Prospective assessment of a nasopharyngeal carcinoma risk score in a population undergoing screening
Genghang Chen, Zhiwei Liu, Mingfang Ji, Ruth M. Pfeiffer, Qihong Huang, Yu‐Qiang Lu, Shang‐Hang Xie, Chuyang Lin, Wen‐Jie Chen, Xiaoxia Chen, Wei Ling, Yuying Fan, Yu Xia, Biaohua Wu, Kuangrong Wei, Hui‐Lian Rao, Xiang Guo, Ming‐Huang Hong, Jun Ma, Qing Liu, Allan Hildesheim, Su‐Mei Cao
Abstract
Despite evidence suggesting the utility of Epstein-Barr virus (EBV) markers to stratify individuals with respect to nasopharyngeal carcinoma (NPC) risk in NPC high-risk regions, no validated NPC risk prediction model exists. We aimed to validate an EBV-based NPC risk score in an endemic population undergoing screening for NPC. This prospective study was embedded within an ongoing NPC screening trial in southern China initiated in 2008, with 51 235 adult participants. We assessed the score's discriminatory ability (area under the receiver-operator-characteristics curve, AUC). A new model incorporating the EBV score, sex and family history was developed using logistic regression and internally validated using cross-validation. AUCs were compared. We also calculated absolute NPC risk combining the risk score with population incidence and competing mortality data. A total of 151 NPC cases were detected in 2008 to 2016. The EBV-based score was highly discriminating, with AUC = 0.95 (95% CI = 0.93-0.97). For 90% specificity, the score had 87.4% sensitivity (95% CI = 81.0-92.3%). As specificity increased from 90% to 99%, the positive predictive value increased from 2.4% (95% CI = 1.9-3.0%) to 12.5% (9.9-15.5%). Correspondingly, the number of positive tests per detected NPC case decreased from 272 (95% CI = 255-290) to 50 (41-59). Combining the score with other risk factors (sex, first-degree family history of NPC) did not improve AUC. Men aged 55 to 59 years with the highest risk profile had the highest 5-year absolute NPC risk of 6.5%. We externally validated the discriminatory accuracy of a previously developed EBV score in a high-risk population. Adding nonviral risk factors did not improve NPC prediction.