Unveiling the relationship between stress–hyperglycemia ratio and cardiometabolic multimorbidity risk using interpretable machine learning

BACKGROUND: Cardiometabolic multimorbidity (CMM) is the simultaneous manifestation of multiple cardiovascular and metabolic diseases, and it has arisen as a substantial worldwide healthcare issue. The stress-hyperglycemia ratio (SHR) represents a novel biomarker that is strongly associated with the prognosis of various diseases; however, its role in CMM remains insufficiently understood. This research aims to examine the association between SHR and CMM risk and assess its clinical utility in risk assessment. METHODS: This cross-sectional study utilized data from the National Health and Nutrition Examination Survey (NHANES), with a total of 12,279 participants meeting the inclusion criteria. A weighted logistic regression model was used to examine the correlation between SHR and CMM. Furthermore, machine learning (ML) techniques were applied to develop a CMM prediction model, and the validity of the findings was confirmed by several sensitivity analysis, including external validation using the China Health and Retirement Longitudinal Survey (CHARLS). In addition, mediation analysis was performed to investigate the potential mediating roles of body mass index (BMI) and waist circumference (WC). RESULTS: A substantial positive relationship was identified between SHR and CMM risk. For 1-unit rise in SHR, the risk of CMM rose by 21.131 (OR = 22.131 10.688, 45.823). Smooth curve fitting analysis indicated a U-shaped correlation between SHR and CMM risk. When SHR is below 0.841, CMM risk decreases as SHR increases (OR = 0.001 0.000, 0.004); however, when SHR exceeds 0.841, CMM risk increases sharply as SHR rises (OR = 116.890 70.086, 194.951). The findings of the mediation study demonstrated that BMI and WC moderated the association between SHR and CMM risk. Furthermore, the gradient boosting machine (GBM) model demonstrated robust predictive performance with an Area Under the Curve (AUC) of 0.880 (95% CI 0.866-0.894), while shapley additive explanations identified age, SHR, and WC as the top three predictors with the most significant impact on the model outcomes. Sensitivity analyses consistently validated the findings, with the external validation in the CHARLS cohort confirming a significant positive association (OR = 1.389, 1.026, 1.880. CONCLUSIONS: This study identified a U-shaped non-linear relationship between SHR and CMM risk, underscoring the potential clinical value of SHR in the early diagnosis and risk assessment of CMM. The predictive model developed using machine learning methods can significantly aid clinicians in conducting personalized risk assessments for CMM.