Promise and Peril of Clinical Decision Support

> “Prediction is very hard, especially about the future.”> > – Niels Bohr The amount of clinical research available to clinicians today has never been more extensive or complex. By one estimation, almost 2100 scientific publications, 75 clinical trials, and 11 systematic reviews are generated daily.1 This explosion in information greatly exceeds a clinician’s cognitive ability to integrate the full body of literature when considering a specific clinical situation or patient. Accordingly, clinical decision support, defined as a system that integrates patient information with a computerized database of clinical research and guidelines, is one of the more exciting potential applications of electronic health records to clinical medicine.2 However, the potential of clinical decision support is not without peril.Article see p 27 Clinical decision support relies on the fundamental assumption that medical evidence can be precisely translated to the individual patient. This translation is not as simple as it may initially appear. Clinical trial results are reported as the average effect of a therapy among the trial population, but these aggregated results rarely correspond to an individual’s response to the therapy.3 Patient characteristics can significantly exaggerate or attenuate its impact. For example, a therapy that reduces myocardial infarction by 10% would have significantly greater effects in a patient whose baseline risk for myocardial infarction was 20% compared with a patient whose risk was only 1%. In addition, therapy effects often occur in a skewed distribution, where a relatively small number of patients derive a significant benefit, whereas most experience little or no effect. Thus, the mean measure of effect in the population fails to describe a typical patient’s expected response to the therapy. Finally, because most therapies usually entail at …