Stroke risk prediction models: A systematic review and meta-analysis.

Area of Science:

• Medical Informatics
• Epidemiology
• Biostatistics

Background:

• Stroke risk-score-prediction models (SRSMs) are crucial for identifying high-risk individuals for timely intervention.
• Limited evidence exists on the performance and generalizability of these SRSMs across diverse populations.
• This study systematically evaluates the performance and identifies weaknesses of existing SRSMs.

Purpose of the Study:

• To examine the performance of existing stroke risk-score-prediction models (SRSMs).
• To identify weaknesses in current SRSMs.
• To determine if SRSM performance varies by population and geographical region.

Main Methods:

• A systematic literature search was conducted in PubMed, EMBASE, and Web of Science until February 2022.
• The Prediction Model Risk of Bias Assessment Tool assessed the quality of included studies.
• Performance was meta-analyzed using pooled C-statistics derived from a random-effects model.

Main Results:

• 17 articles involving over 6.3 million participants were included, with most SRSMs developed from cohort studies.
• The overall pooled C-statistic for SRSMs was 0.78, indicating fair predictive performance.
• Subgroup analyses by region showed similar performance: Asia (0.81), Europe/UK (0.76), and US (0.75).

Conclusions:

• SRSM performance demonstrated considerable variability.
• The pooled C-statistics suggest fair predictive accuracy for SRSMs.
• A significant limitation is the lack of validation for most SRSMs in independent, diverse global populations.