Risk prediction models for incident heart failure: a systematic review and meta-analysis

Early identification of patients at risk of heart failure (HF) provides opportunities for preventative management. Though models have been developed to predict HF incidence, their validation remains unclear. Our objective was to summarise the performance, as observed in validation studies, of risk prediction models for incident HF.

In addition to articles from three previous systematic reviews, a search in Medline and Embase from 2014 to 2025 identified derivation or validation studies for incident HF prediction models. Performance was assessed in models validated in ≥1 cohort, with random-effects meta-analyses used to pool discrimination measures, and calibration descriptively summarised. We used the Prediction Model Risk Of Bias Assessment Tool to assess risk of bias in individual studies and the Grading of Recommendations, Assessment, Development and Evaluation approach to assess certainty in inferences drawn from the evidence.

From 24 531 publications identified, 76 studies representing 238 models were included. Risk of bias was high in 82.9% of assessments. With moderate to high certainty, among 64 models validated in at least one cohort, four models had moderate and eight models had high discrimination. In patients with low predicted risk, calibration may have been adequate. The Predicting Risk of CVD EVENTs (PREVENT), Atherosclerosis Risk in Communities (ARIC), and Multi-Ethnic Study of Atherosclerosis (MESA) models were most promising for further validation and impact studies. Among externally validated models, 14 were derived using machine learning, six incorporated novel biomarkers such as proteomics and polygenic risk scores, and nine included measures of social determinants of health as predictors.

The PREVENT, ARIC and MESA risk scores demonstrate promising performance and should be prioritised for further validation and progression to impact studies.

CRD42021266756.