Real-World Data Study for a
Medication Adherence Program

RESEARCH QUESTION

What the Study Set Out to Ask

This real-world data study example follows a community pharmacy research network that has dispensing records from several pharmacies. The pharmacies offer a refill-synchronization program, and the network wants to know whether enrollment is associated with better adherence to statin therapy among patients starting treatment.

Patients choose whether to enroll, and those who do are probably more engaged with their care. The question is therefore framed as an association in routinely collected data, to be estimated with its uncertainty, not as proof that the program causes better adherence. The analysis plan fixes the population, the time at which follow-up starts and the outcome definitions before any adherence values are calculated.

STUDY DESIGN AND DATA

Dispensing Records for About 12,000 Patients

A retrospective cohort built from pharmacy dispensing records for about 12,000 patients who filled a first statin prescription after a year with no statin fills, about one third of them later enrolling in the program. Each record holds fill dates, days supplied, hospital stays where available, demographics and the pharmacy. Patients enroll some months after their first fill, so follow-up starts at a landmark six months after it, and enrollment status is defined at that point.

METHOD AND ASSUMPTIONS

Why Matching, and What Must Hold

Enrolled and non-enrolled patients differ at baseline, so a raw comparison mixes the program with who joins it. Matching on a propensity score built from covariates measured before the first statin fill, and so before anyone could enroll, makes the groups comparable on those covariates, using a caliper and balance judged by standardized differences, not by significance tests. The new-user design and the landmark give both groups the same time zero, in the spirit of a target trial emulation, so the program is not credited with time before patients enrolled.

Patients with few comparable non-enrollees go unmatched, and the plan documents who they are and how the estimate narrows to the matched population. Days in hospital, when medication comes from the ward rather than the pharmacy, are removed from the observation period used for proportion of days covered, and the grace period between fills is varied in a sensitivity analysis. Variance estimation accounts for matching and for clustering by pharmacy. An E-value shows how strong an unmeasured confounder, such as motivation, would need to be to explain the association, and the text says plainly that matching cannot adjust for what the records do not hold.

OUTPUT AND INTERPRETATION

How the Results Are Read

The output has a flow diagram from all patients to the matched cohort, a balance table and plot before and after matching, an overlap figure, and the adjusted difference in proportion of days covered with its 95% confidence interval. The proportion at or above 0.80 is reported beside it, since a threshold outcome and a continuous one can tell different stories.

Sensitivity analyses cover the grace period, a different landmark time, censoring at enrollment for comparison patients who join the program after the landmark, and a stricter set of covariates. The interpretation is worded as an association between enrollment and adherence in these pharmacies, with its precision, and it states the population to which the estimate applies. Whether the program should be extended is a decision for the network and the pharmacies, informed by the estimate and its limits.

REPORTING

What the Researcher Received

A written report structured to RECORD-PE, with statistical wording for the methods, matched to the output, for the team to check and incorporate. The research question, the clinical interpretation and the conclusions remain with the research team.