Impact of data feedback on physician practice patterns

Data were collected from computerized clinical and financial information systems at LDS Hospital in Salt Lake City, Utah. The data were statistically analyzed to display interphysician differences for a variety of practice parameters. Additional analyses indicated if individuals' variations from the group mean were greater than could be expected due to chance alone. The data were fed back to the physicians and a before-after analysis was carried out to determine if the data feedback had any impact on the physicians' subsequent practice patterns. Two medical procedures were the subject of study: well-baby delivery and coronary artery bypass graft (CABG) surgery. Four parameters were studied in each domain. Minimal to moderate changes in physician practice were noted following the data feedback in obstetrics. The group rate for use of antibiotics in vaginal delivery decreased from 8.21% to 6.77% (p = 0.298). Two physicians decreased their antibiotic use dramatically; as a result there was decreased variability among the practitioners following the feedback. The group elective induction rate decreased from 9.86% to 6.62% (p = 0.028) although the reasons for the change are unclear. Laceration rates did not change, and sample sizes in use of antibiotics in cesarean section were too small to be meaningful. The changes in CABG surgery were more dramatic. Surgical time dropped from 215 to 205 minutes (p = 0.045), length of stay dropped from 9.35 to 9.01 days (p = 0.3339), and postoperative length of stay dropped from 7.57 to 7.14 days (p = 0.1413). Despite drops in surgical time and length of stay, adjusted total charges rose by 3%. On several occasions, individual physicians changed by statistically significant amounts for the various parameters. Several techniques of continuous quality improvement were attempted in this project: separation of random from assignable variation, graphical display of data, user determination of variables to examine, and analysis of the process of care. Although the graphical display of data seemed to work well, analysis of process and user determination of variables to examine were difficult to implement due to limited physician availability and the complexity of medical care. The objective results of the project are encouraging; however the complexity of medical care presents significant limitations to the widespread use of these techniques.