Performance of CT radiomics in predicting the overall survival of patients with stage III clear cell renal carcinoma after radical nephrectomy

Purpose To investigate the performance of CT radiomics in predicting the overall survival (OS) of patients with stage III clear cell renal carcinoma (ccRCC) after radical nephrectomy. Materials and methods The 132 patients with stage III ccRCC undergoing radical nephrectomy were collected, and the patients were divided into training set ( n = 79) and validation set ( n = 53). The ccRCC was segmented and 396 radiomics features were extracted. After dimensionality reduction, radiomics score (RS) was obtained. COX regression was used to construct Model 1 (clinical variables + CT findings) and Model 2 (clinical variables + CT findings + RS) in the training set to predict the OS of patients, and then, the performance of the two models in the two data sets was compared. Results In the training set, Akaike information criterion, C-index, and corrected C-index were 295.51, 0.744, and 0.728 for Model 1, and 271.78, 0.805, and 0.799 for Model 2, respectively. In the validation set, the corresponding values were 185.68, 0.701, and 0.699 for Model 1, and 175.99, 0.768, and 0.768 for Model 2. The calibration curves showed that both models had good calibration degrees in the validation set. Compared with Model 1, the continuous net reclassification index and integrated discrimination improvement index of Model 2 in the two data sets were positively improved. Conclusion The two prediction models showed high performance in the evaluation of OS of stage III ccRCC patients after radical nephrectomy, among which Model 2 based on ISUP grade and RS was more concise and efficient.