The Debate over the Concurrency Hypothesis by Larry Sawers American University, Washington, DC
[email protected] and Eileen Stillwaggon Gettysburg College, Gettysburg, PA DRAFT: do not cite without authors’ permission November 28, 2011 Abstract This article is a review of the recent debate over the concurrency hypothesis—the proposition that overlapping sexual partnerships explain sub-Saharan Africa’s extraordinary HIV epidemics. Early efforts to model HIV epidemic dynamics and sexual networks substantially overstated the importance of concurrency. Recent models, including our own, show that concurrent sexual partners cannot explain contemporary HIV epidemics in eastern and southern Africa or track the early dramatic rise in HIV prevalence during the 1980s and 1990s. Survey data, including from recent surveys, show that the prevalence of concurrency is too low to explain the growth of HIV epidemics. The prevalence of concurrency and HIV do not correlate within or across countries or regions. Recent attempts to demonstrate such a correlation using the share of outside infections among total seroconversions in stable couples draws inferences that are not valid. Participants in the recent debate over concurrency disagree over the importance of coinfections in HIV epidemics in sub-Saharan Africa. Twenty-one recently published articles address the issue of concurrency. None them provides evidence for the concurrency hypothesis and many of them seriously erode the already weak case for the hypothesis. That finding has important implications for future research and HIV-prevention and treatment policy.
Electronic copy available at: http://ssrn.com/abstract=1951029
2
This article addresses the key issues that have emerged in recent contributions to the debate over the role of concurrent sexual partnering in HIV epidemics in sub-Saharan Africa [1-21]. For more than two decades, the conventional explanation for subSaharan Africa’s extraordinarily high HIV prevalence has centered on presumed differences in sexual behavior between Africa and elsewhere. Survey research, however, has found that most forms of sexual behavior that are associated with higher risk of HIV are no more common, or are less common, in sub-Saharan Africa than in numerous countries with far lower HIV prevalence [22-26]. Those behaviors include early age of sexual debut, high prevalence of pre- or extra-marital sex, having many lifetime or previous-year partners, having any partner, or patronage of commercial sex workers. In responding to that evidence, supporters of the sexualbehavior explanation for the region’s HIV epidemics narrowed their focus to a single form of sexual behavior—overlapping sexual partners or concurrency. The assertion that the high prevalence of concurrency is the principal explanation for subSaharan Africa’s hyper-epidemics of HIV is called the concurrency hypothesis, which is the focus of this article. The empirical evidence and argumentation marshaled in support of the concurrency hypothesis have been examined closely and found deficient, first by Lurie and Rosenthal [27,28] in 2009 and 2010 and subsequently by our own critiques [5,6,29]. The objective of the present article is to examine contributions to the debate over the concurrency hypothesis that have appeared since our systematic review in September 2010 [6]. We focus on four critical issues raised in the recent debate over concurrency: models of HIV epidemic dynamics, measures of the prevalence of concurrency, analysis of the correlation between HIV and concurrency, and clarification of the role of coinfections. The article ends with a discussion of the research agenda and HIVprevention policy that flows from the analysis presented in this article.
Modeling does not support the concurrency hypothesis For the concurrency hypothesis to be valid, concurrency must spread HIV more effectively than multiple partnering in general. Since mathematical models are used to demonstrate that
Electronic copy available at: http://ssrn.com/abstract=1951029
3
proposition, much of the controversy over the concurrency hypothesis has centered on modeling HIV epidemics. Realistic model parameterization leads to rejection of concurrency hypothesis
The model published by Eaton, Hallett, and Garnett [9] in September 2010 is an important contribution to the literature and plays a key role in showing that the concurrency hypothesis is not correct. Eaton and colleagues adapt an early, influential model created by Morris and Kretzschmar [30] by using staged transmission rates (different transmission rates during each stage of infection), much lower and more realistic average transmission rates, and vital dynamics, that is births and deaths, which allows modeling epidemics for extended periods. Eaton et al. conclude that “with staged transmission and up to 8% of individuals having concurrent partnerships, HIV fails to spread” [9]. In their model, when point prevalence of concurrency, that is, measured at a point in time for men and women together, is even slightly less than 8 percent, simulated epidemics of HIV fall from the initial 1 percent prevalence toward zero prevalence, that is, their basic reproduction number is less than 1 and they move to extinction. No nationally representative survey in a sub-Saharan African country using currently accepted measures of concurrency has found point prevalence as high as Eaton et al.’s 8 percent. As we discuss below, the Demographic and Health Surveys (DHS) in Lesotho and Malawi are the only national surveys that use the UNAIDS recommended questionnaire design to measure concurrency. Those surveys report point prevalence of concurrency to be 5.1 percent in Lesotho (our calculation from DHS datasets) and 2.0 percent in Malawi [31]. Other recent nationally representative surveys in sub-Saharan Africa find similar results. In other words, Eaton et al. show that concurrency is not effective in spreading HIV at the rates of concurrency that prevail in subSaharan Africa, even if one assumes that women substantially under-report their concurrency. Eaton et al.’s model shows that at the highest level of concurrency they considered (point prevalence of 14 percent, meaning that two-thirds of partnerships overlap), HIV prevalence rises in the first ten years of their simulations from 1 percent to 1.4 percent. In some eastern and southern African countries, HIV prevalence grew from less than 1 percent to over 10 percent (and in a few countries, over 20 percent) during the 1990s. Thus, Eaton et al.’s model of sexual networks shows that the exponential growth in HIV prevalence in the early years of the epidemic in eastern and southern Africa cannot be explained by sexual behavior alone,
4
even when assuming that rates of concurrency were far higher in the 1990s than at present. Eaton et al.’s model actually overstates the importance of concurrency. The average partnership duration in the model is only 200 days (following Morris and Kretzschmar [30]). High partnership turnover leads to the rapid spread of HIV [27]. Evidence shows that average partnership duration in sub-Saharan Africa is much longer than 200 days [3,18,32,33]. Second, Eaton et al. themselves point out that their transmission risk could have been confounded by coinfections that promote the transmission of HIV. Boily et al. [7] agree, concluding that “the role of concurrency in Africa may have been overestimated because of the high prevalence of HIV cofactors.” If coinfections produced an upward bias in Eaton et al.’s measures of transmission risk, then their simulations represent the combined impact of concurrency and coinfections, not the result of concurrency per se, that is, their results overstate the importance of concurrency. Moreover, Eaton et al. assume gender symmetry in rates of concurrency and transmission risks, leading to upward biases in the effect of concurrency on simulated HIV epidemics [14]. By far the most important way that Eaton et al.’s parameterization exaggerates the importance of concurrency is by not incorporating coital dilution. Both Morris and Kretzschmar [30] and Eaton et al. [9] assume that adding a second partner doubles one’s coital frequency. A third partner triples one’s sexual activity, and so on. Both empirical evidence and common sense indicate that the assumption is incorrect [5,13]. In reality, two, three, or four partnerships may allow an individual to have more sex than a single partnership, but not often two, three, or four times as much. Sawers, Isaacs, and Stillwaggon [5] modified Eaton et al.’s model by incorporating coital dilution (lower average coital frequencies in secondary partnerships). Simulating Eaton et al.’s model with the level of coital dilution that Morris et al. report in Rakai, Uganda [3] produces HIV epidemics that move rapidly from the initial 1 percent HIV prevalence toward zero prevalence at every level of concurrency. Indeed, they move to extinction more rapidly at higher levels of concurrency than at lower levels or serial monogamy, that is, concurrency can be protective against HIV. Even with much higher coital frequencies in secondary partnerships than reported by Morris et al., concurrency cannot sustain HIV epidemics. In sum, recent modeling by Eaton et al. and Sawers et al. demonstrates the fallacy of the concurrency hypothesis.
5
Those models also shed light on Epstein and Morris’s argument that coital frequency and transmission risk during acute infection are “high enough for HIV to propagate via concurrency” [1]. They say the acute-infection transmission risk could be as much as 3.6 percent (citing Pinkerton [34]), which is high enough to produce a substantial number of seroconversions in initially discordant couples during the acute-infection window. Nevertheless, epidemics are composed of multiple chains of transmission and cannot be understood by looking at only the first link in one chain. Before one can determine the ability of concurrency to accelerate the spread HIV, one needs to know whether the partner of the index partner has a partner and, if so, how frequently they have sexual contact. One also needs to know how many of the partners’ partners have partners and the coital frequency in those partnerships, and so on. The appropriate methodology for exploring the epidemic spread of HIV takes into account every link in every chain of transmission and incorporates information about all partnerships (whether overlapping or not) including their duration and coital frequencies. That methodology is the kind of modeling discussed above, and it shows that HIV epidemics are unsustainable at levels of concurrency found in actual countries (Eaton et al. [9]) or at any level of concurrency (Sawers et al. [5]). Epstein and Morris’s discursion on acute infection-transmission risk and coital frequency only confuses the discussion. Eaton et al. [9] show that modeling staged transmission rates in place of an unstaged transmission rate (that is the weighted average of the staged rates) actually slows the simulated spread of HIV by a substantial degree and results in lower endemic HIV prevalence at all levels of concurrency. Eaton et al.’s results show that with unstaged transmission risks (that is, with no special allowance for acute infection) and point prevalence of concurrency less than 8 percent, HIV prevalence eventually exceeds 15 percent. In contrast, with staged transmission risks and concurrency less than 8 percent (higher than found in any country using proper survey instruments), HIV prevalence approaches 0 percent as the epidemics move to extinction. That erodes the rationale for the concurrency hypothesis that has been repeated over and over again since 2004 [35-37], and is rehearsed in the recent debate [1,2,9]. In discussing the results of their modeling, Eaton et al. argue that using staged rather than unstaged transmission rates enhances the importance of concurrency because HIV epidemics diverge more sharply with staged rather than unstaged transmission risks. Nevertheless, all of that divergence is at levels of concurrency irrelevant to an understanding of any actual country’s epidemic. In their Figure 1 (b) [9], the graph lines for the epidemics with point
6
prevalence of concurrency below 8 percent lie almost on top of each other. Their modeling shows that in the real world, the interaction between concurrency and acute infection has almost no impact on the spread of HIV. Early models overstated the impact of concurrency on HIV
Epstein and Morris assert that the early model of Morris and Kretzschmar produces “an underestimate, not an overestimate, of the effect of concurrency” [1], and Goodreau et al. make the same assertion [10]. Morris and Kretzschmar's model played a pivotal role in launching the concurrency hypothesis [30,32,38]. Their model—to the exclusion of all others—was repeatedly cited by the most prominent proponents of the concurrency hypothesis [35,36,39,40], who were in turn cited in several hundred publications. Morris and Kretzschmar’s model is foundational. Eaton et al.’s and Sawers et al.’s models are modified versions of their original model. Morris and Kretzschmar’s work still frames the concurrency debate and the realism of their simulations continues to be an important and contested issue. Epstein and Morris say that adding vital dynamics to Morris and Kretzschmar’s model would greatly increase “the estimated impact of concurrency because in the ‘serial monogamy’ scenario—but not in the concurrency scenario—most infected individuals die before they can infect at least one other person” [1]. That is not correct since the average partnership (whether concurrent or not) in their model lasts only 200 days and new partnerships quickly form after old ones dissolve. Furthermore, the model’s daily transmission risk is so high that transmission occurs in almost every discordant couple whether concurrent or sequentially monogamous. In the model, when originally discordant partnerships dissolve, both partners are likely to be infected and thus both are likely to spread the infection to their next partner. That is true whether the couples are serially monogamous or concurrent. The reason that vital dynamics increases HIV prevalence is that the death of a partner allows the surviving partner to form a new partnership sooner than if their partner had not died. Increasing the rate of partner turnover for any reason accelerates the spread of HIV. Eaton et al. modified Morris and Kretzschmar’s model in several ways, but we explore the impact of reversing all of those changes except for the incorporation of vital dynamics. We find that Morris and Kretzschmar’s model with vital dynamics included generates HIV prevalence after five years (the length of Morris and Kretzschmar’s original simulations) that is about 25 to 40 percent
7
higher—depending on the level of concurrency—than without vital dynamics. Thus, adding vital dynamics to Morris and Kretzschmar’s model does accelerate the spread of HIV. Nevertheless, when we continue the simulations for six more years, we find that HIV prevalence reaches 99 percent at every level of concurrency including serial monogamy, a result that is obviously unrealistic. The reason why Morris and Kretzschmar’s model is so unrealistic is not because of the failure to include vital dynamics, but because of their high daily transmission risk coupled with rapid partner turnover. Their 5 percent daily transmission risk produces a 99.996 percent chance of transmission in serodiscordant partnerships that last 200 days, which is the average partnership duration in their model. After a partnership dissolves, the former partners quickly find new partners (in 100 days on average), so that after five years, the average individual in their model has had six partners. In Morris and Kretzschmar’s model, the virus is trapped in neither serially monogamous nor concurrent partnerships due to the short duration of each partnership and the brief interlude between partnerships. To show just how much Morris and Kretzschmar’s 5 percent daily transmission risk exaggerates the importance of concurrency, we exchange their transmission risk for the rate used by Eaton et al., which is based on the calculations of Hollingsworth et al. [41], who rework Wawer et al.’s data [42] from a study in Rakai in the 1990s. Morris and Kretzschmar’s 0.05 daily transmission risk is 89 times the size of Eaton et al.’s unstaged daily transmission risk of 0.00056 (which is the weighted average of the staged transmission risks). The smaller transmission risk produces only a 10.6 percent chance of transmission in 200 days, not the nearly 100 percent chance of transmission that Morris and Kretzschmar’s assumption produces. Even though Eaton et al. assume the same unrealistically rapid partner turnover that Morris and Kretzschmar do, the lower daily transmission risk produces a dramatically slower growth path of HIV prevalence. In five years of simulations, HIV prevalence rises from the initial 0.05 percent to 0.06 percent for all levels of concurrency and for serial monogamy. In eleven years, HIV prevalence rises to either 0.07 or 0.08 percent depending on the level of concurrency, not the 99 percent that Morris and Kretzschmar’s transmission rate produces. Morris and Kretzschmar report that when concurrency is 12 percent, HIV prevalence after five years of simulations is 45 percent, which “is 10 times as large as under sequential monogamy” ([30] from the abstract). In contrast, their model—but with Eaton et al.’s lower transmission risk and concurrency still at
8
12 percent—takes over a century for HIV prevalence to rise from the initial 0.05 percent prevalence to 3 percent (not 45 percent), at which point it is only three times (not ten times) as large as under serial monogamy. If one follows Eaton et al.’s lead and increases the realism of the model one step further by using different transmission rates for different stages of HIV infection, the simulated epidemics grow even more slowly or move to extinction depending on the level of concurrency. The parameter values that Morris and Kretzschmar use produce an exponential spread of HIV, quickly infecting nearly 100 percent of the population and making concurrency irrelevant. If they had used a more realistic transmission risk, however, their model would take more than 60 years to generate HIV prevalence as high as 1 percent even at the highest considered level of concurrency (14 percent point prevalence) and even including vital dynamics that accelerate the spread of HIV. Epstein and Morris’s talk of vital dynamics is thus a distraction. As has been repeatedly stated by critics of the concurrency hypothesis [6,27,28], the reason for Morris and Kretzschmar’s unrealistic results is their transmission risk, an issue that Epstein, Morris, and Kretzschmar have never effectively addressed. It is inappropriate to compare concurrency only with serial monogamy
Embedded in the discourse over concurrency during the last two decades and continuing in recent contributions to the debate is a default counterfactual—serial monogamy—to which concurrency is almost always explicitly or implicitly compared. Of course, there are only two kinds of multiple partnering—with and without overlapping partnerships—so the dichotomous contrast is analytically useful. Nevertheless, in no country does multiple partnering take only the form of serial monogamy, nor in any country is point prevalence of concurrency as high as 8 percent. The comparison of high levels of concurrency with zero concurrency (serial monogamy) leads to an exaggerated perception of concurrency’s importance. Instead, using a sexual network model to compare a country like Lesotho (with 5.1 percent point prevalence of concurrency) to the United States (with 3.5 percent point prevalence [3]) generates only a trivial difference in simulated HIV prevalence, not the forty-fold difference in actual HIV prevalence. Other recent models do not support the concurrency hypothesis
We have already discussed two recent models (Eaton et al. [9] and Sawers et al. [5]) that invalidate the concurrency hypothesis. We now examine other recent models (published since 2009) that
9
various contributors to the recent debate believe confirm the importance of concurrency. As the discussion has made clear so far, parameterization that assumes, for example, no coital dilution, excessive levels of concurrency or daily transmission risks, and too rapid partnership turnover will generate unrealistic levels of HIV prevalence. Much of what follows, therefore, is a critique of the unrealistic parameter values used in recent models. Leclerc et al. (2009) [43] state in their abstract that “all parameters were derived from empirical population-based data. Results show that basic parameters could not explain the dynamics of the HIV epidemic in Zambia.” The transmission risk in the model was based on Wawer et al. [42] and commercial sex patronage by men was determined from the Zambian DHS. Only when those parameter values were increased to empirically unsupported levels could the modellers track the actual epidemic in Zambia. Eaton et al. (2010) [9], in a supplement to their article discussed earlier, present simulations using parameter values based on a study in one district in Zimbabwe ([44] page 1897). Because the model’s parameterization in the article supplement is not based on a nationally representative sample of the country’s population, the simulations shed little or no light on national or regional HIV epidemics. Furthermore, the Zimbabwe study measured concurrency using a question that is now widely understood to overstate concurrency [45] and thus the modeling presented in the supplement exaggerates the role of concurrency. Morris et al.(2009) [46] try to use concurrency to explain racial differences in HIV prevalence in the United States and thus their study is indirectly relevant to the concurrency hypothesis. They find that when compared to serial monogamy, concurrency produces a larger “epidemic potential”, that is, a larger number of individuals who are potentially at risk of HIV under their assumption that every sero-discordant partnership transmits the virus upon partnership formation. The authors consider such individuals to be in the “reachable path” of the initial infection. Real-world HIV transmission risks in the absence of coinfections, however, are so low that many chains of transmission, perhaps almost all of them, quickly break off. Thus, Morris et al.’s “reachable path” is a misnomer since it does not refer to an outcome that is possible in an actual HIV epidemic. The potential of any actual epidemic to grow depends on transmission risk, so assuming a 100 percent sexual transmission risk tells us nothing about epidemic potential in the real world. Johnson et al.(2009) [47] divide the population into two groups based on different propensities for high-risk sex, which
10
they define as commercial sex and concurrency. They say men’s concurrency rates are 12 to 18 percent and women’s are 1 to 3 percent, citing Shisana et al. [48]. Johnson et al. ignore the concurrency rates from Shisana et al. because they assume respondents do not tell the truth, and assign 35 percent of men and an astonishing 25 percent of women to the risky sex group. Even so, the concurrency rates reported by Shisana et al. likely overstate actual concurrency. The study reports rates of multiple partnering that are far lower than rates of concurrency, so the figures cannot both be correct. Concurrency is a subset of multiple partnering and therefore must be smaller, perhaps much smaller than rates of multiple partnering. Furthermore, the only data on concurrency in Shisana et al. are only for those aged 15 to 24, not for all adults. The study does not report the question or questions used to measure concurrency, so there is no way to judge the reliability of the reported concurrency data. Johnson et al. also assume, on the basis of two articles published in 1990 (when HIV transmission was poorly understood), that per-act transmission rates are five or six times higher (depending on the gender of the index partner) in nonspousal partnerships than in spousal partnerships. We have not found any publication since 1990 other than Johnson et al. that asserts a causal relationship between per-act transmission rates and the number of sex acts. (See [5] Additional File 1 for an extended discussion of this issue.) Not surprisingly, they conclude on the basis of the model’s results that most HIV transmission in South Africa occurs in non-spousal partnerships and that concurrency accounts for “roughly three quarters of new HIV infections” in South Africa. Goodreau, Morris, and colleagues (2010) [10] developed a model that uses parameter values calibrated from a Zimbabwean study that recruited rather than sampled participants who were all 18–30 years old and thus the study was neither nationally representative nor a study of all adults. They say that, “the [point] prevalence of concurrency in this population [7.3 percent] is relatively high, compared to levels observed in the United States,” but they do not note that the rates of concurrency they use in their model are also high compared with those found in Zimbabwe itself and in other countries in sub-Saharan Africa. A 2008 national survey in Zimbabwe reports 5.3 percent point prevalence of concurrency among married or cohabiting adults [49] and nationally representative surveys in Zambia, Lesotho, and Malawi find even lower point prevalence of concurrency [31,50]. Their model may have some relevance to the part of Zimbabwe they studied, but cannot be generalized to the country or region as a whole.
11
Despite their use of an excessively high concurrency rate, Goodreau et al. find that “the epidemic in Zimbabwe is very close to the persistence threshold—small changes in either behavior [concurrency] or infectivity [transmission risk] may be enough to push it into eventual extinction” [10]. Although finding the HIV epidemic in Zimbabwe teetering on the brink of extinction, Goodreau et al. say “as one moves from the early ‘proof of concept’ models to models that are more realistically parameterized, the effects of concurrency become larger, not smaller.” Morris and Kretzschmar’s proof of concept to which Goodreau et al. referred found epidemics taking off like rockets, not verging on extinction. The effects of concurrency become much smaller or vanish completely with more realistic parameter values. Delva (2010) [11] argues that it is difficult to track the rapid spread of HIV in South Africa in the 1990s with a model that includes serial monogamy but not concurrency. His modeling is problematic for several important reasons. First, Delva does not include concurrency in his model. He only argues that since serial monogamy cannot produce a sufficiently rapid growth of HIV, concurrency must be the explanation. Even if there were no other problems with his model, that research strategy can provide only the weakest support for his hypothesis. Second, Delva assumes HIV can only be transmitted sexually, ruling out iatrogenic or other non-sexual transmission. If Delva’s objective were to explore exclusively the sexual spread of HIV, that could be an appropriate assumption. But Delva’s objective is to build a model that can track an actual epidemic in South Africa in the early 1990s, which doubtlessly had a non-trivial amount of non-sexual transmission at the time. Third, as in most models of HIV and sexual network dynamics, Delva seeds his model only once with “a small cohort of recently infected individuals” (page 107 in [51]). Nevertheless, there is considerable evidence that the epidemic entered South Africa in multiple ways on multiple occasions, as infected contract workers, immigrants, returning travelers or emigrants, traders, tourists, and other visitors poured across the country’s borders [5261]. (See [29] for additional citations.) The epidemic in South Africa was almost surely reseeded repeatedly by substantial numbers of cross-border travelers. Since Delva attempts to track that epidemic, he should have seeded his model repeatedly, and that would have accelerated the simulated growth of HIV prevalence. The usual practice of seeding only once is inappropriate given Delva’s research objective and the reality of South Africa’s epidemic. Delva also assumes a 0.0092 transmission rate during acute infection (with sex every 3.5 days over 91 days) that he says is
12
based on the Rakai data [42]. Two of the three articles he cites as evidence for his transmission rate use the raw data from Rakai to estimate much higher transmission risk than Delva assumes. Pinkerton [62] finds a 0.036 per-act transmission rate (with sex every 2.86 days over 49 days) and Hollingsworth et al. [41] estimate a daily transmission rate of .00732 (over 88 days). Delva’s transmission rate produces a 21 percent risk of transmission during acute infection, but Pinkerton et al.’s and Hollingsworth et al.’s produce a 47 or 48 percent risk. By ignoring non-sexual exposures, by failing to recognize the repeated waves of HIV-infected individuals bringing the disease to South Africa, and by underestimating the risk of transmission during acute infection by more than half, Delva has not effectively modeled the HIV epidemic in South Africa. Leung (2011) [21] tries to determine if polygyny protects against HIV. He presents a model using ordinary differential equations that contrasts the effects of STIs such as HIV on sexual networks with and without polygyny. The author concludes, “there is no simple yes or no answer to the question: ‘Does polygyny have a protective effect?’” ([21], page 77). The model does not allow for coital dilution, which is thought to be an important reason why polygyny is protective against HIV [13]. In sum, early models of HIV epidemics and sexual network dynamics produce wildly unrealistic results that greatly overstate the importance of concurrency. Moreover, recent models show that concurrency at levels prevailing in sub-Saharan Africa produce only epidemic extinction [5,9] or can avoid epidemic extinction only with unrealistic parameterization [10,11,43,46]. In sum, recent modeling only reinforces the conclusion that concurrency cannot be an important driver of the HIV epidemics in eastern and southern Africa.
Concurrency is not especially prevalent in subSaharan Africa For the concurrency hypothesis to be correct, concurrency must not only be more effective in spreading HIV. Concurrency must also be far more prevalent in sub-Saharan Africa than elsewhere, but only two of the participants in the recent debate address the issue. Mah and Shelton argue that “strictly comparable behavioral data on concurrency are just not available on a wide cross-national basis” [2], leaving them unable to address one of the two essential propositions underlying the concurrency hypothesis. In contrast, Epstein and Morris discuss rates of concurrency at some length [1].
13
We have argued [6] that in order to find empirical support for the concurrency hypothesis, the appropriate evidence about prevalence of concurrency must come from country-wide representative surveys of all adults, not from surveys in which respondents are recruited instead of sampled, not from narrow age brackets, and not from surveys of towns, districts, neighborhoods or other geographic areas not representative of the country as a whole. The concurrency hypothesis is about the many countries in sub-Saharan Africa that have dramatically higher HIV prevalence than any other country, so concurrency and HIV prevalence must be measured at the country level. Recent surveys show that rates of concurrency in subSaharan Africa are low
Since most country-level surveys of sexual behavior in subSaharan Africa have been carried out by the Demographic and Health Surveys (DHS), Epstein and Morris properly focus their attention on the DHS. After a lengthy discussion of the problems with questionnaire design in previous DHS surveys, Epstein and Morris say that the “problems [with earlier DHS surveys] appear to have been fixed in 2009 [in] the DHS from Lesotho that uses the corrected questionnaire module” ([1] page 3; see also [63]). The official report of the 2009 Lesotho DHS survey [64] does not report concurrency rates, so we obtained from Measure DHS – IFC Macro the datasets on which the report is based. Those data show that point prevalence of concurrency was 7.8 percent for men and 2.3 percent for women, averaging 5.1 percent. Past surveys using flawed questionnaires [45] found concurrency in Lesotho to be the highest of any country ever surveyed [25]. The 2010 Malawi DHS [31] also uses the methodology for measuring concurrency approved by UNAIDS [45] and Epstein and Morris. Point prevalence of concurrency there was 3.8 percent for men and 0.1 percent for women, averaging 2.0 percent. Two other recent nationally representative surveys report point prevalence of concurrency close to the level found in Lesotho. One is a 2008 study in Zimbabwe that found point prevalence of concurrency among married or cohabiting adults to be 5.3 percent [49], and the other is a 2003 study in Zambia reporting point prevalence of concurrency among those aged 15–49 to be 4.2 percent [50]. One reason why the recent DHS are so important is that we now have an accurate measure of point prevalence of concurrency (having two or more partners at a point in time) that allows us to match survey data with models of HIV epidemic dynamics. To designate different levels of concurrency, modelers use point prevalence of concurrency. Some surveys in sub-Saharan Africa
14
report point prevalence but measured with a questionnaire design that is now understood to exaggerate the true level of concurrency [45]. Most surveys in the region, however, report concurrency over the previous year. In the Lesotho and Malawi DHS and other surveys [20] (though not in [19]), previous-year concurrency is roughly double the point prevalence. Furthermore, numerous studies report concurrency as a percent of sexually active or sexually experienced respondents rather than for all adults. Those who are familiar with survey data on rates of concurrency are therefore accustomed to seeing numbers much larger than the point prevalences reported in the recent DHS from Lesotho and Malawi. The lack of correspondence in the ways concurrency has been measured and modeled has made it difficult for both producers and consumers of models to understand their lack of realism. Another reason why the Lesotho 2009 DHS is so important is because it suggests the approximate size of the errors produced by questionnaire design in earlier DHS surveys in other countries. In 2009 in Lesotho, men’s one-year concurrency was 16.1 percent, not much higher than the 14.1 percent reported in an earlier DHS. (The DHS did not measure women’s concurrency until 2009.) One cannot not know the actual size of the error in each country’s pre2009 DHS survey until the new questionnaire module is used more widely, but the results of the Lesotho 2009 DHS suggest that rates of concurrency found in other DHS surveys in sub-Saharan Africa are underestimated, but that the errors will turn out to be small. The most comprehensive source of concurrency data in sub-Saharan Africa at the country level is Mishra and BignamiVan Assche’s compilation of DHS surveys [65]. In that study, average one-year concurrency in the 14 DHS surveys in subSaharan Africa (12 surveys for women) was 8.5 percent for men and 0.8 percent for women. If those numbers are too low by the same amount as in Lesotho, then one-year concurrency in the region would average approximately 10 and 1 percent for men and women respectively. In the US, either 11 percent [66] or 13.0 percent of men [3] and 6.1 percent of women [3] report previousyear concurrency. In six European countries, 10.0 percent of men and 3.1 percent of women (unweighted country averages) report being in a “steady relationship” of at least a year’s duration and having at least one other partner during the previous year [67]. (If the European surveys had measured concurrency using the same questions as the DHS in sub-Saharan Africa, they would have found higher levels of concurrency since the European surveys asked about concurrency among those in regular partnerships of at least a year’s duration, not about overlap of any length.) Thus, the DHS in Lesotho and Malawi suggest that concurrency in sub-
15
Saharan Africa is lower than in the US and probably much lower than in Europe. In sum, the concurrency hypothesis requires concurrency to be more prevalent in sub-Saharan Africa than elsewhere, but the measurement of concurrency has posed perplexing difficulties. The recent surveys from the DHS in Lesotho and Malawi surmount those measurement problems. Recent modeling of HIV and sexual networks [9] shows that, at levels of concurrency far higher than has been found in Lesotho, Malawi, and other countries in the region, epidemics of HIV are unsustainable, that is, they move to extinction. Moreover, levels of concurrency in sub-Saharan Africa are lower than in the US and in Europe. Since concurrency is not especially high in sub-Saharan Africa, the concurrency hypothesis cannot be correct. Qualitative studies do not show that surveys understate concurrency
Recent surveys find rates of concurrency in sub-Saharan Africa so low that they disprove the concurrency hypothesis, but Epstein and Morris [1] say we should not believe those surveys. They assert, “qualitative studies of small population samples consistently find that respondents report engaging in concurrent partnerships at rates that are often many times higher than in behavioral surveys” [1]. To support their assertion about qualitative research, Epstein and Morris provide seven citations (one of which is to a survey, that is, quantitative research, [68] reporting that point prevalence of concurrency was 4.2 percent of adults in Botswana [6]). Another study [69] reports on interviews with participants younger than 30 years who all had concurrent partnerships. One of them said that it was “normal” among his friends and two others said it was “common” to find young women with many partners, though it was not clear if they meant overlapping partners. None of the many other respondents said anything about the prevalence of concurrency. Two other cited studies did not provide any information about the prevalence of concurrency [70,71]. Three additional citations are to studies from the 1970s and 1980s that cannot tell us anything about what HIV-prevention policy should be three or four decades later [72-74]. People recruited to talk about sex in a group of friends and neighbors do not produce reliable data on rates of concurrency in the population at large and within their own community. Moreover, none of the six qualitative studies cited by Epstein and Morris reports any rates of concurrency and thus could not have
16
reported rates of concurrency that are “many times higher than in behavioral surveys.” Social desirability bias in both focus groups and surveys can produce estimates of sexual behavior prevalence that are too high (or too low) if respondents are boastful (or embarrassed) about their behavior. As is clear from the sources cited by Epstein and Morris, there is no evidence that reporting errors in surveys are so large that true rates of concurrency are “many times higher” than reported rates. Even if they were, those sources also make it clear that reporting errors are found in surveys everywhere, not just in sub-Saharan Africa [51,62]. Thus any downward bias in women’s reported concurrency or upward bias in men’s would not affect comparisons of concurrency between sub-Saharan Africa and other regions. An arguable position would be to say that neither surveys nor qualitative studies are reliable. Mah and Shelton appear to accepting that position (and thereby come close to abandoning the concurrency hypothesis). The only other participants in the recent debate to raise the issue of rates of concurrency are Epstein and Morris. Their assertion that nationally representative surveys in sub-Saharan report high levels of concurrency is not correct. Their assertion that qualitative studies show that those surveys substantially underreport concurrency is not supported by the sources they cite as evidence.
Still no evidence of correlation between concurrency and HIV The concurrency hypothesis is an assertion that there is a correlation between HIV prevalence and concurrency at the population level, but among sub-Saharan African countries or among all countries, there is no such correlation. A recent study by Morris et al. [3] says there is an “alignment” in concurrency and HIV prevalence among Thailand, Uganda, and the United States. Nevertheless, “alignment” is a word without statistical meaning. The authors’ repeated use of the word suggests a relationship that the evidence cannot support. The study, moreover, compares data from low-income men in two of Thailand’s 75 provinces, from Thai truckers, from only one of Uganda’s 35 districts at the time of the survey, and from a national survey in the United States. Data from such disparate sources, only one of which is a nationally representative survey, show almost nothing. Our comments attached to their published article explain other reasons why their study cannot even demonstrate
17
“alignment.” At the population level, as the prevalence of concurrency increases, some individuals have more partnerships and others have fewer partnerships. That could actually lower the population-level risk of HIV if coital frequencies are lower in secondary partnerships than in primary partnerships [5]. Reniers and Tfaily [13] show that at the population level, HIV prevalence and polygyny are inversely correlated. Many concurrent partnerships, even though not sanctioned by law, religion or custom, function as though they were formal polygynous unions, so Reniers and Tfaily’s findings are relevant to a broader discussion of concurrency. Researchers have also tried to determine if concurrency raises individual risk of HIV infection. Even if such a link could be established, it would have no direct bearing on the concurrency hypothesis, which is about the concurrency-HIV link at the population level, not the individual level. An individual’s risk of HIV acquisition from having a concurrent partner is the same as from having a non-concurrent partner [3,75]. Not surprisingly, surveys (for a recent example, see [15]) do not find a correlation between own concurrency when properly measured and own HIV infection. Concurrency, if it has any effect, does not raise the risk of HIV acquisition of the individual who has concurrent partners, but raises the risk to those partners. Previous studies have not found convincing statistical evidence for that hypothesized correlation. Two recent contributions to the concurrency debate also find no correlation between HIV and partner’s or neighbors’ concurrency. Tanser et al.’s [12] study finds no correlation between female HIV incidence and the prevalence of male concurrency in the woman’s neighborhood (but does find a strong correlation with number of partners of men in the neighborhood). Maher et al. find no correlation between men reporting concurrency (measured using the UNAIDS-approved method [45]) and HIV prevalence among their wives [19]. Boily et al. present an extended discussion of the difficulties in establishing the HIV-concurrency correlation [7]. The most obvious reason for the failure to find a correlation— given the evidence presented in the first two sections of this paper—is that there is no important connection between concurrency and HIV, a possibility that Boily et al. acknowledge. Several recent works offer what their authors present as a new way to measure the link at the individual level between HIV and concurrency [1-3,17]. Epstein and Morris claim that “whether new infections arise from inside or outside the couple” can show whether “concurrency is a key driver of HIV epidemics in generalized epidemics in Africa [1].” They say that the share of
18
incident infections in stable couples that come from outside the partnership is 60 to 84 percent in sub-Saharan Africa, which they take as confirmation of the importance of concurrency. The share of incident infections in stable couples that comes from outside the partnership (which we call the outside-infections share), however, cannot demonstrate a correlation between HIV and concurrency or tell us anything about the validity of the concurrency hypothesis for the following six reasons. 1. Information about the outside-infections share pertains to individual risk and has no direct relevance to the concurrency hypothesis, which is an assertion about concurrency and HIV prevalence at the population level. Furthermore, providing information about the outside-infections share only for subSaharan Africa cannot even shed light indirectly on the concurrency hypothesis since the hypothesis is inherently comparative. To make their case, those who stress the importance of the outside-infections share must measure that share in other countries and show that it has something to do with explaining differences in HIV prevalence among countries. A recent study in India shows just how difficult that task will be. It argues that the driving force for the HIV epidemic there is men who bring the infection into stable couples via concurrency. Nevertheless, HIV prevalence in India (0.3 percent of adults in 2009 ) is far lower than in sub-Saharan Africa [16,76] even though the outsideinfections share is roughly similar in the two regions. 2. The outside-infections share in stable discordant couples depends upon the HIV transmission rate within discordant couples. If the intra-discordant-couple transmission rate falls (for example, because couples counseling reduces transmission), then the share of incident infections in discordant couples that comes from outside the couple must rise, other things equal. Thus, the outsideinfections share can move independently of any change in concurrency, that is, it does not necessarily tell us anything about the link between concurrency and HIV. Celum et al.’s study of discordant stable couples [77] illustrates the point. They used genetic sequencing to show that 29 percent (38 out of 132) of incident infections in initially discordant couples came from outside the couple. Both partners in all 3360 couples were counseled on how to prevent HIV transmission and given condoms every three months. Both partners were treated regularly for sexually transmitted infections. From the outset of the study, all infected participants who met national guidelines for the initiation of anti-retroviral therapy (ART) were referred to local HIV clinics. (ART inhibits HIV transmission [78].) Transmission occurred between partners in 91 (or 2.7 percent) of the 3360
19
couples in the two-year study. Without those efforts to reduce transmission, we estimate that transmission between partners would have occurred in 19.1 percent of the couples. That percentage is based on the daily transmission risk in initially discordant stable couples who report no outside partners from Hollingsworth et al. [41], who use data that predate ART. (We use Hollingsworth et al.’s asymptomatic-infection transmission risk because the Celum et al. study recruited couples in a way that appears to have ruled out index partners who were in acute infection.) With incident infections in 19.1 percent of couples (641 out of 3360) over a period of two years and the same number of outside infections that Celum et al. reported, those outside infections would have been only 6 percent of total transmissions (38 out of 679), not 29 percent (38 out of 132). Thus, Celum et al.’s successful efforts to reduce intra-couple transmission produced a sharp increase in the outside-infections share in discordant couples independent of any changes in concurrency. 3. As noted in item 2., a fall in transmission within discordant couples will decrease total incident infections among all discordant couples. Reducing the number of transmissions in discordant couples means that incident infections in concordant negative couples will become a larger share of incident infections among all stable couples (which is the sum of discordant couples and concordant negative couples). Since all incident infections in concordant negative couples come from outside the couple, that would raise the outside-infections share among all stable couples. Again, the outside-infections share can change without any change in concurrency. 4. Epstein and Morris’s estimate of the outside-infections share among all stable couples in sub-Saharan Africa (60 to 84 percent) is much too high since it assumes that the outsideinfections share among discordant couples in sub-Saharan Africa is 29 percent. Unlike the couples in Celum et al.’s study, only a fraction of discordant couples in sub-Saharan Africa receive couples counseling or ART. Thus, the transmission rate within discordant couples in the region is much higher than among participants in Celum et al.’s study. For the reasons given in item 2., higher transmission within discordant couples will reduce the outside-infections share among discordant couples. For the reason given in item 3., that will also reduce the outside-infections share for all stable couples. Since Epstein and Morris chose an outsideinfections share among discordant couples that is too high, they have overestimated the outside-infections share for all stable couples for both reasons.
20
5. A change in the outside-infections share can be caused by events other than changes in transmission rates in sexual exposures or changes in patterns of concurrency. The outsideinfections share could rise because of an increase in non-sexual transmission from, for example, a scale-up in male circumcision without sufficient precautions to prevent iatrogenic transmission. It could also change from the maturing of an epidemic. In the early stages of an epidemic, most incident infections come from outside the couple, but as the epidemic matures, an increasing fraction will occur within stable couples even if rates of concurrency are stable. 6. The outside-infections share can tell us nothing about the level of concurrency or the ability of concurrency to enhance the spread of HIV. For the concurrency hypothesis to be valid, concurrency must be especially prevalent in sub-Saharan Africa and must be especially efficient in spreading HIV. The outsideinfections share gives us no information about either of the two essential propositions underlying the concurrency hypothesis. In sum, a country could have a high outside-infections share if it pursued a successful couples counseling program, scaled up ARV distribution, had an immature epidemic, and/or had a significant amount of non-sexual transmission, none of which has anything to do with concurrency. The inability to find a statistical correlation between HIV and concurrency undermines the concurrency hypothesis. The share of incident infections in stable couples that come from outside the couple cannot serve as a surrogate for the elusive correlation between HIV and concurrency that no one can find. Since the outside-infections share depends on many factors, it has no stable link to concurrency and cannot be used to support the concurrency hypothesis.
The importance of coinfections Several recent contributions to the debate over concurrency raise the issue of coinfections. If not properly controlled for, coinfections that enhance sexual transmission or acquisition of HIV can bias estimates of transmission risk upward [2,7,9]. That can exaggerate the importance of concurrency in both empirical studies and in modeling exercises. Alternatively, coinfections compete with concurrency as explanations for sub-Saharan Africa’s extraordinarily high HIV prevalence [24]. Boily et al. repeatedly emphasize how coinfections can confound efforts to link concurrency to HIV [7]. In contrast, Mah and Shelton [2] raise the issue of coinfections only to dismiss their importance. Similarly, Epstein and Morris[1] argue that coinfections (which they erroneously identify as “non-sexual
21
drivers of the epidemic”) cannot explain the patterns of HIV prevalence in sub-Saharan Africa. Epstein and Morris’s conflation of coinfections that promote sexual and vertical transmission with iatrogenic and other blood exposures that promote non-sexual transmission takes a sinister turn in their next-to-last paragraph where they say, “over the three decades since the AIDS pandemic first emerged, the field has been plagued by highly publicized ‘controversies’ driven by ideological advocates, some of whom have proposed that nonsexual drivers [emphasis added] associated with poverty explain the extreme disparities in HIV prevalence within and between countries.” The argument that certain coinfections can enhance transmission and/or acquisition of HIV is not ideologically driven—but is based on hundreds of scientific studies. Contributors to this literature do not claim that poverty causes HIV, but instead they study the connections between HIV and certain coinfections that enhance sexual and vertical transmission of HIV and whose prevalence is increased by poverty. Readers should find these remarks of Epstein and Morris chilling. There are hundreds, perhaps thousands, of research scientists, epidemiologists, and clinicians studying infectious and parasitic diseases that affect poor, tropical populations and who are trying to understand the ways that those diseases interact with transmission, progression, and treatment of HIV. Epstein and Morris paint those working in this field of inquiry as “ideological” and their work as “a dangerous distraction”. They link them in a not very subtle way to South Africa’s terrible episode with AIDS denialism that produced tragic consequences for the country. The goal of research must be to solve the puzzle of HIV in the very complex setting of sub-Saharan Africa, and it will take the skills and insights of people in multiple fields to achieve that goal. Censoring inquiry that lies outside the narrow field of sexual behavior by labeling it ideological, unscientific, or dangerous is an obstacle to finding the answers needed.
Concluding remarks Most of the recent debate over concurrency muddles two distinct though related issues. The first (labeled the concurrency hypothesis) is the explanation for sub-Saharan Africa’s extraordinary HIV epidemics. For the hypothesis to be valid, the prevalence of concurrency must be substantially higher in subSaharan Africa than in countries with low HIV prevalence and concurrency must be substantially more effective in spreading HIV than other forms of multiple partnering. The evidence supports
22
neither proposition, and thus the hypothesis that rests on them cannot be correct. That is so even if causal connections between concurrency and HIV within sub-Saharan Africa could be established. For over two decades, the prevailing explanation for sub-Saharan Africa’s hyper-epidemics of HIV has been unusually prevalent or aberrant sexual behavior. In recent years, concurrency is all that has kept the sexual-behavior explanation alive. The demise of the concurrency hypothesis, therefore, will have profound effects on how HIV in the region and elsewhere is understood since there is no longer a broadly accepted, evidencedbased explanation for African HIV. The second issue in the concurrency debate is about individual-level risk of HIV rather than population-level risk. Most of the flood of recent articles about concurrency sidestep any effort to explain the difference between HIV epidemics in sub-Saharan African and elsewhere. Instead, they address the second issue. Though empirical evidence is as yet elusive, it is difficult to imagine that an individual’s concurrency does not raise—if even by a modest amount—the risk of HIV acquisition of his or her partner(s). Despite the worries of some [8], finding a way to recognize that elevated risk in an HIV-prevention message could be as simple as “Don’t put your loved ones at risk.” That message or something similar might be easily understood and possibly effective in encouraging people to reduce both concurrency and the number of their partners. It is also clear that having a concurrent partner instead of a non-concurrent partner does not affect one’s acquisition risk [75]. Furthermore, if concurrency has any effect on HIV incidence at the population level, it more likely reduces rather than enhances HIV transmission [5,13]. Given the small and complex impact of concurrency on HIV, campaigns such as the one recently launched by SADC [79] that give primary emphasis to concurrency are clearly misguided. It would add light and reduce the heat of the debate over concurrency if participants would clearly distinguish between individual-level and population-level effects of concurrency. Interest in concurrency over the last 15 years has been motivated by the desire to explain Africa’s extraordinary HIV epidemics. Those who write about concurrency need to acknowledge that that effort has failed and to articulate a new motivation for their research. Otherwise, the muddled thinking over concurrency will continue. The finding that the concurrency hypothesis is incorrect raises two issues: First, what direction should research take and what are the implications for HIV-prevention policy?
23 Research Program
The key research priority is to discover what can explain sub-Saharan Africa’s hyper-epidemics. Sexual behavior cannot explain what is driving the epidemics in eastern and southern Africa and the embargo on research on other potential drivers must end. Moving beyond the failed sexual-behavior paradigm must not be labeled as ideological, unscientific, or dangerous. Modeling can be a powerful tool for elucidating HIV epidemic dynamics, but only if modelers adopt a new attitude toward model parameterization. Every model of HIV epidemics, even our own model [5], overestimates the impact of concurrency. Two decades of modeling have disproved the concurrency hypothesis and clarified the risks of various sexual behaviors to individuals. We doubt that more research can do much to clarify further those risks. Modelers of HIV epidemic dynamics would do well to follow Boily et al.’s [7] advice and use their models, as others have done [80-83], to examine the effect of risk factors that are not sexual behaviors but impinge upon sexual transmission. Since Africans’ sexual behavior cannot explain the hyperepidemics of the region, since sexual behaviors that increase the risk of HIV acquisition are already well understood, and since supporters of the concurrency hypothesis often refuse to believe what respondents tell them anyway, there are few as yet unanswered questions about African sexual behavior that would appear to warrant the cost of the research. Survey research could be very helpful, however, in elucidating the role of coinfections and non-sexual transmission in African HIV epidemics and the potential difficulties in implementing programs to reduce them. How many people with incident HIV had a recent circumcision? How would one design a campaign to persuade villagers to build and use latrines? How does one get nurses and doctors to wash their hands or women to switch from birth control injections to tablets? What is most needed to unravel Africa’s epidemics is not behavioral researchers, but bench and field scientists and epidemiologists to explore fully the possible coinfections and noninfectious diseases including nutritional deficiencies that promote HIV transmission. We need other scientists to find out why, according to the DHS, 31 percent of children in Mozambique or 22 percent of children in Swaziland have HIV-negative parents [84] or to trace genetically all of the HIV infections in a village to determine finally how the infection spreads and how much of it spreads non-sexually. There is much work to do and many eager to help. Too much time and money has been spent on asking people
24
the date of first sex with their third to last partner. It is time to move on. HIV-prevention programming
One could argue that disproving the concurrency hypothesis has no implications for HIV-prevention programming until it is determined what are the true drivers of HIV epidemics in subSaharan Africa. Nevertheless, likely candidates to replace concurrency are coinfections that raise transmission or acquisition efficiency and non-sexual exposures. Since treating and/or preventing many of the coinfections that could (or have already been shown to) promote the spread of HIV are inexpensive, have few side effects, and are desirable in their own right and reducing iatrogenic and nosocomial transmission of any disease is clearly the correct thing to do, there is no reason to wait until there is irrefutable evidence of their connection to HIV. The debate over whether or not to include concurrency in HIV-prevention messages, misses the far more important point that prevention policy is already too narrowly focused on sexual behavior and adding concurrency to the message only takes it further in the wrong direction. In most countries of sub-Saharan Africa, HIV prevention policy has had little impact, not because of inability to convey information or lack of personal agency on the part of the listeners, but because risky sexual behavior is only one dimension of personal risk, only a single aspect of peoples’ very complicated lives. Messages about sexual behavior change are compatible with and are reinforced by messages about other health-promoting behaviors. People need information that simple, safe, effective, and inexpensive medications can eliminate worms, heal genital sores of schistosomiasis—if initiated early enough—and cure certain STIs that promote HIV infection. They also need information and encouragement to demand safe and effective medical care, safe cosmetic treatments, such as tattoos, and safe informal medical care, such as vitamin injections. All of these biomedical and behavioral interventions support HIV prevention and treatment at low cost because they promote good health and greater productivity in multiple ways. Such messaging makes safe sex part of a broad health promotion program that encourages personal agency and responsibility and empowers people to demand both safe sex and safe and effective medical care. And, it is probably the best way to get people to practice safe sex because it addresses the whole person instead of one narrow aspect of their lives.
25
Works Cited 1.
Epstein H, Morris M: Concurrent partnerships and HIV: An inconvenient truth. Journal of the International AIDS Society 2011, 14:1-11.
2.
Mah TL, Shelton J: Concurrency revisited: Increasing and compelling epidemiological evidence. Journal of the International AIDS Society 2011, 14.
3.
Morris M, Epstein H, Wawer M: Timing is everything: International variations in historical sexual partnership concurrency and HIV prevalence. PLoS ONE 2010, 5:1-8.
4.
Goodreau S: A decade of modelling research yields considerable evidence for the importance of concurrency: A response to Sawers and Stillwaggon. Journal of the International AIDS Society 2011, 14:1-7.
5.
Sawers L, Isaac AG, Stillwaggon E: HIV and concurrent sexual partnerships: Modelling the role of coital dilution. Journal of the International AIDS Society 2011, 14:1-9.
6.
Sawers L, Stillwaggon E: Concurrent sexual partnerships do not explain the HIV epidemics in Africa: A systematic review of the evidence. Journal of the International AIDS Society 2010, 13:13-34.
7.
Boily M-C, Alary M, Baggaley RF: Neglected issues and hypotheses regarding the impact of sexual concurrency on HIV and sexually transmitted infections. AIDS and Behavior 2011.
8.
Padian NS, Manian S: The concurrency debate: Time to put it to rest. The Lancet 2011, 378:203-204.
9.
Eaton J, Hallett T, Garnett G: Concurrent sexual partnerships and primary HIV infection: A critical interaction. AIDS and Behavior 2010, 15:687-692.
10. Goodreau S, Cassels S, Kasprzyk D, Montano D, Greek A, Morris M: Concurrent partnerships, acute infection and HIV epidemic dynamics among young adults in Zimbabwe. AIDS and Behavior 2010 in press. 11. Delva W: Sexual behaviour and the spread of HIV – statistical and epidemiological modelling applications. Ghent: International Centre for Reproductive Health; 2010.
26 12. Tanser F, Bärnighausen T, Hund L, Garnett GP, McGrath N, Newell M-L: Effect of concurrent sexual partnerships on rate of new HIV infections in a high-prevalence, rural south African population: A cohort study. Lancet 2011, 378:247-255. 13. Reniers G, Tfaily R: Polygyny, partnership concurrency and HIV transmission in sub-Saharan Africa. Demography in press. 14. Santhakumaran S, O'Brien K, Bakker R, Ealden T, Shafer LA, Daniel RM, Chapman R, Hayes RJ, White RG: Polygyny and symmetric concurrency: Comparing long-duration sexually transmitted infection prevalence using simulated sexual networks. Sexually Transmitted Infections 2010, 86:553-558. 15. Steffenson AE, Pettifor AE, Ill GRS, Rees HV, Cleary PD: Concurrent sexual partnerships and human immunodeficiency virus risk among south African youth. Sexually Transmitted Diseases 2011, 38. 16. Arora P, Nagelkerke N, Sgaier SK, Kumar R, Dhingra N, Jha P: HIV, hsv-2 and syphilis among married couples in india: Patterns of discordance and concordance. Sexually Transmitted Infections 2011. 17. Arora P: Does concurrency explain the heterosexual HIV epidemic in sub-Saharan Africa? Lessons from sero-discordant couples (comment on Sawers and Stillwaggon, concurrent sexual partnerships do not explain the HIV epidemics in Africa: A systematic review of the evidence). Journal of the International AIDS Society 2010, 13. 18. Powers KA, Hoffman IF, Ghani AC, Hosseinipour MC, Pilcher CD, Price MA, Pettifor AE, Chilongozi DA, Martinson FEA, Cohen MS, Miller WC: Sexual partnership patterns in Malawi: Implications for HIV/STI transmission. Sexually Transmitted Diseases 2011, 38:657-666. 19. Maher D, Waswa L, Karabarinde A, Baisley K: Concurrent sexual partnerships and associated factors: A cross-sectional population-based survey in a rural community in Africa with a generalised HIV epidemic. BMJ Public Health 2011, 11:1-14. 20. Xu H, Luke N, Zulu EM: Concurrent sexual partnerships among youth in urban kenya: Prevalence and partnership effects. Population Studies 2010, 64:247-261.
27 21. Leung KY, Transmission of infection along a dynamic sexual network with star-shaped components, in Department of Mathematics, Faculty of Science2011, Utrecht University: Utrecht. p. 84. 22. Wellings K, Collumbien M, Slaymaker E, Singh S, Hodges Z, Patel D, Bajos N: Sexual behaviour in context: A global perspective. Lancet 2006, 368:1706-1728. 23. Cleland J, Ferry B, Caraël M: Summary and conclusions. In Sexual behaviour and AIDS in the developing world. Edited by J Cleland, B Ferry. London: Taylor and Francis for the World Health Organization; 1995: 208-228. 24. Stillwaggon E: AIDS and the ecology of poverty. New York: Oxford University Press; 2006. 25. Caraël M: Sexual behavior. In Sexual behavior and AIDS in the developing world. Edited by London: Taylor and Francis and WHO; 1995: 75-123. 26. Carael M, Slaymaker E, Lyerla R, Sarkar S: Clients of sex workers in different regions of the world: Hard to count. Sexually Transmitted Infections 2006, 82:iii26-iii33. 27. Lurie MN, Rosenthal S: Concurrent partnerships as a driver of the HIV epidemic in sub-Saharan Africa? The evidence is limited. AIDS and Behavior 2010, 14:17-24; discussion 25-8. 28. Lurie MN, Rosenthal S: The concurrency hypothesis in subSaharan Africa: Convincing empirical evidence is still lacking. Response to Mah and Halperin, Epstein, and Morris. AIDS and Behavior 2010, 14:17-24. 29. Sawers L, Stillwaggon E: Understanding the southern African ‘anomaly’: Poverty, endemic disease and HIV. Development and Change 2010, 41:195–224. 30. Morris M, Kretzschmar M: Concurrent partnerships and the spread of HIV. AIDS 1997, 11:641-648. 31. National Statistical Office, ICF Macro: Malawi Demographic and Health Survey 2010. Zomba, Malawi and Calverton, Maryland; 2011.
28 32. Morris M, Kretzschmar M: A microsimulation study of the effect of concurrent partnerships on the spread of HIV in Uganda. Mathematical Population Studies 2000, 8:109-133. 33. Xu H, Luke N, Zulu EM: Concurrent sexual partnerships among youth in urban kenya: Prevalence and partnership effects. Population Studies, 64:247-261. 34. Pinkerton SD: Probability of HIV transmission during acute infection in Rakai, Uganda. AIDS and Behavior 2008, 12:677-684. 35. Halperin D, Epstein H: Concurrent sexual partnerships help to explain africa's high HIV prevalence: Implications for prevention. Lancet 2004, 364:4-6. 36. Halperin D, Epstein H: Why is HIV prevalence so severe in southern Africa? Southern African Journal of HIV Medicine 2007:19-25. 37. Mah TL, Halperin DT: Concurrent sexual partnerships and the HIV epidemics in Africa: Evidence to move forward. AIDS and Behavior 2010, 14:11-16. 38. Kretzschmar M, Morris M: Measures of concurrency in networks and the spread of infectious disease. Mathematical Bioscience 1996, 133:165-195. 39. Epstein H: The invisible cure. New York: Farrar, Strauss and Giroux; 2007. 40. Epstein H: AIDS and the irrational. BMJ 2008, 337:a2638. 41. Hollingsworth TD, Anderson R, Fraser C: HIV-1 transmission, by stage of infection. Journal of Infectious Diseases 2008, 198:687693. 42. Wawer MJ, Gray RH, Sewankambo NK, Serwadda D, Li X, Laeyendecker O, Kiwanuka N, Kigozi G, Kiddugavu M, Lutalo T, Nalugoda F, Wabwire-Mangen F, Meehan MP, Quinn TC: Rates of HIV-1 transmission per coital act, by stage of HIV-1 infection, in Rakai, Uganda. Journal of Infectious Diseases 2005, 191:14031409. 43. Leclerc PM, Matthews AP, Garenne ML: Fitting the HIV epidemic in zambia: A two-sex mlcro-slmulatlon model. PLoS ONE 2009, 4:e5439.
29 44. Gregson S, Nyamukapa CA, Garnett GP, Mason PR, Zhuwau T, Carael M, Chandiwana SK, Anderson RM: Sexual mixing patterns and sex-differentials in teenage exposure to HIV infection in rural Zimbabwe. Lancet 2002, 359:1896-903. 45. UNAIDS Reference Group on Estimates MaP: Consultation on concurrent sexual partnerships. UNAIDS; Nairobi, Kenya; 2009. 46. Morris M, Kurth A, Hamilton D, Moody J, Wakefield S: Concurrent partnerships and HIV prevalence disparities by race: Linking science and public health practice. American Journal of Public Health 2009, 99:1023-1031. 47. Johnson LF, Dorrington RE, Bradshaw D, Pillay-VanWyk V, Rehle TM: Sexual behaviour patterns in South Africa and their association with the spread of HIV: Insights from a mathematical model. Demographic Research 2009, 21:289-340. 48. Shisana O, Rehle T, Simbayi L, Parker W, Zuma K, Bhana A, Connolly C, Jooste S, Pillay V: South African national HIV prevalence, HIV incidence, behaviour and communication survey, 2005. Cape Town, South Africa: HSRC Press; 2005. 49. Taruberekera N, Jafa K, Mushayi W: Multiple concurrent partnerships in Zimbabwe: Determinants and monitoring indicators. PSI Research Division, Population Services Internationa.; Washington, DC; 2008. 50. Sandøy IF, Dzekedzeke K, Fylkesnes K: Prevalence and correlates of concurrent sexual partnerships in zambia. AIDS Behavior 2010, 14:59-71. 51. Langhaug LF, Sherr L, F.M. Cowan: How to improve the validity of sexual behaviour reporting: Systematic review of questionnaire delivery modes in developing countries. Tropical Medicine and International Health 2010, 15:362-381. 52. Brummer D: Labour migration and HIV/AIDS in southern Africa. Pretoria: International Organization for Migration, Regional Office for Southern Africa; 2002. 53. Clark SJ, Collinson MA, Kahn K, Drullinger K, Tollman S: Returning home to die: Circular labour migration and mortality in South Africa. Scandinavian Journal of Public Health 2007, 35:35–44.
30 54. Crush J, Williams V, Peberdy S: Migration in southern Africa. Policy Analysis and Research Programme of the Global Commission on International Migration.; Geneva; 2005. 55. Hargreaves JR, Bonell CP, Morison LA, Kim JC, Phetla G, Porter JD, Watts C, Pronyk PM: Explaining continued high HIV prevalence in South Africa: Socioeconomic factors, HIV incidence and sexual behaviour change among a rural cohort, 2001-2004. AIDS 2007, 21 Suppl 7:S39-48. 56. Marks S: An epidemic waiting to happen? The spread of HIV/AIDS in South Africa in social and historical perspective. African Studies 2002, 61:13–26. 57. Thahane TT: International labor migration in southern Africa. In The unsettled relationship: Labor migration and economic development. Edited by DG Papademetriou, PL Martin. New York: Greenwood Press; 1991: 65–88. 58. Lurie MN, Williams BG, Zuma K, Mkaya-Mwamburi D, Garnett GP, W. SA, Sweat MD, Gittlesohn J, Karim AS, The impact of migration on HIV-1 transmission in South Africa, 2002, South African Medical Research Council, HIV Prevention and Vaccine Research Unit: Durban. 59. Zuma K, Gouws E, Williams B, Lurie M: Risk factors for HIV infection among women in carletonville, South Africa: Migration, demography and sexually transmitted diseases. Int J STD AIDS 2003, 14:814-7. 60. Lurie M, Williams BG, Zuma K, Mkaya-Mwamburi D, Garnett G, Sweat MD, Gittelsohn J, Karim SS: Who infects whom? HIV-1 concordance and discordance among migrant couples in South Africa. AIDS 2003, 17:2245-2252. 61. Lurie MN, The epidemiology of migration and AIDS in South Africa, in Working Paper2004, University of Oxford, Centre on Migration, Policy and Society: Oxford, UK. 62. Poulin M: Reporting on first sexual experience: The importance of interviewer-respondent interaction. Demographic Research 2010, 22:237-288. 63. Morris M, Leslie-Cook A, Nelson SJ:Evaluating concurrent partnership data from the 2005-2008 demographic and health surveys (DHS).[XVIII International AIDS Conference:Vienna, Austria.
31 64. Ministry of Health and Social Welfare [Lesotho] and ICF Macro: Lesotho demographic and health survey 2009. Ministry of Health and Social Welfare and ICF Macro; Maseru, Lesotho and Calverton, Maryland, USA; 2010. 65. Mishra V, Bignami-Van Assche S, Concurrent sexual partnerships and HIV infection: Evidence from national population-based surveys, in DHS Working paper2009, USAID. 66. Adimora AA, Schoenbach VJ, Doherty IA: Concurrent sexual partnerships among men in the United States. American Journal of Public Health 2007, 97:2230-2237. 67. Leridon H, van Zessen G, Hubert M: The europeans and their sexual partners. In Sexual behaviour and HIV/AIDS in europe: Comparisons of national surveys. Edited by M Hubert, N Bajos, T Sandfort. London: UCL Press; 1998: 165-196. 68. Meyerson B, Robbins A, Koppenhaver T, Fleming D: TCM exposure and HIV-related knowledge, attitudes, and practices from the 2003 makgabaneng listenership survey in Botswana. BOTUSA; 2003. 69. Rweyemamu D: One love connect. Protect. Respect: Multiple and concurrent sexual partnerships among youth in Tanzania. A research study commissioned by femina hip in preparation for a regional youth mcp campaign 2008. University of Dar es Salaam; Dar es Salaam, Tanzania; 2008. 70. Parker W, Connoly C: Namibia: HIV/AIDS community survey report rundu, walvis bay, keetmanshoop and oshakati. NawaLife Trust; Windhoek; 2007. 71. Parker W, Makhubele B, Ntlabati P, Connolly C: Concurrent sexual partnerships amongst young adults in South Africa. Cadre (Centre for AIDS Development, Research and Evaluation); Johannesburg and Grahamstown, South Africa; 2007. 72. Standing H, Kisekka M: Sexual behavior in sub-Saharan Africa. London: Overseas Development Administration: A Review and Annotated Bibliography; 1989. 73. Obbo C: HIV transmission through social and geographic networks in Uganda. Social Science and Medicine 1993, 36:949955.
32 74. Talle A: Desiring difference: Risk behavior among young maasai men. Young people at risk: Fighting AIDS in northern Tanzania. Copenhagen, Denmark: Scandinavian University Press; 1995. 75. Morris M: Barking up the wrong evidence tree. Comment on lurie & rosenthal, "concurrent partnerships as a driver of the HIV epidemic in sub-Saharan Africa? The evidence is limited". AIDS and Behavior 2010, 14:31-3; discussion 34-7. 76. UNAIDS: Global report: Unaids report on the global AIDS epidemic, 2010. Joint United Nations Programme on HIV/AIDS; Geneva; 2010. 77. Celum C, et al.: Acyclovir and transmission of HIV-1 from persons infected with HIV-1 and hsv-2. N Engl J Med 2010, 362:427-39. 78. Reynolds SJ, Makumbi F, Nakigozi G, Kagaayi J, Gray RH, Wawer M, Quinn TC, Serwadda D: HIV-1 transmission among HIV-1 discordant couples before and after the introduction of antiretroviral therapy. AIDS 2011, 25:473-477. 79. Shelton JD: Why multiple sexual partners? Lancet 2009, 374:367–369. 80. Mushayabasa S, Bhunu CP: Modeling schistosomiasis and HIV/AIDS codynamics. Computational and Mathematical Methods in Medicine 2011, 2011. 81. Gibson LR, Bingtuan L, Remold SK: Treating cofactors can reverse the expansion of a primary disease epidemic. BMC Infectious Diseases 2010, 10. 82. Cuadros D, Branscum A, Crowley P: HIV-malaria co-infection: Effects of malaria on the prevalence of HIV in east sub-Saharan Africa. International Journal of Epidemiology 2011, 39:1-9. 83. Nagelkerke N, de Vlas SJ, Jha P, Luo M, Plummer FA, Kaul R: Heterogeneity in host HIV susceptibility as a potential contributor to recent HIV prevalence declines in Africa. AIDS 2009, 23:125-130. 84. Grimm M, Class DM: The fight against HIV / AIDS must be brought into balance. KfW, German Development Bank; Frankfurt am Mein; 2011.