Assessing Diagnostic Accuracy of Quantitative Data in Biomedical Studies Using Descriptive Statistics and Standardized Mean Difference

Abstract

ROC analysis is the most used method for analyzing the diagnostic accuracy of quantitative data in biomedical research. ROC analysis generates a curve describing the frequencies of true positive and false positive results for different degrees of the analyzed variable. However, in many publications devoted to the application of quantitative diagnostic methods, this analysis is not carried out: researchers report only analysis of statistical significance for the groups difference. In meta-analyses, the estimated parameter is the effect size expressed through standardized mean difference. The article describes the approach, which allows performing ROC analysis using cumulative normal distribution functions for studied and controlling groups. The proposed approach can be used to evaluate the diagnostic accuracy of quantitative variables on the base of one of the sets of descriptive statistics (mean and standard deviation, or median and quartiles) or the value of standardized mean difference. Examples of application of the proposed approach on model data, on data from literature sources, as well as on the authors' own observations are given as an example of assessment of diagnostic accuracy of quantitative variables analyzed in the microcirculation studies in various diseases. The results presented in the article can be used by medical and biological specialists to assess the diagnostic accuracy of various quantitative variables without access to primary data.