Predicting pharmacy choice for managed care network design.

Area of Science:

• Health Services Research
• Pharmacoepidemiology
• Health Informatics

Background:

• Pharmacy type selection is crucial for medication access and adherence.
• Traditional logistic regression may oversimplify complex patient characteristics influencing pharmacy choice.

Purpose of the Study:

• To identify patient characteristics associated with pharmacy type selection.
• To compare the predictive performance of logistic regression against ensemble machine learning models for pharmacy choice.

Main Methods:

• Cross-sectional analysis of 1,502 adults from the 2021 National Consumer Survey on Medication Experience.
• Compared logistic regression with random forest and XGBoost using 5-fold cross-validation and held-out test data.
• Assessed model discrimination using the area under the receiver operating characteristic curve.

Main Results:

• Ensemble models, particularly XGBoost, showed superior discrimination compared to logistic regression.
• Key predictors of pharmacy type included prior mail-order use, chronic conditions, income, and US region.
• Ensemble models effectively captured nonlinear relationships between patient factors and pharmacy choice.

Conclusions:

• Pharmacy selection is influenced by patient demographics and health status, varying by pharmacy type.
• The analytical approach significantly impacts conclusions drawn about pharmacy choice.
• Findings underscore the importance of advanced modeling for understanding medication access and informing future research.