# Machine learning-based association analysis of triglyceride-glucose index with melanoma prevalence and all-cause mortality: insights from cross-sectional NHANES 1999–2018 data and an external hospital-based dataset

## TL;DR

This study explores how the triglyceride-glucose index relates to melanoma prevalence and mortality using data from 1999–2018 and machine learning models.

## Contribution

The study introduces a U-shaped relationship between the TyG index and melanoma mortality and identifies key predictors using machine learning.

## Key findings

- The TyG index shows a U-shaped association with all-cause mortality in melanoma patients.
- Ridge regression model outperformed other models in predicting melanoma risk with an AUROC of 0.85.
- Race, age, and serum phosphorus levels were identified as key predictors of melanoma risk via SHAP analysis.

## Abstract

Insulin resistance has been associated with melanoma, however, the relationship between the triglyceride-glucose (TyG) index and this condition remains unclear. This study aims to investigate the relationship between the TyG index and melanoma.

This study included 21,360 participants from the 1999–2018 National Health and Nutrition Examination Survey (NHANES). We used weighted logistic regression for the TyG index's link with melanoma prevalence, weighted Cox regression for mortality, restricted cubic spline for dose-response, and subgroup analyses to verify robustness. An optimal predictive model was constructed using seven machine learning algorithms, and Shapley additive explanations (SHAP) values visualization was conducted. A total of 475 patients with primary non-metastatic acral melanoma from three tertiary hospitals were included for descriptive analysis.

After demographic adjustment, the third TyG tertile exhibited elevated all-cause mortality risk (HR: 1.329, 95% CI: 1.201–1.458, P < 0.01). However, after further adjustment for all covariates, this association was no longer statistically significant (P > 0.05). RCS demonstrated a U-shaped relationship between the TyG index and all-cause mortality. Similar results were observed across most subgroup analyses. The ridge regression model (AUROC = 0.85) performed optimally, with SHAP analysis identifying race, age, and serum phosphorus levels as key predictors of melanoma risk.

The TyG index shows a U-shaped association with all-cause mortality in patients with melanoma. The ridge regression model demonstrated the best predictive performance for melanoma risk in internal validation, with SHAP analysis identifying race, age, and metabolic markers as key influencing factors.