Performance of the heart failure risk scores in predicting 1 year mortality and short‐term readmission of patients

AIMS: The aim of this study was to assess the performance of these main scores in predicting prognosis in patients with heart failure (HF). METHODS AND RESULTS: A total of 2008 patients who were admitted to the Fourth People's Hospital of Zigong, Sichuan, from December 2016 to June 2019 and diagnosed with HF were included in the study. We compared the prognostic predictive performance of Seattle Heart Failure Model (SHFM), Meta‐Analysis Global Group in Chronic Heart Failure (MAGGIC‐HF) risk score, Get With the Guidelines‐Heart Failure programme (GWTG‐HF), Acute Study of Clinical Effectiveness of Nesiritide in Decompensated Heart Failure (ASCEND) risk scores, the Acute Decompensated Heart Failure National Registry (ADHERE) model, Barcelona Bio‐Heart Failure (BCN‐Bio‐HF) risk calculator, and Gruppo Italiano per lo Studio della Streptochinasi nell'Infarto Miocardico‐Heart Failure (GISSI‐HF) for the endpoints. The primary endpoint was 1 year all‐cause mortality and the secondary endpoint was the incidence of 28 day readmission post‐discharge. At 1 year follow‐up, 44 (2.21%) patients with HF died. Discrimination analyses showed that all risk scores performed reasonably well in predicting 1 year mortality, with areas under the receiver operating characteristic curve (AUCs) fluctuating between 0.757 and 0.822. GISSI‐HF showed the best discrimination with the AUC of 0.822 (0.768–0.876), followed by MAGGIC‐HF, BCN‐Bio‐HF, ASCEND, SHFM, GWTG‐HF, and ADHERE with AUCs of 0.819 (0.756–0.883), 0.812 (0.758–0.865), 0.802 (0.742–0.862), 0.787 (0.725–0.849), 0.762 (0.684–0.840), and 0.757 (0.681–0.833), respectively. All risk scores were similarly predictive of 28 day emergency readmissions, with AUCs fluctuating between 0.609 and 0.680. Overestimation of mortality occurred in all scores except the ASCEND. The risk scores remained with good prognostic discrimination in patients with biventricular HF and in the subgroup of patients taking angiotensin‐converting enzyme inhibitor/angiotensin II receptor blocker. CONCLUSIONS: Currently assessed risk scores have limited clinical utility, with fair accuracy and calibration in assessing patients' 1 year risk of death and poor accuracy in assessing patients' risk of readmission. There is a need to incorporate more patient‐level information, use more advanced technologies, and develop models for different subgroups of patients to achieve more practical, innovative, and accurate risk assessment tools.