Abstract WP242: Modeling Prevention, Prediction or Explanation of Stroke Risk: Insights From Meta-Analysis of Studies With Models Predicting Stroke or Composite Outcomes

Abstract

Objective: Models predicting stroke suffer from several problems: they are not accurate, contain mostly non-modifiable factors, and they explain more than they predict. We performed a meta-analysis of frequently used stroke models. Our goal was to estimate the discriminative ability of the concordance statistic by establishing confidence and predictive intervals. Methods: Studies with most representative predictive methods were used in our analysis. Inputs in analysis were study-reported c-index values and corresponding 95% CI. Subgroup analysis, separating survival and logistic regression models, and separating models by outcome (predicting stroke or composite outcome) were executed. Combined effect sizes with the random model, test for heterogeneity, and publication bias were considered. Egger’s test was used to assess for funnel asymmetry. Results: Fifteen models were included (patients= 13177; LR=7, Cox= 8; only stroke =9, composite=6) in the analysis. Models predicting composite outcomes performed worse (0.62; 95% CI: 0.59, 0.65) than model predicting only stroke (0.70; 95% CI: 0.64, 0.75); combined mean c-index was 0.66 (95% CI: 0.62, 0.70; 95% predictive interval: 0.55, 0.78). Test of heterogeneity showed high variation between studies (I 2 =74.9%). Egger’s test intercept was 1.59 (95% CI: –0.80, 3.98, P > .17). Conclusion: At the moment, there is no good quality predictive or explanatory model available for stroke risk. Current models do not indicate towards risk factors exclusive only to stroke. Because of this, models predicting stroke mostly have non-modifiable risk factors. In addition to this, models orienting toward explanation are borrowing heavily from models developed for predicting composite cardiovascular diseases.