Neurology · 2026
Development of a Prognostic Model for Poststroke Dementia Using Multiple International Cohorts: A STROKOG Collaboration Study.
СтатьяНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Lo Jessica W, Crawford John D, Desmond David W, Godefroy Olivier, Roussel Martine, Bordet Régis, Dondaine Thibaut, Mendyk Anne-Marie, Bae Hee-Joon, Lim Jae-Sung, Ojagbemi Akin, Bello Toyin, Chen Christopher P L H, Chong Eddie J, Venketasubramanian Narayanaswamy, Klimkowicz-Mrowiec Aleksandra, Traykov Latchezar, Mehrabian Shima, Chung Chih-Ping, Chi Nai-Fang, Lau Gary Kui Kai, Liu Dillys Xiaodi, Welberry Heidi, Brodaty Henry, Sachdev Perminder S, Stroke and Cognition (STROKOG) Consortium
Аннотация
BACKGROUND AND OBJECTIVES: Dementia risk prediction models developed for the general population perform poorly in stroke cohorts. Existing stroke-specific models are few and limited by short prediction horizons or reliance on neuroimaging. The aim of this study was to develop a clinically practical model for predicting 5-year dementia risk after stroke using commonly available variables and individual participant data from the Stroke and Cognition Consortium (STROKOG).
METHODS: Data were pooled from 12 studies across 10 countries. Dementia was diagnosed mainly by expert panel consensus and algorithmic classification. Fine-Gray subdistribution hazard models estimated dementia probability, accounting for death as a competing event. Candidate predictors included routinely collected baseline clinical and stroke-related variables, selected through backward stepwise elimination. Model performance was evaluated using discrimination (C-index) and calibration for prediction up to 5 years after stroke. Internal-external cross-validation (IECV) assessed generalizability across studies, regions, and study periods.
RESULTS: A total of 2,663 participants (mean age 67.0 years [SD 11.1]; 40% female) were followed for a median of 2.0 years (IQR 1.0-5.0), during which 655 developed dementia (8.7 per 100 person-years). The final model included age, sex, education, history of previous stroke, diabetes, stroke severity, 2 interactions (age × sex; age × stroke severity), and study-level variables including national current health expenditure. An Excel-based risk calculator is available in the Supplement (eAppendix 1). The model demonstrated strong discrimination (C-index: 0.81; 95% CI 0.75-0.87) and excellent calibration in the full data set used for development. In IECV, discrimination was acceptable across individual studies (pooled C-index: 0.70 [0.67-0.73]) and higher in recent (post-2010; 0.79 [0.76-0.82]) and European (0.74 [0.71-0.78]) cohorts. Risks were slightly overestimated in Asian cohorts. Case numbers were too small for reliable assessment in other regions.
DISCUSSION: We developed and internally-externally validated a 5-year dementia risk model for stroke survivors using routinely available clinical variables. The model showed strong performance in the full development data set and generalized well to recent and European cohorts, although external validation in diverse populations is needed. This tool can help identify high-risk individuals for targeted cognitive monitoring and follow-up. By informing clinical decision making and resource planning, it offers a practical means to improve long-term outcomes.
METHODS: Data were pooled from 12 studies across 10 countries. Dementia was diagnosed mainly by expert panel consensus and algorithmic classification. Fine-Gray subdistribution hazard models estimated dementia probability, accounting for death as a competing event. Candidate predictors included routinely collected baseline clinical and stroke-related variables, selected through backward stepwise elimination. Model performance was evaluated using discrimination (C-index) and calibration for prediction up to 5 years after stroke. Internal-external cross-validation (IECV) assessed generalizability across studies, regions, and study periods.
RESULTS: A total of 2,663 participants (mean age 67.0 years [SD 11.1]; 40% female) were followed for a median of 2.0 years (IQR 1.0-5.0), during which 655 developed dementia (8.7 per 100 person-years). The final model included age, sex, education, history of previous stroke, diabetes, stroke severity, 2 interactions (age × sex; age × stroke severity), and study-level variables including national current health expenditure. An Excel-based risk calculator is available in the Supplement (eAppendix 1). The model demonstrated strong discrimination (C-index: 0.81; 95% CI 0.75-0.87) and excellent calibration in the full data set used for development. In IECV, discrimination was acceptable across individual studies (pooled C-index: 0.70 [0.67-0.73]) and higher in recent (post-2010; 0.79 [0.76-0.82]) and European (0.74 [0.71-0.78]) cohorts. Risks were slightly overestimated in Asian cohorts. Case numbers were too small for reliable assessment in other regions.
DISCUSSION: We developed and internally-externally validated a 5-year dementia risk model for stroke survivors using routinely available clinical variables. The model showed strong performance in the full development data set and generalized well to recent and European cohorts, although external validation in diverse populations is needed. This tool can help identify high-risk individuals for targeted cognitive monitoring and follow-up. By informing clinical decision making and resource planning, it offers a practical means to improve long-term outcomes.