A predictive model and survival analysis for local recurrence in differentiated thyroid carcinoma

BACKGROUND: Local recurrence (LR) is associated with poor outcome in patients with differentiated thyroid carci-noma (DTC). The aim of this study was to explore potential risk factors for LR and build a predictive model.METHODS: The medical data of patients who were diagnosed with DTC after initial surgery in three medical centers (2000-2018) were reviewed. Detailed clinicopathologic characteristics of all cases were identified.RESULTS: Multiple factors, including extrathyroidal extension (ETE), histology, symptoms, multifocality, and tumor diameter, were significantly different between the LR and no evidence of disease groups in univariate and multivariate analysis (P<0.05). Tumor diameter, symptoms, and ETE made the greatest contributions to prognosis according to deci-sion tree analysis and random forest algorithm. The predictive model constructed from these data achieved 98.7% ac-curacy of classification. A five-fold cross-validation confirmed that the model has 84.7-89.7% accuracy of classification. Additionally, symptoms and ETE were independent predictors on survival analysis (P<0.05).CONCLUSIONS: This study optimized the weight of risk factors, including tumor diameter, symptoms, ETE, and mul-tifocality, in predicting LR in patients with DTC. Our predictive model provides a strong tool to distinguish between high-risk and low-risk DTC.