Validity of drug and diagnosis data in pharmacoepidemiology

Accurate pharmacoepidemiologic study conclusions require assessment of valid data, regardless of whether the data originate from questionnaires, administrative claims, or electronic health records. This chapter provides an overview of current methodologic concerns in validation studies, including types and detection of measurement error, effects of error on the point estimate of association, and correction options. Validity issues for diagnosis and medication data, and how the accuracy of these data varies by patient and other characteristics, are also discussed. Two examples are provided to illustrate validation efforts for questionnaires and electronic data: nonsteroidal antiinflammatory drugs (NSAIDs) and myocardial infarction and NSAIDs and gastrointestinal bleeding. Current solutions include steps to consider when designing and analyzing validation studies using either questionnaire or electronic data. The chapter concludes with a discussion of the continued need for validation studies, especially as research moves toward incorporation of new data sources, increase in data linkage, and use of unstructured data for pharmacoepidemiology research.