WHEN ADHERENCE ESTIMATES DIFFER: THE IMPACT OF DATA SOURCE AND GAP HANDLING IN REAL-WORLD MEDICATION USE ASSESSMENT
Author(s)
Yifan Zheng, PharmD, Corey A. Lester, PharmD. PhD.;
University of Michigan College of Pharmacy, Department of Clinical Pharmacy Translational Science, Ann Arbor, MI, USA
University of Michigan College of Pharmacy, Department of Clinical Pharmacy Translational Science, Ann Arbor, MI, USA
OBJECTIVES: Medication adherence metrics are used to identify patients for intervention, evaluate quality of care, and inform value-based payment decisions. However, estimates may vary with data sources and how medication use gaps are handled. We examined how these choices impact adherence estimates using real-world data.
METHODS: We conducted a retrospective cohort study using longitudinal e-prescribing data (2009-2019) with written, fill, and sold dates. Adults (N=7,226) from an academic medical center with ≥12 months follow-up and ≥3 prescriptions within a therapeutic class (ACE-inhibitors, ARBs, calcium channel blockers, diuretics) were included. Adherence was measured using the proportion of days covered (PDC) under two conceptualizations: (1) exposure-based PDC, which treats all gaps as non-exposure, and (2) adherence-based PDC, which leverages personal historical use patterns to dynamically adjust gaps. PDC was calculated using various timestamps, and patients were classified as adherent (PDC≥0.80). We compared adherence estimates and classification stability across approaches.
RESULTS: Patients contributed 230,445 fills with a mean longitudinal follow-up of approximately 3-4 years. Exposure PDC varied across data sources (median 0.747-0.794; p<0.001), with written dates yielding the lowest values and sold dates the highest. Adherence PDC was uniformly high and less source-sensitive (median 0.959-0.977; p<0.001). At the 0.80 threshold, Adherence PDC classification showed >94% concordance across data sources, whereas Exposure PDC was sensitive to source choice; switching from fill to written dates led to substantial reclassification, with >15% of patients changing adherence status.
CONCLUSIONS: Data source and gap handling significantly influence adherence estimates and patient classification. Approaches that incorporate personal past fill history yield more stable classification, which may be preferable for longitudinal outcomes research and quality programs such as Medicare Star Ratings where misclassification can affect plan-level performance and payment. Future studies should assess whether these measures differ in their associations with cardiovascular outcomes to inform value-based care decisions.
METHODS: We conducted a retrospective cohort study using longitudinal e-prescribing data (2009-2019) with written, fill, and sold dates. Adults (N=7,226) from an academic medical center with ≥12 months follow-up and ≥3 prescriptions within a therapeutic class (ACE-inhibitors, ARBs, calcium channel blockers, diuretics) were included. Adherence was measured using the proportion of days covered (PDC) under two conceptualizations: (1) exposure-based PDC, which treats all gaps as non-exposure, and (2) adherence-based PDC, which leverages personal historical use patterns to dynamically adjust gaps. PDC was calculated using various timestamps, and patients were classified as adherent (PDC≥0.80). We compared adherence estimates and classification stability across approaches.
RESULTS: Patients contributed 230,445 fills with a mean longitudinal follow-up of approximately 3-4 years. Exposure PDC varied across data sources (median 0.747-0.794; p<0.001), with written dates yielding the lowest values and sold dates the highest. Adherence PDC was uniformly high and less source-sensitive (median 0.959-0.977; p<0.001). At the 0.80 threshold, Adherence PDC classification showed >94% concordance across data sources, whereas Exposure PDC was sensitive to source choice; switching from fill to written dates led to substantial reclassification, with >15% of patients changing adherence status.
CONCLUSIONS: Data source and gap handling significantly influence adherence estimates and patient classification. Approaches that incorporate personal past fill history yield more stable classification, which may be preferable for longitudinal outcomes research and quality programs such as Medicare Star Ratings where misclassification can affect plan-level performance and payment. Future studies should assess whether these measures differ in their associations with cardiovascular outcomes to inform value-based care decisions.
Conference/Value in Health Info
2026-05, ISPOR 2026, Philadelphia, PA, USA
Value in Health, Volume 29, Issue S6
Code
PT26
Topic
Real World Data & Information Systems
Topic Subcategory
Data Protection, Integrity, & Quality Assurance, Health & Insurance Records Systems
Disease
SDC: Cardiovascular Disorders (including MI, Stroke, Circulatory)