An exploration of machine learning to predict medication waste amongst rheumatoid arthritis patients.
Summary
Background: In case of treatment failure, rheumatoid arthritis (RA) patients switch their relatively expensive biological disease- modifying anti-rheumatic drug (bDMARD) therapy, which could lead to waste. Machine learning has potential to be used in pharmacy to predict medication waste.
Aim: To explore the application of machine learning to identify patterns in patient, clinical and medication factors that lead to medication waste.
Methods: In a retrospective cohort study, patient, clinical and medication data was collected from a Dutch outpatient pharmacy and hospital information system of patients (≥ 18 years) who received at least one bDMARD prescription, dispensed between January 2015 and December 2020. Medicine waste was defined as a treatment switch before its expected end date. A random forest model was used to identify predictors.
Results: The database included 1996 patients, of which 285 wasted at least 1 syringe of bDMARD. During the five-year study period, a total of 719 units were unused, with an economic value of €237,692. Out of 32,484 prescriptions, 324 lead to waste. The random forest model had a positive prediction value of 0.21, with total cost of a prescription, age, disease duration, as the highest predictors for medication waste.
Conclusion: bDMARD waste occurs when rheumatoid arthritis patients switch therapies. Machine learning has the potential to be used in waste preventing activities. With improvements to the model, such as down sampling, reducing features and correcting for correlations, it can be used to identify patterns which lead to medication waste.