263 Clustering in Central Disorders of Hypersomnolence by resampling with random subsets of the entire database (Appendix A) showed that the EU-NN database is adequately sized with solid cluster (and biomarker) reproducibility for people without cataplexy. Separate post hoc analyses were performed to test whether our methodologic choices could have influenced the results. Clustering was repeated three times by excluding those with cataplexy and/or hypocretin deficiency (<110 pg/mL), with uniform weightings for all variables, and by recoding the polysomnography REM latency variable to polysomnography SOREMP presence. All three analyses demonstrated grouping and differentiating variables for subgrouping of those without cataplexy similar to those in the full dataset (sleep drunkenness, subjective difficulty in awakening, and weekend-week sleep length difference). These robustness checks further validate that the differentiating variables for those without cataplexy are reproducible. Within the EU-NN database, no data entries are available for people with other conditions that could lead to daytime sleepiness complaints such as chronic sleep deprivation, myalgic encephalomyelitis/chronic fatigue syndrome, and circadian rhythm disorders. These disorders could function as control groups in future clustering analyses on central hypersomnolence disorders to test the specificity of our results. Narcolepsy type 1 could also be considered a control group with a distinct cataplectic phenotype. The fact that the clustering algorithm recognized narcolepsy type 1 as separate clusters while we were blinded for the current diagnosis provides an important argument that the algorithm is able to identify distinct, clinically relevant subgroups. Future studies should focus on external validation of our clustering results in a substantially sized independent dataset and prove internal validation by standardized follow-up data. Both approaches will validate the clinical impact of our clustering results by assessing how cluster assignments relate to clinical decisions such as treatment planning, prognosis, and mechanisms of disease. We report an exceptionally sized quantitative subgroup assessment in people with central hypersomnolence disorders using the full range of clinical and diagnostic variables. Our study further illustrates the urgent need for new biomarkers in central hypersomnolence disorders that allow robust subclassification and improve our understanding of disease aetiology. The main finding is not the number of clusters but the fact we found subgrouping consistent with current diagnosis of narcolepsy type 1, not type 2 or idiopathic hypersomnia. Instead, people with narcolepsy type 2 and idiopathic hypersomnia were divided over 2 distinctly separated clusters, differing mainly on clinical variables related to quality of awakening, including presence of sleep drunkenness and feeling refreshed after daytime sleep, weekend-week 9
RkJQdWJsaXNoZXIy MjY0ODMw