Internal validation of a residual thyroid recurrence model after PTC

Abstract

Background At present, the risk factors contributing to the recurrence of papillary thyroid carcinoma (PTC) in residual thyroid tissue post-surgery are not yet fully understood. This study sought to identify multivariable predictors (MVPs) associated with such recurrence and to develop a dynamic, individualized nomogram. The objective is to establish a stratification framework to inform precision “risk-intervention” decision-making for patients at elevated risk of recurrence. Methods A cohort of 2,512 patients diagnosed with thyroid cancer and treated at the First Affiliated Hospital of Bengbu Medical University between January 2014 and December 2024 was analyzed. The cohort was stratified into a training set (n = 1,760) and a validation set (n = 752) using stratified random sampling with a 7:3 ratio. To address the class imbalance between recurrent and non-recurrent cases in the residual thyroid, the Random Over-Sampling Examples (ROSE) method was employed on the training set. Variable selection was conducted using univariate and multivariate logistic regression, LASSO regression, and standard stepwise regression based on the Akaike Information Criterion (AIC). A nomogram was subsequently developed by integrating the coefficients of the selected variables from the logistic regression model. The performance of the nomogram was assessed in both cohorts in terms of discrimination, calibration, and clinical utility. Results In a study of 2,512 thyroid cancer patients who received primary treatment, 146 experienced recurrence in the remaining thyroid tissue. Using univariate and multivariate logistic regression and LASSO regression for feature selection, four independent predictors were identified for the final predictive model: maximum tumor diameter, number of metastatic lymph nodes, lymph node positivity rate, and age. The nomogram showed strong discriminatory ability, with an AUC of 0.7518 in the training set and 0.8347 in the validation set. It also demonstrated excellent calibration and clinical utility across all groups. The best cut-off value for the nomogram score to identify patients with residual thyroid recurrence was 0.245, with scores above this linked to a significantly worse postoperative outcome. Conclusions We have developed and validated an innovative clinical prediction model specifically designed to assess the risk of recurrence in residual thyroid tissue. This model serves as a valuable tool for clinicians, facilitating personalized decision-making and representing a significant advancement in the postoperative management of papillary thyroid carcinoma.