Identification of target groups and individuals for adherence interventions using tree-based prediction models

Area of Science:

• Health Economics
• Data Science in Healthcare
• Pharmacoeconomics

Background:

• Medication adherence is critical for managing chronic illnesses and controlling healthcare expenditures.
• Traditional regression models may not capture the nuanced relationship between adherence and costs in all patient populations.
• Identifying subgroups with varying adherence effects is essential for optimizing intervention strategies.

Purpose of the Study:

• To estimate subgroup-specific and personalized medication adherence effects on healthcare costs.
• To identify target patient groups for adherence-focused interventions.
• To evaluate the utility of advanced modeling techniques (model-based trees and random forests) for this purpose.

Main Methods:

• Analysis of German claims data (2012-2015) for patients with type 1 diabetes, type 2 diabetes, and hyperlipidemia.
• Estimation of the association between medication adherence (proportion of days covered) and total healthcare costs.
• Comparison of linear regression, model-based trees, and model-based random forests to identify heterogeneous and personalized adherence effects.

Main Results:

• Linear models indicated a positive association between adherence and costs across all cohorts.
• Model-based trees identified patient subgroups, particularly in type 2 diabetes and hyperlipidemia, with negative adherence effects on costs.
• Model-based random forests revealed personalized adherence effects, with a significant proportion of patients experiencing decreased costs (up to -8.31 Euro) with higher adherence.

Conclusions:

• Tree-based modeling approaches effectively identify patient subgroups with heterogeneous medication adherence effects.
• Personalized adherence effect estimation is feasible and measurable, enabling the identification of patients for targeted interventions.
• This methodology can be extended to other health outcomes, such as hospitalization risk, to maximize intervention impact.