Key result
AutoML model integrating PROMs predicts 12-month heart failure readmission with ~0.89 AUC.
Why the study?
Traditional readmission risk models for chronic heart failure have limitations in processing multidimensional physiological-psychological-social data and low clinical translation efficiency.
Does an automated machine learning model integrating patient-reported psychosocial outcome measures accurately predict 12-month all-cause readmission in patients with chronic heart failure?
Cohort (n=552)
No
Does an automated machine learning model integrating patient-reported psychosocial outcome measures accurately predict 12-month all-cause readmission in patients with chronic heart failure?
Effect estimate: AUC 0.8909 (95% CI 0.8392-0.9426)
Absolute Event Rate: 0.8909% vs 0.6064%
Integrating patient-reported psychosocial factors into an automated machine learning framework demonstrates high accuracy in predicting 12-month readmission risk in patients with chronic heart failure.
May support AutoML readmission tools integrating PROMs in chronic HF; leaves open prospective validation before practice adoption.
Objective: To construct a readmission risk prediction model for heart failure based on an Automated Machine Learning (AutoML) framework and develop a Clinical Decision Support System (CDSS) integrating Patient-Reported Outcome Measures (PROM), aiming to address the limitations of traditional models in processing multidimensional physiological-psychological-social data and their low clinical translation efficiency. Methods: A retrospective cohort study design was employed, enrolling 552 chronic heart failure patients (training-to-testing set ratio of 8:2). Multidimensional features were collected via electronic medical records and the Chinese version of the Heart Failure-Specific PROM Scale (CHF-PROM). Missing data were handled using Multivariate Imputation by Chained Equations (MICE). An AutoML framework driven by an Improved Pied Kingfisher Optimizer (IPKO) algorithm, integrating algorithms including Logistic Regression, Support vector machine (SVM), XGBoost, and LightGBM, was developed. Model performance and interpretability were evaluated using the Area Under the Receiver Operating Characteristic Curve (AUC), Decision Curve Analysis (DCA), and SHapley Additive exPlanations (SHAP). A visualized decision-making tool was implemented on the MATLAB platform. Results: Testing set showed the AutoML model achieved significantly enhanced predictive efficacy [AUC = 0.8909, 95% Confidence Interval (CI): 0.8392 to 0.9426]. SHAP analysis revealed key multidimensional factors influencing prognosis, among which milder depressive symptoms, higher levels of social support, and favorable appetite-sleep status were important protective factors reducing readmission risk, while female sex and advanced age were identified as risk predictors. Decision curve analysis further confirmed that applying this model yielded positive net clinical benefit across a wide risk threshold range of 1.2% to 78.6%. Conclusion: This study achieved efficient translation of PROM data into clinical decisions and innovatively demonstrated the independent predictive value of psychosocial factors for readmission. The developed CDSS system provides a feasible pathway for the proactive prevention and management of heart failure.
No takes yet. Share an insight, caveat, or question.
Chen et al. (2026) conducted a cohort in Chronic heart failure (n=552). AutoML predictive model integrating Patient-Reported Outcome Measures (PROM) vs. Logistic regression model was evaluated on Predictive accuracy for 12-month all-cause readmission (ROC-AUC) (AUC 0.8909, 95% CI 0.8392-0.9426). An Automated Machine Learning (AutoML) prediction model integrating patient-reported outcome measures demonstrated high predictive efficacy for 12-month all-cause readmission in patients with chronic heart failure (AUC 0.8909).
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: