Red flags and safety concerns

TL;DR — A vascular label must not end diagnostic thinking. Rapid progression, fluctuation, fever, seizures, headache, focal change, medication toxicity, hydrocephalus, subdural collection, inflammatory disease, and metabolic disturbance require targeted reassessment (Hermann 2022, PMID 35508635; Day 2022, PMID 35678409). Microbleeds and CAA complicate anticoagulation but do not create an automatic rule; ischaemic and haemorrhagic risks must both be estimated (Charidimou 2017, PMID 29117953). Intensive BP treatment can benefit selected populations while orthostatic hypotension increases falls and may impair cognition (Bhanu 2021, PMID 34752479; Iseli 2019, PMID 30825549). Driving and capacity are task-specific functional judgments, not consequences of a diagnosis alone.

Diagnostic red flags

Red flag	Urgent alternatives
hours–days decline	stroke, delirium, seizure, toxic-metabolic cause
weeks–months rapid progression	autoimmune, infection, prion, neoplasm, inflammatory CAA
new focal deficit	acute stroke/haemorrhage
headache plus cognitive change	haemorrhage, vasculitis, mass, inflammatory CAA
fever/systemic illness	infection/delirium
gait + incontinence + ventricular enlargement	hydrocephalus differential
falls/trauma on antithrombotics	subdural/ICH
marked fluctuation/hallucinations/parkinsonism	Lewy-body or medication differential

Rapidly progressive dementia reviews emphasize broad etiologic search because several causes are treatable and delay is consequential (Hermann 2022, PMID 35508635; Geschwind 2016, PMID 27042906).

The treatable mimic that looks like accelerated vascular dementia

CAA-related inflammation is the one condition on this list that presents as subacute cognitive decline with white-matter change in a patient who already has an MRI consistent with small-vessel disease, and that responds to immunosuppression. Pooling 21 cohort studies with 378 patients (mean age 71.5, 52% women), the presenting features were cognitive decline 70% (95% CI 54–84), focal neurological deficits 55% (40–70), encephalopathy 54% (39–68), seizures 37% (27–49) and headache 31% (22–42). Imaging was near-universal for T2/FLAIR-hyperintense white-matter lesions (98%, 93–100) and lobar microbleeds (96%, 92–99), with gadolinium-enhancing lesions in 54% (42–66), cortical superficial siderosis 51% (34–68) and lobar macrohaemorrhage 40% (11–73). Homozygous APOE ε4/ε4 was present in 34% (17–53) — far above population frequency, and the closest thing to a genetic flag (Theodorou 2023, PMID 36453271).

Diagnosis does not require biopsy. Against pathologically confirmed cases (17 CAA-ri, 37 non-inflammatory CAA controls), prespecified clinicoradiological criteria for probable CAA-ri — requiring asymmetric white-matter hyperintensity extending into subcortical white matter — gave sensitivity 82% and specificity 97%, whereas dropping the asymmetry requirement ("possible" CAA-ri) held sensitivity at 82% but collapsed specificity to 68%, because 11 of 16 non-inflammatory CAA controls without ICH met it (Auriel 2016, PMID 26720093). The asymmetry criterion is doing nearly all of the discriminating work; a symmetric leukoencephalopathy in a CAA patient is much more likely to be ordinary small-vessel disease. The concept has since been extended into proposed CAA-ri spectrum criteria (Charidimou 2025, PMID 39673239). Treatment response and prognosis are covered on the CAA page.

Reversible amplifiers

Contributor	Check
medication	anticholinergic/sedative load, dosing, interactions
mood	depression, anxiety, apathy
sleep	apnea, insomnia, circadian disruption
sensory	hearing and vision
metabolic	thyroid, B12, electrolytes, renal/hepatic status
infection/pain	delirium triggers
seizures	episodic confusion or decline

“Reversible” usually means a contributor can be treated, not that established infarction or SVD disappears.

Anticholinergic load is the amplifier with the most direct vascular evidence. In 39,107 people with dementia and no prior stroke in the Swedish Dementia Registry followed a mean 2.31 years, with drug exposure modelled time-varying and death treated as a competing risk, an Anticholinergic Cognitive Burden score of 1 gave HR 1.09 (95% CI 1.04–1.14) and a score of ≥2 gave HR 1.20 (1.14–1.26) for the composite of stroke or death versus non-users — a dose–response relationship that remained significant in the vascular dementia and mixed dementia strata specifically, not only in Alzheimer disease (Tan 2018, PMID 30056424). The outcome here is stroke and death, not test scores; the observational association motivates a deprescribing trial but does not establish deprescribing as a vascular intervention.

The prognostic case is broader than that single registry and the deprescribing case is much weaker. A Cochrane prognostic-factor review of 18 studies covering 102,684 older adults with pre-existing dementia or cognitive impairment assessed anticholinergic burden against further cognitive decline, neuropsychiatric disturbance, mortality, physical function and institutionalisation, noting explicitly that the available burden scales agree poorly with one another — which is why a "high anticholinergic burden" finding is partly a statement about which scale was used (Taylor-Rowan 2022, PMID 35994403). The intervention side is nearly empty: a companion Cochrane review of randomized deprescribing trials with cognitive outcomes found three trials totalling 299 participants, all in cognitively mixed populations, with outcomes at one to three months, data unsuitable for meta-analysis, and only a single reported improvement (Digit Symbol Substitution Test) (Taylor-Rowan 2023, PMID 38063254). Reducing anticholinergic load is therefore a well-founded prognostic inference and an almost untested intervention — the same structure as most of the vascular-risk recommendations on the prevention page.

Anticoagulation and microbleeds

Microbleeds predict increased ICH risk during anticoagulation after stroke/AF, especially with higher burden and CAA patterns, but anticoagulation prevents cardioembolic stroke (Charidimou 2017, PMID 29117953; Zhao 2024, PMID 38560733).

Factor	Haemorrhage concern	Ischaemic concern
prior event	lobar ICH	embolic stroke
imaging	lobar microbleeds/siderosis	infarcts
BP	uncontrolled hypertension	undertreated vascular risk
indication	weak/uncertain	high-risk AF
function	falls/poor adherence	inability to manage monitoring

In 1,552 AF-stroke patients, CMB presence raised ICH odds (OR 2.68, 1.19–6.01) and ≥5 CMBs raised them further (OR 5.50, 2.07–14.66); annual ICH incidence rose from 0.30% without CMBs to 0.81% with any CMB and 2.48% with ≥5 (Charidimou 2017, PMID 29117953). TFNEs in CAA can mimic TIA and herald lobar ICH, so antithrombotics given for presumed TIA can be harmful (Smith 2021, PMID 34016709). Do not use an MRI marker as a stand-alone prohibition; document the competing risks and uncertainty.

The absolute risks are smaller than the hazard ratios suggest

CROMIS-2 prospectively followed 1,447 people with AF and recent ischaemic stroke or TIA who started anticoagulation, over 3,366 patient-years. Symptomatic intracranial haemorrhage occurred at 9.8 per 1,000 patient-years (95% CI 4.0–20.3) with microbleeds versus 2.6 per 1,000 patient-years (1.1–5.4) without (adjusted HR 3.67, 95% CI 1.27–10.60) (Wilson 2018, PMID 29778365). A tripled hazard on a baseline of roughly 0.3% per year is still well under 1% per year — for most patients an order of magnitude below the ischaemic stroke risk anticoagulation prevents. The same study exposes how poorly the standard bleeding score performs in this setting: HAS-BLED alone had a C-index of 0.41 (95% CI 0.29–0.53), i.e. worse than chance, rising to 0.66 (0.53–0.80) when microbleeds were added and 0.74 (0.60–0.88) with microbleeds, diabetes, and anticoagulant type (Wilson 2018, PMID 29778365). Using HAS-BLED to justify withholding anticoagulation in this population is not defensible on its own performance.

A post-hoc CROMIS-2 analysis found that basal-ganglia perivascular spaces — a marker with no cognitive correlate in population cohorts (see SVD) — carried the strongest multivariable association with symptomatic ICH: >10 BGPVS gave HR 8.96 (95% CI 2.41–33.4, P=0.001), alongside diabetes (HR 3.91, 95% CI 1.34–11.4); centrum-semiovale perivascular spaces did not (Best 2020, PMID 32934168). Only 14 symptomatic ICH events occurred, so the interval is wide and the finding is hypothesis-generating.

After an intracerebral haemorrhage: what the randomized evidence actually shows

Trial / synthesis	Design	Result
SoSTART (PMID 34487722)	203 randomized at a median 115 days after intracranial haemorrhage, CHA₂DS₂-VASc ≥2, 67 UK hospitals, start vs avoid oral anticoagulation, median 1.2 y follow-up	Non-inferiority not established: recurrent intracranial haemorrhage in 8/101 (8%) starting vs 4/102 (4%) avoiding (adjusted HR 2.42, 95% CI 0.72–8.09; P=0.152). Deaths 22 (22%) vs 11 (11%). Recurrence rates were lower than the trial had assumed
COCROACH IPD meta-analysis (PMID 37839434)	412 participants pooled from SoSTART, APACHE-AF, NASPAF-ICH, and an ELDERCARE-AF subgroup; 75% aged ≥75	Any stroke or cardiovascular death 29/212 (14%) starting vs 43/200 (22%) avoiding (pooled HR 0.68, 95% CI 0.42–1.10; I²=0%). Ischaemic major adverse cardiovascular events fell sharply: 9/212 (4%) vs 38/200 (19%), HR 0.27 (95% CI 0.13–0.56). Haemorrhagic events 15/212 (7%) vs 9/200 (5%), HR 1.80 (95% CI 0.77–4.21). All-cause death HR 1.29 (0.78–2.11); death or dependence at 1 year OR 1.12 (0.70–1.79)

The pattern is consistent and clinically usable even though the primary outcomes were inconclusive: anticoagulation after intracranial haemorrhage clearly reduces ischaemic events (a 15-percentage-point absolute reduction in the pooled data) while the increase in haemorrhagic events is smaller in absolute terms and statistically uncertain, and net functional outcome is unchanged. What no trial has resolved is the subgroup question that matters most here — whether probable CAA, lobar haemorrhage, or heavy microbleed burden reverses that balance — because the completed trials were not powered for it and the pooled analysis found no subgroup effect it could trust (Al-Shahi Salman 2023, PMID 37839434). Ongoing trials are due to complete in 2028.

ARIA during anti-amyloid therapy is a related safety phenotype: ARIA-E occurred in 12.6% on lecanemab and 24.0% (52/205 events symptomatic) on donanemab in phase 3 Alzheimer trials (van Dyck 2023, PMID 36449413; Sims 2023, PMID 37459141).

Blood pressure and orthostasis

BP lowering modestly reduces cognitive outcomes in trials, but drug-induced orthostatic hypotension and symptomatic cerebral hypoperfusion can create falls, syncope, renal injury, and functional loss (Hughes 2020, PMID 32427305; Bhanu 2021, PMID 34752479). Orthostatic hypotension is associated with poorer cognition in meta-analysis, although direction of causality is uncertain (Iseli 2019, PMID 30825549).

The reverse manoeuvre — stopping antihypertensives in frail older adults — has now been examined randomly. Pooling four RCTs with 2,173 participants comparing antihypertensive deprescribing with usual care, there was no statistically significant difference in all-cause mortality (RR 1.02, 95% CI 0.93–1.12), cardiovascular mortality (RR 1.11, 0.80–1.55), all-cause hospitalisation (RR 0.95, 0.85–1.05), major adverse cardiovascular events (RR 1.09, 0.90–1.33), stroke (RR 1.12, 0.66–1.89), serious adverse events (RR 1.08, 0.90–1.30) or falls (RR 1.00, 0.89–1.13) (Alsubaiei 2026, PMID 41550131). The confidence intervals and evidence base do not establish equivalence or absence of harm; the trials also detected no reduction in falls. Neither deprescribing nor intensification trials measured cognition as a primary outcome, so the U-shape described above remains unresolved from both directions.

Safety monitoring includes seated/standing BP, symptoms, falls, hydration, renal function, medication timing, and intercurrent illness.

Cognitive drugs

Donepezil 10 mg and galantamine increased adverse events in pooled VCI trials (OR 1.95 and 1.57); bradycardia, syncope, GI effects, weight loss, and interactions matter in vascular populations (Battle 2021, PMID 33704781). Atypical antipsychotics increase all-cause death versus placebo in dementia trials (118/3,353 [3.5%] vs 40/1,757 [2.3%]; OR 1.54, 95% CI 1.06–2.23) over generally 10–12 weeks (Schneider 2005, PMID 16234500). Pooled risperidone/olanzapine dementia RCTs also raised cerebrovascular-event reporting (Herrmann 2005, PMID 15697324). A treatment trial requires a defined benefit target and stopping rule.

Driving

Driving depends on attention, speed, visuospatial skill, judgment, motor control, vision, and insight. Consensus statements support individualized assessment rather than a diagnosis-only decision (Wheatley 2014, PMID 24754761).

Office cognitive tests predict on-road performance, but weakly. Meta-analysis of 30 reviewed and 16 pooled studies found statistically significant but weak-to-moderate associations (r ≈ 0.3–0.4) between cognitive measures and on-road driving outcome for every cognitive domain except language, in patients with mild dementia, and concluded that a composite battery outperforms any single domain test (Rashid 2020, PMID 31018100). An r of 0.3–0.4 explains roughly 10–16% of variance: enough to justify referral for on-road assessment, nowhere near enough to license or revoke on a test score. This is the quantitative basis for the individualized-assessment position rather than a threshold rule — and it applies to mild dementia broadly, with no vascular-specific estimate available.

Trigger for reassessment	Example
crash/near miss	new safety event
getting lost	navigation failure
slowed responses	processing/motor issue
family concern	observed change
recurrent stroke/seizure	new medical risk
medication change	sedation or hypotension

Local legal reporting obligations vary and must be checked in the relevant jurisdiction.

Capacity and safeguarding

Capacity is decision-specific and can fluctuate. Evaluate understanding, appreciation, reasoning, and ability to communicate a choice, with supported communication after stroke. Financial vulnerability, medication errors, cooking hazards, wandering, firearms, and caregiver stress require proportionate safeguards. Global cognitive scores do not independently determine capacity (Barbas 2001, PMID 11794448).

Financial capacity deserves separate treatment because impairment can create exploitation risk. The research literature on financial exploitation of older adults is described by its own reviewers as only beginning to emerge, with empirical work and clinical guidelines lagging the scale of the problem and structural barriers having limited interventions (Wood 2017, PMID 28452630). For a condition characterised by executive and processing-speed impairment with relatively preserved social presentation, this is a safety domain where the cognitive profile can mislead observers: conversational fluency or an adequate screen does not by itself establish capacity to detect financial manipulation.

Structured instruments help but do not settle it. A systematic review of tools published 2000–2017 identified nine meeting UK legal requirements, varying widely in design and in available reliability and validity data; some can be tailored to a specific decision, others to a patient group. The review's central conclusion is that no gold standard for capacity assessment exists — which is itself why comparative evaluation of approaches is impossible — and that the most robust available practice is a trained assessor's opinion combined with a structured tool, attentive to personal, cultural and disease-specific factors (Pennington 2018, PMID 30010696). After stroke, the disease-specific factor is usually communication: an aphasic patient may retain capacity that no standard instrument can demonstrate (see assessment).

Escalation after change

1. Define onset and trajectory.
2. Check vital signs, glucose, oxygenation, infection, and medication changes.
3. Perform neurological examination.
4. Obtain urgent imaging for focal/acute/head-trauma patterns.
5. Evaluate seizure and inflammatory/infectious causes where indicated.
6. Revisit the original etiologic formulation.

Open questions

• If anticholinergic burden is prognostic across 102,684 people but deprescribing has been tested in only 299, is the recommendation to deprescribe evidence-based? (Taylor-Rowan 2022, PMID 35994403; Taylor-Rowan 2023, PMID 38063254)
• Antihypertensive deprescribing did not reduce falls (RR 1.00) — what is its remaining rationale in frail vascular cognitive impairment? (Alsubaiei 2026, PMID 41550131)
• Should financial-capacity assessment be routine rather than reactive in executive-predominant vascular cognitive impairment? (Wood 2017, PMID 28452630)
• Which microbleed/CAA model best predicts net benefit from anticoagulation? (Charidimou 2017, PMID 29117953)
• What BP de-intensification criteria prevent orthostatic harm without increasing stroke? (Bhanu 2021, PMID 34752479)
• Which driving assessments predict real-world safety in VCI? (Wheatley 2014, PMID 24754761)
• How often is rapid decline misclassified as progression of vascular dementia? (Hermann 2022, PMID 35508635)
• If HAS-BLED discriminates worse than chance (C-index 0.41) in anticoagulated AF-stroke patients, what should replace it? (Wilson 2018, PMID 29778365)
• Are basal-ganglia perivascular spaces a real ICH risk marker (HR 8.96 on 14 events) or a small-numbers artifact? (Best 2020, PMID 32934168)
• Does probable CAA or heavy lobar microbleed burden reverse the favourable ischaemic/haemorrhagic balance seen after ICH overall? (Al-Shahi Salman 2023, PMID 37839434; SoSTART 2021, PMID 34487722)
• How often is CAA-related inflammation missed as "rapid progression of vascular dementia", given that 70% present with cognitive decline? (Theodorou 2023, PMID 36453271)
• Should APOE genotyping be part of the workup for subacute cognitive decline with lobar microbleeds, given the 34% ε4/ε4 rate in CAA-ri? (Theodorou 2023, PMID 36453271)
• Would deprescribing anticholinergics reduce stroke and death in vascular dementia, as the dose–response cohort association predicts? (Tan 2018, PMID 30056424)
• If cognitive tests explain only ~10–16% of variance in on-road driving, what would a defensible fitness-to-drive pathway for VCI actually contain? (Rashid 2020, PMID 31018100)
• Can capacity assessment be evaluated at all without a gold standard, and what would serve as one? (Pennington 2018, PMID 30010696)

Related pages

• CAA — haemorrhage risk.
• Prevention — BP/AF evidence.
• Treatment — adverse effects.
• Patient experience — autonomy and burden.

References

1. Hermann P, et al. Rapidly progressive dementias: aetiologies, diagnosis and management. Nat Rev Neurol. 2022. PMID 35508635
2. Day GS. Rapidly progressive dementia. Continuum. 2022. PMID 35678409
3. Geschwind MD. Rapidly progressive dementia. Continuum. 2016. PMID 27042906
4. Charidimou A, et al. Microbleeds, anticoagulation, hemorrhage risk. Neurology. 2017. PMID 29117953
5. Zhao B, et al. Anticoagulation after microbleeds and stroke. Front Neurol. 2024. PMID 38560733
6. Hughes D, et al. BP lowering and cognitive outcomes. JAMA. 2020. PMID 32427305
7. Bhanu C, et al. Drug-induced orthostatic hypotension. PLoS Med. 2021. PMID 34752479
8. Iseli R, et al. Orthostatic hypotension and cognition. Exp Gerontol. 2019. PMID 30825549
9. Battle CE, et al. Cholinesterase inhibitors for VCI. Cochrane Database Syst Rev. 2021. PMID 33704781
10. Wheatley CJ, et al. Consensus statements on driving for persons with dementia. Occup Ther Health Care. 2014. PMID 24754761
11. Barbas NR, Wilde EA. Competency issues in dementia: medical decision making, driving, and independent living. J Geriatr Psychiatry Neurol. 2001. PMID 11794448
12. Schneider LS, et al. Risk of death with atypical antipsychotic drug treatment for dementia. JAMA. 2005;294:1934-43. PMID 16234500
13. Herrmann N, Lanctôt KL. Do atypical antipsychotics cause stroke? CNS Drugs. 2005;19:91-103. PMID 15697324
14. Smith EE, et al. CAA-related transient focal neurologic episodes. Neurology. 2021;97:231-238. PMID 34016709
15. van Dyck CH, et al. Lecanemab in early Alzheimer's disease. N Engl J Med. 2023;388:9-21. PMID 36449413
16. Sims JR, et al. Donanemab in early symptomatic Alzheimer disease. JAMA. 2023;330:512-527. PMID 37459141
17. Wilson D, et al. Cerebral microbleeds and intracranial haemorrhage risk in patients anticoagulated for atrial fibrillation after acute ischaemic stroke or transient ischaemic attack (CROMIS-2). Lancet Neurol. 2018;17:539-547. PMID 29778365
18. Best JG, et al. Association of enlarged perivascular spaces and anticoagulant-related intracranial hemorrhage. Neurology. 2020;95:e2192-e2199. PMID 32934168
19. SoSTART Collaboration. Effects of oral anticoagulation for atrial fibrillation after spontaneous intracranial haemorrhage in the UK (SoSTART). Lancet Neurol. 2021;20:842-853. PMID 34487722
20. Al-Shahi Salman R, et al. Effects of oral anticoagulation in people with atrial fibrillation after spontaneous intracranial haemorrhage (COCROACH): prospective individual participant data meta-analysis. Lancet Neurol. 2023;22:1140-1149. PMID 37839434
21. Theodorou A, et al. Clinical, neuroimaging, and genetic markers in cerebral amyloid angiopathy-related inflammation: a systematic review and meta-analysis. Stroke. 2023;54:178-188. PMID 36453271
22. Auriel E, et al. Validation of clinicoradiological criteria for the diagnosis of cerebral amyloid angiopathy-related inflammation. JAMA Neurol. 2016;73:197-202. PMID 26720093
23. Charidimou A. Cerebral amyloid angiopathy-related inflammation spectrum disorders: introduction of a novel concept and diagnostic criteria. Ann Neurol. 2025;97:470-474. PMID 39673239
24. Tan ECK, et al. Anticholinergic burden and risk of stroke and death in people with different types of dementia. J Alzheimers Dis. 2018;65:589-596. PMID 30056424
25. Rashid R, et al. Systematic review and meta-analysis of association between cognitive tests and on-road driving ability in people with dementia. Neuropsychol Rehabil. 2020;30:1720-1761. PMID 31018100
26. Pennington C, et al. Tools for testing decision-making capacity in dementia. Age Ageing. 2018;47:778-784. PMID 30010696
27. Taylor-Rowan M, et al. Anticholinergic burden for prediction of cognitive decline or neuropsychiatric symptoms in older adults with mild cognitive impairment or dementia. Cochrane Database Syst Rev. 2022;8:CD015196. PMID 35994403
28. Taylor-Rowan M, et al. Anticholinergic deprescribing interventions for reducing risk of cognitive decline or dementia in older adults with and without prior cognitive impairment. Cochrane Database Syst Rev. 2023;12:CD015405. PMID 38063254
29. Alsubaiei AA, et al. Efficacy and safety of antihypertensive drugs deprescribing in older adults: a systematic review and meta-analysis of randomized controlled trials. Int J Cardiol Cardiovasc Risk Prev. 2026;28:200570. PMID 41550131
30. Wood S, Lichtenberg PA. Financial capacity and financial exploitation of older adults: research findings, policy recommendations and clinical implications. Clin Gerontol. 2017;40:3-13. PMID 28452630