Objective: To assess patient recruitment and quality of data in a randomized database study.
Study design and setting: A randomized database study was conducted in the Integrated Primary Care Information (IPCI) general practice research database. Software was built to allow for automated patient identification and recruitment, and randomization. As an application, we compared gastrointestinal tolerability in persons treated with diclofenac and celecoxib for osteoarthritis. The outcomes were assessed in the IPCI database. To assess accuracy of exposure and outcome, we also collected information by self-administrated patient questionnaires. For all eligible subjects, we assessed the main reason for noninclusion. Physicians were interviewed to evaluate the study and to identify the major obstacles.
Results: Forty-two general practice physicians collaborated with the study and 7,127 potential study subjects were identified. Among these subjects, 170 were eligible for recruitment and 20 (11.8%) were randomized. Of the eligible patients, 96 (56.5%) were not recruited because the physician was too busy or the patient was treated by another healthcare provider and 54 (31.8%) were not recruited because of exclusion criteria.
Conclusion: Concordance between questionnaires and IPCI data and the outcome was good (kappa=0.7; SD=0.14). The physicians reported that recruitment during routine visits was too time-consuming, in particular because of the need for informed consent. Although a randomized database study is feasible, patient recruitment during routine consultations should be avoided.