Cluster-Based Analysis of Sleep Reveals Metabolic and Lifestyle Determinants Beyond Aggregate Sleep Score

Authors: Madison Gangi , Amina Haq, Anna Delaj, Rania Khan, Sara Saleh, Nema Sayeed, Maxime Argenson

Field of Study: Biological Sciences

Program Affiliation: Metabolic and Exercise Physiology Lab

Faculty Mentors: Daniel Miller, Lina Begdache

Easel: 81

Timeslot: Midday

Abstract: Poor sleep quality is an increasingly recognized modifiable risk factor for chronic diseases, particularly in college-aged populations. The Pittsburgh Sleep Quality Index (PSQI) is widely used to assess self-reported sleep quality; however, it may inadequately measure sleep as a multidimensional biological process. A total of 129 participants aged 18-24 were included in phenotype analyses, with 98 participants used for component-level modeling. Multivariable regression and unsupervised clustering were used to assess associations between seven PSQI components, PSQI total score, and primary predictors. Smoking status was the strongest predictor of sleep duration (p < 0.01), and predicted membership in a short sleep latency phenotype when combined with elevated respiratory quotient and reduced fat oxidation (p < 0.01). Domain-based models yielded larger effect sizes and greater stability compared to PSQI totals. Findings suggest sleep quality is linked to metabolic and lifestyle factors, and cluster-based approaches provide greater biological insight than aggregate sleep scores.