Sex-Stratified Prediction Models for 5-Year Nonalcoholic Fatty Liver Disease Risk in Thyroid Cancer Patients: A Nationwide Cohort Study.

Abstract

Background/Objectives: Nonalcoholic fatty liver disease (NAFLD) is a significant complication among survivors of thyroid cancer; however, existing prediction models for NAFLD remain inadequate. Our objective was to develop survival prediction models for 5-year risk of NAFLD in patients diagnosed with thyroid cancer. Methods: Utilizing the Korean National Health Insurance Service claims database, we selected 3644 post-thyroidectomy patients with thyroid cancer between 2004 and 2014. Following a 7:3 stratified division into training and test datasets, we developed sex-stratified survival models using random survival forest (RSF) and Cox proportional hazards regression (Cox). The evaluation of prediction models was performed using Harrell's concordance index (C-index), time-dependent area under the curve (AUC), and risk stratification analysis. Results: In the female cohort, the Cox model exhibited a superior C-index of 0.67 (95% CI 0.61-0.72), surpassing the RSF model, which had a C-index of 0.62 (95% CI 0.57-0.68). Notably, age-stratified Cox models for females demonstrated enhanced performance compared to the unstratified female Cox model. Conversely, male-specific models did not show significant performance in NAFLD. Risk stratification analysis revealed that the female-specific models effectively categorized patients into low- and high-risk groups, with statistical significance (p < 0.001). Conclusions: This study constructed well-performing time-to-event prediction models for NAFLD of female patients with thyroid cancer, which is significant in risk stratification.