Purpose: This study aimed to develop a clinically predictive nomogram model to predict the survival probability of differentiated thyroid carcinoma patients and compare the value of this model with that of the eighth edition AJCC cancer staging system.
Methods: We selected 59,876 differentiated thyroid carcinoma patients diagnosed between 2004 and 2015 from the SEER database and separated those patients into a training set (70%) and a validation set (30%) randomly. We used Cox regression analysis to build the nomogram model (model 1) and the eighth edition AJCC cancer staging model (model 2). Then we compared the predictive accuracy, discrimination, and clinical usage of both models by calculating AUC (Area under the curve), C-index, as well as analyzing DCA (Decision Curve Analysis) performance respectively.
Results: AUCs of all predicted time points (12-month, 36-month, 60-month, and 120-month) of model 1 were 0.933, 0.913, 0.879, and 0.868 for the training set; 0.933, 0.926, 0.916, and 0.894 for the validation set. As for model 2, data were 0.938, 0.906, 0.866, and 0.847 for the training set; 0.924, 0.925, 0.912, and 0.867 for the validation set. C-indices of model 1 were higher than those of model 2 (0.923 vs. 0.918 for the training set, 0.938 vs. 0.930 for the validation set). DCA comparison showed that the net benefit of model 1 was bigger when comparing with that of model 2.
Conclusions: Model 1 provided with both better predictive accuracy and clinical usage compared with those of model 2 and might be able to predict the survival probability of differentiated thyroid carcinoma patients visually and accurately with a higher net benefit.
Keywords: AJCC cancer staging system; Cox regression analysis; Differentiated thyroid carcinoma; Nomogram; Prediction model; SEER database.
© 2021. The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature.