Descriptive analyses of the integrity of a US Medicaid claims database.

PURPOSE: To examine the integrity of six Medicaid databases for use in pharmacoepidemiology research.
METHODS: We performed descriptive analyses to examine four categories of potential data errors: incomplete claims for certain time periods; absence of an accurate indicator of inpatient hospitalizations; missing hospitalizations for those aged 65 years and over; and diagnostic codes in demographic groups in which those conditions should be rare.
RESULTS: Prescription claims appeared to be missing intermittently in some states. No valid marker of inpatient hospitalizations could be found for three of six states. Hospitalizations appeared to be missing to varying degrees for those aged 65 years and over. Gross errors in diagnostic codes and demographic data did not appear to be widespread.
CONCLUSIONS: Whenever possible, investigators using administrative data should perform macro-level descriptive analyses on the parent data set. In particular, researchers should examine the number of medical and pharmacy claims over time, looking for gaps. Validity of markers of hospitalization should be assessed. The accuracy of diagnosis and demographic data should be examined. Such a descriptive macro-level approach should be used to supplement, and perhaps precede validation of study outcomes using clinical records.