascular dysfunction predicts white-matter damage locally but not globally (Pham 2026, PMID 40832975). The specific unmet requirement is a demonstration that an intervention's effect on any of these markers mediates its effect on cognition. No study has attempted it.
Design: Multi-trial validation linking intervention effects on imaging to cognition/function.
OQ-8 — Which cognitive rehabilitation strategy transfers to independent daily function?¶
Why open: Exercise and rehabilitation studies are heterogeneous, and scale improvement may not generalize to medication, finance, or participation (Masserini 2023, PMID 37539725; Ye 2024, PMID 39666076).
Design: Adaptive trial of goal-based compensatory strategies with blinded performance outcomes.
OQ-9 — How should Boston v2.0 be used in non-haemorrhagic cognitive clinics?¶
Why open: Pathology-validated accuracy is strong in selected presentations, but cognitive-only populations need external validation (Charidimou 2022, PMID 35841910; Switzer 2024, PMID 38710005).
Design: Consecutive memory-clinic cohort with standardized MRI and pathology/longitudinal reference.
OQ-10 — What service model best joins stroke, memory, primary, and caregiver care?¶
Why open: Qualitative work identifies unrecognized problems and help-seeking barriers, while current guidelines call for longitudinal pathways (Tang 2019, PMID 30452613; Swartz 2025, PMID 39822128).
Design: Stepped-wedge integrated VCI clinic model measuring function, recurrence, caregiver burden, and cost.
Tier 2 — requires enabling measurement or biology¶
OQ-11 — Can individual vascular and Alzheimer attributable fractions be estimated?¶
Autopsy establishes common copathology but not precise living-person attribution (Schneider 2007, PMID 17568013).
OQ-12 — Is BBB leakage upstream, downstream, or bidirectional?¶
Human BBB biomarkers are heterogeneous and not validated for causal staging (French 2025, PMID 40145342).
OQ-13 — Does impaired perivascular clearance connect SVD to amyloid accumulation?¶
Fluid-dynamic imaging is advancing, but enlarged perivascular spaces are not direct proof of impaired clearance (Solé-Guardia 2025, PMID 40289686).
OQ-14 — Which blood panel distinguishes active vascular injury from accumulated scar?¶
NfL and endothelial markers lack etiologic specificity (Zhao 2019, PMID 31026486; Beran 2024, PMID 39428719).
OQ-15 — Can lesion-network mapping operationalize “strategic infarct”?¶
Location matters, but existing syndromic labels are not quantitatively standardized (O'Brien 2015, PMID 26595643).
OQ-16 — Which monogenic SVD biomarkers translate to sporadic disease?¶
CADASIL offers known onset biology but differs from multifactorial age-related SVD (Chabriat 2009, PMID 19539236; Whittaker 2022, PMID 35699195).
OQ-17 — Are vascular–Alzheimer effects additive or synergistic?¶
Neuropathology supports coexistence, but interaction estimates require large continuous-burden datasets (Jellinger 2007, PMID 17324442).
OQ-18 — What is a patient-important cognitive change in VCI?¶
Drug trials show statistically significant 1–2-point changes with inconsistent function (Battle 2021, PMID 33704781).
OQ-19 — How can trials include aphasia and severe disability without invalid measurement?¶
Post-stroke assessment is distorted by language and motor deficits, and excluded groups carry high burden (McMahon 2024, PMID 38226361).
OQ-20 — Can CAA and ARIA risk be modeled as one vessel-wall continuum?¶
CAA criteria and anti-amyloid safety literatures examine overlapping vessel pathology but use different endpoints (Charidimou 2022, PMID 35841910; Cordonnier 2025, PMID 40721902).
Tier 1 additions from the 2026-08-31 deepening pass¶
OQ-21 — Are the Asia-Pacific pharmacological leaders real, or risk-of-bias artifacts?¶
Why open: A Bayesian network meta-analysis of 194 randomized trials across 21 drugs, drawing on CNKI, Wanfang, and grey literature, ranks butylphthalide, huperzine A, and edaravone above rivastigmine and memantine on MMSE, and huperzine A and butylphthalide first on activities of daily living (Dang 2024, PMID 39239652). A second network ranks the herbal compound sailuotong first on ADAS-Cog (MD −3.00, 95% CI −4.50 to −1.50) while calling its own network sparse (Li 2025, PMID 41404461). Restricting to English-language, low-risk-of-bias trials leaves 16 studies and nothing recommendable (Cheng 2026, PMID 42635820). If the Asian estimates are real they are practice-changing; if they are bias they are among the largest bodies of misleading evidence in dementia therapeutics. No independent cross-region replication of any of these agents exists.
Design: A single adequately powered, independently monitored, placebo-controlled trial of butylphthalide or huperzine A in subcortical VCI, run outside the region of origin, with the ADAS-Cog and a functional coprimary — the design that produced Jia 2016's ambiguous signal (PMID 26086183) executed at confirmatory scale.
OQ-22 — Would a confirmatory trial reproduce the Cerebrolysin and Actovegin effect sizes?¶
Why open: Cerebrolysin produced an ADAS-Cog+ least-squares mean difference of −6.17 points (P<0.0001) with combined responder rates of 67.5% vs 27.0% in 242 patients (Guekht 2011, PMID 20656516) — several times any cholinesterase-inhibitor effect. ARTEMIDA gave actovegin a treatment difference of −2.3 ADAS-Cog+ points (95% CI −3.9 to −0.7; P=0.005) in 503 post-stroke patients (Guekht 2017, PMID 28432265). Cochrane identified six Cerebrolysin trials (597 participants) but the cognitive pooling is three trials/420 people (SMD 0.36, 95% CI 0.13–0.58) and the global-response pooling is two trials/379 (RR 2.69, 95% CI 1.82–3.98), both very low quality, with no new eligible trial appearing between the 2013 and 2019 reviews and industry sponsorship wherever funding was reported (Cui 2019, PMID 31710397). Absence of confirmation is not evidence of absence of effect; it is evidence that nobody has funded the confirmation.
Design: Investigator-initiated, independently funded, placebo-controlled replication with the original primary endpoint and prespecified futility boundaries.
OQ-23 — Do imaging endpoints that respond to treatment mediate cognitive benefit?¶
Why open: Intensive BP control moves a validated global SVD factor with a clear dose-response (Charisis 2026, PMID 42614618) and reduces WMH growth (SPRINT MIND Investigators 2019, PMID 31408137), while ESO–EAN concludes cognitive benefit of vascular risk treatment after stroke is unproven (Quinn 2021, PMID 34746430) and the Cochrane dementia estimate is null (Cunningham 2021, PMID 34028812). This is the field's central measurement crisis and it is presently arbitrated by which endpoint a guideline happens to prioritise.
Design: Pooled mediation analysis across completed BP trials with paired imaging and cognition, prespecified as a formal surrogate-validation exercise rather than a secondary analysis.
OQ-24 — Does raw-scale or standardized-effect framing give the right reading of cholinesterase-inhibitor benefit?¶
Why open: Cochrane reports donepezil 5 mg at MD −0.92 ADAS-Cog points (95% CI −1.44 to −0.40); a network meta-analysis of overlapping trials reports SMD −1.11 (95% CI −1.88 to −0.34) for the same comparison, and credits donepezil 10 mg with executive and global benefit (CDR-SB −0.25; EXIT25 −1.47) that Cochrane does not endorse (Battle 2021, PMID 33704781; Shi 2022, PMID 35048806). Since the underlying trials largely coincide, the divergence is methodological. It decides whether these drugs are marginal or moderately useful, and therefore whether the guideline split between "insufficient evidence" and "conditionally recommended" is substantive or semantic (Quinn 2021, PMID 34746430; Kim 2025, PMID 39944527).
Design: Re-analysis of the shared trial set with individual participant data, reporting raw units, standardized units, and responder proportions against an anchor-based minimal important difference.
OQ-25 — Is single-substrate cerebrovascular pathology consequential at all?¶
Why open: Among 1,474 autopsied decedents, single cerebrovascular pathologies — one vessel disease or one type of tissue injury — showed no faster cognitive decline than no cerebrovascular pathology, while mixed profiles did, with atherosclerosis-plus-arteriolosclerosis combinations worst and different profiles mapping to different cognitive domains (Lamar 2022, PMID 34601898). Nearly every cohort, criterion, and trial in this knowledge base codes cerebrovascular disease as present or absent, which averages a null group together with a rapidly declining one.
Design: Re-analysis of existing autopsy and imaging cohorts using combination profiles rather than binary coding, and prospective trial stratification on profile rather than presence.
Tier 2 additions¶
OQ-26 — What determines NOTCH3 penetrance?¶
Genetic prevalence of cysteine-altering NOTCH3 variants is 1 in 277 in 129,933 sequenced Korean adults, against a clinical CADASIL prevalence of roughly 2–11 per 100,000 (Cho 2025, PMID 40982447; Razvi 2005, PMID 15834040; Moreton 2014, PMID 24840674). Variant position explains part of it — high-risk EGFr domains carry stroke OR 10.81 (95% CI 5.46–21.37) versus low-risk — and the risk metric tracks NOTCH3 aggregation load (P=0.006) rather than signalling activity (P=0.88) (Hack 2023, PMID 36535904). What explains the remaining variance, and how incidentally-identified carriers should be counselled, is unresolved.
OQ-27 — Why does executive function alone not decline after stroke?¶
Beyond one year post-stroke, global cognition and every domain except executive function declined faster than in stroke-free controls (Lo 2022, PMID 34775838). Executive dysfunction is the canonical vascular signature, so either it is lesion-fixed while other domains reflect superimposed degeneration, or it is subject to a floor effect at entry. The two explanations imply different targets for treatment.
OQ-28 — By how much does invisible microinfarction bias imaging-based SVD burden?¶
Chronic cortical microinfarcts were present in 35% of SVD patients at baseline, and 21 acute microinfarcts detected by monthly DWI in 7 of 54 participants all disappeared on follow-up MRI (ter Telgte 2020, PMID 32065609). Autopsy-visible ischaemic injury therefore exceeds MRI-visible injury by an unquantified factor, which propagates into every marker-to-cognition association in this knowledge base (Huang 2024, PMID 37470314).
OQ-29 — Should enlarged perivascular spaces be weighted in burden scores?¶
Across five harmonized population cohorts (3,575 people), total enlarged-perivascular-space counts showed no association with MMSE or a general cognitive factor, unchanged by adjustment or by region-specific analysis (Hilal 2018, PMID 30068634) — yet >10 basal-ganglia perivascular spaces carried the strongest multivariable association with anticoagulant-related symptomatic ICH in CROMIS-2 (HR 8.96, 95% CI 2.41–33.4, on 14 events) (Best 2020, PMID 32934168). A marker that is cognitively inert and haemorrhagically informative should not be summed into a single "total SVD burden" score.
OQ-30 — Do the two independent genetic nominations of calcium-channel blockade justify a trial?¶
Drug-target Mendelian randomization nominates calcium-channel blockade for SVD imaging markers in two independent analyses (Koohi 2025, PMID 39661645; Ma 2024, PMID 38818733), an exploratory network meta-analysis ranks calcium-channel blockers highest for dementia prevention among the four adjudicated placebo-controlled trials (SUCRA 94.7%) (Ding 2026, PMID 41962914), and the largest antihypertensive dementia effect ever reported came from a nitrendipine-based regimen (Forette 2002, PMID 12374512). Four independent lines converge on one drug class and none is a randomized cognitive endpoint.
Tier 1 additions from the 2026-09-02 deepening pass¶
OQ-31 — Should exercise be trialled head-to-head against cholinesterase inhibitors in subcortical VCI?¶
Why open: The only two randomized trials that recruited subcortical vascular cognitive impairment specifically and tested exercise produced cognitive effects at least as large as the drugs. Six months of thrice-weekly progressive aerobic training gave ADAS-Cog −1.71 (95% CI −3.15 to −0.26, P=0.02) — larger than donepezil 5 mg (−0.92) and comparable to galantamine (−2.01) in the Cochrane pooling — with added gains in 6-minute walk distance and diastolic pressure and no gastrointestinal harm; the effect was gone six months after the programme stopped (Liu-Ambrose 2016, PMID 27760869; Battle 2021, PMID 33704781). Twelve months of progressive resistance training against an active balance-and-tone comparator gave a much smaller ADAS-Cog-Plus effect (−0.18, −0.35 to −0.01) that was present in women and absent in men, with a concurrent CRP reduction (Liu-Ambrose 2026, PMID 41795685). The AAN mild-cognitive-impairment guideline already recommends exercise (Level B) while permitting clinicians not to offer cholinesterase inhibitors (Level B) — and it is retired (Petersen 2018, PMID 29282327).
Design: Three-arm randomized trial in mild-to-moderate subcortical VCI — supervised aerobic training, donepezil 10 mg, and both — with ADAS-Cog and a function co-primary, 12-month intervention and 12-month post-cessation follow-up to test durability, prespecified sex-stratified analysis, and an adherence-cost endpoint.
OQ-32 — Does anticholinergic deprescribing change any outcome in vascular cognitive impairment?¶
Why open: Anticholinergic burden is prognostic at scale — dose-response for stroke or death in 39,107 people with dementia including the vascular and mixed strata specifically (ACB ≥2: HR 1.20, 95% CI 1.14–1.26) (Tan 2018, PMID 30056424), with a Cochrane prognostic-factor review covering 102,684 older adults and noting that burden scales agree poorly with one another (Taylor-Rowan 2022, PMID 35994403). The intervention evidence is three randomized trials totalling 299 participants, in cognitively mixed populations, with outcomes at one to three months and data unsuitable for meta-analysis (Taylor-Rowan 2023, PMID 38063254). A near-universal clinical recommendation rests on a 299-person evidence base.
Design: Pharmacist-led deprescribing randomized against usual care in an SVD-enriched cognitive population, powered for stroke-or-death rather than test scores, with a prespecified scale-comparison analysis so the result does not depend on which burden instrument was used.
OQ-33 — Which perivascular-space measurement, if any, carries prognostic information?¶
Why open: Three studies of the same STRIVE marker disagree in a structured way. Counted perivascular spaces at the top quartile in both basal ganglia and centrum semiovale predicted incident dementia over 8 years (OR 2.91, 95% CI 1.43–5.95) independent of WMH, microbleeds and lacunes (Paradise 2021, PMID 33504642); harmonized counts across five population cohorts (n=3,575) showed no cross-sectional cognitive association (Hilal 2018, PMID 30068634); and segmented PVS volume in 120 patients with lacunar stroke and confluent WMH predicted neither marker progression nor dementia, while the diffusion index intended to measure perivascular function predicted dementia strongly (HR 0.328, 0.183–0.588) but lost its longitudinal signal once conventional diffusion metrics were entered (Hong 2024, PMID 38465597). Meanwhile the same marker showed the strongest multivariable association with symptomatic ICH during anticoagulation (>10 basal-ganglia PVS: HR 8.96, 2.41–33.4) on 14 events (Best 2020, PMID 32934168).
Design: One cohort, four measurements — visual count, segmented volume, DTI-ALPS, and free water — against both cognitive decline and haemorrhage, with a neuropathological subsample.
Tier 2 additions from the 2026-09-02 deepening pass¶
OQ-34 — Why is incident vascular dementia six times higher in Hisayama than in Rotterdam or NEDICES?¶
Why open: Hisayama reports 9.5 incident vascular dementias per 1,000 person-years with 91.2% of cases morphologically evaluated and 164 autopsied, against 1.5 in Rotterdam and 1.4 in NEDICES (Matsui 2009, PMID 18977814; Ott 1998, PMID 9521184; Bermejo-Pareja 2008, PMID 17727890). Stronger ascertainment should have reduced, not raised, the vascular count. Candidate explanations — higher stroke incidence and hypertension prevalence, lower Alzheimer detection displacing cases, and the environmental gradient implied by Japanese-American migrants sitting between Japanese and European-ancestry rates (White 1996, PMID 8805729) — have never been decomposed. Hisayama's vascular incidence also flattened after 85 while Alzheimer incidence accelerated.
Design: Harmonized re-analysis of the surviving cohorts under one criteria set with common imaging definitions, decomposing the gap into ascertainment, competing-diagnosis, and exposure components.
OQ-35 — Is astroglial activation the earliest measurable vascular-cognitive signal?¶
Why open: In 5,270 UK Biobank participants aged 40–60, baseline plasma GFAP predicted WMH volume, fractional anisotropy and mean diffusivity nine years later and predicted 3-year progression of the diffusion markers, while plasma NfL predicted none of them (Prapiadou 2026, PMID 41461058). In established disease the ordering reverses: NfL level tracks SVD burden cross-sectionally and NfL rate of change predicts progression (Qu 2021, PMID 33517704). If astroglial activation genuinely precedes measurable white-matter injury by about a decade, GFAP is a risk-stratification analyte and NfL a progression-monitoring one, and they should not be compared head-to-head. But GFAP also rises with Alzheimer pathology and is the one plasma marker that tracks both (Dewenter 2026, PMID 42252508), so a decade-early GFAP signal may be reading a shared upstream process rather than a vascular one.
Design: Nested case-control within an existing midlife cohort with serial GFAP, NfL, p-tau217 and imaging, testing whether GFAP's predictive value for SVD markers survives adjustment for concurrent amyloid status.
OQ-36 — Are haemorrhagic and non-haemorrhagic CAA one disease with two criteria sets, or two diseases?¶
Why open: The two presentations differ genetically (APOE ε2 for siderosis and bleeding, OR 2.42, 95% CI 1.48–3.95; ε4 for the non-haemorrhagic amyloid-heavy form), pathologically (neurofibrillary tangles 87% vs 42%), and diagnostically — Boston v2.0 probable-CAA sensitivity falls from 74.5% at the autopsy standard to 28.6% (13.2–48.7) in non-haemorrhagic cognitive presentations (Charidimou 2019, PMID 31243071; Charidimou 2015, PMID 25716356; Switzer 2024, PMID 38710005). A criteria set with 74.5% and 28.6% sensitivity in two populations is either one instrument used outside its range or two diseases sharing a name. Iatrogenic CAA adds a third form with a defined seeding event and a 30–40-year latency, now reported in adults aged ≥65 where it is indistinguishable from sporadic disease except by surgical history (Panteleienko 2024, PMID 38511868; Banerjee 2019, PMID 30597599).
Design: Prospective memory-clinic cohort with susceptibility imaging, APOE genotyping, amyloid PET, and a systematic neurosurgical-history questionnaire, testing whether separate criteria for the non-haemorrhagic phenotype outperform one set.
Questions the deepening pass closed or downgraded¶
| Question | Status change | Basis |
|---|---|---|
| Does anticoagulation after intracranial haemorrhage do more harm than good? | Largely resolved for the ischaemic side: pooled randomized data show ischaemic events fall from 19% to 4% (HR 0.27, 0.13–0.56) with an uncertain haemorrhagic increase and unchanged death-or-dependence | Al-Shahi Salman 2023, PMID 37839434 |
| Are enlarged perivascular spaces cognitively meaningful in the general population? | Downgraded to probably not, on harmonized data from five cohorts | Hilal 2018, PMID 30068634 |
| Do general-population dementia risk models work in stroke survivors? | Closed: they do not (CAIDE AUC 0.53). A stroke-specific replacement now reaches C 0.70 under internal-external validation | Tang 2020, PMID 32568644; Lo 2026, PMID 41525568 |
| Does lipid lowering prevent dementia, or harm cognition? | Closed on both counts: OR 0.96 (0.74–1.26) across 15 trials and 139,169 participants | Reddin 2025, PMID 40794911 |
| Does lowering blood pressure prevent dementia? | Largely closed in the affirmative for all-cause dementia when the achieved contrast is large: RR 0.85 (0.76–0.95) on a 22.0 mm Hg systolic separation over 48 months in 33,995 adults. Open for the target, for people already treated, and for adjudicated vascular dementia | He 2025, PMID 40258956 |
| Are the criteria for vascular dementia irredeemably insensitive? | Downgraded: sensitivity 0.20–0.25 is a property of conjunctive criteria in hospital autopsy series; population-based disjunctive criteria reached 0.75 sensitivity and 0.81 specificity | Knopman 2003, PMID 12707071; Gold 2002, PMID 11772694 |
| Is there a registered phase-3 successor to LACI-2? | Closed: yes. IMPACT (NCT07252544), a 3,156-participant 2×2 factorial of isosorbide mononitrate and butylphthalide in acute lacunar stroke, was found on the 2026-09-02 registry sweep. The 2026-08-31 "none found" was a query-coverage artifact | ClinicalTrials.gov, 2026-09-02 |
| Does antihypertensive deprescribing reduce falls in older adults? | Closed in the negative: RR 1.00 (0.89–1.13) across four RCTs and 2,173 participants, with no change in mortality, MACE or serious adverse events either | Alsubaiei 2026, PMID 41550131 |
| Is vascular-dementia incidence sex-patterned? | Closed: no. Cumulative risk to age 95 is 0.04 in both sexes, against 0.22 vs 0.09 for Alzheimer disease | Andersen 1999, PMID 10599770 |
| Can PSMD and free water be measured reproducibly across sites? | Closed: yes — instrumental and biological validation complete; clinical validation is what remains | Maillard 2022, PMID 35382232; Luckey 2024, PMID 39569745 |
Dots not yet connected¶
| # | Dot A | Dot B | The missing junction | Powers |
|---|---|---|---|---|
| D1 | CADASIL presymptomatic imaging | sporadic SVD trial endpoints | validate the same marker trajectories across etiologies | OQ-7, OQ-16 |
| D2 | CAA Boston v2.0 burden | anti-amyloid ARIA monitoring | prospective shared vessel-risk model | OQ-6, OQ-20 |
| D3 | acute post-stroke cognition | secondary-prevention infrastructure | cognitive endpoint embedded in stroke prevention | OQ-3, OQ-4 |
| D4 | BBB leakage imaging | blood endothelial markers | temporally aligned multimodal signature | OQ-12, OQ-14 |
| D5 | Alzheimer biomarker positivity | cholinesterase response in VCI | biomarker-treatment interaction | OQ-5, OQ-11 |
| D6 | lesion-network mapping | individualized rehabilitation goals | anatomy-to-strategy assignment | OQ-8, OQ-15 |
| D7 | orthostatic safety monitoring | intensive BP cognitive prevention | dynamic net-benefit target | OQ-2 |
| D8 | caregiver-reported daily failures | trial cognitive endpoints | patient-important responder definition | OQ-10, OQ-18 |
| D9 | retinal microvasculature | brain SVD progression | accessible validated surrogate | OQ-7, OQ-14 |
| D10 | mixed-pathology autopsy | living multimodal biomarkers | calibrated attributable fractions | OQ-11, OQ-17 |
| D11 | drug-target Mendelian randomization nominating calcium-channel blockade (PMID 39661645, 38818733) | Syst-Eur's nitrendipine dementia result (PMID 12374512) | a randomized CCB-versus-other-class trial with a cognitive endpoint | OQ-30, OQ-2 |
| D12 | person-specific attributable fractions ranging 22–100% at autopsy (PMID 29244218) | plasma NfL/p-tau217/GFAP dissociation in living patients (PMID 42252508) | in-vivo estimation of individual attributable fractions | OQ-11, OQ-5 |
| D13 | MarkVCID markers responsive to intervention (PMID 39569745, 36523847) | completed BP trials with paired imaging and cognition (PMID 31408137, 42614618) | formal surrogate-validation mediation analysis | OQ-7, OQ-23 |
| D14 | cortical superficial siderosis as an ICH-recurrence marker (PMID 41197104, 38020625) | CAA's tau-mediated cognitive pathway (PMID 35759327) | one vessel-wall model predicting both bleeding and cognition | OQ-20, OQ-6 |
| D15 | rodent OPC-recruitment failure as the rate-limiting repair step (PMID 27663285, 36529961) | human perilesional WMH penumbra with reduced reactivity and perfusion (PMID 41656632) | a human remyelination-target experimental-medicine trial | OQ-12, OQ-7 |
| D16 | CADA-PRO patient-reported outcome validated in CADASIL (PMID 39234671) | sporadic SVD trials with no validated patient-reported measure | a sporadic-SVD PRO development study | OQ-18, OQ-8 |
| D17 | national health expenditure improving a post-stroke dementia model (PMID 41525568) | clinician-reported VCI service fragmentation (PMID 39240640) | health-system exposure as a modifiable dementia risk factor | OQ-10, OQ-3 |
| D18 | midlife vascular risk explaining 22–44% of dementia by 80 but 2–8% after 80 (PMID 40455489) | prevention trials that recruit at a mean age above 70 (PMID 18614402, 41870419) | an age-window-matched prevention trial recruiting under 65 | OQ-2, OQ-3 |
| D19 | aerobic exercise ADAS-Cog −1.71 in subcortical VCI (PMID 27760869) | cholinesterase-inhibitor ADAS-Cog −0.92 to −2.01 (PMID 33704781) | a head-to-head randomized comparison, never performed | OQ-31, OQ-18 |
| D20 | plasma GFAP predicting SVD markers nine years ahead (PMID 41461058) | the 40–60 age band where attributable fractions are highest (PMID 40455489) | biomarker-triggered midlife vascular prevention | OQ-35, OQ-2 |
| D21 | microinfarcts in 62% of vascular dementia brains and largely MRI-invisible (PMID 22234334, 32065609) | imaging-based SVD severity scales used for trial entry (PMID 40953349) | a pathology-calibrated imaging burden scale | OQ-7, OQ-28 |
| D22 | delirium predicting dementia independently of white-matter disease while infection predicts it only with white-matter disease (PMID 38310893) | delirium prevention interventions with no cognitive-endpoint follow-up | a delirium-prevention trial with 5-year dementia as an endpoint | OQ-3, OQ-4 |
| D23 | 15.5% incident covert ischaemic lesions at 6 months post-stroke (PMID 39417418) and 5.5% new covert infarcts in anticoagulated AF (PMID 35171989) | secondary-prevention trials using clinical stroke as the endpoint | covert infarction as a randomized-trial endpoint | OQ-3, OQ-6 |
| D24 | 32–37% cumulative dementia at 5 years after intracerebral haemorrhage (PMID 38640161) | the absence of any post-ICH trial with a cognitive endpoint | a cognitive-endpoint trial in ICH survivors | OQ-6, OQ-3 |
| D25 | tract-specific WMH outperforming total WMH volume for post-stroke cognition (PMID 38651756) | trials using whole-brain WMH volume as their imaging endpoint (PMID 31408137) | a tract-based imaging endpoint in an SVD trial | OQ-7, OQ-15 |
| D26 | cognitive reserve moderating the WMH–cognition slope independently of lesion volume (PMID 27951523) | SVD trials that stratify on lesion burden but not education | reserve-stratified randomization | OQ-7, OQ-18 |
| D27 | iatrogenic CAA with a defined seeding event and 30–40-year latency (PMID 30597599, 38511868) | sporadic CAA of unknown initiation (PMID 35841910) | a seeded-progression model of vascular amyloid deposition | OQ-36, OQ-20 |