267 Clustering in Central Disorders of Hypersomnolence Cluster distinctness The almost unanimously positive silhouette coefficients of clusters 5 and 6 indicate cluster distinctness compared to the other clusters and to each other (Supplementary Figure 1B). The lower silhouette coefficients in other clusters indicate that these clusters are more similar to each other. Clustering reproducibility Visual inspection of the resampling repetitions showed similar distinguishing factors for individuals without cataplexy as the original clustering (i.e., sleep drunkenness and weekend-week sleep length difference) or merger into one larger cluster in 77% of repetitions. In Supplementary Figure 1C, it can be seen that there is relatively frequent mixing among individuals in clusters 1-4 and 7. This means that different subdivisions or merging of clusters of people with cataplexy was regularly seen. Clusters 5 and 6 were relatively robustly subdivided from other clusters, further highlighting the ability of the clustering algorithm to distinguish people with cataplexy from those without. Visual inspection of the resampling iterations showed that roughly half of the 42% mixing between clusters 5 and 6 was because these two clusters were already merged together (similar to the six clusters result when clustering the entire database). The remaining mixing between people in clusters 5 and 6 was caused by subdivision based on other differentiating variables, mainly the presence of hypnagogic hallucinations and sleep paralysis (16%). 9
RkJQdWJsaXNoZXIy MjY0ODMw