Characteristics of the Study Population
The summary of the distributions of selected characteristics of cases and controls is presented in Table 1. There were no significant differences in the distributions of age and sex between cases and controls. The average age was 58.8 years for the cases (median, 58; range, 40-91) and 58.7 years for the controls (median, 59; range, 40-90). Of all the subjects, 66.2% of cases and 64.2% of controls were male, 56.7% of cases and 50.1% of controls were ever smokers and 67.3% of cases and 61.7% of controls were ever drinkers. There were more current smokers (32.1%) and current drinkers (33.8%) in cases than in controls (17.5% and 31.9%, respectively). The primary HNSCCs of 337 patients included the oral cavity (67, 19.9%), oropharynx (150, 44.5%), hypopharynx and larynx (120, 35.6%).
Differences in NER Protein Expression Levels between Cases and Controls
The cases showed lower relative mean expression levels in six of the nine core NER proteins analyzed than did controls, except for XPC, XPG, and ERCC1 (Table 2). In Wilcoxon rank-sum test for differences in NER protein expression levels between cases and controls, only XPA levels were statistically significantly lower in cases than in controls (P= 0.001; Fig 1A). Because the expression levels of the nine NER proteins were measured at the same time, they were likely to be correlated with each other. As shown in Supplementary Table 1, expression levels of XPA were statistically significantly correlated with XPB, XPC, XPD, and ERCC1(P = 0.019, P = 0.050, and P < 0.001, and P = 0.012, respectively).
Stratification Analyses of Expression Levels of XPA by Selected Variables
Stratification analyses of XPA expression levels revealed that patients in subgroups of the age ≤ 59, age > 59, male, female, former and current smokers, and former and current drinkers exhibited significantly lower mean expression levels of XPA than did controls (All the P < 0.001, respectively, Table 3). In cases, women had lower expression levels of XPA than did men, but in controls, women had higher expression levels of XPA than did men, and the sex differences in the expression levels were insignificant in both case and control groups (P = 0.249 and P = 0.889, respectively, Table 3). Moreover, both ever smokers and drinkers had significant lower expression levels of XPA than did never smokers and drinkers, respectively (All the P < 0.001, respectively, Table 3). There were no significant differences in the expression levels of XPA by tumor sites, suggesting that expression levels of XPA may not be different among tumors of HNSCCs (Supplementary Table2).
Associations between NER Protein Expression Levels and Risk of HNSCCs
To estimate HNSCC risk, the relative expression levels were grouped into median values of the controls (Table 4). The crude ORs for HNSCC risk associated with lower relative expression levels of XPA were 1.43 (95% CI, 1.04-1.97), compared with the high expression levels of XPA. After adjusting for age, sex, smoking status and alcohol consumption in multivariate logistic regression analysis, the OR of XPA remained essentially unchanged. When continuous expression values were used in the logistic regression model with adjustment for all covariates, there was also a dose-response relationship between the reduced expression levels of XPA and the increased HNSCC risk (Ptrend= 0.031).
Interactions between XPA Expression Levels and Selected Variables
We further assessed possible interactions on a multiplicative scale between expression levels of XPA and selected variables listed in Table 1. The multiplicative interaction was tested when we included the interaction term (i.e., relative expression levels of XPA × each of the risk factors) in a multivariate regression model that also included the main effects of NER protein expression levels and other covariates. We found that smoking status as well as drinking status had significantly multiplicative interactions with relative expression levels of XPA (P = 0.005 and P = 0.044, respectively, Table 3), in association with HNSCC risk. To further unravel these multiplicative interactions, we stratified the adjusted ORs by smoking status and drinking status. It was apparent that ORs for the relative expression levels of XPA by median in groups of ever smokers were greater than those of never smokers (Fig. 1B). And the ORs for the relative expression levels of XPA by medians in groups of ever drinkers were greater than those of never drinkers (Fig. 1C).
We further assessed the prediction performance of models integrating demographic variables and protein expression levels on HNSCCs using the ROC curves that measure the effect of XPA expression levels in two dimensions. The AUC was significantly improved in the model that included the effect of XPA expression levels, compared with the model that did not (Fig. 2A, P = 0.004). Furthermore, the AUC was significantly improved in former and current smokers that included the effects of XPA expression levels, compared with the model that did not (Fig. 2C and 2D, P < 0.001 and P < 0.001, respectively), but insignificantly improved in never smokers (Fig. 2B, P = 0.462). The AUC was significantly improved in former and current drinkers that included the effects of XPA expression levels, compared with the model that did not (Supplementary Fig. 1B and 1C, P = 0.001 and P = 0.001, respectively), but insignificantly improved in never drinkers (Supplementary Fig. 1A, P = 0.404).