Abstract
Background: The prevalence of type 2 diabetes (T2D) is on the rise, and insulin resistance (IR) is one of the key risk factors for developing T2D. This paper seeks to identify risk factors for IR in women with normal menstrual cycles (NM) and early menopausal women (EM). Methods: EM women between 30 and 50 years old were compared with an NM control group. Four machine learning (ML) methods were trained using comprehensive physiological and lifestyle data to estimate a homeostasis model for insulin resistance (HOMA-IR dependent variable). Traditional multiple linear regression (MLR) was used as a benchmark for comparison. Results: A total of 948 participants were enrolled (NM: 410, EM: 538). On average, ML outperformed MLR, identifying the six key risk factors in the EM group (from most to least important) as waist-hip ratio (WHR), triglyceride (TG), glutamic-pyruvic transaminase (GPT), glutamic oxaloacetic transaminase (GOT), HDL-Cholesterol (HDL-C), and lactic dehydrogenase (LDH). Rankings differed in the NM group, with WHR identified as the leading risk factor, followed by C-reactive protein (CRP), HDL-C, total bilirubin (TBIL), diastolic blood pressure (DBP), and white blood cell count (WBC). Conclusions: Using ML, we found that WHR and HDL-C are the common denominators in both EM and NM women, with additional correlations with TG, liver enzymes and LDH for EM women. These results clearly indicate the importance of estrogen protection, suppressing less important factors (TG, liver enzyme, and LDH), and only the stronger inflammatory markers become important (CRP, TBIL, and WBC). Once estrogen's protection disappears, the suppression of CRP, TBIL, and WBC would become weaker. Since these 3 features are significantly correlated with body weight, for women under 50, reducing body weight is the most important factor in preventing hyperglycemia.