Traditional Machine Learning Outperforms Automated Machine Learning for Postpartum Readmission Prediction: A Comprehensive Performance and Health-Economic Analysis

Automated machine learning (AutoML) promises to democratize predictive modeling in healthcare by automating algorithm selection and hyperparameter optimization. Our objective was to compare the performance, clinical utility, and health-economic implications of traditional machine learning versus AutoML frameworks for predicting 14-day postpartum readmission using a large national cohort. We analyzed data from 8,774 participants in the nuMoM2b (Nulliparous Pregnancy Outcomes Study) with complete readmission data. Three traditional ML algorithms (logistic regression, random forest, gradient boosting) were compared with two leading AutoML frameworks (FLAML, TPOT). Models were evaluated on discrimination (ROC-AUC), clinical utility (sensitivity/specificity), and computational efficiency. Health-economic analysis included development costs, computational resources, clinical impact valuation, and return on investment using current healthcare costs. Overall, 1.8% of participants experienced 14-day readmission. Logistic regression achieved the highest ROC-AUC (0.7465) and was the only model with clinically meaningful sensitivity (51.6%), correctly identifying 16 of 31 readmissions. Both AutoML frameworks showed moderate discrimination (FLAML: 0.7156, TPOT: 0.6556) but had zero sensitivity, missing all readmissions. Training times favored traditional ML (logistic regression: 0.07s vs. FLAML: 143.6s vs. TPOT: 303.4s). Health-economic analysis showed logistic regression yielded a 3,492% return on investment and $325,080 in annual clinical benefit per 10,000 deliveries, whereas AutoML approaches offered no clinical benefit despite higher implementation costs. Traditional logistic regression outperformed AutoML frameworks for postpartum readmission prediction, providing the only clinically actionable screening capability at a fraction of the cost. For critical healthcare applications requiring high sensitivity and interpretability, domain-informed traditional machine learning approaches deliver superior clinical and economic value compared with automated systems. Traditional logistic regression outperformed two leading AutoML frameworks, uniquely identifying high-risk patients and demonstrating that more complex algorithms do not guarantee clinical utility. Simple traditional models were both cost-saving and outcome-improving, whereas AutoML yielded no clinical benefit despite consuming resources, underscoring the economic and implementation advantages of parsimonious methods. As one of the first studies to jointly evaluate performance, cost-effectiveness, and runtime, this work clarifies when AutoML adds value and shows that transparent, fast traditional models may be better suited for real-world clinical decision support.