Figure 4
Summary of all RFC model results depicted as a forest plot. The diamonds correspond to the point estimates (median) of each model: white for out-of-bag classification error; grey for placebo group classification error; black for statin group classification error, whereas the bars represent the estimated 95% confidence interval of the corresponding classification error. The vertical dashed line represents the 50% classification error, which is considered as random. OOB error and estimated 95% confidence interval below 50% indicates that the classifiers perform better than random. The serum lipid and lipoprotein lipids (SL) after the statin intervention performs extremely well as the classifier, whereas the intraprostatic tissue lipidome (IPL) performs barely below the 50% line. Remarkably, in each intraprostatic lipidome model, the statin arm in much better classified in comparison to the placebo arm. Intraprostatic lipidome (IPL) does not discriminate the study arms among men with Gleason 7-and-above, whereas it does separate the study arms among men with Gleason 7-and-below. The intraprostatic lipidome (IPL) cannot make a distinction between the study arms in below-27-days (lo-expose) statin use sub-population, while it does discriminate the study arms among men who used stains more than 27 days (hi-expose).
