Skip to content

Post-stroke cognitive impairment

TL;DR — Cognitive impairment after stroke is common, heterogeneous, and dynamic: it can predate stroke, appear acutely, improve, persist, or progress (Pendlebury 2009, PMID 19782001). Hospital studies estimate any post-stroke neurocognitive disorder at 53.4%, mild disorder at 36.4%, and major disorder at 16.5%, but timing and thresholds materially change rates (Barbay 2018, PMID 30504699). Baseline cognitive impairment is the strongest pooled predictor of later post-stroke dementia (RR 3.10, 95% CI 2.77–3.47) and of later post-stroke cognitive impairment (RR 2.00, 95% CI 1.66–2.40) (Filler 2024, PMID 38101426). Screening must be accessible to aphasia and motor disability and tied to action; a one-time score during delirium is not a dementia diagnosis.

Trajectory model

Trajectory Candidate explanation
impairment before stroke prior SVD, silent infarcts, neurodegeneration
acute deficit then recovery focal injury plus resolving network dysfunction
persistent stable deficit established lesion burden
delayed decline recurrent stroke, progressive SVD, mixed pathology
fluctuation delirium, seizures, illness, medication, mood

Quantitative anchors

Estimate Value Source
pre-stroke dementia, population cohorts 9.1% (6.9–11.3) Pendlebury 2009, PMID 19782001
pre-stroke dementia, hospital cohorts 14.4% (12.0–16.8) same
early dementia after first stroke, pre-stroke dementia excluded 7.4% (4.8–10.0) same
dementia after recurrent stroke with pre-stroke cases included 41.3% (29.6–53.1) same
any post-stroke NCD 53.4% (46.9–59.8) Barbay 2018, PMID 30504699
mild post-stroke NCD 36.4% (29.0–43.8) same
major post-stroke NCD 16.5% (12.1–20.8) same

What the trajectory actually looks like

The Stroke and Cognition (STROKOG) consortium has quantified the shape of the post-stroke cognitive curve using individual participant data, and the answer contradicts both the "steady decline" and the "recover and plateau" pictures.

Year one is dominated by heterogeneity, not by mean change. Latent-class growth analysis of 1,149 patients from nine hospital cohorts (63% male, mean age 66.4) using global cognition at ~3.6 months and at 1 year identified three trajectory groups: low-performance (−3.27 SD, 17%), medium-performance (−1.23 SD, 48%), and high-performance (+0.71 SD, 35%). Only the high-performance group changed significantly, and it improved (0.22 SD/year, 95% CI 0.07–0.36); the low- and medium-performance groups were flat (−0.10 SD/year, 95% CI −0.33 to 0.13; and 0.11 SD/year, 95% CI −0.08 to 0.24) (Lo 2023, PMID 37072222). Membership in the low- versus high-performance group was predicted by age (RRR 1.18 per year, 95% CI 1.14–1.23), education (RRR 0.61 per year, 95% CI 0.56–0.67), diabetes (RRR 3.78, 95% CI 2.08–6.88), large-artery versus small-vessel stroke (RRR 2.77, 95% CI 1.32–5.83), and moderate/severe stroke (RRR 3.17, 95% CI 1.42–7.08). The practical finding is deflationary: the trajectory groups predicted cognition at ~3.2 years no better than the single ~3.6-month score did.

After year one, decline begins and exceeds normal ageing. Restricting analysis to beyond 1 year post-stroke in 1,488 patients (98% ischaemic) followed a median 2.68 years, global cognition declined at −0.053 SD/year (95% CI −0.073 to −0.033) after adjustment for age, sex, education, vascular risk factors, and stroke characteristics — in every domain except executive function — and significantly faster than stroke-free controls (difference −0.078 SD/year, 95% CI −0.11 to −0.045). Recurrent stroke and older age accelerated the decline (Lo 2022, PMID 34775838). The exemption of executive function is unexplained and sits awkwardly with the standard claim that executive dysfunction is the signature of vascular cognitive impairment; one reading is a floor effect (executive scores were already low at entry), another is that executive deficits are lesion-driven and static while the declining domains reflect ongoing degeneration.

Taken together: post-stroke cognition improves for about a year, then declines faster than ageing, and the level reached at 3–6 months carries most of the prognostic information.

Where the infarct is

Location is the predictor most often invoked and least often quantified. The Meta VCI Map consortium pooled individual patient data from 12 acute ischaemic stroke cohorts — 2,950 patients, mean age 66.8, of whom 1,286 (43.6%) had PSCI — with 86.9% lesion coverage of the brain, and produced the first comprehensive voxel-wise map. Infarcts in the left frontotemporal lobes, left thalamus and right parietal lobe carried voxel-wise odds ratios exceeding 20 for PSCI (FDR q<0.01). A 5-point "location impact score" derived from the map predicted PSCI on leave-one-cohort-out cross-validation, and — the result with the most practical weight — the location score alone performed similarly to a multivariable model that added age, sex, education, assessment interval, stroke history and total infarct volume. Clinicians rating the score visually reproduced it well (weighted κ 0.88–0.92 against the computed score, 0.85–0.87 between raters, 0.95 within a rater) (Weaver 2021, PMID 33901427). A predictor that needs only the scan and can be read by eye is unusual in this field.

The same method exposes an instrument problem. Mapping MMSE impairment (<5th percentile) in 1,198 patients, the significant clusters lay almost entirely in the left middle cerebral artery territory and thalamus (highest ORs >15 in thalamus and superior temporal gyrus), whereas domain-specific impairments on full neuropsychological testing mapped across both hemispheres — left medial temporal lobe for verbal memory, right parietal lobe for visuospatial function (Weaver 2021, PMID 33472574). MMSE does not sample the brain evenly; a right-hemisphere stroke can leave it intact while producing real deficits, which is a mechanistic explanation for the pooled accuracy figures in assessment.

Background white-matter burden acts independently of the infarct. In 1,568 patients from 9 cohorts with MRI and multidomain testing within 15 months of stroke, WMH volume showed a dose-dependent inverse association with every cognitive domain — coefficients from −0.09 (SE 0.04, P=0.01) for verbal memory to −0.19 (SE 0.03, P<0.001) for attention and executive function — independent of acute infarct volume, lacunes, old infarcts, and largely independent of infarct type (de Kort 2023, PMID 37901947). The infarct and the pre-existing small-vessel disease are therefore additive contributors, not competing explanations.

Infarct composition matters beyond volume. In 1,026 ESCAPE-NA1 participants with visible infarcts at 24 hours, total infarct volume predicted worse 90-day MoCA (adjusted common OR 1.05 per 10 mL, 95% CI 1.04–1.06), and after adjusting for it, mixed grey-and-white-matter involvement (acOR 1.92, 1.37–2.69), white-matter infarct volume (acOR 1.36 per 10 mL, 1.18–1.58) and a territorial rather than scattered pattern (acOR 1.65, 1.15–2.38) each independently predicted worse cognition (Ospel 2024, PMID 38511330). White matter carries disproportionate cognitive cost per millilitre — the same principle that makes small-vessel disease cognitively expensive at low lesion volumes.

Delirium as an acute marker and possible mechanism

Delirium is the most common acute confounder of post-stroke cognitive assessment and also a candidate contributor. In 527 consecutive stroke-unit patients screened with the Confusion Assessment Method during the first week, 62 (11.8%) became delirious; independent risk factors were pre-existing cognitive decline (IQCODE >50: OR 2.6, 95% CI 1.2–5.7), infection (OR 3.4, 1.7–6.8), right-hemisphere stroke (OR 2.0, 1.0–3.0), anterior-circulation large-vessel stroke (OR 3.4, 1.1–10.2), highest NIHSS tertile (OR 15.1, 3.3–69.0) and brain atrophy (OR 2.7, 1.1–6.8) (Oldenbeuving 2011, PMID 21307355). In 263 consecutive ischaemic stroke patients aged 55–85 followed for 10 years, delirium in the first 7 days occurred in 50 (19.0%) and was associated with dementia diagnosed at 3 months; median survival was 6.1 versus 9.1 years, but in Cox regression age (HR 1.08) and stroke severity (HR 1.83), not delirium itself, predicted mortality (Melkas 2012, PMID 21560162). Whether delirium injures the brain or merely reveals a brain already vulnerable is unresolved by these data — pre-existing cognitive decline and atrophy being risk factors argues for revelation, while the association with 3-month dementia after adjustment leaves room for contribution.

A population-based design has since separated the two exposures and produced the most discriminating result available. Among 1,369 OXVASC patients with TIA or minor stroke (mean age 72, 364 with moderate/severe white-matter disease on baseline imaging) followed 5 years, 209 (15%) developed dementia. Hospitalisation without delirium or infection did not predict dementia at all (HR 1.01, 95% CI 0.86–1.20) — which removes admission itself, and the illness severity that prompts it, as the explanation. Hospitalisation with delirium predicted dementia independently of infection in patients both with and without white-matter disease (HR 2.64, 1.47–4.74 and 3.41, 1.91–6.09), and the association was strongest in those with unimpaired baseline cognition (HR 4.01, 2.23–7.19 and 3.94, 1.95–7.93). Hospitalisation with infection, by contrast, predicted dementia only in patients with moderate or severe white-matter disease (HR 1.75, 1.04–2.94) (Pendlebury 2024, PMID 38310893).

The dissociation is the finding. Delirium's association is independent of pre-existing white-matter burden and largest in those who were cognitively normal — the opposite of what pure revelation-of-vulnerability predicts. Infection's association is entirely conditional on white-matter disease, which is what a vulnerability model predicts. Two different mechanisms are therefore likely operating under what is usually described as one "acute illness increases dementia risk" association.

Covert lesions accumulate after the index stroke

Recurrent clinical stroke is the usual explanation for delayed decline, but most new ischaemic lesions are silent. In a prospective multicentre study of 503 acute stroke patients with registered baseline and 6-month MRI followed to 36 months, 78 (15.5%) had incident ischaemic lesions at 6 months, 72% DWI-positive and 91% clinically covert. Older age and baseline small-vessel-disease burden were the risk factors; the lesions were associated with worse global cognition (β −0.31, 95% CI −0.48 to −0.14), worse modified Rankin score (β 0.36, 0.14–0.58) and a 3.81-fold higher recurrent-stroke hazard (95% CI 1.35–10.69) — and they partially mediated the relationship between small-vessel disease and poor cognition (Fang 2024, PMID 39417418). Incident lesions therefore occurred in about one in six patients and were clinically covert in 91%, the same phenomenon documented in anticoagulated atrial fibrillation.

Haemorrhagic stroke is not the same trajectory

Nearly all of the above is derived from ischaemic cohorts. After spontaneous intracerebral haemorrhage the cumulative dementia incidence is 32.0–37.4% at 5 years, and the proposed mechanisms split into a triggering role for the haemorrhage itself (primary and secondary brain injury) and a contributory role for pre-existing small-vessel and neurodegenerative pathology, reduced reserve, and genetic predisposition, converging on neurovascular-unit dysfunction and disrupted functional connectivity (Zhang 2024, PMID 38640161). The same review is explicit that data and standardized reporting on incident post-ICH dementia remain sparse. Cognitive outcomes have nevertheless been measured after ICH: a post hoc analysis of 291 iDEF trial participants used MoCA at days 7, 30 and 90, with 134/205 (65%) of those assessed at day 90 remaining below 26 (Lioutas 2025, PMID 38583421). Registered intervention studies also include cognition as a secondary outcome or substudy endpoint (PROHIBIT-ICH, NCT03863665; TRIDENT Cognitive Sub-Study, NCT03785067), although neither establishes a primary clinical dementia-prevention endpoint.

A nationwide propensity-matched cohort supports the ordering at the population level. Among 8,236 Taiwanese stroke patients without prior dementia matched 1:1 to controls, adjusted dementia hazard was 1.87 at 5 years and 1.53 at 10; by subtype, intracerebral haemorrhage carried the highest 5-year hazard (aHR 2.14), ahead of TIA (1.92) and ischaemic stroke (1.81), with the three converging by 10 years (1.61, 1.61, 1.49). Low insurance level and region of residence were independent risk factors (Li 2019, PMID 31267646). That socioeconomic terms enter a post-stroke dementia model in a single-payer system is the same signal that puts national health expenditure into the Lo prediction model below.

Risk prediction: the general-population models fail here

Three dementia risk models developed in general populations were applied to four harmonized stroke cohorts (Hong Kong, USA, Netherlands, France; ~12–18 months follow-up). Discrimination collapsed: the ANU Alzheimer Disease Risk Index reached C = 0.66, the Brief Dementia Screening Indicator C = 0.61, and the CAIDE score AUC = 0.53 — barely better than chance (Tang 2020, PMID 32568644). CAIDE, notable because it is the enrichment instrument used to recruit FINGER and its replications (see prevention), carries essentially no information about who will develop dementia after a stroke.

A stroke-specific replacement now exists. Pooling 2,663 participants from 12 studies across 10 countries followed a median 2.0 years, 655 developed dementia (8.7 per 100 person-years); a Fine-Gray model treating death as a competing risk and using age, sex, education, previous stroke, diabetes, stroke severity, two interactions (age × sex, age × severity), and national current health expenditure achieved C = 0.81 (95% CI 0.75–0.87) in development, falling to a pooled C = 0.70 (95% CI 0.67–0.73) under internal-external cross-validation — 0.79 in post-2010 cohorts, 0.74 in European cohorts, with risk overestimated in Asian cohorts (Lo 2026, PMID 41525568). That a national health-expenditure term earns a place in a biological prediction model is itself a finding: post-stroke dementia risk is partly a property of the health system, not only of the patient.

Imaging-derived prediction is at an earlier stage but is being validated across centres: T1-weighted texture features (kurtosis and inverse difference moment) in the hippocampus and entorhinal cortex distinguished impaired from unimpaired patients across four STROKOG datasets (n=327), with a random-forest model reaching ~90% accuracy in the training base and ~77% in external validation (Betrouni 2022, PMID 35862196). The drop from 90% to 77% on unseen scanners is the transportability problem in miniature.

Predictors

Filler et al. synthesized 89 studies/160,783 patients. Baseline cognitive impairment had RR 3.10 (2.77–3.47) for later post-stroke dementia and RR 2.00 (1.66–2.40) for later post-stroke cognitive impairment; diabetes, atrial fibrillation, and WMH were among potentially treatable risk factors (Filler 2024, PMID 38101426).

Predictor group Examples Interpretation
premorbid age, education, prior decline reserve and antecedent disease
brain imaging WMH, atrophy, prior infarcts vulnerability/load
stroke severity, location, recurrence direct injury
vascular diabetes, AF prevention targets and disease markers
acute complications delirium, infection may cause and reveal vulnerability

Diabetes (RR 1.29, 1.14–1.45), AF (RR 1.29, 1.04–1.60), and moderate/severe WMH (RR 1.51, 1.20–1.91) predicted later post-stroke cognitive impairment independent of age and stroke severity; corresponding dementia RRs were 1.38 (1.10–1.72) for diabetes and 1.55 (1.01–2.38) for moderate/severe WMH (Filler 2024, PMID 38101426). In SPS3, nearly half of recent lacunar-stroke participants already had MCI at baseline, including many with minimal physical disability (Jacova 2012, PMID 23034910). SPS3 cognitive follow-up found no CASI effect of dual antiplatelet therapy or of a <130 vs 130–149 mm Hg systolic target over median 3.0 years (Pearce 2014, PMID 25453457). PROGRESS reduced dementia and cognitive decline mainly when those outcomes accompanied recurrent stroke (Tzourio 2003, PMID 12742805).

Prediction rules remain heterogeneous and often lack external validation (Drozdowska 2021, PMID 33817331).

Screening timeline

Time Purpose Caution
acute admission detect major deficits/delirium and plan safety unstable state
discharge/early rehab support communication and rehabilitation goals fatigue/aphasia
around 3–6 months establish post-recovery profile access and attrition
annual/after events detect recurrent/progressive change practice effects

The optimal schedule is not established by one RCT. Canadian guidance supports systematic attention to cognition across the stroke pathway (Swartz 2025, PMID 39822128).

Accessible assessment

Barrier Adaptation
aphasia supported communication/nonverbal measures
neglect screen separately; arrange material
motor deficit oral/eye-gaze response where valid
fatigue shorter sessions and breaks
hearing/vision aids and accessible print
language/education validated norms/interpreter

Patients and carers report that memory problems are under-recognized and help-seeking pathways are unclear (Tang 2019, PMID 30452613; Tang 2020, PMID 32637417). A 2024 qualitative study examined acceptability of post-stroke cognitive testing, supporting the need to make testing understandable and actionable (McMahon 2024, PMID 38226361).

Prevention and management

  • Prevent recurrent stroke with cause-specific secondary prevention.
  • Treat hypertension, AF, diabetes, smoking, lipids, and activity for established vascular benefits.
  • Review delirium, depression, sleep, seizures, and medications.
  • Use cognitive rehabilitation and environmental/task strategies based on deficits.
  • Support return to roles, caregiver education, and safety planning.

No trial has established that a generic cognitive screen alone prevents dementia. Screening benefit depends on effective downstream intervention.

How good is the downstream intervention?

The honest answer is modest and uncertain at the level that matters most. A Cochrane review of occupational therapy for post-stroke cognitive impairment included 24 trials from 11 countries with 1,142 analysed participants (mostly under 50 per trial, mean intervention 19 hours, 20 of 24 using a remediation approach and most of those computer-based). The primary outcome — basic activities of daily living — improved by a mean difference of 2.26 points on the Functional Independence Measure (95% CI 0.17–4.22, P=0.03, I²=0%; 6 studies, 336 participants), rated low-certainty evidence (Gibson 2022, PMID 35349186). Cognitive outcomes fare better than functional ones: pooling 19 RCTs (n=875) of computerized cognitive training against usual care or routine rehabilitation gave moderate-to-high-certainty gains in general cognition (15 trials, SMD 0.46, 95% CI 0.21–0.71, I²=60%), attention (11 trials, SMD 0.45 in magnitude, 95% CI 0.25–0.64, I²=0%), executive function (6 trials, SMD 0.39, 0.12–0.67, I²=0%) and quality of life (9 trials, SMD 0.34, 0.15–0.53), with low- to very-low-certainty and limited effect for memory, language and motor function, and short-term high-frequency training outperforming long-term low-frequency schedules (Gao 2025, PMID 40503808). The pattern across both reviews — trained domains improve, daily function barely moves — is the same transfer problem that limits cognitive training generally, and it is why screening without a functional intervention pathway has an unproven benefit.

Trial design priorities

Design issue Preferred solution
pre-stroke cognition informant measure/records
acute delirium repeated assessment
stroke heterogeneity site/severity/etiology stratification
death competing-risk methods
aphasia exclusion accessible outcomes
mixed pathology AD biomarker subset
recurrence time-varying exposure

Open questions

  • Why does delirium predict post-stroke dementia independently of white-matter disease while infection predicts it only in those with white-matter disease? (Pendlebury 2024, PMID 38310893)
  • Should 6-month follow-up MRI be standard after stroke, given that 15.5% of patients have new, 91%-covert ischaemic lesions that predict cognition and recurrence? (Fang 2024, PMID 39417418)
  • When will post-ICH prevention trials elevate cognition from a secondary or substudy outcome to a primary clinical dementia endpoint? (Lioutas 2025, PMID 38583421; NCT03863665; NCT03785067)
  • Which screening schedule improves patient-centered outcomes? (Swartz 2025, PMID 39822128)
  • Can risk models predict change rather than one-time impairment? (Drozdowska 2021, PMID 33817331)
  • Does aggressive secondary prevention reduce delayed cognitive disorder? (Filler 2024, PMID 38101426)
  • Which rehabilitation strategies improve daily function despite persistent deficits? (Tang 2020, PMID 32637417)
  • Why does every cognitive domain except executive function decline after year one, when executive impairment is supposed to be the vascular signature? (Lo 2022, PMID 34775838)
  • If trajectory group adds nothing beyond the 3–6-month score, is repeated early testing worth its cost? (Lo 2023, PMID 37072222)
  • Why does national health expenditure improve a post-stroke dementia prediction model, and what is it standing in for? (Lo 2026, PMID 41525568)
  • Can imaging-texture prediction survive the 90%→77% drop from internal to external scanners? (Betrouni 2022, PMID 35862196)
  • If a visually-rated infarct location score matches a full multivariable model, what do the clinical predictors add at all? (Weaver 2021, PMID 33901427)
  • Should MMSE be retired from post-stroke use given that its impairment map is confined largely to the left MCA territory? (Weaver 2021, PMID 33472574)
  • Does delirium after stroke cause later dementia or reveal a brain already vulnerable to it? (Oldenbeuving 2011, PMID 21307355; Melkas 2012, PMID 21560162)
  • Why does white-matter infarct volume carry more cognitive cost per millilitre than grey-matter volume? (Ospel 2024, PMID 38511330)
  • Why do cognitive-training gains (SMD ≈ 0.46) not transfer to daily function (FIM MD 2.26, low certainty)? (Gao 2025, PMID 40503808; Gibson 2022, PMID 35349186)

References

  1. Pendlebury ST, Rothwell PM. Pre- and post-stroke dementia. Lancet Neurol. 2009. PMID 19782001
  2. Barbay M, et al. Post-stroke neurocognitive disorders. Dement Geriatr Cogn Disord. 2018. PMID 30504699
  3. Filler J, et al. Risk factors after stroke. Lancet Healthy Longev. 2024. PMID 38101426
  4. Drozdowska BA, et al. Prognostic rules for cognitive syndromes following stroke. Eur Stroke J. 2021. PMID 33817331
  5. Swartz RH, et al. Canadian Stroke Best Practice Recommendations: Vascular cognitive impairment. Alzheimers Dement. 2025. PMID 39822128
  6. Tang EYH, et al. Post-stroke memory deficits and barriers to seeking help. Fam Pract. 2019. PMID 30452613
  7. Tang EYH, et al. Impact of memory problems post-stroke on patients and family carers. Front Med (Lausanne). 2020. PMID 32637417
  8. McMahon D, et al. Acceptability of post-stroke cognitive testing. Cereb Circ Cogn Behav. 2024. PMID 38226361
  9. Jacova C, et al. Cognitive impairment in lacunar strokes: the SPS3 trial. Ann Neurol. 2012;72:351-62. PMID 23034910
  10. Pearce LA, et al. SPS3 cognitive function. Lancet Neurol. 2014;13:1177-85. PMID 25453457
  11. Tzourio C, et al. PROGRESS dementia and cognitive decline. Arch Intern Med. 2003;163:1069-75. PMID 12742805
  12. Lo JW, et al. Short-term trajectories of poststroke cognitive function: a STROKOG collaboration study. Neurology. 2023;100:e2331-e2341. PMID 37072222
  13. Lo JW, et al. Long-term cognitive decline after stroke: an individual participant data meta-analysis. Stroke. 2022;53:1318-1327. PMID 34775838
  14. Tang EYH, et al. Assessing the predictive validity of simple dementia risk models in harmonized stroke cohorts. Stroke. 2020;51:2095-2102. PMID 32568644
  15. Lo JW, et al. Development of a prognostic model for poststroke dementia using multiple international cohorts: a STROKOG collaboration study. Neurology. 2026;106:e214574. PMID 41525568
  16. Betrouni N, et al. Texture features of magnetic resonance images predict poststroke cognitive impairment: validation in a multicenter study. Stroke. 2022;53:3446-3454. PMID 35862196
  17. Weaver NA, et al. Strategic infarct locations for post-stroke cognitive impairment: a pooled analysis of individual patient data from 12 acute ischaemic stroke cohorts. Lancet Neurol. 2021;20:448-459. PMID 33901427
  18. Weaver NA, et al. Post-stroke cognitive impairment on the Mini-Mental State Examination primarily relates to left middle cerebral artery infarcts. Int J Stroke. 2021;16:981-989. PMID 33472574
  19. de Kort FAS, et al. White matter hyperintensity volume and poststroke cognition: an individual patient data pooled analysis of 9 ischemic stroke cohort studies. Stroke. 2023;54:3021-3029. PMID 37901947
  20. Ospel JM, et al. Influence of infarct morphology and patterns on cognitive outcomes after endovascular thrombectomy. Stroke. 2024;55:1349-1358. PMID 38511330
  21. Oldenbeuving AW, et al. Delirium in the acute phase after stroke: incidence, risk factors, and outcome. Neurology. 2011;76:993-9. PMID 21307355
  22. Melkas S, et al. Post-stroke delirium in relation to dementia and long-term mortality. Int J Geriatr Psychiatry. 2012;27:401-8. PMID 21560162
  23. Gibson E, et al. Occupational therapy for cognitive impairment in stroke patients. Cochrane Database Syst Rev. 2022;3:CD006430. PMID 35349186
  24. Gao M, et al. The effectiveness of computerized cognitive training in patients with poststroke cognitive impairment: systematic review and meta-analysis. J Med Internet Res. 2025;27:e73140. PMID 40503808
  25. Pendlebury ST, et al. Infection, delirium, and risk of dementia in patients with and without white matter disease on previous brain imaging: a population-based study. Lancet Healthy Longev. 2024;5:e131-e140. PMID 38310893
  26. Fang R, et al. Risk factors and clinical significance of post-stroke incident ischemic lesions. Alzheimers Dement. 2024;20:8412-8428. PMID 39417418
  27. Zhang Z, et al. Incident dementia after spontaneous intracerebral hemorrhage. J Alzheimers Dis. 2024;99:41-51. PMID 38640161
  28. Li CH, et al. Factors of post-stroke dementia: a nationwide cohort study in Taiwan. Geriatr Gerontol Int. 2019;19:815-822. PMID 31267646
  29. Lioutas VA, et al. Cognitive outcome after acute spontaneous intracerebral hemorrhage: analysis of the iDEF randomized trial. Cerebrovasc Dis. 2025;54:156-164. PMID 38583421