Mortality Prediction in Heart Failure Patients: Machine Learning Versus Get With The Guidelines-Heart Failure (GWTG-HF) and Acute Decompensated Heart Failure National Registry (ADHERE)

Background Predicting the mortality risk in Heart Failure (HF) patients is crucial for identifying high-risk individuals and establishing appropriate treatment strategies. Machine-learning algorithms offer immense potential for the risk prediction of cardiovascular conditions. This study aimed to evaluate and compare the predictive performance of Extreme Gradient Boosting (XGB), a machine-learning algorithm, with two traditional scores, Acute Decompensated Heart Failure National Registry (ADHERE) and Get With The Guidelines-HF (GWTG-HF), for mortality risk in patients admitted to the Cardiac Intensive Care Unit (CICU) with HF. Methods Using data from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database, we selected patients who were 18 years or older, with a diagnosis of HF admitted to CICU. We developed a risk prediction model using the XGB algorithm and computed the ADHERE and GWTG-HF scores for comparison. The predicted variable was the in-hospital mortality rate. The performance of these scores and the XGB model was assessed using the area under the receiver operating characteristic curve (AUC). Results A total of 5602 adult patients were included in the study and randomly divided into a derivation group (n = 4481, 80%) and a validation group (n = 1121, 20%) for analysis. The analysis revealed 346 (6.2%) in-hospital deaths. The performance of our XGB-based model (AUC = 72.1%, 95% CI: 65.23%-79%) was superior to that of GWTG-HF (AUC = 65.5%, 95% CI: 58.7%-72.2%) and ADHERE (AUC = 63.4%, 95% CI: 56.6%-70.2%). Conclusion The XGB-based model demonstrated superior performance to the ADHERE and GWTG-HF models, suggesting its utility for enhancing clinical decision-making. Routine implementation and evaluation in prospective studies are indicated to validate its potential in real-world settings.