FIGURE 4.

Performance of SuStaIn for subtyping individuals. (A) Accuracy of Ordinal SuStaIn for recovering the ground truth subtypes of individuals, (B) the confidence SuStaIn assigned to the estimated subtypes, and (C) the accuracy of the confidence intervals SuStaIn assigned to the estimated subtypes. As in Figure 2, the x-axis shows the experiments in which we varied the simulated number of biomarkers I (orange), number of subjects J (red), number of subtypes C (purple), proportion p with an accurate score (i.e., the categorical probability each test score is accurate; blue), the proportion f of misdiagnosis (i.e., the proportion of individuals that follow randomly chosen alternative progression patterns; green), and the choice of algorithm (either the proposed Ordinal SuStaIn algorithm or the existing Z-score SuStaIn algorithm). The default value for each simulation setting is indicated with an asterisk on the x-axis.