SLEEP DEPRIVATION AND EPIRETINAL MEMBRANE RISK: Machine Learning Findings from a Nation-wide Survey

Abstract

Purpose: This cross-sectional study explored the association between sleep deprivation and epiretinal membrane (ERM) using machine learning applied to data from the Korean National Health and Nutrition Examination Survey 2017 to 2020. Methods: Data from 2018 to 2020 were used for training and internal validation, and from 2017 for external validation. Participants were divided into ERM and non-ERM groups, and their sociodemographic, lifestyle, and clinical characteristics were assessed. Sleep deprivation was defined as sleeping <6 hours on weekdays. Machine learning-based logistic regression was used to model the association between sleep deprivation and ERM, adjusting for confounders. The consistency of the results and importance of each feature were assessed using subgroup analyses and Shapley additive explanations. Results: Data from 15,240 participants were included, with an ERM prevalence of 9.59%. The final adjusted model achieved an area under the ROC curve of 0.763 (95% CI 0.733-0.792) in external validation. Sleep deprivation was significantly associated with increased ERM risk (adjusted odds ratio [OR], 1.247; 95% CI 1.051-1.481), particularly among non-high-risk alcohol consumers (OR 1.216; 95% CI 1.057-1.399) and individuals with diabetes mellitus (OR 1.259; 95% CI 1.069-1.481). Sleep deprivation was the fourth most influential predictor (5.3%), after age, cataract surgery, and dyslipidemia. Conclusion: Sleep deprivation was significantly associated with a 1.25-fold increase in the prevalence of ERM, especially among non-high-risk alcohol consumers and those with diabetes. Weekday sleep deprivation may be a modifiable risk factor for ERM. Prospective studies are warranted to confirm causality and explore the underlying mechanisms.