Methods for constructing multivariate predictive models for diagnosing diseases for which test methods are individually inadequate, including: (a)
performing laboratory tests on a statistically significant test
population of individuals; (b) generating a
score function from a linear combination of
test panel results; (c) performing a
receiver operating characteristic (ROC) regression or alternative regression technique of the
score function using those tests and β coefficients calculated simultaneously to maximize the
area under the curve (AUC) of the function and chosen simultaneously to generate the largest area below that portion of the ROC curve for the (1−t0) quantile of individuals without
disease, where t0 represents the maximum acceptable false-positive rate; (d) calculating individual pre-test
disease odds; generating a diagnostic likelihood ratio of
disease by determining the frequency of each individual's test
score in the diseased
population relative to the control
population; and multiplying pre-test
odds by the likelihood ratio to determine individual post-test disease
odds; (e) converting a set of posttest odds into posttest probabilities for each potential multivariate methodology and creating an ROC curve for each methodology by altering posttest probability cutoff values; (f) comparing partial ROC areas generated by one or more regression techniques to determine the optimal methodology; and (g) dichotomizing the optimal methodology by finding that point on the ROC curve tangent to a line with slope (1−p) C / p·B, where p is population
prevalence of disease, B is regret associated with failing to treat patients with disease and C is regret associated with treating a patient without disease.