The Role of Machine Learning in Predicting Hospital Readmissions Among General Internal Medicine Patients: A Systematic Review.

Hospital readmissions contribute significantly to healthcare costs. While traditional regression models for predicting 30-day readmission risk offer modest accuracy, machine learning (ML) presents an opportunity to capture complex relationships in healthcare data, potentially enhancing predictions. This review assesses the role of ML in predicting 30-day readmissions for general internal medicine admissions in the U.S. Following the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) guidelines, a literature search of PubMed (2014-2023) was conducted using the keywords "artificial intelligence," "machine learning," and "readmission." The review focused on ML models predicting readmissions in general internal medicine patients in the U.S. Nine studies were reviewed, covering conditions like acute myocardial infarction (AMI), heart failure (HF), pneumonia (PNA), chronic obstructive pulmonary disease (COPD), and other general internal medicine cases. ML models such as artificial neural networks (ANN), random forests (RF), gradient boosting, logistic regression, and natural language processing (NLP) were used. ANN and RF models outperformed traditional regression methods, while NLP-based approaches showed limited success. Subgroup modeling provided marginal improvements in predictive accuracy. In conclusion, ML offers significant potential for improving 30-day readmission predictions by overcoming the limitations of traditional models. ANN and RF are particularly effective in predicting readmissions in general internal medicine. To advance predictive capabilities, future research should refine NLP, subgroup modeling, and focus on model generalizability, integration of diverse data sources, and the development of explainable AI for clinical adoption. Addressing these challenges could transform healthcare delivery, improve patient outcomes, and reduce costs.