Improving diagnostic accuracy: simple statistical nomograms to interpret medical literature.

Abstract

Authors of medical diagnostic literature frequently report sensitivity and specificity as measures of the quality of an evaluative study. However, these representations are easily misinterpreted by clinicians to be indicative of the prospective value of a test as predictive of the presence (positive predictive value, PPV) or absence of disease (negative predictive value, NPV). Although these phenomena are related, the mathematical expression and, therefore, the conclusions are more complex. Using algebraic methods, we derived simplified formulas to determine PPV, NPV, and accuracy (A). These general terms were solved by constraining individual variables, resulting in the development of curves that may be used routinely to analyze medical diagnostic literature. Equations for PPV, NPV, and A were generated by using sensitivity, specificity, and incidence/prevalence as the dependent variables. These equations have been employed to generate representative graphs of PPV, NPV, and A and to clarify trends in these features with respect to commonly reported data. These simplified equations allow clinicians to determine the utility of diagnostic studies in prospect, despite having only sensitivity, specificity, and incidence or prevalence of disease.