Impact of prescriber nonresponse on patient representativeness

In pharmacoepidemiology studies where patients are selected by prescribers, there is concern that the patients of responding prescribers are not necessarily an unbiased sample of all patients. However, this usually cannot be explored. In the CADEUS study, patients and prescribers were independently contacted so that data are available for patients irrespective of whether their prescriber responded or not. Our objective was to compare the characteristics of patients whose prescriber did or did not respond.

The CADEUS study included patients treated with COX-2 inhibitors (celecoxib, rofecoxib) or traditional NSAIDs from September 2003 to August 2004. Redeemed prescriptions were randomly sampled on a monthly basis within the database of the French national healthcare insurance system for salaried persons during 1 year. Patients and prescribers were questioned independently. Data from patients and from the database were used to compare patients whose prescriber responded and those whose prescriber did not.

Of 45,217 patients, 26,618 had prescriber data. Patients whose prescriber responded were similar to patients whose prescriber did not respond for the main study outcomes: age (56.8 +/- 16.3 years vs. 56.1 +/- 16.3 years), sex (66.0% female vs. 64.8%), cardiovascular disease history (52.2% vs. 52.0%), gastrointestinal disease history (39.5% vs. 39.4%), concomitant prescription of gastroprotective agents (22.4% vs. 23.7%), and NSAID indication, prescription type, use, and duration.

We found no evidence for a difference between patients whose prescriber responded and patients whose prescriber did not participate in the study.