We suggest that methodological issues specific to observational database studies may have contributed to the surprisingly large benefits of statin treatment reported by Shalev and colleagues1 when compared with a meta-analysis of secondary prevention trials that showed only a 16% reduction in mortality.2 First, the use of mean proportion of days covered (PDC) over the entire follow-up, as a measure of adherence, may have led to biased results. This measure was computed once for each patient, and the value was applied to all person-moments of follow-up of the patient. This requires the strong assumption of stable prescribing patterns during follow-up, which is unlikely. For example, assuming 30-day prescriptions, patients with a PDC of less than 10% would have needed, by definition, to survive a minimum of 300 days, with no additional prescription than the first. However, patients entering the cohort in the last month of exposure assessment would automatically be classified as having a PDC of 90% or higher. The implications are unclear, particularly since the numbers of deaths and persons are not reported by PDC category. An analysis using a time-dependent cumulative exposure measure that updates the PDC over time would not require an assumption of no changes in prescribing patterns during follow-up.
展开▼
