Machine learning models for predicting postpartum convulsions using clinical indicators from PMA Ethiopia data.

Cancer is a leading cause of mortality worldwide, with increasing incidence and disease burden reported in both developed and developing regions (, ). Despite advances in surgical techniques, chemotherapy, radiotherapy, targeted therapy, and immunotherapy, long-term survival of patients with cancer remains unsatisfactory, largely due to tumor recurrence, metastasis, treatment intolerance, and systemic comorbidities. Among these comorbid conditions, metabolic disorders and nutritional deterioration are two of the most prevalent and impactful issues that affect the entire disease course from diagnosis to survival (, ). Insulin resistance (IR) is a core pathological feature of metabolic disturbance in patients with cancer, characterized by impaired insulin-mediated glucose uptake and utilization in peripheral tissues (, ). Mounting evidence has confirmed that IR is closely associated with tumor initiation, progression, epithelial-mesenchymal transition, angiogenesis, and treatment resistance (, ). Hyperinsulinemia caused by IR can activate the PI3K/Akt/mTOR signaling pathway, promoting tumor cell proliferation and inhibiting apoptosis (, ). Meanwhile, IR induces chronic low-grade inflammation, oxidative stress, and lipid metabolism disorders, further accelerating cancer development (). Traditional IR evaluation indexes such as homeostasis model assessment of insulin resistance (HOMA-IR), quantitative insulin sensitivity check index (QUICKI), and McAuley index require simultaneous detection of fasting insulin and blood glucose, which increases clinical detection costs and is not routinely included in basic laboratory examinations for patients with cancer, limiting their wide application in clinical practice (). In recent years, simple and convenient IR surrogate indicators based on fasting blood glucose and triglycerides have attracted extensive attention (, ). The triglyceride-glucose index (TyG), calculated by fasting triglycerides and blood glucose, has been widely verified in metabolic diseases such as type 2 diabetes, metabolic-associated fatty liver disease, and hypertension as a reliable and stable marker of IR (). Compared with HOMA-IR, TyG has the advantages of easy access to indicators, simple calculation, and good reproducibility (, ). Further studies found that combining TyG with body mass index (BMI) to construct the composite index TyG-BMI can better reflect the interaction between IR, body fat distribution, and nutritional status, and has higher predictive value for metabolic diseases and tumors than a single index ().