Skip to content

Cognitive profile and assessment

TL;DR — VCI often impairs processing speed, attention, and executive control, but no neuropsychological profile reliably separates vascular from Alzheimer or other pathology in an individual (Desmond 2004, PMID 15537510). MoCA usually detects executive/mild impairment better than MMSE, yet cutoffs vary by language, education, setting, and spectrum (Ghafar 2019, PMID 31050033). Assessment must distinguish screening from diagnosis, measure function, and account for aphasia, neglect, motor slowing, depression, fatigue, delirium, hearing, and premorbid ability (Swartz 2025, PMID 39822128). Serial multidomain measurement is more informative than a single total score.

Group-level profile

Domain Common vascular pattern Why it is not specific
Processing speed slow timed performance age, depression, motor deficit
Executive control set-shifting/planning deficits occurs in many dementias
Attention reduced sustained/divided attention delirium, sleep, medication
Memory retrieval may exceed encoding impairment strategic infarcts can impair encoding
Language focal stroke effects aphasia invalidates verbal composites
Visuospatial neglect or posterior-network injury Alzheimer/Lewy disease overlap
Social cognition network-dependent change poorly sampled by screens

Desmond's review concluded that executive dysfunction is common but not a singular defining deficit (Desmond 2004, PMID 15537510). Severe or mixed disease becomes multidomain.

Assessment ladder

  1. Establish change from baseline and time course.
  2. Exclude delirium and unstable medical states.
  3. Obtain patient and informant functional histories.
  4. Use a brief screen suited to language and disability.
  5. Perform domain-level testing where results affect attribution or care.
  6. Pair cognition with MRI and etiologic assessment.
  7. Repeat after recovery or clinical change.

Brief instruments

Instrument Strength Blind spot
MoCA executive, attention, visuospatial sampling education/language; motor/visual demands
MMSE familiarity and longitudinal legacy ceiling effect; limited executive testing
NINDS-CSN 5-minute protocol scalable vascular research screen not full diagnostic battery
BMET executive/processing emphasis fewer validation settings
Oxford Cognitive Screen stroke-domain design, aphasia/neglect sensitivity post-stroke rather than all VCI
Informant questionnaire premorbid and functional change informant bias/availability

A systematic review included 15 studies, eight instrument families, and 4,575 participants (1,015 with VCI); MoCA AUCs were generally >0.90 for VaD versus controls (and MMSE 0.86–0.99) and 0.87–0.93 for vascular MCI, with BMET AUC 0.94 for VMCI versus controls, but heterogeneity prevented pooling (Ghafar 2019, PMID 31050033). Mean ages ranged 51.6–75.5 years.

In one validation of 34 VaD patients, full MoCA AUC was 0.950 (95% CI 0.868–0.988), short MoCA 0.936, and MMSE 0.860. The small homogeneous sample makes its cutoff (<17 full MoCA) inappropriate as a universal threshold (Freitas 2012, PMID 22676901).

Pooled accuracy: the tests differ less than their reputations do

The claim that MoCA outperforms MMSE in stroke is weaker than it is usually stated. Three syntheses, each larger than the last, bracket the question.

Synthesis Studies / n Instrument and threshold Sensitivity Specificity
Lees 2014 (PMID 25190446) 35 articles, 25 tests ACE-R <88/100 (2 studies) 0.96 0.70
MMSE <27/30 (12 studies) 0.71 0.85
MoCA <26/30 (4 studies) 0.95 0.45
MoCA <22/30 (6 studies) 0.84 0.78
Rotterdam-CAMCOG <33/49 (2 studies) 0.57 0.92
Wei 2023 (PMID 37190789) 24 studies, 4,231 patients MoCA, optimal threshold search best accuracy at 21/22, not 26 —
Wei 2025 (PMID 40383729) 9 head-to-head studies, 1,135 patients MoCA 0.80 (0.72–0.86) 0.79 (0.71–0.85)
MMSE 0.76 (0.71–0.81) 0.78 (0.73–0.83)

The head-to-head comparison found no significant difference between MoCA and MMSE for post-stroke cognitive impairment (sensitivity P=0.36, specificity P=0.80), and most included studies were at high risk of bias (Wei 2025, PMID 40383729). Lees found no test clearly superior and no benefit from longer administration time; the MoCA advantage at its conventional threshold is bought almost entirely with specificity — 0.45 at <26, rising to 0.78 at <22 without a proportionate sensitivity loss (Lees 2014, PMID 25190446). Wei's threshold analysis reaches the same place from the other direction: the optimal cut-point in stroke populations is 21/22, and it moves further with region, stroke type and stroke phase (Wei 2023, PMID 37190789). The practical reading is that instrument choice matters less than threshold choice, and that the threshold is not a property of the test but of the population — which is the same conclusion the cross-cultural review reaches below, by a different route.

A fourth synthesis anchors the threshold question at a different cut-point again. Across 12 studies and 2,130 patients, MoCA AUCs for post-stroke cognitive impairment within one month were 0.90 at 20/19, 0.90 at 21/20 and 0.95 at 26/25; with sensitivity and specificity weighted equally the optimal cut-point was 20/19 (Youden index 0.58), MoCA had higher sensitivity and lower specificity than MMSE, and the optimum moved with stroke stage (Shi 2018, PMID 29427168). Placed beside Wei's 21/22 and Lees's 22, the honest summary is that the defensible post-stroke MoCA threshold lies somewhere in 19–22 rather than at the published 26, and that no single number survives across stage, region and screening purpose.

Informant questionnaires do a different job, and do it asymmetrically

Informant tools are the only way to establish the pre-stroke baseline that the diagnosis of post-stroke cognitive disorder requires. Pooled across four studies (n=1,197), IQCODE for diagnosis of post-stroke dementia gave sensitivity 0.81 (95% CI 0.60–0.93) and specificity 0.83 (0.64–0.93) — comparable to direct screens. Used prognostically across five studies (n=837), it inverted: sensitivity 0.60 (0.32–0.83), specificity 0.97 (0.70–1.00). And the review found no papers at all on the test accuracy of informant tools for diagnosing pre-stroke cognitive decline — the use for which they are most often deployed (McGovern 2016, PMID 26683423).

The community-level Cochrane review of IQCODE reaches the same structural conclusion from outside stroke: the accuracy data are dominated by heterogeneity in threshold, in the 26-item versus 16-item short form, and in language of administration, and the "community" evidence base includes selected populations such as stroke survivors that required sensitivity analysis to separate (Quinn 2021, PMID 34278562). Informant instruments therefore share the index problem of direct screens — the threshold is a property of the setting — while being the only tools that address the baseline question at all.

A direct comparison in 137 patient–informant dyads partly filled that gap. At usual cut-points IQCODE-SF and AD8 had identical sensitivity for pre-stroke dementia (both 92%) but very different specificity (82% vs 58%); raising the AD8 threshold to ≥4 improved it. IQCODE-SF also correlated more strongly with generalized and medial-temporal atrophy, neurovascular disease and brain frailty, and predicted 18-month dementia better (AUROC 0.903, 95% CI 0.798–1.00 vs 0.821, 0.664–0.977) (Taylor-Rowan 2022, PMID 35278006).

Remote assessment

Where follow-up cannot be in person, telephone instruments have defensible cut-points from a prospective stroke cohort with full neuropsychological reference testing at 6, 12, 36 and 60 months (DEDEMAS, NCT01334749). Across MCI definitions, AUROC was 0.76–0.83 for TICS and 0.73–0.94 for T-MoCA; for multidomain MCI on a multiple-test definition the optima were TICS <36 and T-MoCA <18, rising to ≈19 for T-MoCA against a Clinical Dementia Rating reference. Validity was consistently lower under single-test MCI definitions (Zietemann 2017, PMID 29042492) — an instrument-independent reminder that the reference standard moves the answer as much as the index test does.

The Hachinski Ischaemic Score: what a 1975 instrument still shows

The HIS predates all modern criteria and is still the most-used bedside vascular-attribution questionnaire, which makes its validation history directly relevant to the false-attribution problem discussed in nosology. Prospective neuropathological validation in 32 autopsied dementia cases found that the score separated primary degenerative from multi-infarct or mixed dementia, but labelled 21% of primary degenerative dementia cases as vascular — and the authors warned specifically against applying it to large epidemiological samples, where that false-positive rate would inflate vascular-dementia counts (Fischer 1991, PMID 1895120). Correspondence analysis of 2,968 well-characterized Canadian Study of Health and Aging cases later showed the 13-item instrument to be redundant: a 7-item binary version classified as well as the original, and a 5-item composite version measuring two dimensions classified better than the original (Hachinski 2012, PMID 21987392). Both results point the same way — the score carries real signal about vascular contribution while being unable to establish it in an individual.

Arteriopathy type changes the profile

The "subcortical executive" profile is specific to a subset of vascular pathology rather than to vascular pathology in general. Comparing 32 patients with cerebral amyloid angiopathy against 39 with hypertension-related microangiopathy on an extensive neuropsychological protocol, the hypertensive group showed the expected attentional/executive pattern — multidomain deficit tracked Symbol Digit Modalities (β = −0.364) and phonemic fluency (β = −0.351) — whereas the CAA group performed worse on MoCA (P=0.001) and semantic fluency (P=0.043), had a higher prevalence of amnestic MCI (68% vs 46%), and multidomain deficit tracked verbal learning instead (RAVLT β = −0.574) (Barucci 2024, PMID 38467658). A memory-led profile therefore does not exclude a vascular cause; it may indicate which vessel disease is present. See CAA.

Apathy is a network phenomenon, not a mood item

Apathy — reduced motivated goal-directed behaviour — is common in stroke and small-vessel disease and is poorly sampled by cognitive screens, which is why it is under-counted in this population. Lesion-based localisation models do not account for it: cerebrovascular pathology produces network changes propagating beyond the damaged territory to structurally or functionally connected regions, and distinct subnetworks appear to support separable components of goal-directed behaviour, which would explain why apathy's presentation and longitudinal trajectory differ between stroke and diffuse SVD (Tay 2020, PMID 32151533). Measuring it requires a dedicated instrument and, given the network framing, one that distinguishes initiation from reward-valuation deficits.

What screens miss entirely: behavioural and psychological symptoms

Cognitive screens do not sample the symptoms that most often determine care needs, and those symptoms differ by vascular subtype. Pooling 35 studies (n=5,805) that used the Neuropsychiatric Inventory:

VCI subtype Most prevalent symptoms (pooled)
Unspecified VCI apathy 54.29%, depression 43.48%, irritability 38.76%
Subcortical VCI apathy 62.01%, depression 52.11%, irritability 44.73%
Mixed dementia apathy 61.65%, depression 45.68%, sleep disturbance 44.63%, hallucinations 26.64%
VCI–no dementia depression 44.97%, irritability 32.75%, anxiety 30.07%

(Santos 2025, PMID 41043167). Three readings follow. Apathy is the single commonest symptom in every dementia-level subtype and is the one least represented on MoCA or MMSE. The hallucination excess in mixed dementia (26.64%) is the subtype signal with the clearest mechanistic interpretation — Lewy or Alzheimer copathology rather than vascular injury — and is a practical reason to reconsider a pure-vascular attribution when hallucinations are prominent. And depression is already near 45% at the pre-dementia stage, which means a depression finding cannot be used to explain away cognitive impairment in this population without circularity.

The cutoff does not transport, and the direction of the error is known

A systematic review of diagnostic-accuracy studies for MMSE (28 studies), MoCA (39), and the Oxford Cognitive Screen (5) across 13 WEIRD and 4 other countries found a specific, actionable asymmetry: optimal MMSE and OCS subtest cutoffs were similar across WEIRD and less-WEIRD populations, whereas optimal MoCA cutoffs were lower in less-WEIRD populations (Gangaram-Panday 2024, PMID 37480233). Applying a North American MoCA threshold in a lower-education or non-Latin-script setting therefore over-diagnoses impairment, and the error is systematic rather than random. Using adjusted scores either shifted the optimal cutoff or preserved it with better accuracy. The review also records the coverage gap plainly: no diagnostic-accuracy studies were found in South American, African, or non-Chinese Asian stroke populations, and most papers reported almost nothing about their sample's cultural background.

The stroke-specific alternative and its blind spots

The Oxford Cognitive Screen was built for the population that invalidates general screens: it is deliberately aphasia- and neglect-friendly, takes 15–20 minutes, covers apraxia and unilateral neglect alongside memory, language, executive function, and number, and returns domain-specific scores rather than a single total. It was normed in 140 neurologically healthy participants and characterized in 208 acute stroke patients within 3 weeks of onset, with an alternate form for retest (Demeyere 2015, PMID 25730165).

An independent psychometric evaluation on 316 consecutive acute-stroke-unit patients is unusually candid about its limits. Impairment rates on memory and receptive-communication subtests were lower than expected, suggesting those subtests are relatively insensitive; patients with aphasia were more often uncategorizable even on nominally non-language tests, implying residual language and dominant-hand demands; and some subtests may index overall ability rather than the domain they name — though several achieved high retest reliability, making them good candidates for tracking change (Murphy 2023, PMID 37186035). A screen designed to survive aphasia still partly fails in aphasia; this is the measurement floor under OQ-19.

Against that, the head-to-head with MoCA is favourable and quantified. In 200 consecutive patients tested within 3 weeks of stroke, 76% were impaired on MoCA and 86% showed at least one impairment on an OCS domain; overall sensitivity was 87% for OCS against 78% for MoCA, and OCS alone detected neglect, apraxia and reading/writing impairment. Critically, MoCA impairment was dominated by left-hemisphere lesions while OCS gave differentiated profiles across hemispheres (Demeyere 2016, PMID 26588918). That lateralization bias is the same artifact that makes the left angular gyrus look "strategic" in lesion-mapping studies — an instrument that samples heavily in language will find language-dominant hemisphere lesions, in both study designs.

Prediction from a stroke-specific outcome

Most post-stroke cognitive prediction models target dementia. A model built instead on the OCS outcome — impaired versus unimpaired across 12 subtasks and six domains at 6 months — was developed in 430 OCS-Recovery participants (400, 93%, completed ≥10 of 12 subtasks) using only predictors available in electronic health records (age, sex, stroke severity, education, hemisphere, acute PSCI, plus data-driven candidates). Optimism-adjusted C-statistic was 0.76 (95% CI 0.71–0.80), holding at 0.74 (0.68–0.80) on external validation in the OCS-Care cohort; performance was best in adults under 60 (0.76) and in those with moderate-to-severe acute impairment (0.72) (Kusec 2026, PMID 41794047). That a routinely-collected-variable model reaches C ≈ 0.75 for cognitive impairment while general-population dementia models collapse to C ≈ 0.53–0.66 for dementia in stroke cohorts (see post-stroke cognitive impairment) suggests the prediction problem is tractable when the outcome is stroke-specific and the horizon is short.

Desmond concluded that executive dysfunction is common but not a singular defining deficit; AD more often impairs posterior-cortical functions including encoding, cueing benefit, and naming, whereas VaD more often affects planning, sequencing, speed, unstructured tasks, attention, syntax, and perseveration (Desmond 2004, PMID 15537510). SPS3 shows that this profile is easy to miss: 47% of 1,636 recent lacunar-stroke participants met MCI criteria, including 41% of those with modified Rankin 0–1 and Barthel 100 (Jacova 2012, PMID 23034910).

NINDS-CSN harmonization

VICCCS endorsed NINDS-CSN cognitive and imaging protocols to improve cross-study comparability (Skrobot 2018, PMID 29055812; Hachinski 2006, PMID 16917086). Harmonization provides common measurement; it does not make test performance culture-free or solve etiologic attribution. In SPS3, 47% of 1,636 English-speaking lacunar-stroke participants met MCI criteria (z ≤ −1.5), including 41% of those with modified Rankin 0–1 and Barthel 100; largest mean deficits were in episodic memory (z −0.65 to −0.92), verbal fluency (−0.89), and motor dexterity (−2.5) (Jacova 2012, PMID 23034910).

Battery level Use Tradeoff
5-minute large cohorts/bedside triage shallow domain coverage
30-minute pragmatic research/clinic moderate burden
60-minute detailed characterization fatigue and accessibility

Confounding after stroke

Confounder Distortion Mitigation
Aphasia verbal tests look globally impaired nonverbal/adapted methods
Neglect missed visual items hemispatial assessment
Hemiparesis drawing/timed tasks slow alternative response mode
Delirium fluctuating global deficit defer stable diagnosis
Depression/apathy low initiation and effort mood assessment and serial testing
Fatigue late-test decline breaks/shorter sessions
Hearing/vision poor encoding optimize access
Education/language norm mismatch appropriate norms

Functional assessment

Major disorder requires interference with independence. Stroke motor disability can create dependence without cognitive loss, so tasks should be decomposed: can the person plan medication but not open packaging; understand finances but not write; choose a route but not walk it? VASCOG's mild/major distinction depends on this functional reasoning (Sachdev 2014, PMID 24632990).

Function Cognitive demands Noncognitive confounders
medication sequencing, memory, judgment vision, dexterity
finance numeracy, fraud resistance prior role
cooking planning, safety mobility
driving speed, attention, visuospatial motor/visual disease
appointments prospective memory transport/access

Longitudinal interpretation

Reliable change should consider practice effects, alternate forms, interval, intercurrent stroke, delirium, and test–retest variability. A one-point total-score change is rarely mechanistically interpretable. Domain trajectories paired with function and imaging are preferable.

Open questions

  • Is the defensible post-stroke MoCA threshold 19, 20, 21 or 22, and should it be reported as an interval rather than a number? (Shi 2018, PMID 29427168; Wei 2023, PMID 37190789; Lees 2014, PMID 25190446)
  • How much of MoCA's apparent sensitivity in stroke is a left-hemisphere detection bias rather than general cognitive sensitivity? (Demeyere 2016, PMID 26588918)
  • Should apathy — present in 54–62% of vascular cognitive impairment — be a required assessment domain rather than an optional add-on? (Santos 2025, PMID 41043167; Tay 2020, PMID 32151533)
  • Do prominent hallucinations (26.6% in mixed dementia versus far less in subcortical VCI) warrant re-opening the etiologic formulation? (Santos 2025, PMID 41043167)
  • Which accessible screen performs best across aphasia, low education, and multilingual populations? (Ghafar 2019, PMID 31050033)
  • Can digital processing-speed measures detect meaningful preclinical SVD change? (Duering 2023, PMID 37236211)
  • What is a patient-important cognitive change in VCI trials? (Skrobot 2018, PMID 29055812)
  • How should motor dependence be separated from cognitive dependence after stroke? (Swartz 2025, PMID 39822128)
  • Should MoCA cutoffs be formally lowered for less-WEIRD populations rather than left to local adjustment? (Gangaram-Panday 2024, PMID 37480233)
  • Can an aphasia-friendly screen ever be genuinely aphasia-independent, given that OCS non-language subtests still fail in aphasia? (Murphy 2023, PMID 37186035)
  • Are the OCS memory and receptive-communication subtests too insensitive for use as outcome measures? (Murphy 2023, PMID 37186035)
  • Why do stroke-specific cognitive-impairment models reach C ≈ 0.75 while general dementia models reach C ≈ 0.53–0.66 in the same populations? (Kusec 2026, PMID 41794047)
  • If MoCA and MMSE do not differ in pooled accuracy, is instrument choice a research question at all, or only a threshold-calibration question? (Wei 2025, PMID 40383729; Lees 2014, PMID 25190446)
  • Should 21/22 replace 26 as the default MoCA threshold in stroke services, and what would the resulting false-negative rate be in mild impairment? (Wei 2023, PMID 37190789)
  • Why has no study measured informant-questionnaire accuracy for pre-stroke cognitive decline, the use for which these tools are most often deployed? (McGovern 2016, PMID 26683423)
  • Does the memory-led CAA profile mean amnestic presentations are being routinely misattributed to Alzheimer disease? (Barucci 2024, PMID 38467658)
  • Would a network-derived apathy instrument separate initiation from reward-valuation deficits well enough to serve as a trial endpoint? (Tay 2020, PMID 32151533)

References

  1. Desmond DW. Neuropsychology of vascular cognitive impairment. J Neurol Sci. 2004. PMID 15537510
  2. Ghafar MZAA, et al. Cognitive screening instruments to identify VCI. Int J Geriatr Psychiatry. 2019. PMID 31050033
  3. Freitas S, et al. MoCA validation for vascular dementia. J Int Neuropsychol Soc. 2012. PMID 22676901
  4. Sachdev PS, et al. VASCOG criteria. Alzheimer Dis Assoc Disord. 2014. PMID 24632990
  5. Skrobot OA, et al. VICCCS guidelines. Alzheimers Dement. 2018. PMID 29055812
  6. Swartz RH, et al. Canadian Stroke Best Practice Recommendations: Vascular cognitive impairment. Alzheimers Dement. 2025. PMID 39822128
  7. Duering M, et al. STRIVE-2. Lancet Neurol. 2023. PMID 37236211
  8. Hachinski V, et al. NINDS-Canadian Stroke Network vascular cognitive impairment harmonization standards. Stroke. 2006;37:2220-41. PMID 16917086
  9. Jacova C, et al. Cognitive impairment in lacunar strokes: the SPS3 trial. Ann Neurol. 2012;72:351-62. PMID 23034910
  10. Gangaram-Panday SG, et al. Screening for post-stroke neurocognitive disorders in diverse populations: a systematic review. Clin Neuropsychol. 2024;38:588-611. PMID 37480233
  11. Demeyere N, et al. The Oxford Cognitive Screen (OCS): validation of a stroke-specific short cognitive screening tool. Psychol Assess. 2015;27:883-94. PMID 25730165
  12. Murphy D, et al. Oxford cognitive screen: a critical review and independent psychometric evaluation. J Neuropsychol. 2023;17:491-504. PMID 37186035
  13. Kusec A, et al. Multidomain post-stroke cognitive impairment: development and validation of a clinical prediction model. Lancet Healthy Longev. 2026;7:100820. PMID 41794047
  14. Lees R, et al. Test accuracy of cognitive screening tests for diagnosis of dementia and multidomain cognitive impairment in stroke. Stroke. 2014;45:3008-18. PMID 25190446
  15. Wei X, et al. Which cutoff value of the Montreal Cognitive Assessment should be used for post-stroke cognitive impairment? A systematic review and meta-analysis on diagnostic test accuracy. Int J Stroke. 2023;18:908-916. PMID 37190789
  16. Wei X, et al. MoCA and MMSE for the detection of post-stroke cognitive impairment: a comparative diagnostic test accuracy systematic review and meta-analysis. J Neurol. 2025;272:407. PMID 40383729
  17. McGovern A, et al. Test accuracy of informant-based cognitive screening tests for diagnosis of dementia and multidomain cognitive impairment in stroke. Stroke. 2016;47:329-35. PMID 26683423
  18. Taylor-Rowan M, et al. Comparative validity of informant tools for assessing pre-stroke cognitive impairment. Int J Geriatr Psychiatry. 2022;37:10.1002/gps.5700. PMID 35278006
  19. Zietemann V, et al. Validation of the Telephone Interview of Cognitive Status and Telephone Montreal Cognitive Assessment against detailed cognitive testing and clinical diagnosis of mild cognitive impairment after stroke. Stroke. 2017;48:2952-2957. PMID 29042492
  20. Fischer P, et al. Prospective neuropathological validation of Hachinski's Ischaemic Score in dementias. J Neurol Neurosurg Psychiatry. 1991;54:580-3. PMID 1895120
  21. Hachinski V, et al. Optimizing the Hachinski Ischemic Scale. Arch Neurol. 2012;69:169-75. PMID 21987392
  22. Barucci E, et al. Cognitive profile in cerebral small vessel disease: comparison between cerebral amyloid angiopathy and hypertension-related microangiopathy. Sci Rep. 2024;14:5922. PMID 38467658
  23. Tay J, et al. Network neuroscience of apathy in cerebrovascular disease. Prog Neurobiol. 2020;188:101785. PMID 32151533
  24. Shi D, et al. Diagnostic test accuracy of the Montreal Cognitive Assessment in the detection of post-stroke cognitive impairment under different stages and cutoffs: a systematic review and meta-analysis. Neurol Sci. 2018;39:705-716. PMID 29427168
  25. Quinn TJ, et al. Informant Questionnaire on Cognitive Decline in the Elderly (IQCODE) for the detection of dementia within community dwelling populations. Cochrane Database Syst Rev. 2021;7:CD010079. PMID 34278562
  26. Demeyere N, et al. Domain-specific versus generalized cognitive screening in acute stroke. J Neurol. 2016;263:306-315. PMID 26588918
  27. Santos AP, et al. Behavioral and psychological symptoms of dementia associated with vascular impairment: a systematic review and meta-analysis. Braz J Psychiatry. 2025;47:e20254225. PMID 41043167