Prognostic models in populations with heart failure: a systematic review and meta-analysis

Background: Heart failure (HF) remains a major cause of morbidity and mortality, highlighting the need for reliable prognostic models. This study provides a systematic review and meta-analysis of prognostic models focused on mortality, hospitalization, and their composite event. Methods: We screened 2271 papers and reviewed 58 prognostic models from 44 studies involving 362,759 HF patients. The predictive performance of these models was assessed, and a meta-analysis was performed for the Seattle Heart Failure Model (SHFM), which focuses on mortality outcomes at 1 year. The models were evaluated via the PROBAST tool for risk of bias and applicability. Results: Of the 58 models, 86% underwent internal and/or external validation in independent cohorts, with statistical models (88%) being more common than machine learning approaches (12%). Clinical data were used in 79% of the models, whereas the remaining models used electronic health records (EHR) or mixed sources of data. Mortality models (n = 40) revealed a 1-year discrimination range between 0.66 and 0.89. The most common predictors included age, renal function, blood pressure, coronary artery disease, and serum sodium. A meta-analysis of 5 studies that applied the SHFM at 1 year revealed a pooled C-statistic of 0.71 (95% CI: 0.64-0.78), with relatively low heterogeneity (τ2 = 0.003). Hospitalization models (n = 9) demonstrated discrimination up to 0.86, and composite event models (n = 9) showed similar predictive power. The risk of bias was high in 88% of the models, largely due to univariable predictor selection and handling/reporting of missing values. Conclusions: This systematic review highlighted the heterogeneity of HF prognostic models and patient populations in terms of severity and symptoms, emphasizing challenges in developing commonly applicable tools. Most studies enrolled patients with reduced ejection fraction (EF), whereas evidence for HF with preserved EF was limited. Despite widespread research, few HF prognostic models meet current standards for clinical implementation. The large majority of the studies did not report calibration and had a poor alignment with contemporary therapies. Future model development should prioritize transparency, methodological rigor, and external validation. Systematic review registration: PROSPERO CRD42023488017.