# Prediction models for incident stroke in the community: a systematic review and meta-analysis of predictive performance

## TL;DR

This study reviews and evaluates stroke prediction models in community settings, finding that most lack sufficient validation and clinical utility for real-world use.

## Contribution

A systematic review and meta-analysis of stroke prediction models, highlighting gaps in validation and clinical applicability.

## Key findings

- Only two models (R-FSRS and Basic IS) showed acceptable discrimination performance with c-statistics around 0.7.
- Most models had high risk of bias and poor calibration reporting.
- Few models were externally validated or tested for clinical utility.

## Abstract

Stroke is the second leading cause of death and the third leading cause of disability worldwide. We performed a systematic review and meta-analysis of multivariable models applicable to the prediction of incident stroke in community cohorts.

Ovid Medline and Embase were searched for studies related to stroke and prediction models from inception to 3 November 2025. Measures of discrimination were extracted and pooled by Bayesian meta-analysis, with heterogeneity assessed through a 95% prediction interval (PI). Risk of bias was assessed using the Prediction model Risk Of Bias Assessment Tool and certainty in effect estimates by Grading of Recommendations, Assessment, Development and Evaluation. Forty-one studies met the inclusion criteria, describing 80 prediction models, with two (R-FSRS and Basic IS) eligible for meta-analysis, including 969 514 participants. Both R-FSRS (summary c-statistic 0.714, 95% CI 0.681–0.747) and Basic IS (0.709, 95% CI 0.647–0.769) showed acceptable discrimination performance. Risk of bias was high in 66% of models, and both models showed reduced performance when excluding development cohorts and studies at high risk of bias (R-FSRS, 0.667, 95% CI 0.604–0.727; Basic IS 0.701; 95% CI 0.583–0.807). Only 43% of studies reported calibration, and no model underwent clinical utility analysis or a clinical impact study.

Many models have been derived for stroke prediction, however, they are rarely externally validated, and studies are limited by a high risk of bias, poor reporting of calibration and a lack of clinical utility analysis or prospective validation. Thus, the evidence base is insufficient to translate these models to clinical practice.

Graphical Abstract