Oxidative Balance Score and Menopausal Status: Insights from Epidemiological Analysis and Machine Learning Models

Yanjun Zhou^1*Chunlin Dong¹Ding Ma²Jinjin Yu¹Ke Gu¹Yaying Lin¹Jing Song¹Yuan Wang¹

• ¹Affiliated Hospital of Jiangnan University, Wuxi, China
• ²Key Laboratory of the Ministry of Education, Cancer Biology Research Center, Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, China, Wuhan, China

Background: Unhealthy lifestyle habits, such as smoking, can impact oxidative stress. During oxidative stress, unnaturalized free radicals can damage DNA, proteins, and lipids, leading to cellular damage and death. A comprehensive measurement of various pro-oxidative and antioxidative exposures can reflect an individual's oxidative stress burden. However, studies on assessing the association between dietary and lifestyle factors related to oxidative stress and menopause were previously lacking.Materials and Methods: A cohort of 2,813 women aged 40 to 60 years from the National Health and Nutrition Examination Survey conducted between 2003 and 2020 was identified as meeting the eligibility criteria. The associations of oxidative balance score (OBS) with the menopausal status were examined via weighted logistic regression models, and the odds ratios (ORs) of menopause onset were calculated with 95% confidence intervals (CIs). Machine learning models were developed and compared to classify the menopausal status based on the OBS and other epidemiological factors, with the interpretability of the models explored using the Shapley Additive Explanations method.Results: Following adjustment for various confounding factors, OBS was reversely associated with menopause (OR: 0.97, 95% CI: 0.94-0.99, p = 0.010). When the OBS was categorized into quartiles, the association with menopause was still significant (p for trend = 0.009). The association of the OBS with menopause remained significant after excluding any each survey year cycles (p for trend < 0.050). The menopause classification models developed using TabFPN, Random Forest, CatBoost, and XGBoost achieved an area under the curve of 0.880, 0.884, 0.884, and 0.862, respectively. Based on the results from epidemiological analysis and machine learning models, the intake of magnesium, zinc, niacin, and vitamin B6 showed a decline in the early postmenopausal period and contributed in the model performance.