Applying Machine Learning Models Derived From Administrative Claims Data to Predict Medication Nonadherence in...

Area of Science:

• Gastroenterology
• Health Informatics
• Machine Learning in Healthcare

Background:

• Adherence to self-administered biologic therapies is crucial for managing Inflammatory Bowel Disease (IBD).
• Nonadherence can lead to disease remission failure and adverse clinical outcomes.
• Predictive models are needed to identify patients at risk of nonadherence.

Purpose of the Study:

• To develop and test machine learning models for predicting nonadherence to self-administered biologic therapies in IBD patients.
• To evaluate the utility of administrative claims data in predicting medication nonadherence.
• To assess the performance of 48 machine learning models using a tertiary academic medical center's electronic medical record data.

Main Methods:

• Extracted administrative claims data from Commercial Claims and Encounters and Medicare Supplemental databases.
• Defined nonadherence as a proportion of days covered ratio <80% at 1 year.
• Trained and tested 48 machine learning models, using the area under the receiver operating characteristic curve (AUC) as the primary performance metric.

Main Results:

• The training dataset comprised 6998 IBD beneficiaries, with 38.3% nonadherent.
• The testing dataset included 285 patients, 47.0% of whom were nonadherent.
• The best-performing models achieved an AUC of 0.55 on test data, indicating poor predictive performance.

Conclusions:

• Machine learning models trained on administrative claims data demonstrated an inability to predict biologic medication nonadherence in IBD patients.
• Future research should incorporate datasets with enriched demographic and clinical information.
• Enhanced data may improve the accuracy of predictive models for medication adherence in IBD.