The importance of assessing effect modification when asserting racial differences in associations between human leukocyte antigen class II alleles and hepatitis C virus outcomes.

To the Editor—In a recent article, Thio et al. [1] sought to determine whether DQB1*0301 and other HLA class II alleles were associated with hepatitis C virus (HCV) clearance and persistence in 3 multiethnic cohorts. The results of their analyses confirmed the previously documented association of DQB1*0301 and other HLA class II alleles with viral clearance [2]. Furthermore, Thio et al. hypothesized that the inconsistencies in the association between HLA class II alleles and HCV clearance found in earlier studies could be due to ethnic differences [1]. They reported that the association of DQB1*0301 and other HLA class II alleles with HCV clearance or persistence differed by ethnic identity. Finally, they emphasized the importance of studying genetic associations in an ethnically diverse cohort. These authors’ hypotheses, conclusions, and recommendations imply that race modifies the association between HLA class II alleles and HCV clearance or persistence, yet they did not present any formal assessment of effect modification. Formal assessment of effect modification is crucial if the differences between stratum-specific estimates are a primary focus of the study. Because of the race-matched case-control design of this study [1], effect modification by race can be assessed only on the multiplicative scale (the heterogeneity of the odds ratios [ORs]) [3]. When strong evidence of heterogeneity is identified, the estimates of the effect measures—ORs in this case—should be presented after stratification by the modifying factor, that is, race [4]. Thio et al. [1] highlighted differences by race but did not show compelling evidence of heterogeneity. In fact, the evidence for racial heterogeneity provided by the ORs in the results of this study is not strong. For example, using Cochran’s Q test of homogeneity [5] to compare the association between DQB1*0301 and HCV clearance in blacks (OR, 0.65; 95% confidence interval [CI], 0.42–1.00) with that in whites (OR, 0.85; 95% CI, 0.54–1.33), we find no evidence of significant heterogeneity of the ORs (Cochran’s Q, 0.70; df, 1; P 1⁄4 :40). Furthermore, the overlap of the CI functions illustrated in figure 1 suggests that the claimed ethnic difference is not clearly apparent. Moreover, if the assessment of race as a potentially important effect modifier of the association between DQB1*0301 and HCV clearance was an anticipated study question, avoidance of matching on race would be preferable [6]. Matching on such a factor precludes further assessment of effect modification on the additive scale in relative terms [3, 6]. Assessment on the additive scale provides better evidence of the extent and direction in which the effect modifier alters the main effect [3]. In addition, we want to emphasize that race is a construct that reflects social stratification of groups according to phenotypic and cultural characteristics [7]. The presumption that observed differences between socially defined groups could reveal genetic/racial traits has been widely criticized [8, 9]. For example, heterogeneity of genetic characteristics within socially defined racial groups is far greater than that between groups [10]. Therefore, to attribute observed outcome differences to biological aspects of race warrants at least some minimal evidence of heterogeneity. We contend that the evidence presented by Thio et al. [1] is not nearly strong enough to justify asserting and publicizing [11] that the association of DQB1*0301 and other HLA class II alleles with HCV clearance differs by ethnicity.