Machine Learning Identifies Risk Factors for Severe Cognitive Decline During Menopause

A research team from Shandong, China, has used machine learning to identify factors contributing to severe subjective cognitive decline (SCD) among nurses undergoing menopause, offering valuable insights for early intervention.

_{©New Africa/stock.adobe.com}

The support vector machine (SVM) model outperformed 6 other machine learning algorithms in identifying individuals at high risk for severe SCD, achieving an area under the receiver operating characteristic curve of 0.846, an accuracy of 78.9%, and a specificity of 80.2%. Menopausal symptoms, stage of menopause, economic status, and sleep satisfaction emerged as the most influential predictors of severe SCD.

The study, led by Xiangyu Zhao, PhD, and colleagues from Shandong University, and published online in the journal Menopause, analyzed data from 1,264 nurses aged 40 to 60 years experiencing the menopause transition. Participants completed detailed questionnaires covering demographic, occupational, and health-related variables, and SCD was assessed using the self-reported SCD-Q9 questionnaire. The questionnaire was condensed into 2 components – overall functional memory and time comparison and activities of daily living. A final score could range from 0 to 9, with higher scores reflecting greater severity. Nurses scoring a score of 7.5 or greater on the SCD-Q9 were categorized as having severe SCD, with 340 participants meeting this criterion.

Machine learning benefits. Current testing for cognitive performance is largely based on models that include laboratory indicators and brain imaging, making them costly and impractical in a clinical setting. Questionnaire-based models offer a streamlined alternative.

By applying machine learning, Zhao et al aimed to improve the identification of SCD risk factors beyond the limitations of traditional statistical methods. Seven models, logistic regression, random forest, SVM, extreme gradient boosting, multilayer perceptron, k-nearest neighbors, and elastic net, were evaluated. “The SVM model demonstrated superior performance because of its ability to handle complex, nonlinear relationships among variables,” the team wrote.

The analysis highlighted 13 significant predictors of SCD severity, refined from an original selection of 24 features using the Boruta algorithm. Domains of the 13 features used for model development and validation were sociodemographic variables (4), menstruation-related variables (3), lifestyle-related variables (2) and mental-health related variables (4). Among these, menopausal symptoms ranked highest in importance, with the SVM model revealing that worsening symptoms were strongly correlated with severe SCD.

SHAP analysis. The investigators employed Shapley Additive Explanations (SHAP) analysis to further understand the relationships between predictors and SCD. “Our SHAP analysis revealed that while the severity of menopausal symptoms significantly increased SCD risk, this effect plateaued at higher levels,” investigators explained, suggesting that symptom management could play a critical role in mitigating SCD risk. Other notable factors included sleep satisfaction, which showed a strong protective effect against SCD, and positive emotions and resilience, which were inversely associated with cognitive decline.

Economic status also played a pivotal role, with nurses who reported insufficient financial resources at greater risk. “Economic stress likely exacerbates the cognitive burden by amplifying psychological and physical stressors,” the authors noted.

The study underscores the unique vulnerabilities of nurses, a group already facing high occupational stress and irregular work hours. The cognitive challenges associated with menopause could impair work efficiency and quality of care, emphasizing the need for workplace policies supporting menopausal health, the team observed.

“This research provides a framework for developing targeted interventions,” they stated. “By addressing modifiable factors such as sleep quality and emotional well-being, [health care] organizations can better support their workforce.”

Among the study's limitations, the researchers acknowledged the cross-sectional design, which precludes causal inferences, and the focus on a single occupational group, which may limit generalizability. They recommend expanding future studies to include diverse occupational and age groups, increasing sample sizes to enhance statistical power and incorporating objective cognitive assessments. Longitudinal designs could also explore causal pathways and track the progression of SCD, they wrote.

Despite these limitations, the study highlights the promise of machine learning in healthcare. Zhao et al concluded, “Our findings not only improve our understanding of cognitive decline during menopause but also provide actionable insights for early intervention.” The use of machine learning in this context demonstrates its growing potential to refine predictive models for complex health outcomes, they added.

References
1. Zhao X, Shen X, Jia F, et al. Using machine learning models to identify severe subjective cognitive decline and related factors in nurses during the menopause transition: a pilot study. Menopause. 2025;32:00–00. doi: 10.1097/GME.0000000000002500
2. AI helps to identify subjective cognitive decline during the menopause transition. News release. The Menopause Society. January 15, 2025. Accessed January 15, 2025. https://menopause.org/press-releases/ai-helps-to-identify-subjective-cognitive-decline-during-the-menopause-transition