A technique for analyzing clinical data to provide patient management guidelines

A. Meyers, D. A. Brand, H. G. Dove and T. F. Dolan Jr

This article describes a technique for analyzing clinical data in order to guide patient management decisions. The technique is illustrated by considering a specific decision problem encountered in the management of possible meningitis, namely, whether or not to administer antibiotics while awaiting the results of a CSF bacterial culture. Data from 303 patients with meningitis are analyzed in order to determine which combination of clinical variables best discriminates between bacterial and aseptic cases. From these variables, a probability tree is constructed that indicates the probability of bacterial meningitis, depending on a patient's clinical characteristics. In addition to identifying the most important variables, the analysis reveals that a number of tests performed routinely on patients with meningitis are of questionable diagnostic value.