2026-03-05 オックスフォード大学
＜関連情報＞
- https://www.ox.ac.uk/news/2026-03-05-stroke-cognition-calculator-could-help-predict-thinking-problems-after-stroke
- https://www.thelancet.com/journals/lanhl/article/PIIS2666-7568(26)00004-8/fulltext
脳卒中後の多領域認知障害：臨床予測モデルの開発と検証 Multidomain post-stroke cognitive impairment: development and validation of a clinical prediction model
Andrea Kusec, PhD ∙ Kym I E Snell, PhD ∙ Prof Nele Demeyere, PhD
The Lancet Healthy Longevity Published: March 4, 2026
DOI:https://doi.org/10.1016/j.lanhl.2026.100820
Summary
Background
Post-stroke cognitive impairment (PSCI) is highly prevalent across multiple domains. Individualised PSCI prognosis has mainly been researched using dementia-specific outcomes instead of stroke-specific outcomes, and existing models often use predictors not routinely available in electronic health records. We aimed to develop and externally validate clinical prediction models for overall PSCI via use of a stroke-specific cognitive outcome, using acute PSCI and data routinely collected in stroke care.
Methods
In this prediction model development and validation study, we used data from a cohort of participants with stroke who were consecutively recruited from the acute stroke ward of the John Radcliffe Hospital (Oxford, UK) for the Oxford Cognitive Screening Programme (OCS-Recovery study). Participants completed the Oxford Cognitive Screen (OCS; comprising 12 subtasks covering six cognitive domains) acutely and at the 6-month follow-up. The outcome was binarised (impaired vs unimpaired). The selected predictors for the logistic regression models were available in electronic health records and conceptually relevant to post-stroke cognition. Logistic regression models were fitted with mandatory clinically relevant predictors (age, sex assigned at birth, stroke severity, education, stroke hemisphere, and acute PSCI) and data-driven predictors (acute mood difficulties, length of stay in acute care, and multimorbidity). We conducted backward elimination on multiply imputed data to remove non-significant (p>0·10) data-driven predictors. Internal validation used bootstrapping to obtain optimism-adjusted performance estimates. The same internal validation procedure was followed for a continuous prediction model, using proportion of OCS tasks impaired as the outcome. For external validation, we used the OCS-Care dataset, comprising data from a stroke cohort with mild severity PSCI. Performance measures included discrimination (eg, C-statistic), calibration, and goodness-of-fit. Overall binary PSCI model performance was further evaluated within subgroups by age range, sex assigned at birth, first versus recurrent stroke, and acute PSCI severity.
Findings
Between March 20, 2012, and March 9, 2020, 430 participants recruited to the OCS-Recovery study completed the OCS acutely and at 6-months after stroke. All participants attempted the OCS, with 400 (93%) completing at least ten of 12 subtasks. The overall binary PSCI model had good optimism-adjusted performance (C-statistic 0·76 [95% CI 0·71–0·80]), with similar external validation performance (0·74 [0·68–0·80]). Model performance did not vary by sex assigned at birth but was best in adults younger than 60 years (0·76 [0·62–0·86]) with moderate-to-severe acute PSCI (0·72 [0·60–0·81]).
Interpretation
Stroke-specific cognition prediction models can offer more meaningful PSCI prognoses than models focused on cognitive decline. Our binary and continuous overall PSCI models show promise in terms of generalisability across different stroke cohorts. Future recalibration of domain-specific models would be beneficial.
Funding
UK National Institute for Health and Care Research.
