Machine learning model better identifies patients for pharmacist intervention to reduce hospitalization risk in a large outpatient population

Background: Medication costs, adverse drug events, and associated avoidable hospitalizations exacerbate the strain on the healthcare system. Precisely selecting patients who are at high risk for hospitalization can benefit from pharmacist intervention by reducing their risk of hospitalization. Fewer hospitalizations can increase care delivery efficiency and may improve patient outcomes. Traditionally, polypharmacy has been the key metrics to select patients for pharmacist intervention. A predictive machine learning (ML) model can incorporate multiple patient metrics and be trained on many patients’ historical data to better determine the risk of hospitalization. This manuscript describes a novel predictive ML model used to improve patient selection for pharmacist intervention via better prediction of hospitalization rate. Methods: Deidentified hospitalization data were collected between December 1, 2021 and August 10, 2022. Deidentified patient data and demographics (n=9,797 patients) from two large outpatient medical groups were used to train and validate an ML model to predict hospitalization risk by incorporating metrics including previous hospitalizations, select drug classes, drug expenditure, and medication adherence. This model identified patients with the highest hospitalization risk using these characteristics. The model-based approach was deployed in a separate population and compared to the standard method of selecting patients based on only polypharmacy, defined as patients with 8 or more current active medications, to determine differences in hospitalization rate observed in each respective group. Results: The incorporation of the predictive ML model was validated during a 3-month collection period. The model demonstrated an 80% accuracy rate [area under the curve (AUC) >0.8]. The ML model predicted high risk patients and improved the true positive hospitalization rate by a factor of 3.5. The mean 90-day hospitalization rate for the top 15% of patients selected based only on polypharmacy was 0.120 hospitalization per patient. The 90-day hospitalization rate of the top 15% of patients as selected by the ML model was 0.285 hospitalizations per patient (P<0.001). Conclusions: An ML approach demonstrated superior efficacy stratifying outpatient hospitalization risk and potential utility for selecting patients for pharmacist intervention. This represents a novel approach to optimizing the deployment of healthcare resources and saving healthcare dollars. Further deployment of this model in a larger, more diverse population may increase generalizability and potential benefits of an ML-based approach to improving healthcare workforce efficiency.