Scientific Publications by FDA Staff

Pharmacoepidemiol Drug Saf 1994 Mar-Apr;3(2):63-71

The impact of clustering in physician and pharmacy services on the accuracy of pharmaceutical exposure risk estimates

Burkhart GA, Hammerstrom T

Record-linkage databases have been used to evaluate the safety of approved drugs. Outcome and exposure data collected in these databases are contingent upon temporal contacts between patients and their physicians and pharmacists. We examined the theoretical effect of temporal clustering in physician and pharmacy services on pharmaceutical risk estimates by generating computer modeled datasets of the dates of provider services. Incidence density ratios were estimated for each dataset by comparing the incidence of the hypothetical outcome during exposure to that during no exposure. The presence of clustering in provider services, the risk associated with exposure, the length of the risk period, the number of refills and the underlying physician visit rates were varied by dataset. Biased estimates of the incidence density ratio were obtained in datasets with clustering. The magnitude of bias decreased with increasing length of the risk period, increasing number of refills and increasing physician visit rate. A sample of an HMO record-linkage database showed evidence of clustering in physician visits at the patient level when the transaction dates for all physician visits were tested for clustering. Researchers using record-linkage databases or data derived from patient interactions with providers to study the acute risk of pharmaceutical exposures, should consider the possibility of biased risk estimates for outcomes that cluster with other outcomes and exposure.