248 Chapter 9 were displayed. The significances barcodes better represent how substantially different the means barcodes are from the entire EU-NN database. We post hoc statistically compared the resulting clusters with a substantial proportion of individuals without cataplexy on all included variables using Mann-Whitney tests. If the variables contained count data, we used χ2 tests instead. Corrected values of p < 0.05 were reported after multiple-comparisons correction with the Benjamini-Hochberg procedure to decrease the false discovery rate to 0.05. Current diagnosis and centres of inclusion Researchers were blinded to the centre of inclusion and current diagnosis of the individuals until the hierarchical clustering was completed. After the clustering was finished, pie charts were generated per cluster representing the current diagnosis (with physician’s diagnostic certainty) and centres of inclusion. Contingency table statistics (sensitivity, specificity, positive predictive value, and negative predictive value) were separately computed for clusters dominated by narcolepsy type 1, and narcolepsy type 2 and idiopathic hypersomnia. Centres of inclusion were shown to check whether these could have influenced the clustering. To better understand the general characteristics of the study population, the characteristics of the current ICSD-3 diagnoses are included in Appendix B. Data availability For this study, we used the newly developed clustering package Bowerbird, which integrates widely used agglomerative hierarchical clustering algorithms with clustering validation methods and intuitive data visualization options. This flexible, open-source clustering package is Python based and available online [326]. The data that support the findings of this study are available from the authors on reasonable request.
RkJQdWJsaXNoZXIy MjY0ODMw