262 Chapter 9 always be people with central hypersomnolence who are difficult to categorize, especially when strict cut-offs are used (e.g., during the MSLT). This impression matches our clinical experience, and our results suggest that the introduction of diagnostic certainties in new diagnostic criteria could better depict confidence levels in diagnosing hypersomnolence subtypes. This idea has recently been proposed by different European experts [28] and should be further investigated in future studies. In addition to direct implications on current classification, the organization of the hypersomnolence disorders in multiple clusters offers opportunities for new hypothesis testing on disease aetiology, disease progression, and treatment effects. Longitudinal studies will provide the opportunity to see whether some individuals are prone to change from one cluster to another due to either the natural course of the disease or the effects of medication. This could provide important insights into pre-stages of narcolepsy and data-driven treatment regimens for newly diagnosed individuals. More frequent HLA-DQB1*0602 positivity and lower hypocretin-1 levels were, for example, seen in cluster 6 than in cluster 5, suggesting a pathophysiologic nature of hypersomnolence complaints closer to narcolepsy type 1. Individuals within cluster 6 may be more likely to show disease progression and should therefore be closely monitored for development of cataplexy. A major strength of the EU-NN database is the harmonized prospective data acquisition protocol, resulting in a true Pan-European collaboration with minimal inclusion site–specific biases. This is supported by the uniform distribution of inclusion sites over the clusters (Figure 3B). We included both adolescents and adults to best incorporate different stages of disease in our clustering analyses. Post hoc testing indicated a similar distribution of age at evaluation over different clusters. Children <13 years of age were, however, not included and should be studied in light of our proposed clusters in future studies. Even though the EU-NN database covers most important hypersomnolence-related aspects, incomplete availability of nonmandatory variables related to vigilance, cognitive functioning, and mood has hindered their full integration into our clustering analyses. Future studies should focus on these variables in relation to our proposed subgroups. Our analyses cannot be considered fully unbiased because agglomerative hierarchical clustering algorithms require manual input of variable weightings. We tried to overcome this issue by carefully designing the analysis strategy with predetermined weighting and grouping of variables, potentially to give every asset of central hypersomnolence disorders a fair chance of influencing the clustering. Post hoc testing to determine the influence of clustering settings
RkJQdWJsaXNoZXIy MjY0ODMw