Predicting ICU Transfer and Short-term Mortality in Emergency Department Atrial Fibrillation Patients: An Enhanced Machine Learning Model Using MIMIC Data

Abstract Atrial fibrillation (AF) is a prevalent condition in emergency department (ED) patients and is associated with an elevated risk of intensive care unit (ICU) transfer and short-term mortality. Early identification of high-risk patients is critical for timely intervention and improved clinical outcomes. We constructed a combined cohort from the MIMIC-IV-ED and MIMIC-IV databases, comprising ED admissions of patients with AF, and developed an interpretable machine learning (ML) framework to predict ICU transfer and mortality within 3, 7, and 30 days using clinical variables obtained at triage. The preprocessing pipeline included imputation of missing data, z-score normalization of numerical features, one-hot encoding of categorical variables, and correction for class imbalance using the Synthetic Minority Over-sampling Technique (SMOTE). A hybrid feature selection strategy combining Recursive Feature Elimination with Cross-Validation (RFECV) and Least Absolute Shrinkage and Selection Operator (LASSO) regression reduced the initial set to 19 clinically relevant predictors. Among six evaluated machine learning algorithms, LightGBM demonstrated the highest performance for ICU transfer (AUROC = 0.7979, 95% confidence interval (CI): 0.7916–0.8041) and for 7- and 30-day mortality (AUROC = 0.8316, 95% CI: 0.8156–0.8476; AUROC = 0.8010, 95% CI: 0.7898–0.8123), while CatBoost achieved the best performance for 3-day mortality (AUROC = 0.8444, 95% CI: 0.8237–0.8644). SHAP(SHapley Additive exPlanations) analysis identified O2sat, acuity, and resprate as key determinants, underscoring the clinical plausibility and interpretability of the models. These findings highlight the potential of interpretable machine learning approaches to enable early, time-sensitive risk stratification and support informed clinical decision-making for AF patients in the ED. Competing Interest Statement The authors have declared no competing interest. Funding Statement The author(s) received no specific funding for this work. Author Declarations I confirm all relevant ethical guidelines have been followed, and any necessary IRB and/or ethics committee approvals have been obtained. Yes The details of the IRB/oversight body that provided approval or exemption for the research described are given below: This study used de-identified, publicly available data from the MIMIC-IV and MIMIC-IV-ED databases. Access to these databases was approved under the data use agreement with the MIT Institutional Review Board (IRB) for research involving human subjects. No direct patient interaction or identifiable data was involved, and the study was conducted in accordance with relevant ethical guidelines. I confirm that all necessary patient/participant consent has been obtained and the appropriate institutional forms have been archived, and that any patient/participant/sample identifiers included were not known to anyone (e.g., hospital staff, patients or participants themselves) outside the research group so cannot be used to identify individuals. Yes I understand that all clinical trials and any other prospective interventional studies must be registered with an ICMJE-approved registry, such as ClinicalTrials.gov. I confirm that any such study reported in the manuscript has been registered and the trial registration ID is provided (note: if posting a prospective study registered retrospectively, please provide a statement in the trial ID field explaining why the study was not registered in advance). Yes I have followed all appropriate research reporting guidelines, such as any relevant EQUATOR Network research reporting checklist(s) and other pertinent material, if applicable. Yes Footnotes ↵¶ Membership list can be found in the Acknowledgments section. Data Availability The data underlying the results presented in this study are available from the publicly accessible MIMIC-IV and MIMIC-IV-ED databases, which are hosted on PhysioNet (https://physionet.org/content/mimiciv/ and https://physionet.org/content/mimiciv-ed/). Access to these databases requires registration, completion of the required human subjects training, and agreement to the data use policies. Researchers who meet these criteria can request and obtain the data directly from PhysioNet. All data necessary to replicate the analyses reported in this manuscript are included within the manuscript and its Supporting Information files, and additional details can be provided upon reasonable request.