Analyzing longitudinal antidiabetic medication patterns: a data-driven clustering framework

Area of Science:

• Diabetes management
• Pharmacotherapy research
• Health services research

Background:

• Glycemic control is crucial for preventing diabetes complications.
• Long-term antidiabetic medication use patterns are not well understood.
• This study addresses gaps in knowledge regarding medication trajectories.

Purpose of the Study:

• To identify common antidiabetic medication use patterns in Medicare beneficiaries.
• To analyze demographic and clinical characteristics across identified patient clusters.
• To understand long-term treatment pathways for diabetes management.

Main Methods:

• Retrospective cohort study of Medicare beneficiaries initiating metformin.
• Defined medication transitions: switch, intensification, de-intensification, discontinuation, re-initiation.
• Applied dynamic time warping and Partitioning Around Medoids clustering to analyze medication sequences.

Main Results:

• Identified 222 distinct medication patterns, grouped into five clusters.
• Most common patterns: continuous metformin, recurrent discontinuation/re-initiation, single discontinuation.
• Disparities noted: Non-Hispanic Black patients more likely to discontinue; females more likely to intensify.

Conclusions:

• Sequence analysis and clustering reveal complex medication use patterns.
• Understanding treatment trajectories and disparities is key for improving adherence.
• Findings support data-driven decisions for equitable diabetes care.