245 Clustering in Central Disorders of Hypersomnolence scaled European Narcolepsy Network (EU-NN) database [22, 159, 325]. The EU-NN is an association of 21 leading European sleep centres that launched the first prospective European web-based database for narcolepsy and idiopathic hypersomnia. One main goal of the EU-NN database is to identify new biomarkers specific to central hypersomnolence disorders and to improve definitions and understanding of its subtypes. The comprehensiveness of variables and large number of individuals across different European countries provide the opportunity to implement unsupervised machine learning algorithms in an unprecedented fashion and allows comprehensive datadriven insights into the different phenotypes of central hypersomnolence disorders. We hypothesized that clustering would result in clear separation of individuals with the current diagnosis of narcolepsy type 1 from those without cataplexy, while mixing people currently diagnosed as narcolepsy type 2 and idiopathic hypersomnia over multiple clusters according to differences in sleep duration and presence of sleep drunkenness. Given the known poor test-retest reliability of narcolepsy type 2 and idiopathic hypersomnia diagnoses [34, 35, 319], our main focus was on grouping those without cataplexy. Methods The analysis steps are divided into core and advanced analyses (Table 1). The core analyses are essential for understanding the clinical implications of the clustering results, whereas the advanced analyses in Appendix A validate why we deem the clustering results trustworthy. EU-NN database Records of 1,078 adults and adolescents (≥13 years old) with central hypersomnolence disorders from 21 European sleep centres were included. In line with diagnostic ICSD-3 recommendations, only data of individuals unmedicated at the time of evaluation (including polysomnography and MSLT) were used. In total, 97 variables were input into the hierarchical clustering (Appendix B). Variables were assessed by sleep experts (e.g., symptom presence), objectively assessed (e.g., sleep tests, hypocretin and HLA-DQB1*0602 positivity), or self-reported through questionnaires (Epworth Sleepiness Scale and Fatigue Severity Score). Except for the questionnaire results that were fully patient rated, other subjective variables were entered by the clinician after the clinical interviews. For the database pre-processing steps, we refer to Appendix C. 9
RkJQdWJsaXNoZXIy MjY0ODMw