Fig. 3: Variable importance of clinical and psychosocial features in association with risk decision making for investigational drugs.

In (a) the x-axis displays the conditional variable importance obtained from random forest regression analyses conducted separately for MSA (red) and PD (blue) patients. Shown are point estimates and 95% confidence intervals obtained from 200 runs. The variable with the highest importance is assigned a value of 100%, and all other variables are expressed as a percentage relative to that value. Any variables with confidence intervals that include zero or negative values are considered to have no predictive power in our model and are assigned a value of zero. In (b–i) bivariate associations between clinical, psychosocial features, and the median accepted risk of drug side effects are presented. MSA and PD patients are represented as red triangles and blue dots, respectively. Pearson correlation coefficients (“r”) were used to analyze continuous variables such as satisfaction with life situation (b), RPS (c), quality of life: nonmotor subscore (d), quality of life: emotional subscore (e), age (f), required social support (g), disease duration (h) and sex (i). Point-biserial correlation (“r”) was used for categorical variables, and Spearman correlation coefficients (“ρ”) were employed for ordinal scaled variables like the degree of required social support. Regression curves and 95% confidence intervals are provided for continuous variables. P values (“p”), both raw and adjusted following the Benjamini-Hochberg procedure, were computed (raw p values are shown in brackets).