Key result
XGBoost model identifies in-hospital heart failure in older AF patients with ~0.74 AUROC.
Why the study?
Real-world machine learning models needed to be developed, internally evaluated, and compared to identify clinically adjudicated HF during hospitalisation in older patients with AF.
Does a machine learning model (XGBoost) accurately identify clinically adjudicated heart failure during hospitalisation in older patients with atrial fibrillation?
Observational (n=1,989)
No
Does a machine learning model (XGBoost) accurately identify clinically adjudicated heart failure during hospitalisation in older patients with atrial fibrillation?
Effect estimate: AUROC 0.737
An XGBoost machine learning model using 16 routine clinical and hematologic variables demonstrated moderate discriminatory ability (AUROC 0.737) for identifying in-hospital heart failure among older patients admitted with atrial fibrillation.
ML models may aid HF identification in older AF inpatients; leaves open external validation and prospective testing before clinical use.
Objective This study aimed to develop and internally evaluate a machine learning (ML) model based on real-world data for identifying clinically adjudicated heart failure (HF) during hospitalisation in older patients with atrial fibrillation (AF) and to compare various models for such an assessment approach. Methods Older patients with AF who were admitted to the General Hospital of Ningxia Medical University between 2023 and 2024 were included. Demographic characteristics, vital signs, comorbidities, medication use, and routine laboratory measurements were extracted from electronic medical records. Least absolute shrinkage and selection operator (LASSO) regression was used to select candidate predictors, after which five ML models were developed and compared. Separate, non-overlapping training and test sets were used for model development and evaluation. Model performance was assessed in terms of discrimination, calibration, potential clinical net benefit, and interpretability. Results This study included 1,989 patients, 11% of whom experienced HF during hospitalisation. Through LASSO regression, 16 key variables were identified, including age, vital signs, comorbidities, and multiple hematologic indicators. Among the five models, XGBoost performed the best on the test set (AUROC 0.737), exhibiting balanced accuracy, sensitivity, and specificity without marked overfitting. The calibration curve demonstrated strong concordance between model predictions and actual risks. The decision curve analysis (DCA) demonstrated high net benefit within the 5%–40% threshold range. SHAP analysis identified age, RDW series indicators, anemia indicators, and infection-related variables as the most significant features, aligning with the pathophysiology of HF. Conclusion The XGBoost model for identifying clinically adjudicated HF among hospitalised older patients with AF showed moderate discriminatory ability, calibration, and exploratory findings from decision curve analysis, as well as high interpretability. This model may serve as an exploratory admission-based assessment approach for identifying patients who meet the prespecified criteria for HF during hospitalisation. Prospective external validation is required before its potential clinical application can be determined.
No takes yet. Share an insight, caveat, or question.
Chen et al. (2026) conducted an observational in Atrial fibrillation with hospitalisation-associated heart failure (n=1,989). XGBoost machine learning model vs. Four other machine learning models was evaluated on Clinically adjudicated heart failure during hospitalisation (AUROC 0.737). An XGBoost machine learning model achieved the best discrimination (AUROC 0.737) among five models for identifying in-hospital heart failure in older patients with atrial fibrillation.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: