Evaluation of sexual networks as a cause for disparate HIV prevalence between blacks and whites

Abstract

We read with great interest findings from the second round of the National Behavioral Surveillance System's NHBS among Men Who Have Sex With Men (NHBS-MSM2), which was conducted to better understand disparities in HIV infection between black and white MSM in the United States [1]. Specifically, the aim of this study was to assess several hypotheses about the cause of marked racial disparities among MSM [2], and of particular interest to our group, those related to sexual networks. We raise two major aspects of the described study that require additional consideration: sampling limitations that distort network characterization; and intranetwork mixing patterns as an explanation for the disparities between white and black MSM. A major limitation of analysis presented was the failure to assign weights to participants. In venue-based, time-space sampling, the ‘visit’ is the unit of analysis and not the participant and, thus, weighting of participants is required. MSM who participate in venues more frequently are overrepresented in the sample. This observation is not a new one for NHBS-MSM [3], and the resulting skew could be important to network characterization because there is likely to be wide variation in the number of visits by regular and occasional visitors. When network comparisons are then made between races, we do not know whether we are measuring actual network differences or differences in frequency and type of venue affiliation generated by the sampling methodology used. Another distortion of the sexual network occurs because attribute data are collected only on the last male partner. Such data are limited to specific dyadic subnetworks and, thus, cannot be configured to provide meaningful data to answer hypotheses about black MSM sexual networks. For example, sampling an individual's last male sex partner privileges both recency and men, and it is then inferred that the subnetwork generated is typical of the unobserved MSM network. This is still an empirical question. Further distortions in sex networks would exist if differential casual partnering patterns existed in this specific sample with both male and female partners, and could compound existing differences observed in the sample such as age (blacks were younger) and venue recruitment type (most whites were recruited from bars). To illuminate the networks further, several follow-up questions emerge: are the venue distributions of the analytic subsample similar to that depicted in the sample presented in Table 1 of [1]? Are the index cases evenly distributed across metropolitan areas? Does this subsample of newly infected MSM map to specific venues/regions? Fundamentally, the boundaries of networks analyzed are unclear; and subanalysis of an unobserved larger ‘network’ violates core principles of network analysis [4]. The authors cite our work [5] several times to support the fact that MSM partner according to homophily on race (e.g., blacks with blacks; whites with whites). Indeed, homophilous race partnering was one of the findings; the most significant finding from this work, however, was the measured intranetwork mixing pattern. Disassortative mixing (high-risk with low-risk) was found to be much more common among black compared to white sex networks [5]. Disassortative mixing can be conceptualized as mixing within a core-periphery structure, with, for example, a higher risk core more likely to mix with a lower risk periphery. In the study by Oster et al.[1], the limitations in sampling combined with a unilateral focus on one type of dyadic subnetwork limit any analyses of disassortativity on age, drug-use, incarceration history, sex exchange, concurrency, and HIV status. Interestingly, this last variable, ‘unknown HIV status of last partner’ was found to be significantly associated with black race in bivariate analysis. This variable, however, is a partner attribute, and thus has been misclassified into a category separate from ‘network hypotheses’ in this and previous work [2]. Furthermore, the index client's own HIV status is not included in any of the current analyses and we are unable to observe potentially important network differences in serosorting (mixing by serostatus). Differences in serosorting within networks of black MSM could be an explanation for the disparity in HIV infection observed. In a recent sex network mixing analysis of 352 black MSM [6], using standard assortativity measures [7], we found that mixing by HIV serostatus was highly assortative; yet this finding is contingent upon perceptions of partners’ HIV status and one's own HIV status. The increased likelihood of disassortative mixing in black communities could magnify any ‘error’ in serosorting resulting in lower serosorting efficacy. Serosorting efficacy would be contingent upon each individual in the network knowing the HIV status of a partnered other and acting on that information, an important area of future network research. Acknowledgements Conflicts of interest There are no conflicts of interest.