Prediction of treatment outcome in nasopharyngeal carcinoma patients.

BACKGROUND: Currently, the American Joint Committee on Cancer staging systems do not reliably predict outcomes in nasopharyngeal carcinoma as patients with similar tumour stages and treatments often experience divergent outcomes. This study aimed to develop prediction models for overall survival (OS), progression-free survival (PFS), locoregional control (LRC), and distant control (DC) based on routinely collected clinical parameters in nasopharyngeal carcinoma patients treated with (chemo)radiation to support personalised treatment strategies. METHODS: Data from 956 patients treated with intensity-modulated radiotherapy at Hunan Cancer Hospital were used for model training, with 409 patients reserved for internal testing. An external cohort of 484 patients from the Cancer Hospital of Shantou University Medical College was used to validate the OS model. Eleven clinical variables were included with multivariate imputation applied to missing values. Highly correlated variables were excluded through pre-selection, followed by stepwise forward Cox regression modelling. Model performance was compared to AJCC staging for each endpoint. RESULTS: The OS model achieved good performance in both the training set (C-index = 0.73, 95% CI [0.68–0.77]), internal test set (0.71 [0.64–0.77]) and the external test set (0.67 [0.61–0.72]). Both OS and PFS models performed significantly better than the AJCC models (DeLong’s test: both p < 0.01). CONCLUSIONS: Models for OS, PFS and DC effectively stratified patients into low and high-risk groups with significantly different outcomes, demonstrating their potential to guide individualised treatment decision.