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

Why It Matters

The findings suggest metabolic health, captured by the TyG index, may influence melanoma outcomes and inform risk‑stratification models, guiding clinicians toward more personalized screening strategies.

Key Takeaways

• U‑shaped TyG‑mortality link in melanoma patients
• Ridge regression achieved highest AUROC (0.85) among models
• Race, age, phosphorus identified as top SHAP predictors
• Full covariate adjustment nullified TyG’s significance for prevalence
• Metabolic comorbidities amplify mortality risk with high TyG

Pulse Analysis

The triglyceride‑glucose (TyG) index, a surrogate for insulin resistance, has emerged as a potential biomarker linking metabolic dysfunction to cancer biology. In this study, researchers leveraged two decades of NHANES data to explore whether TyG correlates with melanoma incidence and overall survival. Although crude models suggested a positive relationship between higher TyG levels and mortality, comprehensive adjustment for demographic, lifestyle, and biochemical factors revealed a nuanced U‑shaped curve, indicating that both low and high extremes of the index may confer risk. This pattern underscores the complexity of metabolic pathways in melanoma progression and cautions against simplistic interpretations of single‑marker analyses.

To translate these epidemiological insights into clinical utility, the authors applied seven machine‑learning algorithms to predict melanoma risk, integrating TyG with a suite of clinical variables. Ridge regression emerged as the top performer, achieving an AUROC of 0.85 while maintaining calibration across validation folds. The model’s robustness stems from its regularization strength, which mitigates overfitting in the context of a rare outcome (0.6% prevalence). Nevertheless, precision‑recall metrics highlighted the challenges of class‑imbalanced data, suggesting that further refinement—perhaps through synthetic minority oversampling or ensemble techniques—could enhance real‑world applicability.

Interpretability was addressed through SHAP values, which pinpointed race, age, and serum phosphorus as the most influential features, alongside the TyG index itself. These results align with established melanoma risk factors while introducing metabolic markers into the risk equation. For healthcare systems, incorporating such a model could streamline early‑detection pathways, prioritize high‑risk individuals for dermatologic evaluation, and ultimately reduce downstream treatment costs. Future research should validate these findings in larger, multi‑ethnic cohorts and examine whether interventions targeting insulin resistance can modify the observed U‑shaped mortality relationship.