Area under curves (AUC) as a measure of predictive strength for risk-prediction models based on different indicators^a.

aThe environmental model is based on consumption of salted fish and preserved vegetables, and cumulative amount of smoking. The family history of NPC model includes family history of NPC only. The epidemiological model combines both environmental and family history of NPC predictors. The genetic risk score model includes a score derived from seven SNPs identified in the Cantonese GWAS. The inclusive model integrates all data on epidemiological and genetic predictors. bχ2 statistic and P value was calculated from the Hosmer–Lemeshow Goodness-of-Fit test, a model with χ2 statistic <20 (P>0.01) is considered as a good calibration. cAUC of the models were compared with a nonparametric approach, and P value was obtained from the comparison of the inclusive model with the other models.