Validating a Medication Adherence Index in Large Urban Population
Author(s)
Jasjot Saund, MBChB, MSc, iBSc, Isobel Weinberg, MD, PhD, Dan Stein, MD, Lawrence Adams, MD, Joe Zhang, MD.
Artificial Intelligence Centre for Value-Based Healthcare, London, United Kingdom.
Artificial Intelligence Centre for Value-Based Healthcare, London, United Kingdom.
OBJECTIVES: Understanding medication adherence has the potential to inform disease optimisation and identify patients requiring targeted support. Existing adherence measures vary in accuracy and practicality with limited validation. This study aims to validate Proportion of Days Covered (PDC), derived from pharmacy refill data, as an informative metric.
METHODS: Data was extracted from primary care electronic health records from a 2.2 million population in North-East London. PDC was defined as the proportion of prescribed medication days covered by actual supply dispensed to the patient. First, PDC was evaluated in patients with a clinician-documented adherence code (“good” or “poor”), for 7 classes of long-term medications. PDC was calculated over multiple time windows prior to code recording. Logistic regression was used to assess association with adherence code presence.
Second, we evaluated clinical validity by examining three disease-specific cohorts: patients on lipid-lowering drugs, antihypertensives, and antidiabetics. For each cohort, PDC was calculated per patient-drug, and corresponding biomarkers of disease control (LDL-C, systolic blood pressure, HbA1c) were compared pre- and post-treatment. Linear regression was used to test association of PDC with biomarker trajectory.
RESULTS: All PDC formulations showed significant associations with clinician-recorded adherence codes, with the largest effect size belonging to PDC scores in the year prior to a coded status (n=300,843; OR: 9.31, CI:8.61 to 10.07). In the lipid-lowering cohort, higher PDC was associated with greater LDL-C reduction (mmol/L) (n= 86,051, β: -.21, CI: -.24 to -.18). Similarly, in the antihypertensive and antidiabetic cohorts, higher PDC corresponded to improved blood pressure (mmHg) (n=424,289, β: -2.39, CI: -2.56 to -2.21) and HbA1c (mmol/mol) (n=149,984, β=-3.38, CI: -3.70 to -3.06).
CONCLUSIONS: PDC was significantly associated with clinician-recorded adherence codes, and strongly associated with improvements in relevant biomarkers. This supports its potential use as a pragmatic tool for assessing long-term medication adherence in routine clinical practice, post-market surveillance, and research.
METHODS: Data was extracted from primary care electronic health records from a 2.2 million population in North-East London. PDC was defined as the proportion of prescribed medication days covered by actual supply dispensed to the patient. First, PDC was evaluated in patients with a clinician-documented adherence code (“good” or “poor”), for 7 classes of long-term medications. PDC was calculated over multiple time windows prior to code recording. Logistic regression was used to assess association with adherence code presence.
Second, we evaluated clinical validity by examining three disease-specific cohorts: patients on lipid-lowering drugs, antihypertensives, and antidiabetics. For each cohort, PDC was calculated per patient-drug, and corresponding biomarkers of disease control (LDL-C, systolic blood pressure, HbA1c) were compared pre- and post-treatment. Linear regression was used to test association of PDC with biomarker trajectory.
RESULTS: All PDC formulations showed significant associations with clinician-recorded adherence codes, with the largest effect size belonging to PDC scores in the year prior to a coded status (n=300,843; OR: 9.31, CI:8.61 to 10.07). In the lipid-lowering cohort, higher PDC was associated with greater LDL-C reduction (mmol/L) (n= 86,051, β: -.21, CI: -.24 to -.18). Similarly, in the antihypertensive and antidiabetic cohorts, higher PDC corresponded to improved blood pressure (mmHg) (n=424,289, β: -2.39, CI: -2.56 to -2.21) and HbA1c (mmol/mol) (n=149,984, β=-3.38, CI: -3.70 to -3.06).
CONCLUSIONS: PDC was significantly associated with clinician-recorded adherence codes, and strongly associated with improvements in relevant biomarkers. This supports its potential use as a pragmatic tool for assessing long-term medication adherence in routine clinical practice, post-market surveillance, and research.
Conference/Value in Health Info
2025-11, ISPOR Europe 2025, Glasgow, Scotland
Value in Health, Volume 28, Issue S2
Code
RWD191
Topic
Clinical Outcomes, Health Service Delivery & Process of Care, Real World Data & Information Systems
Disease
Cardiovascular Disorders (including MI, Stroke, Circulatory), Diabetes/Endocrine/Metabolic Disorders (including obesity)