Primary Submission Category: Causal Discovery
Understanding the Comorbidity of Epilepsy and Depression on Sleep Disorder by Analyzing National Health and Nutrition Examination Survey (NHANES) data
Authors: Vasundhara Acharya, Bulent Yener, Madeline C Fields, Lara Marcuse,
Presenting Author: Vasundhara Acharya*
Epilepsy, sleep disorders (SD), and major depressive disorder (MDD) often co-occur, complicating care. MDD, common in epilepsy, may mediate its impact on sleep, while some antiseizure medications (ASMs) can worsen mood or sleep disturbances. Age, race, and Body Mass Index (BMI) further modulate these effects. Traditional logistic regression and pairwise analyses cannot capture such complexities. We propose a Structural Causal Model (SCM) learned from NHANES 2015–2020 data. Categorical variables are encoded using a Variational Autoencoder to improve on one-hot encoding. Bootstrapping is applied to each causal structure learning algorithm to compute edge confidence (EC) scores. Flipped-edge conflicts are resolved via EC comparison and significance tests (z-test). Aggregated edges and scores form an ensemble graph representing part of SCM, benchmarked against domain knowledge. Our causal estimations assume causal sufficiency (no unobserved confounders). Our initial analysis shows that MDD is a major driver of SD (EC=0.52), and epilepsy also contributes (EC=0.34). Age increases the likelihood of ASM use, with an overall average treatment effect (ATE) of 0.0217 (p<0.001). Conditional estimates reveal that in the highest BMI category (≥25), the ATE of age on medication intake is 0.02135 for Black patients versus 0.02942 for White patients, warranting longitudinal validation. MDD’s ATE of 0.205 on SD (p<0.001) underscores the need to address depression to improve sleep outcomes.