Figure 3
The association between PFS and the predicted level of sensitivity for the combined REGS for CHO in the merged IDRC and LLMPP cohort. In panel A the probability of being sensitive (one minus the probability of being resistant) according to the REGS classifier is plotted for each patient. Based on the probabilities the patients are categorised into tertiles with those deemed sensitive, intermediate, and resistant indicated by green, blue, and red. Kaplan-Meier curves for PFS are shown in panel D. Panel B shows estimated log HR versus predicted resistance index modelled by an RCS-model with four knots for the REGS predictor for CHO adjusted for IPI. Panel E shows the corresponding survival curves generated by the fitted Cox proportional hazards regression. The survival curves are generated for the values marked by arrows in Panel B. Panels C and F illustrate an analysis of ROC curves for prediction of the combination therapy CHO where all curves are shown with 95% CI. Panel C shows AUC under the ROC curves plotted against time for the CHO REGS classifier (green) and predictor (blue). Panel (F) shows the difference in AUC plotted against time.