The Association of Obesity and Lipid-Related Indicators with All-Cause and Cardiovascular Mortality Risks in Patients with Diabetes or Prediabetes: A Cross-Sectional Study Based on Machine Learning Algorithms

Zhaoqi Yan Xing Chang Zhiming Liu ^* Ruxiu Liu ^*

• Guang’anmen Hospital, China Academy of Chinese Medical Sciences, Beijing, China

Objective: This study aims to explore the associations between various obesity and lipid-related indicators in patients with diabetes or prediabetes. Specifically, the indicators examined include the triglyceride-glucose index (TyG), along with its derived metrics: TyG-BMI, TyG-WHtR, TyG-WWI, TyG-WC, lipid accumulation product (LAP), visceral adiposity index (VAI), and abdominal obesity index (ABSI), resulting in a total of eight indicators. Methods: This study utilizes data from the NHANES conducted from 1999 to 2018, analyzing a cohort of 4,058 patients diagnosed with diabetes/prediabetes. We utilized multivariable Cox regression models to evaluate the impact of these indicators on both all-cause and cardiovascular mortality rates. Additionally, we compared the predictive performance of eight machine learning (ML) algorithms regarding mortality risk and used the SHAP method to clarify the significance of obesity and lipid-related indicators in mortality prediction.The results of the multivariable Cox regression analysis reveal significant associations between TyG, TyG-WWI, and ABSI with all-cause mortality among patients with diabetes/prediabetes. Compared to baseline levels, the HR for TyG in the fourth quartile (Q4) was 1.49, while for TyG-WWI (Q4), the HR was 1.52. Furthermore, ABSI was associated with increased all-cause mortality risk in groups Q3 and Q4, presenting risk ratios of 1.80 and 1.68, respectively. Notably, TyG (Q4) was also significantly associated with cardiovascular mortality risk, with an HR of 1.98. RCS analysis indicated a linear trend between TyG, TyG-WWI, and all-cause mortality, whereas ABSI displayed a non-linear trend. Among the ML algorithms evaluated, the XGBoost model exhibited the strongest predictive capability. The SHAP analysis indicated that the indicators with the greatest impact on all-cause mortality in patients with diabetes/prediabetes were ranked as follows: TyG > ABSI > TyG-WWI. Furthermore, sex-based subgroup analysis indicated that VAI was positively associated with cardiovascular mortality in male patients with diabetes/prediabetes, exhibiting a linear trend. Conclusion: TyG, TyG-WWI, ABSI, and VAI are closely linked to mortality risk in diabetes/prediabetes patients. Among these, TyG is significantly associated with both all-cause and cardiovascular mortality, showing superior predictive capability. We recommend long-term monitoring of these indicators and their inclusion in management strategies to effectively inform diabetes/prediabetes patients about their mortality risks.