USE OF COMPUTERIZED STATISTICAL MODELS TO EDUCATE MEDICAL HOUSESTAFF IN ESTIMATING CLINICAL OUTCOMES.
The suggestion that computerized diagnostic and prognostic predictions from statistical models would improve the ability of medical housestaff to estimate outcomes in patients with coronary artery disease (CAD) was experimentally tested. Interns made estimates of the likelihood of significant CAD, three-vessel or left main CAD, and left main CAD. Then, they received immediate feedback from computerized statistical models that had been developed from a database of over 10,000 patients. A control group of interns made similar predictions without exposure to the computerized model estimates. In addition, all interns made predictions on a set of historical case summaries both before and after the rotation. Despite the superior predictive ability of the models, the experimental and control groups were similar in all aspects of their predictive abilities. For the historical case summaries, both groups improved their predictions on the more prevalent outcome of significant disease, but did not improve their estimation of severe disease during their one-month rotation on the cardiology service.
