Detecting and Predicting Post-stroke Cognitive Impairment Using the Oxford Cognitive Screen

Aim of the study:

The main aim of this study is to investigate the diagnostic accuracy of the OCS in subacute and chronic stroke compared to a comprehensive matched neuropsychological battery. In addition, the study aims to investigate the predictive validity of the Oxford Cognitive Screen (OCS), as well as to estimate updated prevalence rates of post-stroke cognitive impairment (PSCI).

Primary research questions:

• How accurately do OCS subtests identify cognitive impairment compared with matched comprehensive neuropsychological tests at 0-3 weeks and 6-9 months post-stroke? Hypothesis: OCS subtests will detect fewer impairments than the matched comprehensive tests at both timepoints, reflecting the greater sensitivity of detailed neuropsychological measures for detecting mild deficits.

Secondary research questions:

• How well does early OCS performance predict chronic cognitive impairment? Hypothesis: There will be a significant association between scores on the OCS in the subacute stage and performance on the detailed neuropsychological test battery in the chronic stage.
• What is the prevalence of PSCI as measured by neuropsychological tests at both subacute and chronic stages of stroke care in a contemporary, inclusive stroke cohort? Hypothesis: The prevalence of PSCI will differ from current estimates due to more sensitive cognitive assessments, changes in stroke aetiology, population characteristics and demographics, and advances in acute stroke treatment, which may differentially affect cognitive outcomes across patient groups.

Review of literature and rationale:

Stroke is a leading global health challenge, affecting approximately one in four adults during their lifetime (World Health Organization, 2025), and frequently results in cognitive impairments across domains such as language, memory, attention, visuospatial processing, and executive functioning (Nys et al., 2007; Rost et al., 2022). These impairments are associated with reduced participation and independence in daily activities (Stolwyk et al., 2021), lower health related quality of life (Stolwyk et al., 2024) and increased caregiver burden (Tziaka et al., 2024). Although some spontaneous recovery is common, a large proportion of patients continue to experience PSCI throughout both the subacute and chronic phases of recovery, making accurate early identification clinically essential (Barbay et al., 2018). Importantly, post-stroke cognitive impairments are highly heterogeneous and frequently span multiple cognitive domains, and growing evidence suggests that impairments may reflect dysfunction in distributed neural systems rather than isolated focal deficits (Demeyere \& Gillebert, 2019; Gajardo-Vidal et al., 2018; Rost et al., 2022).

Cognitive screening in stroke care is still evolving, with ongoing debate about how to best capture the complexity of PSCI (Kosgallana et al., 2019). For decades, global cognitive tools such as the MMSE and MoCA originally designed for the field of dementia have dominated clinical stroke practice and research, despite providing only a single summary score that often fails to detect the multidimensional, domain specific deficits characteristic of stroke (Stolwyk et al., 2014; Sujanthan et al., 2025). The Oxford Cognitive Screen (OCS) was developed to address these limitations. It is a brief, stroke specific tool assessing five cognitive domains individually, designed to minimise cross contamination between deficits and to allow valid assessment even in patients with stroke-typical deficits as aphasia or neglect (Demeyere et al., 2015). This is particularly important as many commonly used cognitive assessments impose linguistic or other task demands that can systematically exclude patients with aphasia or severe impairments. It has been translated into multiple languages and validated against the global screens, such as MMSE and MoCA (Hong et al., 2018; Huygelier et al., 2020; Kong et al., 2016; Mancuso et al., 2016; Ramos et al., 2018; Robotham et al., 2020; Shendyapina et al., 2019; Valera-Gran et al., 2019). Despite its widespread adoption, there is still limited knowledge about the tool's ability to predict long term cognitive outcomes as well as the tool's diagnostic accuracy when evaluated against a comprehensive neuropsychological gold standard (Godefroy et al., 2018).

A major limitation in PSCI research is the systematic exclusion of patients with the most severe strokes, who are often unable to provide informed consent or cannot be adequately assessed using standard cognitive tests, particularly due to severe aphasia, communication impairments, or other cognitive deficits that interfere with test performance (Brady et al., 2013; Burton \& Tyson, 2015; Sexton et al., 2019; Vaughan \& Manning, 2023). This dual exclusion, driven by both consent procedures and limitations of standard assessment methods, results in biased samples and likely underestimation of PSCI prevalence. Moreover, earlier prevalence estimates do not reflect the characteristics of contemporary stroke cohorts treated with modern acute interventions such as thrombolysis and endovascular therapy cohorts (Humphrey et al., 2024; López-Cancio et al., 2017; Vestergaard et al., 2023).

The present project addresses these gaps by evaluating the diagnostic accuracy and predictive validity of the OCS against a matched, comprehensive neuropsychological battery at both subacute and chronic stages, and by using an inclusive recruitment strategy that allows inclusion of patients who are initially unable to provide informed consent through use of routinely collected clinical data. This approach enables more accurate and contemporary prevalence estimates of PSCI.

Methods:

This study is a prospective, longitudinal observational cohort investigation with two assessment timepoints: the subacute phase T1 (0-21 days post stroke) and the chronic phase T2 (6-9 months post stroke). 180 stroke patients are assessed with a stroke specific cognitive screening (OCS), a matched comprehensive neuropsychological assessment battery, validated self report measures, extraction of routinely collected clinical and imaging data from the medical record, and a brief exploratory qualitative component assessing patient experiences of cognitive assessment and related healthcare interactions. We expect to screen records from approximately 800 patients to include 180 stroke patients in the study. We expect approximately 50% of included patients to complete assessment at T2.

Ethical approval The project was not subject to review by the Capital Region of Denmark Research Ethics Committee (F-26025299). The project was approved by the Research Ethics Committee at the Department of Psychology, University of Copenhagen (IP-EC-29062026-1).

Setting and Procedure Participants will be recruited pseudo-consecutively from the Department of Neurology at Amager and Hvidovre Hospital Glostrup and Bispebjerg Hospital on predefined weekdays to ensure representativeness of the clinical stroke population. To ensure high participation rates, there will be a large flexibility of location for assessment: testing at T1 will be conducted at the bedside or in dedicated neuropsychological testing facilities depending on clinical stability, or if the patient is discharged early from hospital the testing will take place in the participants home or at experimental facilities at the University of Copenhagen depending on the patients preference - T2 assessment may take place either in the hospital, rehabilitation centres, experimental facilities at University of Copenhagen or in the participant's home if preferred.

At T1, all participants will complete the Oxford Cognitive Screen (OCS) and additional detailed neuropsychological tests. To reduce testing burden in the subacute phase (considering the high likelihood of their mental fatigue) while still ensuring that all detailed tests are represented across the full sample, the participants will be randomly allocated to one of three subsets of the detailed neuropsychological tests. Randomisation will be performed using a computer-generated block randomisation procedure with a block size of three, ensuring balanced allocation across the three subsets over time. Within each block, the order of subset assignment will be randomised. Participants will be assigned sequentially according to this predefined allocation list. The allocation procedure is independent of clinical judgement and participant characteristics, thereby minimising the risk of selection bias. The subsets are organized by cognitive domain. Subset 1 includes: language and visual perception; Subset 2 includes: memory and praxis; Subset 3 includes: executive functions, visual attenton and numerical cognition. A brief self-report questionnaire assessing fatigue, mood, and functional ability and participation will also be administered, and participants will be invited to take part in a brief optional qualitative reflection on their experiences of how cognitive functioning is addressed in healthcare interactions. As mental fatigue is very common in the subacute stage, assessments may be split across two shorter sessions, and breaks will be offered as needed. The estimated time of the T1 data collection is approximately 45 minutes.

At T2, participants will undergo a full comprehensive cognitive assessment with the battery we have developed, in addition to repeating the OCS and completing an expanded battery of psychosocial questionnaires. Again here, participants are invited to take part in a brief optional qualitative reflection on their experiences of how cognitive functioning is addressed in healthcare interactions. As at T1, the testing may be divided into multiple sessions if the participant experiences fatigue. The estimated time of the T2 data collection is approximately 2-3½ hours.

All assessments consist exclusively of standard paper-and-pencil neuropsychological tasks and validated questionnaires routinely used in clinical practice. Structural neuroimaging (CT/MRI) already acquired as part of routine clinical care will be retrieved from the medical record to enable detailed characterization and analysis of lesions when possible.

Measures The project will rely exclusively on non-invasive clinical, cognitive, and self-report data. Cognitive assessments will consist primarily of paper-and-pencil neuropsychological tasks, such as marking stimuli, naming items, and writing/drawing, in accordance with standard clinical practice for stroke patients.

Oxford Cognitive Screen:

The Oxford Cognitive Screen (OCS) assesses five cognitive domains commonly affected after stroke: (1) Language, (2) Attention, (3) Memory, (4) Number Processing, and (5) Praxis. The instrument includes 10 subtasks (Picture naming, Semantics, Orientation, Visual field, Sentence reading, Number writing \& calculation, Broken hearts, Imitation, Recall and recognition, Executive task) giving a total of 14 individual measures (as Number writing \& calculation, Broken hearts, Recall and recognition, and Executive task gives two scores each), designed to minimize confounding effects of aphasia and neglect, enabling domain specific classification of impairment (Demeyere et al., 2015).

Matched neuropsychological tests:

To evaluate the diagnostic accuracy and predictive validity of the Oxford Cognitive Screen (OCS), each of the 10 OCS-subtasks was matched to one or more detailed neuropsychological tests assessing the same or closely related cognitive constructs. The selection of matched tests was guided by four overarching principles.

First, tests were chosen to reflect standard neuropsychological practice in stroke assessment, based on correspondence with neuropsychologists at Danish stroke units and rehabilitation centres, and informed by international consensus initiatives aimed at harmonising post-stroke cognitive assessment, such as the Neuropsychological Minimal Dataset after Stroke (NMDS) (Legrand et al., 2023). This ensured that the reference battery represents how comprehensive cognitive assessments are actually conducted in subacute and chronic stroke populations.

Second, correspondence in task format between OCS subtests and the matched tests was prioritised whenever possible (e.g., picture naming, cancellation tasks). When no direct equivalent exists as a detailed neuropsychological test with appropriate normative data, functional correspondence was used, whereby tests capturing the closest related cognitive processes as assessed in routine clinical practice were selected as reference standards. In addition, the comprehensive battery includes supplementary neuropsychological tests recommended in clinical guidelines and expert consensus that are not directly assessed by the OCS, enabling systematic identification of cognitive domains and processes that may be underrepresented or not captured by the OCS.

Third, the matched battery was designed to be robust to communication impairments. Verbal neuropsychological tests were included where they represent standard clinical practice and valid assessment of the target. However, for a limited number of tests in which performance is particularly confounded by aphasia or language comprehension difficulties, alternative assessment strategies were included when well established equivalents exist. This mirrors routine neuropsychological assessment practice and prevents systematic exclusion of patients with aphasia from the reference standard.

Fourth, inclusion of tests with established normative data was prioritised wherever possible, as normative reference values are essential for defining cognitive impairment and enabling direct comparison with OCS-based impairment classifications. Danish normative data were preferred when available, but when not available, well-established international normative data were used.

The resulting matched battery covers all major cognitive domains typically assessed after stroke, including language, memory, attention (including neglect and working memory), executive functioning, numerical cognition, perceptual processing, praxis. The battery draws on established standardized instruments commonly used in clinical neuropsychological assessment after stroke, including CAT (Swinburn et al., 2004), RBANS (Randolph, 2012), WAIS IV (Wechsler, 2008), WMS IV (Wechsler, 2009), BIT (Wilson et al., 1987), the Trail Making Test (Reitan, 1958) and Color Trails (D'Elia et al., 1996), RAVLT (Schmidt, 1996), ROCFT (Osterrieth, 1944; Rey, 1964) , as well as standardized measures of verbal fluency (Borkowski et al., 1967), design fluency (Regard et al., 1982), serial subtraction (Smith, 1967), and computerized visual field assessment (Nordfang et al., 2019). At T2 a small subset of patients will also be assessed with the Oxford Visual Perceptual Screen (Vancleef, K., 2025). Together, these measures constitute a comprehensive reference standard against which the diagnostic accuracy and predictive validity of the OCS subtests can be evaluated, as well as for estimating the impairment based prevalence of post stroke cognitive deficits.

Self-report questionnaires In addition to objective cognitive testing, participants will complete a structured set of validated self report questionnaires at both timepoints assessing non cognitive factors that are known to influence cognitive performance and functional outcomes after stroke. These factors include fatigue, mood, functional ability and participation, perceived quality of life, sleepiness, and cognitive reserve. Additionally, the participants will answer a project-specific questionnaire about the type and amount of rehabilitation they received and if they had any neuropsychological testing, that was not reported in their medical records. The inclusion of these self report measures allows for a broader characterisation of post stroke outcomes, facilitates interpretation of neuropsychological test results, and enables analyses of the relationships between objective cognition, subjective experience and everyday functioning.

At T1, self report questionnaires are used to characterise fatigue, mood, functional limitations and participation restrictions in the subacute phase after stroke. The focus at T1 is on impairment and function related outcomes that can be meaningfully reported during hospitalisation and early recovery, while minimising assessment burden. The total duration of the self report questionnaire battery at T1 is approximately 5-15 minutes depending on patient fatigue and need for supported administration. At T2, the questionnaire battery is expanded to additionally capture perceived quality of life, sleepiness and cognitive reserve, alongside a more comprehensive assessment of functional ability and participation, with an estimated duration of 35-45 minutes.

Fatigue is assessed using the Fatigue Severity Scale (FSS; Krupp et al., 1989). Mood symptoms are assessed using the Hospital Anxiety and Depression Scale (HADS; Zigmond \& Snaith, 1983). Functional ability and participation are assessed using the World Health Organization Disability Assessment Schedule 2.0 (WHODAS 2.0; World Health Organization, 2010). The 12 item version is used at T1 to minimise respondent burden in the subacute phase, while the full 36 item version is used at T2 to allow a more detailed assessment of functioning and participation at follow up. Perceived quality of life is assessed using the World Health Organization Quality of Life Brief Version (WHOQOL BREF; The WHOQOL Group, 1998). Daytime sleepiness is assessed at T2 using the Epworth Sleepiness Scale (ESS; Johns, 1991). Cognitive reserve is assessed at T2 using the Cognitive Reserve Index Questionnaire (Nucci et al., 2012).

As there is no validated self-report instrument that captures rehabilitation exposure across heterogeneous care pathways, rehabilitation and neuropsychological assessments received between T1 and T2 will be assessed using a brief, structured set of study-specific self-report items. Information on neuropsychological assessment between T1 and T2 is collected when available to identify potential exposure to cognitive testing that may contribute to retest or practice effects at follow-up. Information on the type and amount of rehabilitation received between T1 and T2 is collected to characterise post-stroke care and, where appropriate, included as covariates in statistical analyses to account for variability in rehabilitation exposure. These variables are not treated as psychometric outcome measures, and their limitations will be considered in interpretation of the results.

All remaining self report instruments included in the study are widely used in clinical and research settings and have demonstrated validity and clinical relevance in neurological populations. When required, questionnaires may be administered with reading aloud or supported communication strategies to accommodate aphasia, visual impairment or other stroke related difficulties.

Clinical Measures The study will also draw on existing clinical information extracted from the medical record, to characterise the stroke cohort and to contextualise cognitive and patient reported outcomes. All clinical measures are derived exclusively from routinely collected clinical data and do not involve any additional procedures beyond standard care.

Clinical variables extracted from the medical records include at T1: demographic variables (age and sex), stroke characteristics (date of stroke and stroke type, defined as ischaemic or haemorrhagic), measures of stroke severity and functional status (National Institutes of Health Stroke Scale and modified Rankin Scale), and details of acute stroke treatment (intravenous thrombolysis and/or mechanical thrombectomy). Relevant medical history will be extracted, including prior stroke, documented mood disorders, and other major comorbid medical conditions. Information on length of stay in the acute setting and level of independence prior to admission, including the need for formal or informal assistance, will also be recorded where documented in the medical record.For participants who are unable to provide informed consent, only a restricted subset of clinical variables may be obtained, depending on applicable approvals and regulatory requirements.

Clinical comorbidity will be assessed using the Charlson Comorbidity Index (CCI) based on information from medical records (Charlson et al., 1987). The CCI will be used to quantify overall comorbidity burden, including conditions such as dementia and other neurodegenerative diseases, allowing these factors to be accounted for analytically rather than through strict exclusion criteria.

Structural neuroimaging acquired as part of routine clinical care will be retrieved when available. From routine radiology reports and clinical documentation, coarse lesion characteristics combining lesion laterality and broad anatomical location will be extracted and included for exploratory analyses. Imaging-based measures are included for exploratory purposes only and are not required for study inclusion. At T2, medical records will be reviewed for updated clinical information, including newly documented medical or psychiatric conditions, updated structural neuroimaging when available, and any neuropsychological assessments conducted as part of routine clinical care.

The clinical measures serve several purposes within the project. First, they provide a detailed clinical description of the study population and support estimation of representative prevalence rates of post stroke cognitive impairment. Second, they allow examination of associations between stroke severity, treatment characteristics and cognitive outcomes across timepoints. Third, selected clinical variables are used as covariates or contextual factors in analyses of cognitive and patient reported outcomes, while imaging based measures are included for exploratory analyses only.

All clinical measures will be handled in accordance with applicable data protection legislation and institutional guidelines, and only information relevant to the objectives of the study will be extracted and analysed.

Structural neuroimaging data (exploratory analyses) For participants with available imaging data, analyses will focus on characterising brain-behaviour relationships using quantitative lesion-based approaches. This approach is motivated by large-scale lesion-symptom mapping studies demonstrating the value of linking structural brain variation to behavioural outcomes in unselected clinical stroke samples (Haupt et al., 2017; Moore et al., 2021). Where feasible, lesion delineation will be performed using semi-automated or manual procedures, and images may be spatially normalised to a standard anatomical space to enable comparison across participants, consistent with established preprocessing pipelines in voxel-based lesion mapping studies.

Imaging-derived variables such as lesion volume, hemispheric laterality, and more fine-grained anatomical location will be extracted and used in statistical modelling. Prior research has shown that global measures of lesion burden, particularly lesion volume, are often among the strongest predictors of behavioural outcome across domains. In addition to these summary measures, voxel-wise analyses may be conducted to examine the relationship between localised structural damage and variation in cognitive and patient-reported outcomes. Depending on the nature of the outcome variables, both group-based comparisons and continuous modelling approaches will be considered.

In addition to acute lesion characteristics, markers of pre-existing brain pathology will be extracted where available, including white matter hyperintensities (Fazekas score), global cerebral atrophy, prior infarcts, and microbleeds. These markers will be used to derive an index of brain frailty, conceptualised as reflecting cumulative cerebrovascular and neurodegenerative burden prior to stroke. Where feasible, standardized visual rating scales will be applied to ensure consistency across patients. Relevant vascular imaging findings, such as large vessel stenosis or occlusion (e.g., internal carotid artery), will also be recorded where available to further characterise cerebrovascular burden.

The primary aim is to identify candidate brain-behaviour relationships that can inform future hypothesis-driven research. Given the use of routine clinical neuroimaging, variability in scan modality, resolution, and timing relative to clinical events is expected.

Qualitative data (exploratory) To complement the quantitative assessments, a brief qualitative data collection component will be included to capture participants' experiences of how cognitive functioning is addressed in their interactions with healthcare professionals after stroke. This component is exploratory and intended to contextualise and extend the interpretation of the quantitative findings.