Figure 4.

Logistic regression and ROC analysis of lymphoblast respiration as an ME/CFS biomarker. (a) Box plot showing the distribution of the propensity score from logistic regression of participant type against 5 key respiratory parameters. The resulting regression coefficients are as indicated. The boxes show the median and the 25^th and 75^th percentiles, so that the height of the box is the interquartile range (IQR). The whiskers extend to the most extreme observations (largest and smallest) falling within ± 1.5 × IQR of the box. The ME/CFS and control sample sizes were 49 and 20 individuals, respectively. The Mann–Whitney significance probability tests the hypothesis that the scores in ME/CFS samples are greater than in control samples. Each point represents a single individual. Scores greater than 0.5 lead to classification of a sample as ME/CFS in the “confusion matrix”. Other relevant statistics showing that the logistic regression model provided a good fit to the data were: Hesmer–Lemeshow goodness of fit = 0.14; Pseudo R² = 0.56; χ² p = 6.7 × 10⁻⁴. (b) ROC analysis of the propensity score, plotting sensitivity (proportion of true positives) against specificity (proportion of true negatives) with 95% confidence limits (cyan shading). The fractional area under the curve (AUC) is shown with 95% confidence limits. The “best” threshold for the propensity score (0.59) is shown, together with the specificity (0.70) and sensitivity (0.90) at that threshold (in parentheses).