Figure 5

Estimation of population-wide cumulative probability (bottom) and counting process (top) curves for MS relapse (left) and HF development (right) using the predictions of SAMGEP trained with \(n=100\) labeled patients. “Actual” curves were estimated using all available labels. Only labeled sets were used to generate curve estimates in order to enable unbiased comparison between the SAMGEP-predicted and actual curves. 95% confidence intervals were empirically estimated by bootstrapping with 100 replicates. More extensive results are displayed in Supplementary Fig. S7.