Background: When calculating medication adherence using healthcare claims data, various decisions about data cleaning and standardizing need to be made. However, information about the effects of various decisions on study result is limited.
Objectives: This study aimed to assess medication adherence by various decisions about analytical approaches and compare each of their sensitivity.
Methods: This is retrospective cohort study using Korean National Health Insurance Service-National Sample Cohort database. Patients aged 20 or over who were diagnosed with hypertension and prescribed antihypertensive drugs were included. The medication adherence calculations used the PDC (Proportion of Days Covered), a method recommended for retrospective observation studies, and the DPPR (Daily Polypharmacy Possession Ratio), a new method of calculation that takes into account multi-drug use. Eight decisions were defined that could be applied in the assessment of medication adherence, the conditions commonly used in each decision were defined as base-case, and the sensitivity of adherence was assessed by changing each setting value singly.
Results: Among the categories of sensitivity analysis, decisions related to “Inclusion/Excluding Criteria” and “Observation Period” revealed to have a significant impact on medication adherence. In particular, the "Handling of the end-of-period" decision in “Observation Period” category showed statistically significant differences in adherence, despite little change in the number of subjects by changes in settings. In other categories of “Data Scrubbing” and “Handling of Sequential Prescription”, there was not much difference in adherence.
Conclusions: Among the methodological decisions used to calculate medication adherence using health insurance claims data in Korea, it appeared that there are decisions that significantly affect adherence and those that do not. Therefore, researchers should carefully define the detailed decisions related to “Inclusion/Exclusion Criteria” and “Observation Period” in evaluating adherence, and the sensitivity analysis for the variables in decisions can reduce uncertainty and increase robustness of the results. In particular, it appeared that the decision of “handling of the end-of-period” should defined more carefully. This study is meaningful in that it reported transparent results by various decisions in the evaluation of medication adherence, and may be used as an objective evidence for researchers to consider in determining the detailed method for evaluating the medication adherence.