Using machine learning to predict pharmaceutical interventions during medication prescription review in a hospital setting.

Abstract

Disclaimer: In an effort to expedite the publication of articles, AJHP is posting manuscripts online as soon as possible after acceptance. Accepted manuscripts have been peer-reviewed and copyedited, but are posted online before technical formatting and author proofing. These manuscripts are not the final version of record and will be replaced with the final article (formatted per AJHP style and proofed by the authors) at a later time.

Objective: Medication errors are a worldwide public health issue. Reducing inappropriate medication use is a daily challenge for clinical pharmacists. Computerization of the medication process and the rise of artificial intelligence make it possible to develop tools to detect inappropriate prescriptions. Our main goal was to compare the performance of two machine learning models capable of predicting the probability of a prescription requiring pharmaceutical intervention (PI) using hospital data.

Methods: The study was conducted in a single hospital, with data collected over 4 years, including 2,059,847 prescription lines ([INSERT DEFINITION]) associated with 260,611 PIs. Two tree-based binary classification machine learning models were tested: the Light Gradient Boosting Machine (LGBM) model and the Random Forest (RF) model. The dataset was split (70% for training and 30% for testing), and training and testing were performed on the global dataset and on data stratified by medical care department.

Results: For the global dataset, the LGBM model outperformed the RF model in most metrics: accuracy (86% vs 85%), precision (80% vs 42%), specificity (97% vs 89%), area under the curve (83% vs 71%) and F1-score (58% vs 47%). However, the RF model had superior recall (53% vs 46%). Furthermore, the LGBM model trained on the global database was generally more effective than models trained on the care departments' databases.

Conclusion: The LGBM model showed superior performance in detecting inappropriate prescriptions, potentially improving the thoroughness and efficiency of prescription review. While further studies are needed to confirm these findings, the model holds significant promise for advancing hospital clinical pharmacy and enhancing patient care through optimized prescription management.