—Hochman and Lumb have raised an important point concerning the stability of likelihood ratios across various surgical procedures. Likelihood ratios, like the related properties of sensitivity and specificity, are usually assumed to be stable across populations with different pretest probabilities. The common example of variations in pretest probabilities results from variations in the prevalence of disease. However, for the clinical problem of assessing operative risk for cardiac patients, it is reasonable to think of the variations in pretest probabilities as variations in cardiac complication rates across categories of surgery.
In performing the hypothesis-testing study of our modified index, which is reported elsewhere,1 we were also concerned about the stability of likelihood ratios across the various surgical categories. From the outset, we recognized that we would not have a large enough sample size in each of the surgical categories to estimate likelihood ratios separately for each procedure. We did, however,