Key result
Random Forest model outperforms conventional logistic regression in predicting 365-day readmission with ~0.77 AUC.
Why the study?
Unplanned one-year readmission after PCI for STEMI poses serious clinical and economic challenges, motivating development of predictive models for risk stratification.
Cohort (n=1,836)
No
A random forest machine learning model accurately identifies ST-segment elevation myocardial infarction patients at risk of one-year unplanned readmission following primary PCI, offering better overall discrimination than traditional logistic regression.
Answers are grounded in this paper and cite their sources.
May support ML readmission prediction after STEMI PCI; leaves open prospective validation and outcome impact.
Introduction: Unplanned one-year readmission after percutaneous coronary intervention (PCI) for ST-segment elevation myocardial infarction (STEMI) poses serious clinical and economic challenges. This study developed and validated a random forest (RF) model to predict one-year all-cause unplanned readmission in STEMI-PCI patients and identify key risk factors. Study Design: A single-center retrospective cohort study. Methods: This single-center retrospective cohort study was conducted at Farshchian Heart Hospital from September 2021 to September 2024. Overall, 1,836 STEMI patients undergoing primary PCI were analyzed to predict 365-day unplanned readmission. The primary outcome was 365-day unplanned readmission (binary: readmitted or not). A RF classifier was developed on an 80/20 training-test split, optimized through repeated cross-validation with SMOTE for class imbalance. A SMOTE-RF model was also employed to mitigate the significant class imbalance inherent in the 17.6% readmission rate. Finally, key predictors were identified using mean decrease in Gini impurity. Results: Operating at an optimized threshold, the RF model demonstrated moderate discriminative ability (AUC: 0.765) with a sensitivity of 62.5% and specificity of 78.5%. Moreover, the RF model outperformed logistic regression in terms of AUC (0.695 vs. 0.765) and specificity (78.5% vs. 52%), although logistic regression achieved higher sensitivity (78.5% vs. 62.5%). Multivessel disease, smoking, length of stay, hypertension, ejection fraction, and age were key predictors of readmission identified by the RF. Conclusion: The RF model exhibited acceptable predictive performance for one-year hospital readmission, enabling clinically interpretable risk stratification and identifying key clinical risk factors to inform tailored post-discharge care. However, external validation is required to confirm its generalizability and clinical applicability.
No takes yet. Share an insight, caveat, or question.
Farhadian et al. (2026) conducted a cohort in STEMI undergoing primary PCI (n=1,836). Random Forest (RF) machine learning model vs. Multivariable logistic regression model was evaluated on 365-day all-cause unplanned hospital readmission (prediction AUC). A Random Forest machine learning model predicted 365-day unplanned readmission with moderate discriminative ability (AUC 0.765), outperforming a conventional logistic regression model (AUC 0.695).
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: