Fig. 1

ROC classification and misclassification. a Every patient’s weight in week 20, expressed as a proportion of EBW (x-axis), against the cumulative fraction of patients (y-axis). The horizontal dashed lines at 0.25, 0.50, and 0.75 give the lower quartile, median, and upper quartile of the distribution. The average and median %EBW by 20 weeks is roughly 95%, with about half the patients being higher, and half lower. b The ROC curves using weight change in the first 8 weeks of treatment from lighter (week 1) to darker (week 8) shades of blue. The curve with the highest AUC is shown in orange, and corresponds to week 5 of treatment. c The AUC for the ROC models using weight change in weeks 1 through 19 of treatment. Aside from the initial increase in predictive performance, the ROC model’s AUC hovers around 0.70 (dark blue line) with wide confidence intervals (shaded blue ribbons). d The empirical cumulative distribution of weight change in week 5, grouped by patients who reached 95% of their target (blue) and those who did not (orange). The vertical dashed line represents the weight gain cutpoint determined by the ROC analysis for week 5, such that anyone over 6.2 was predicted to reach 95% EBW, and anyone under was predicted not to. e The average (+−se) weight change from admission in each week for patients grouped by whether the ROC analysis correctly or incorrectly classified them using the cutpoint in week 5. The dark blue line are patients who reached the target, but were predicted not to, and the dark orange line are patients who did not reach the target, but were predicted to. The lighter lines are cases where the ROC algorithm correctly predicted the outcome. The number of patients in each category is printed. (f) plots the average (+−) weight in each treatment week as a proportion of EBW. The horizontal dashed line gives 95% EBW