Benign concurrency in context and practice: a response to Epstein and Stanton

Epstein and Stanton's [1] correspondence is a valuable contribution to the important task of deciphering the implications of partnership concurrency for the spread of HIV [1]. The authors correctly note a paradox between the benign effect of polygyny at the ecological level [2] and other studies (including one of ours) that report positive individual-level correlations, and ask ‘how can polygyny be a risk factor at the individual level, but not at the community level?’ Their own explanation highlights the higher prevalence of polygyny in western Africa, where women are under greater surveillance than in southern Africa. Although these regional cultural differences are probably important, our study demonstrates that they are not sufficient because the neutral or negative association of polygyny with HIV prevalence persists within countries. Before turning to possible explanations for this paradox, we restate it using a multilevel model with individual-level HIV status as the outcome (Fig. 1). The set of estimates at the top are odds ratios for a pooled analysis followed by estimates for groups of countries stratified by HIV prevalence. In Fig. 1a we see that living in a survey cluster with more polygyny is negatively associated with HIV status of both men and women. Turning to the individual-level association between union type and HIV status in Fig. 1b we find that there is no relationship between polygyny status and HIV status for men. The odds ratio does, however, approach statistical significance in high HIV prevalence settings. Similarly, first wives are no more likely to be HIV positive than wives of monogamous men. Conversely, junior wives of polygynous men are more often HIV positive than spouses of monogamous men.Fig. 1: Ecological- and individual-level associations between polygyny and HIV status (odds ratios, all countries, and countries stratified by HIV prevalence, Demographic and Health Surveys 2003 -2007). Estimates from a three-level (individual, survey cluster, and country) hierarchical logistic model. (a) Odds ratios for a one percent point increase in the cluster-level prevalence of polygyny. (b) Odds ratios by individual-level union type whereby monogamous marriages serve as the reference category. Data sources and variable definitions are described in detail in [2]. Because not all the Demographic and Health Surveys and AIDS Indicator Surveys distinguish between wife's rank, we restricted the analysis to Niger, Senegal, Democratic Republic of Congo, Mali, Burkina Faso, Ghana, Guinea, and Liberia (low HIV prevalence stratum: HIV% <2%), Ethiopia, Rwanda, and Cameroon (medium HIV prevalence stratum) and Malawi, Zambia, Zimbabwe, and Swaziland (high HIV prevalence stratum: HIV% > 10%). Estimates are adjusted for individual age, and the survey cluster percentage of male circumcision, median age at first marriage (by sex), median age at first sex (by sex), the percentage self-reporting sexually transmitted infection symptoms (by sex), the percentage of married individuals reporting sex with someone else than the spouse in the last 12 months (by sex), and the urban/rural character of the survey cluster. The cluster-level prevalence of polygyny is defined as the average of the percentage of married men and the percentage of married women who are in a polygynous union. CI, confidence interval.These results invoke the obvious question whether or not the elevated prevalence among junior wives (as well as other, more subtle, variations across HIV prevalence strata) is the result of a concurrency effect, but that is unfortunately difficult to establish with cross-sectional data. Preliminary analysis suggests, however, that the relatively high HIV prevalence among junior wives is at least partly driven by the selection of widows and divorcees into polygynous unions [3]. The maintenance of polygynous marriage systems via the rapid remarriage and disproportionate recruitment of divorcees and widows into polygynous unions has been described well before the advent of the large-scale HIV epidemic [4,5] (see also [6]), and it is particularly relevant today because HIV prevalence correlates with marriage order. These findings suggest possible directions to pursue in accounting for the positive association between polygyny and HIV at the individual level, but our understanding of the benign effect of polygyny at the ecological level also remains incomplete. In our paper, we highlighted the sex asymmetric sexual network structure characteristic of polygyny, and the reduction in coital frequency in conjugal dyads of polygynous unions (i.e., a coital dilution effect). In conjunction with the selection effect described above, the coital dilution effect ensures that the women who are most likely to transmit HIV to someone else have less intercourse, and that could explain both the individual-level and ecological-level associations. Many readers will realize that there are a number of other – sometimes countervailing – mechanisms that have not been given much consideration. First, it is plausible that the interval between infection and the formation of new (polygynous) marriages is longer than in informal partnerships. The remarriage of widows and divorcees into polygynous unions is, therefore, less likely to occur during the highly infectious window shortly after seroconversion. Second, the practice of polygyny might restrict young men's access to sex because older men monopolize the women in their community. Third, and this could negate the previous, we find that communities with polygyny are also characterized by lower ages at first marriage and a lower occurrence of premarital sex (unpublished results). Late marriage and a protracted interval between first sex and first marriage may contribute to the high HIV prevalence rates in many parts of south-eastern Africa [7]. An even more exhaustive treatment of sexual network and partnership attributes that mediate the relationship between HIV and polygyny should probably include the rate of partner acquisition and partnership dissolution, as well as socially regulated behaviors such as the duration of postpartum sexual abstinence, the age difference between spouses, etc. Some of these are intimately related to the practice of polygyny itself, and, to use Epstein and Stanton's words, culture could thus indeed serve as a confounder. We wish to argue, however, that it is largely irrelevant whether or not the polygyny effect is causal. This comment dovetails with responses to our article by Kretzchmar, White and Caraël [8], Boily [9], and Aral [10] who emphasize that concurrency is not an uncorrelated or exogenous characteristic of a sexual network, an insight that can be traced back to the early modeling work [11]. This is important because it suggests that even though one might try to isolate the effect of a single sexual network characteristic – such as partnership concurrency – in simulations, in practice it may never present itself in isolation from other network or partnership attributes that matter for the spread of an sexually transmitted infection. In conclusion, we emphasize that one needs to keep the implied reference group in mind when interpreting our results. In a population where informal concurrency is the norm, for example, – some may recognize similarities to Epstein and Stanton's description of southern Africa – polygyny as an institutionalized form of partnership concurrency may be the least conducive to spreading HIV. Reality is unlikely to be that extreme, but more empirical work is necessary to understand the prevalence of different manifestations of concurrency, how these vary across cultural and social contexts, and whether or not they correlate with other important network characteristics and sexual practices. By no means do we wish to claim that our study provides the definitive answer about the implications of polygyny for the spread of HIV, but at least until modeling and empirical research have advanced further, the conclusion, ad interim, is that polygyny is rather benign indeed. Acknowledgements We thank Ann Swidler and Jimi Adams for insightful discussions and comments on the manuscript.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.