Development of a Prognostic Model for Poststroke Dementia Using Multiple International Cohorts: A STROKOG...

Area of Science:

• Neurology
• Epidemiology
• Biostatistics

Background:

• General dementia risk models perform poorly in stroke survivors.
• Existing stroke-specific models are limited in scope and accessibility.
• Need for a clinically practical dementia risk prediction tool for post-stroke patients.

Purpose of the Study:

• To develop a clinically practical model for predicting 5-year dementia risk after stroke.
• Utilize commonly available variables and individual participant data.
• Leverage data from the Stroke and Cognition Consortium (STROKOG).

Main Methods:

• Pooled data from 12 international studies.
• Employed Fine-Gray subdistribution hazard models accounting for competing risks (death).
• Utilized backward stepwise elimination for predictor selection and internal-external cross-validation (IECV) for generalizability assessment.

Main Results:

• Developed a model including age, sex, education, prior stroke, diabetes, stroke severity, and interactions.
• The model demonstrated strong discrimination (C-index: 0.81) and excellent calibration in the development dataset.
• IECV showed acceptable discrimination across studies (pooled C-index: 0.70), performing better in recent and European cohorts.

Conclusions:

• A validated 5-year dementia risk model for stroke survivors was developed using routine clinical variables.
• The model shows good generalizability, particularly in recent and European cohorts.
• This tool can facilitate targeted cognitive monitoring, improve clinical decision-making, and enhance long-term patient outcomes.