Comparison of health insurance claims and patient interviews in assessing drug use: data from the Three-City (3C) Study.

PURPOSE: Precise determination of drug exposure is fundamental in pharmacoepidemiology. Drug exposure is often presumed from health insurance claims but this may not correspond exactly to what subjects actually take. This study was designed to investigate French reimbursement databases in assessing drug use.
METHODS: Between 1999 and 2001, 9294 subjects were included in the Three-City (3C) Study, a French cohort studying the relationship between vascular risk factors and dementia. Of these, 4112 subjects had data available from both clinical interviews and the reimbursement databases of the French national health insurance system. Agreement between drugs reported as used at interview and drugs reimbursed during the previous 30 or 60 days was measured with kappa coefficients. Using calculations of sensitivity (Se), specificity (Sp), positive predictive values (PPVs) and negative predictive values (NPVs), the validity of reimbursement data for the 30 or 60 days preceding the interview was investigated taking drugs reported at interview as the 'gold standard'.
RESULTS: Declared drug use at interview was less well predicted by 30-day than by 60-day reimbursement data. Agreement between reimbursement data and interviews as well as validity of reimbursement data with reference to interviews were substantial for drugs used in cardiovascular diseases, diabetes, rheumatic disorders or neuropsychiatric conditions and were poor for laxatives, vitamins, vasculoprotectives, first and second line analgesics, anti-infective products or dermatologicals.
CONCLUSIONS: Reimbursement data with an appropriate time frame and interviews estimate exposure to chronically used drugs similarly. Self-medication was better described with interviews whereas reimbursement data seem more useful for drugs used topically or intermittently.