iSPHYNCS: Unsupervised clustering in questionnaires and metadata reveals distinct subtypes in the narcolepsy borderland

ObjectiveThe international Swiss Primary Hypersomnolence and Narcolepsy Cohort Study (iSPHYNCS) is a multicenter study aimed at identifying novel biomarkers for central disorders of hypersomnolence (CDH). We analyzed questionnaires and metadata to uncover distinct clusters of participants and explore phenotypic variability within CDH. MethodsData were collected from 227 patients with CDH and 33 healthy controls. Participants completed validated clinical questionnaires and study-specific questions addressing CDH-related symptoms such as excessive daytime sleepiness, fatigue, cataplexy, disrupted sleep, and sleep paralysis. Demographic metadata (age, gender, BMI) were included. After excluding participants with missing over 30% of data (n = 40), missing values were imputed using a multiple random forest algorithm. A robust clustering pipeline was employed: (1) random sampling of 60% of the dataset, (2) dimensionality reduction via UMAP, (3) K-means clustering, and (4) consensus clustering across 500 iterations. Post hoc analysis was performed to identify biomarkers in data not used for clustering. ResultsWe identified four distinct clusters. One predominantly comprised healthy controls, while another primarily contained individuals with narcolepsy type 1 (NT1). Two clusters represented predominantly the narcolepsy borderland group (NBL), with one distinctly characterized by higher symptom severity and psychiatric comorbidities. ConclusionsThe clustering pipeline produced reproducible results, with the NT1 and healthy control clusters serving as internal validation. The differentiation between the two NBL clusters aligns with prior studies, suggesting a possible NBL subtype marked by increased fatigue and psychiatric comorbidities. These findings emphasize the phenotypic heterogeneity of CDH and the potential for cluster-based approaches in management. Statement of SignificanceCentral disorders of hypersomnolence (CDH) are complex and heterogeneous, challenging conventional diagnostic frameworks. This study employed an innovative clustering pipeline to identify four distinct phenotypic clusters within a large, multicenter cohort. The identification of a subtype in the narcolepsy borderland group with a high psychiatric burden adds critical insights into the variability of CDH and underscores the need for personalized diagnostic and therapeutic strategies. The internal validation of clusters through healthy controls and NT1 phenotypes enhances confidence in the pipelines robustness. Furthermore, the study highlights the potential for integrating clinical data to define disease severity and developing therapeutic strategies. These findings pave the way for novel approaches to understanding and managing CDH.