244 Chapter 9 Introduction The classification of central disorders of hypersomnolence has been a topic of debate for decades and has been revised multiple times. This is due mainly to insufficient knowledge about the pathophysiology, reflected in a lack of validated and reliable biomarkers within this group of disorders, apart from narcolepsy with cataplexy. Different opinion articles have recently been published, all stressing the need for revision of the current classification because its application causes problems for physicians and patients when applied in daily practice [27-29, 164, 317, 318]. The current version of the International Classification of Sleep Disorders (ICSD-3) is based largely on a consensus of expert opinion and describes 3 different categories of chronic central disorders of hypersomnolence: narcolepsy type 1 (almost completely overlapping the former category called narcolepsy with cataplexy), narcolepsy type 2 (almost completely overlapping the former category called narcolepsy without cataplexy), and idiopathic hypersomnia. The disorders share the symptom of excessive daytime sleepiness, and in the absence of cataplexy and hypocretin-1 deficiency, the multiple sleep latency testing (MSLT) results and possible increased sleep duration differentiates between them. Only narcolepsy type 1 is a clinically distinct phenotype because of the specific presence of cataplexy and its strong correlation with hypocretin-1 deficiency (<110 pg/mL in the CSF) [8]. Despite these apparently clear and distinct criteria, it often proves difficult to differentiate reliably between narcolepsy type 2 and idiopathic hypersomnia. Recent studies have shown that test-retest reliability of the MSLT is poor in the absence of cataplexy, and diagnostic crossover of up to 53% was seen for narcolepsy type 2 and 75% for idiopathic hypersomnia [34, 35, 319]. Narcolepsy type 2 may also evolve over time in some individuals; for example, individuals in whom daytime sleepiness is the sole initial manifestation may develop cataplexy many years later and thereby transition into narcolepsy type 1 [48, 49]. More reliable biomarkers are needed to better differentiate between individuals with central hypersomnolence disorders, specifically in those without cataplexy. As a data-driven approach, agglomerative hierarchical clustering has previously proved useful in other diseases, objectively identifying subgroups and corresponding divisive variables by grouping people with similar characteristics in clusters [320-324]. In this study, we used an unsupervised machine learning approach, agglomerative hierarchical clustering, to identify clusters of clinically distinct central hypersomnolence disorders. We used the advantageously large-
RkJQdWJsaXNoZXIy MjY0ODMw