Improving the Diagnostic Utility of Clinical Biomarkers by Discarding Dichotomous Cutoffs

Abstract

The current standard of practice for use of clinical biomarkers is to set a dichotomous cutoff on a continuous variable to distinguish between "positive" or "negative" results. As the majority of infection biomarker tests are fundamentally quantitative, the practice of assigning a cutoff discards a significant amount of information that could contribute to improved diagnostic utility. METHODS:We developed a new method to calculate precise post-test probabilities directly from experimental data and then applied the method to patient data sets. The data of interest included measures of C-reactive protein (CRP), procalcitonin (PCT), and nasopharyngeal lytA gene PCR (lytA) along with the gold standard etiologic criteria for patients. RESULTS:We found that any dichotomous cutoff was essentially arbitrary, as a large range of values serve similarly well as cutoffs. We then computed post-test probabilities for every possible biomarker value. We found that our method provided more accurate post-test probability estimates than any optimal dichotomous cutoff. In addition, we found that our method allowed computing the probability of various categories of outcome variables: for example, the probability of Gram positive or Gram negative, or the probability of a specific bacterium. CONCLUSIONS: Using the described method in clinical practice is a cheap and easy means of calculating exact post-test probabilities for a range of important possible outcomes. This approach avoids dichotomous cutoffs and thereby avoids discarding valuable quantitative information.CLINICAL IMPLICATIONS: Our findings suggest dichotomous cutoffs limit the value of biomarker data, and that data obtained from a number of tests can be better analyzed to guide diagnosis and clinical decision making.