Systematic review and meta-analysis of prediction models of incident heart failure and heart failure hospitalisation in type 2 diabetes
A Mircescu, W Allan, M Haris, B Hurdus, K Raveendra, A Helbitz, W Ginks, J Wu, N Sattar, M C Petrie, R Ajjan, R Nadarajah, C P GaleAbstract
Background
Heart failure (HF) is a frequent and serious complication of type 2 diabetes, associated with increased hospitalisation and mortality. Accurate prediction models could enable targeted screening and early intervention, yet their performance and clinical applicability remain unclear.
Purpose
We conducted a systematic review and meta-analysis to synthesise evidence on models predicting incident HF and HF hospitalisation (HFH) in individuals with type 2 diabetes. Our objective was to assess discrimination, robustness, and readiness for clinical implementation, addressing gaps in prior reviews.
Methods
MEDLINE and EMBASE were searched from inception to December 2025 for studies developing or validating HF or HFH prediction models in diabetes-prevalent cohorts. Discrimination metrics from models validated in ≥3 cohorts were pooled using Bayesian meta-analysis. Risk of bias was assessed with PROBAST, and heterogeneity quantified using 95% prediction intervals.
Results
Sixty-five studies describing 56 HF and 44 HFH models were included. Following exclusion of studies at high risk of bias, four HF models (RECODe (0·711, 95% CI (0·651-0·767), PARFAIT (W) (0·791, 95% CI 0·785-0·797), PARFAIT(M) (0·772, 95% CI 0·768-0·776), and DMRS (0·758, 95% CI 0·684-0·827) and two HFH models (WATCH-DM 2022 (r) (0·705, 95% CI, 0·624-0·795), WATCH-DM 2022(i) (0·718, 95% CI 0·587-0·850)) had acceptable prediction performance, while one HFH model had good performance (DM-CURE (0·837, 95% CI 0·757-0·913). No model predicted short-term (<10 years) risk, and none were prospectively tested; only three reported clinical utility analyses.
Conclusion
Several models demonstrate acceptable or good long-term discrimination for HF and HFH in type 2 diabetes, but lack of prospective evaluation and absence of short-term prediction horizons remain major barriers to clinical translation. Future research should focus on prospective testing and development of short-term, or phenotype-specific models to confirm clinical utility and improve targeted prevention strategies.Meta-analysis of c-stats for HF modelsFor image description, please refer to the figure legend and surrounding text.Meta-analysis of c-stats for HFH modelsFor image description, please refer to the figure legend and surrounding text.