Prediction models for incident heart failure and hospitalization in type 2 diabetes, such as DM-CURE (AUC 0.837; 95% CI 0.757–0.913), demonstrated acceptable to good long-term discrimination.
Meta-Analysis
Do multivariable prediction models accurately predict incident heart failure and heart failure hospitalization in individuals with type 2 diabetes?
Several prediction models for incident HF and HF hospitalization in type 2 diabetes show acceptable to good long-term discrimination, but their clinical translation is limited by a lack of prospective testing and clinical utility analyses.
Effect estimate: AUC 0.837 (DM-CURE) (95% CI 0.757-0.913)
Abstract Aims Heart failure (HF) is a common sequela of type 2 diabetes (diabetes), and models have been developed for its prediction. We aimed to synthesize the available evidence by performing a systematic review and meta-analysis of models predicting incident HF and HF hospitalization (HFH) in individuals with diabetes. Methods We searched MEDLINE and EMBASE for multivariable models predicting HF or HFH in patients with diabetes from inception to 2 December 2025. Discrimination metrics from models validated in ≥3 cohorts were pooled using Bayesian meta-analysis. Heterogeneity was assessed using 95% prediction intervals (PI), and risk of bias with PROBAST. Results In total, 65 studies describing 56 HF and 44 HFH prediction models were included. Following exclusion of studies at high risk of bias, two HF models (RECODe (0.711, 95% CI 0.651–0.767) 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). Of these, none reported a prediction horizon of less than ten years. For all models, regardless of risk of bias, none were prospectively tested, and only three underwent clinical utility analysis. Conclusion While some HF and HFH models in diabetes show acceptable/good long-term discriminative performance, the scarcity of utility analyses and prospective testing limits translation to clinical settings. Prospective evaluation of these models is required to establish clinical utility.
Mircescu et al. (Tue,) conducted a meta-analysis in Type 2 diabetes. Prediction models for incident HF and HFH was evaluated on Discrimination metrics for predicting incident HF and HFH (AUC 0.837 (DM-CURE), 95% CI 0.757-0.913). Prediction models for incident heart failure and hospitalization in type 2 diabetes, such as DM-CURE (AUC 0.837; 95% CI 0.757–0.913), demonstrated acceptable to good long-term discrimination.