OBJECTIVE: Reducing preventable hospital readmissions (HR) could potentially improve outcomes and decrease healthcare costs. This study aimed to build machine learning (ML) models to predict readmissions within 12 months after first hospitalization.

METHODS: Design: A retrospective non-interventional study using population-based health administrative database. Setting: payer-provider healthcare organization. Participants: 52,329 beneficiaries, with 8,033 clinical hospital admissions in three different periods, according to the year of occurrence: 2018(P1), 2019(P2), and 2020(P3). The main outcome to be predicted was any hospital readmission within 12 months of the first admission. Variables: age, gender, Charlson Comorbidity Index (CCI), cancer diagnosis, the average length of stay (ALOS), and number of avoidable admissions, ICU admissions, emergency room visits, therapies.

RESULTS: Of the total number of admissions, 57.0% were women (mean age 68.0 years). The average rate of hospital readmission within 12 months was 19.4% (P1), 22.3% (P2), and 21.0% (P3). ML models tested were logistic regression, decision trees, random trees, and support vector machines, from the Scikit Learn Python package. All algorithms were able to predict HR with a ROC curve greater than 0.70 from data. A ranking among the algorithms showed that the Support Vector Machines had a larger area under the curve (0.81) and greater accuracy (0.82) in the model. The independent variables that most influenced these results were obtained through logistic regression: Admissions Sensitive to Primary Care Conditions Index, ALOS and CCI.

CONCLUSIONS: When test data at a base was applied to our trained model, it had an overall 82% accuracy in targeting patients who had a readmission. Identifying patients at high risk with these models can enable early discharge planning and transitional care to prevent readmissions. Further studies should include additional features that may enable further sensitivity in identifying patients at a risk of early unplanned readmissions and eligible for Case Management Programs.