Observing versus Predicting: Initial Patterns of Filling Predict Long-Term Adherence More Accurately Than...

Area of Science:

• Health Informatics
• Pharmacoeconomics
• Clinical Pharmacy

Background:

• Medication adherence prediction is crucial for effective treatment outcomes.
• Existing prediction models using patient data show limited accuracy.
• There is a need for improved methods to predict patient adherence to medications.

Purpose of the Study:

• To evaluate novel approaches for enhancing medication adherence prediction.
• To assess the utility of early adherence indicators in predicting future adherence.
• To compare prediction model performance based on baseline and post-baseline data.

Main Methods:

• Utilized Medicare beneficiary data for statin initiators.
• Extracted 643 baseline variables from claims and Census data.
• Incorporated post-baseline adherence indicators and logistic regression models.

Main Results:

• Baseline variables alone yielded poor prediction (C-statistics ≤0.606).
• Early adherence indicators substantially improved prediction accuracy (C=0.827).
• Combining early indicators with baseline data further enhanced prediction (C=0.842).

Conclusions:

• Observed early adherence is a strong predictor of future adherence.
• This finding is particularly relevant for patients with shorter initial prescription durations.
• Improved adherence prediction can inform targeted interventions.