Rethinking the prognosis model of differentiated thyroid carcinoma

Background The prediction efficiency of long-term cancer-specific survival (CSS) in guiding the treatment of differentiated thyroid carcinoma (DTC) patients is still unsatisfactory. We need to refine the system so that it more accurately correlates with survival. Methods This is a retrospective study using the Surveillance, Epidemiology, and End Results (SEER) database, and included patients who underwent surgical treatment and were diagnosed with DTC from 2004 to 2020. Patients were divided into a training cohort (2004–2015) and validation cohort (2016–2020). Decision tree methodology was used to build the model in the training cohort. The newly identified groups were verified in the validation cohort. Results DTC patient totals of 52,917 and 48,896 were included in the training and validation cohorts, respectively. Decision tree classification of DTC patients consisted of five categorical variables, which in order of importance were as follows: M categories, age, extrathyroidal extension, tumor size, and N categories. Then, we identified five TNM groups with similar within-group CSS. More patients were classified as stage I, and the number of stage IV patients decreased significantly. The new system had a higher proportion of variance explained (PVE) (5.04%) and lower Akaike information criterion (AIC) (18,331.906) than the 8th TNM staging system (a PVE of 4.11% and AIC of 18,692.282). In the validation cohort, the new system also showed better discrimination for survival. Conclusion The new system for DTC appeared to be more accurate in distinguishing stages according to the risk of mortality and provided more accurate risk stratifications and potential treatment selections.