Excessive daytime sleepiness prevention using causality network driven by score-based Bayesian network structure learning algorithms

Sadeghi, Reza

doi:10.1016/j.smhl.2026.100655

Excessive daytime sleepiness prevention using causality network driven by score-based Bayesian network structure learning algorithms

2026-03-01

0

PRE

AI

DOI：10.1016/j.smhl.2026.100655

Original

Original request for help

Save

Abstract

En 中文

Objectives: Excessive Daytime Sleepiness (EDS) is a primary symptom of several sleep disorders, many of which exacerbate cardiovascular complications. While knowledge of factors associated with EDS has expanded due to extensive research in recent decades, its causal chain remains largely unknown. By establishing a clear cause-and-effect relationship, medical professionals and public health officials can intervene effectively to prevent or mitigate future harm caused by EDS. Methods: This study uncovers etiologic factors leading to EDS by score-based Bayesian network structure learning algorithms. The likelihood of various contributing factors to EDS is examined by the causal chains produced by network models. The causality networks and conditional probabilities are constructed based on clinical and biomedical data of 1881 participants in the Stanford Technology Analytics and Genomics in Sleep study. It should be noted that this dataset contains uneven demographic distributions, particularly regarding ethnic distribution, that may have an effect on the generalizability of the causal network. Results: The Bayesian Network constructed by Tabu algorithm performed the best with goodness of fit measures, yielding a BIC score of-64844.62 versus scores of-63921.19 and-64526.64 for Hill-Climb and Max-Min Hill-Climb, respectively. Causalities of EDS produced by the graphical model include age, smoking, hypercholesterolemia, sex, psychiatric/mental health problems, depression, fatigue, and alcohol consumption, in addition to/in tandem with known causalities such as number of people in household, street/recreational drug use, insomnia, sleep duration, narcolepsy, exercise, body mass index, and apnea. Conclusion: The Causality Bayesian network we developed offers a tool for examining EDS risk factors in contexts ranging from public health and policy to diagnosis, prevention, and research. The results from our model demonstrate the capability of learning Bayesian network structure

Keywords:

Preventive healthcare

Excessive daytime sleepiness

Causal discovery

Score-based Bayesian network structure learning algorithms

Excessive daytime sleepiness prevention using causality network driven by score-based Bayesian network structure learning algorithms

Abstract

Journal

Organization