Computer-Assisted Design of Studies Using Routine Clinical Data

To facilitate the analysis of routine, longitudinal, clinical data, we developed a computer program called the RX Study Module. Our prototype uses a small online knowledge base of medicine and biostatistics to help create and execute a detailed statistical study design. The program identifies possible confounding variables, selects methods for controlling them, creates a statistical model, determines patient eligibility criteria, and retrieves data from records. We used the program to examine the hypothesis that daily prednisone administration elevates serum cholesterol. Data from 49 patients with chronic rheumatologic disorders were analyzed from a database of 1787 patients. A regression model was fitted to each patient's record. Changes in cholesterol were significantly correlated (p = 10(-5)) with changes in prednisone after a lag of at least 1 week and after recorded confounders were controlled: delta cholesterol = 18.4 loge(prednisone). Routinely collected patient data may become an important resource for generating and studying new medical hypotheses.