A comprehensive risk score for effective risk stratification and screening of nasopharyngeal carcinoma

Abstract

Using Epstein-Barr virus (EBV)-based markers to screen populations at high risk for nasopharyngeal carcinoma (NPC) is an attractive preventive approach. Here, we develop a comprehensive risk score (CRS) that combines risk effects of EBV and human genetics for NPC risk stratification and validate this CRS within an independent, population-based dataset. Comparing the top decile with the bottom quintile of CRSs, the odds ratio of developing NPC is 21 (95% confidence interval: 12-37) in the validation dataset. When combining the top quintile of CRS with EBV serology tests currently used for NPC screening in southern China, the positive prediction value of screening increases from 4.70% (serology test alone) to 43.24% (CRS plus serology test). By identifying individuals at a monogenic level of NPC risk, this CRS approach provides opportunities for personalized risk prediction and population screening in endemic areas for the early diagnosis and secondary prevention of NPC.

Conflict of interest statement

The authors declare no competing interests.

© 2021. The Author(s).

Figures

Fig. 1. Workflow of comprehensive risk score (CRS) construction and its application in NPC screening.

The CRS was developed using EBV, host genetic and epidemiological risk factors for NPC risk prediction with a discovery dataset from Guangdong province, and validated in an independent dataset from Guangxi province. The CRS was combined with EBV serology tests for NPC screening.

Fig. 2. NPC risk stratified by comprehensive risk score (CRS).

a Receiver operating characteristic (ROC) curve analysis of the validation dataset. The area under the ROC curve (AUC) for each model is indicated. b Distribution of patients in the validation dataset stratified into five categories, 0–20%, 20–40%, 40–60%, 60–80%, and 80–100%, according to the CRS percentile among controls in the training dataset. c Association between the CRS and NPC risk in the validation dataset (Cases: n = 427, Controls: n = 751). The validation dataset was stratified into categories (lower four quintiles and top two deciles) according to the CRS percentile among controls in the training dataset, and participants in the bottom quintile of CRS served as the reference group. The odds ratio (OR) of developing NPC was estimated using logistic regression analysis with each group included as a categorical variable. The blue squares represent the odds ratios of each category, and the error bars represent the 95% confidence intervals. CI, confidence interval. Source data of (b) are provided in the Source Data file.