Partnership concurrency, individuals having 2 or more overlapping sexual partnerships, has the potential to increase the spread of HIV in a population. Concurrency increases the opportunity for HIV to spread through a mechanism known as “protective sequencing.”1 Under sequential monogamy, an individual can only pass HIV acquired from one partner to future partners, and not to other current partners. If partnerships are concurrent, the virus can be passed to both earlier and later partners. Sequential monogamy also slows the spread of HIV by trapping the virus within a monogamous relationship until its end. It is argued that this is important, as there is a short period of high infectivity after infection. For HIV to have the opportunity to spread during this period under sequential monogamy, newly infected individuals must end the partnership within which they became infected, and start a second, within this short period. This is not necessary, if individuals have concurrent partnerships.
Most modeling studies report that concurrency, in contrast to serial monogamy, increases the initial spread and endemic prevalence of HIV in a population,2–8 but that those results are all highly sensitive to their assumed parameter values, and their ability to reflect actual epidemics in the real world is unclear. Empirical evidence for a link between concurrency and the spread of HIV in a population is also equivocal.9–11 Although studies have used mathematical modeling to investigate the potential role of concurrency in the initial spread of African HIV epidemics,7,8,12–15 little work has been done on the possible effects of an intervention to reduce concurrency in well-established, generalized epidemics, typical of countries in Eastern and Southern Africa.
Despite this, interventions targeting partnership concurrency have been proposed as a way of reducing HIV transmission in many countries, and some have been implemented. A meeting on HIV prevention hosted by the Southern African Development Community in 2006 concluded that “Multiple and concurrent partnerships … are a key driver of the epidemic in the region,” and recommended that reducing multiple and concurrent sexual partnerships should be a key priority in HIV prevention.16 National campaigns to reduce partner concurrency have been planned or launched in Botswana, Kenya, Lesotho, Malawi, Mozambique, Namibia, South Africa, Swaziland, Tanzania, Zambia, and Zimbabwe.17
Southern African Development Community's use of the phrase “multiple and concurrent partnerships” also highlights a key difficulty in investigating the effect of reducing concurrency. Unlike the relatively weak empirical evidence for a link between concurrency and HIV transmission, stronger empirical and theoretical evidence exists on the role of multiple sexual partners in HIV acquisition and transmission.18–23 The results of empirical studies targeting concurrency are therefore likely to be confounded by a reduction in partnership incidence,24,25 and therefore, empirical studies are unlikely to be able to provide clear evidence on whether it is worth complicating behavior change messages by including concurrency. A modeling study is useful because it can separate the likely impact of reducing concurrency from the impact of changing (some) other aspects of sexual behavior. This study investigated the potential impact of an intervention against partnership concurrency on the incidence of HIV in rural Uganda, and other (higher-concurrency) sub-Saharan African populations, keeping the “amount of sex” (measured by the mean per capita incidence of sex acts and partnerships) constant.
MATERIALS AND METHODS
Data Sources and Analysis
The model was fitted to data on the demography, sexual behavior, and HIV prevalences of the Masaka general population cohort in rural South-West Uganda. This open cohort was established in 1989 and consists of 25 villages. There were 7067 residents of age 15 to 54 years in 2008. Data collection methods are reported in full elsewhere,26–28 but in brief, every year basic demographic information is updated, blood samples are collected and tested for HIV, and a behavioral questionnaire is conducted using face-to-face interviews. An average of 76% of adults have a confirmed HIV status each year.29 All data from the cohort were age-standardized to the 2008 eligible population size by gender, unless stated otherwise.
Concurrency and Number of Partners
Data on sexual behavior were collected in 2009–2010. Data on 14 of 25 villages were available when this study was initiated. The response rate for 15 to 54 year olds in the 14 villages was 61.3% (2684/4377).
Sexual partnerships were classified as long duration (spouses and ex-spouses) and short duration (all other partnerships). Spouses were defined by the survey team as “partners who are in a consistent sexual relationship and are living together or perceive themselves as married” (see Supplemental Digital Content, available at: http://links.lww.com/OLQ/A41, for full details). Partnership data were available from 1214 men and 1470 women aged 15 to 54 years. Concurrency was defined as “overlapping sexual partnerships where sexual intercourse with one partner occurs between two acts of intercourse with another partner.” Point prevalences of concurrency at a point, 6 months before the interview, were calculated using the method recommended by the UNAIDS Reference Group on Estimates, Modeling, and Projections17 (see Supplemental Methods, online only, available at: http://links.lww.com/OLQ/A41).
Using this definition, 3 different indicators of concurrency were calculated by gender:
1. Long-duration partnership concurrency: the proportion of individuals with 2 or more concurrent long-duration partnerships.
2. Short-duration partnership concurrency: the proportion of individuals with 2 or more concurrent short-duration partnerships.
3. Total concurrency: the proportion of individuals with 2 or more concurrent partnerships of either type.
After age standardizing, 3.8% of men reported long-duration partnership concurrency, 1.9% reported short-duration partnership concurrency, and 9.6% reported total concurrency. In all, 0.07% of women reported long-duration partnership concurrency, no women reported short-duration partnership concurrency, and 0.20% reported total concurrency (Table S1, online only, available at: http://links.lww.com/OLQ/A41).
A dynamic, stochastic, network model was used to represent sexual partnerships between males and females in a simulated population, and HIV transmission within partnerships. Events such as births, deaths, partnership formation/dissolution, and HIV transmission were modeled using time-dependent rates. Demographic, behavioral, and epidemiologic characteristics of the model population were varied by changing the values of input parameters. Model output was used to measure outcomes, including incidence and prevalence of partnerships and HIV.
In this article, “contact rate” was used to describe the model input partnership acquisition rate that codetermines partnership formation in the model (before adjustment for concurrency, partnership balancing, and so on). The term “partnership incidence” was used to describe the actual rate of formation of new partnerships, both estimated from empirical data and simulated using the model.
Each individual born into the model was assigned to one of 2 sexual activity groups (“high” or “low”) and one of 2 concurrency groups.
In this study, we maintained overall partnership incidence in each activity group while allowing the prevalence of concurrency to change over time. This was achieved as follows (see Supplemental Digital Content, online only, available at: http://links.lww.com/OLQ/A41).
Each individual in gender, g, and concurrency group, k, had a parameter θgk associated with them. θgk varied between 0 and 1. θgkp determined how likely it was that a person with p current sexual partners could take another partner. When θgk = 1, the likelihood of an individual taking another partner is unchanged for any number of current partners. If θgk <1, it becomes less likely that an individual would take another partner. Finally, if θgk = 0, then individuals are constrained to having at most 1 current partner.
Partnership incidence was maintained by sexual activity group (and overall) according to the equation later in the text, by increasing the contact rate among individuals with lower numbers of current partners, and reducing the contact rate among individuals with higher numbers of current partners:
Equation (Uncited)Image Tools
where cgskp is the desired per-person contact rate in the group of gender (g), sexual activity (s), concurrency parameter (k), and number of current partners (p); Ngs is the number of simulated individuals of gender g and sexual activity s, K is the number of concurrency groups (2 in both genders in this study); and P is the number of current partner groups (3[0,1, and 2+] in both genders in this study). To model an intervention reducing concurrency, we specified a time to reduce θgk.
When a simulated partnership was formed, its duration was determined from the activity group of one of the partners selected at random: short duration if they were in the high-activity group, and long duration if they were in the low-activity group. As partnership incidence in each activity group was maintained as θ varied, so was the overall ratio of long to short partnerships and, therefore, the per-capita sex act rate.
Untreated HIV infection was crudely categorized into 4 sequential stages: primary HIV infection, postprimary CD4+ count >200 cells/μL, pre-AIDS CD4+ count <200 cells/μL, and an AIDS stage. ART was implemented by simulating the removal from the pre-AIDS CD4+ count <200 cells/μL and AIDS stages into on-treatment stages (see Supplemental material, online only, available at: http://links.lww.com/OLQ/A41).
Two thousand model runs were averaged to reduce stochastic variation below the precision of the results shown in this study.
We simulated 3 main baseline scenarios of male and female concurrency (Table 1). The first scenario was our “best-estimate” of the prevalence of concurrency in this population. The prevalences of each type of male concurrency were based on the reported prevalences. Women reported a low prevalence of concurrency (0.20% in total). It is likely that this reflects strong social pressures against women reporting multiple partners. We estimated that the prevalence of concurrency in women in the cohort may be around 25% of that in men. Therefore, in this best-estimate scenario, we assumed female concurrency to be 25% of the prevalence in men.
The 2 other scenarios represented lower and higher plausible bounds for the prevalence of concurrency in this population. The lower-concurrency scenario simulated a prevalence of male total concurrency equal to 50% of the reported prevalence, and no female concurrency. The higher-concurrency scenario simulated a prevalence of male total concurrency equal to 150% of the reported prevalence and a prevalence of female total concurrency equal to 25% of the modeled male prevalence. In addition to providing a higher plausible bound on the impact of the intervention in Uganda, the higher-concurrency scenario also provides an estimate of the potential impact of the intervention in countries with higher prevalences of concurrency. Further details of the scenarios are given in the Supplemental Digital Content (online only, available at: http://links.lww.com/OLQ/A41).
In addition to the 3 main scenarios, a fourth scenario simulated the reported prevalences of concurrency in men and no female concurrency (“reported-concurrency” scenario; Supplemental Digital Content, online only, available at: http://links.lww.com/OLQ/A41).
Plausible Ranges for Model Inputs and Outputs
The model input variables and fitting targets used in the model are summarized in Table 2. Plausible ranges for all model inputs and outputs are summarized in supplemental context section “Summary of plausible ranges for model inputs and outputs” (online only, available at: http://links.lww.com/OLQ/A41).
Table 2-a. Summary o...Image Tools
Table 2-b. Summary o...Image Tools
Twenty Percent or 50% Reduction in Total Concurrency.
For each of the 3 aforementioned baseline scenarios, 2 intervention scenarios were created to estimate the impact of a 20% or 50% reduction in the prevalence of total concurrency. The intervention scenarios were identical to the baseline scenarios except that the values of the male and female concurrency parameters (θg) were reduced in 2010 to obtain a 20% or 50% reduction in the prevalence of total concurrency by 2020 in men (and women, if applicable). The fitted reductions in θ are shown in Table S3 (online only, available at: http://links.lww.com/OLQ/A41).
Concurrency Prevalence Reduction by Gender.
We also explored the sensitivity of our findings to a wider range of variation in the reduction of the prevalence of concurrency in each gender separately. For each of the baseline scenarios, we simulated 9 intervention scenarios in which the value of θ for males was reduced by 10% to 90% (with no change in the female concurrency parameter) and 9 intervention scenarios in which the value of θ for females was reduced by 10% to 90% (with no change in the male concurrency parameter). In these scenarios, the impact of the change in θ on the prevalence of concurrency was not fitted to a prespecified value.
In all intervention scenarios, the “amount of sex” (measured by the mean per capita incidence of sex acts and partnerships) in the population was kept constant according to Equation S1 (online only, available at: http://links.lww.com/OLQ/A41), to prevent confounding by these risk factors. A priori, we expect the proportion sexually active to rise if we reduce concurrency while maintaining partnership incidence and partnership duration.
The primary outcome measure was the relative reduction in HIV incidence in 2020 in the intervention scenario compared with the baseline scenario, overall and by gender.
Baseline Simulated Scenarios
A summary of the values of the fitting constraints and the values of the input parameters and outputs in each scenario is given in the Tables S2, S4, and S5 (online only, available at: http://links.lww.com/OLQ/A41). Partnership durations of 10 years for long-duration partnerships and 0.43 years for short-duration partnerships were estimated by fitting the model to partnership incidence data. All demographic, behavioral, and epidemiologic outcomes were within the fitting constraints. Male and female population sizes were within 0.7% of the target population sizes in all scenarios. Table 1 shows the prevalence of concurrency in each scenario. Figure 1 shows the target and simulated partnership distributions by gender in 2008. The simulated distributions were a good fit to the target distributions in all scenarios. Figure 1 also shows the simulated HIV prevalence trends and the fits to HIV prevalence in 1992, 2001, and 2007. In all scenarios, HIV prevalence increases between 2004 and 2009, after the introduction of ART.
Impact on Sexual Behavior.
The desired reductions in concurrency were achieved within the fitting constraints for both sexes in all fitted intervention scenarios (Table S6, online only, available at: http://links.lww.com/OLQ/A41). Annual partnership incidence and the average annual number of sex acts per person were kept constant (Figure S6/7, online only, available at: http://links.lww.com/OLQ/A41). As expected, this led to an increase in the proportion of the population sexually active, as shown in Table S7 (online only, available at: http://links.lww.com/OLQ/A41).
Impact on HIV
Twenty percent or 50% reduction in total concurrency.
Results are presented as the reduction in the best-estimate concurrency prevalence scenario (reduction in the lower-concurrency prevalence scenario − reduction in higher-concurrency prevalence scenario). Reducing the prevalence of concurrency by 20% reduced HIV incidence in 2020 by 4.1% (0.3%–5.7%) in men, 9.2% (2.1%–16.8%) in women, and 7.1% (1.3%–12.4%) overall (Fig. 2). Reducing concurrency by 50% reduced HIV incidence by 6.0% (1.4%–10.8%) in men, 16.2% (6.3%–23.4%) in women, and 11.9% (4.2%–18.3%) overall.
In all scenarios, the interventions caused a larger reduction in HIV incidence in women than in men (2.1–7.1 times larger). The HIV incidence trend between 2010 and 2020 in the best-estimate baseline scenario and the corresponding 50% concurrency reduction intervention scenario is illustrated in Figure S2 (online only, available at: http://links.lww.com/OLQ/A41).
Concurrency prevalence reduction by gender.
There was a nonlinear relationship between the reduction in concurrency and reduction in overall HIV incidence in all scenarios (Fig. 3). In all scenarios, reducing concurrency in only one gender resulted in larger reductions in HIV incidence in the other gender.
Reducing the prevalence of concurrency by 20% resulted in a 7.1% (1.3%–12.4%) reduction in HIV incidence after 10 years. Reducing concurrency by 50% resulted in a 11.9% (4.2%–18.3%) reduction in HIV incidence. In all scenarios, when concurrency was reduced in 1 gender only, reductions in HIV incidence were greater in the “other gender.” In all scenarios, reducing concurrency had a nonlinear impact on HIV incidence. As the reductions in concurrency were increased, the size of the additional impact on HIV incidence reduced.
Limitations of the data collection methods are unlikely to have introduced bias into the measurement of concurrency (see Supplemental Results, online only, available at: http://links.lww.com/OLQ/A41). However, the empirical estimate of concurrency may have been biased by individuals over- or underreporting concurrency. In addition, it may have been biased by the relatively low response rate in the survey used to estimate concurrency, in particular, if people who work outside the study area are both less likely to have responded and more likely to have concurrent partners. We accounted for this to some extent by simulating scenarios with higher and lower prevalence of concurrency.
Sexual behavior data were available from only 14 of 25 cohort villages. This is unlikely to have introduced bias, as a later analysis of all study villages found concurrency prevalences that were similar to the prevalences used in this study.30
The simulated prevalence of concurrency did not immediately decrease to 20% or 50% of its original level after the start of the intervention. Instead, it decreased gradually over a number of years. This is because the simulated intervention reduced the rate of formation of concurrent partnerships. We believe this is more plausible than an intervention that results in the immediate dissolution of existing partnerships, but we may have underestimated impact if an intervention did have this effect.
There are 3 main model limitations. First, sexually transmitted infections other than HIV were not included in the model. This could have biased our projected impact downward or upward.24 Second, condom use was not simulated. In all, 15.9% of men with concurrent ongoing partnerships reported using a condom the last time they had sex compared with only 9.9% of men with a single ongoing partnership (see Supplemental Results, online only, available at: http://links.lww.com/OLQ/A41). This is likely to have biased the projected impact of the intervention upward.
Finally, we did not simulate coital dilution (lower coital frequency/partner with increasing numbers of concurrent partners). A recent modeling study2 suggested even small amounts of coital dilution may reduce the role of concurrency as a population risk factor for HIV transmission. If true, we may have overestimated the role of concurrency as a risk factor for HIV transmission, and therefore may have overestimated the impact of the intervention.
In addition to these limitations, it should be noted that any model of sexual behavior, including this one, simulates a simplified version of the true pattern of sexual networking in a population. This will have inevitably introduced some uncertainty into the estimate of the effect of the intervention. We believe that we have reduced this uncertainty as far as possible by fitting the model to all available data; however, some uncertainly will remain.
We observe exactly what Morris et al10 predicted, that the “other gender” would benefit more from reductions in concurrency. As our and other data suggest that reported concurrency is more common among men than women, if an effective intervention could be identified, reducing levels of concurrency in men may be a useful intervention to reduce HIV infection among women.
An alternative approach would have been to allow partnership incidence to decrease as concurrency decreased and to compare it with a baseline scenario where partnership incidence decreases by the same amount, but the prevalence of concurrency remains constant. However, this would have required making assumptions about the size of reductions in partnership incidence that may or may not accompany reductions in concurrency.
Some of the countries where interventions against concurrency are being planned or implemented have higher prevalences of concurrency than rural Uganda.31,32 Our higher-plausible bound for the prevalence of concurrency in rural Uganda also gives an indication of the possible impact of an intervention against concurrency in higher concurrency settings.
In this setting, interventions against concurrency have the potential to reduce HIV incidence and may have a higher impact in women than in men. As such, if an effective intervention could be identified, it may be a useful intervention to protect women against infection. However, large simulated changes in concurrency resulted in relatively moderate decreases in HIV incidence, and because of the lack of condom use, coital dilution, and age structure in the model, the findings reported here may have overestimated the impact of the intervention. This study does not provide strong support for the prioritization of concurrency as a target for behavior change interventions. However, it may be more useful in higher concurrency settings and for reducing HIV incidence in women.
1. Morris M, Goodreau S, Moody J. Sexual networks, Concurrency and STD/HIV. In: Holmes KK, ed. Sexually Transmitted Diseases 4th ed. New York, NY: McGraw-Hill, 2008:119–125.
2. Sawers L, Isaac AG, Stillwaggon E. HIV and concurrent sexual partnerships: modelling the role of coital dilution. J Int AIDS Soc 2011; 14:44.
3. Goodreau SM, Cassels S, Kasprzyk D, et al.. Concurrent Partnerships, Acute Infection and HIV Epidemic Dynamics Among Young Adults in Zimbabwe. AIDS Behav 2012; 16:312–322.
4. Kretzschmar M, Morris M. Measures of concurrency in networks and the spread of infectious disease. Math Biosci 1996; 133:165–195.
5. Doherty IA, Shiboski S, Ellen JM, et al.. Sexual bridging socially and over time: A simulation model exploring the relative effects of mixing and concurrency on viral sexually transmitted infection transmission. Sex Transm Dis 2006; 33:368–373.
6. Johnson AM, Mercer CH, Erens B, et al.. Sexual behaviour in Britain: Partnerships, practices, and HIV risk behaviours. Lancet 2001; 358:1835–1842.
7. Morris M. Sexual networks and HIV. AIDS 1997; 11:S209–S216.
8. Eaton J, Hallett T, Garnett G. Concurrent sexual partnerships and primary HIV infection: A critical interaction. AIDS Behav 2011; 15:687–692.
9. Lurie MN, Rosenthal S. Concurrent partnerships as a driver of the HIV epidemic in sub-Saharan Africa? The evidence is limited. AIDS Behav 2010; 14:17–24.
10. 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 Behav 2010; 14:31–33.
11. Halperin DT, Epstein H. Concurrent sexual partnerships help to explain Africa's high HIV prevalence: Implications for prevention. Lancet 2004; 364:4–6.
12. Morris M, Kretzschmar M. Concurrent partnerships and the spread of HIV. AIDS 1997; 11:641–648.
13. Santhakumaran S, O'Brien K, Bakker R, et al.. Polygyny and symmetric concurrency: Comparing long-duration sexually transmitted infection prevalence using simulated sexual networks. Sex Transm Infect 2010; 86:553–558.
14. Morris M, Kretzschmar M. A microsimulation study of the effect of concurrent partnerships on the spread of HIV in Uganda. Mathematical Population Studies: Int J Mathematical Demogr 2000; 8:109–133.
15. Goodreau S, Cassels S, Kasprzyk D, et al.. Concurrent partnerships, acute infection and HIV epidemic dynamics among young adults in Zimbabwe. AIDS Behav 2010; 1–11.
16. SADC. Expert think tank meeting on HIV prevention in high-prevalence countries in Southern Africa, Maseru, 2006.
17. UNAIDS. Consultation on concurrent sexual partnerships. Recommendations from a meeting of the UNAIDS reference group on estimates, modelling and projections, Nairobi, 2009.
18. Wawer MJ, Sewankambo NK, Berkley S, et al.. Incidence of HIV-1 infection in a rural region of Uganda. BMJ 1994; 308:171–173.
19. Kiwanuka N, Gray R, Serwadda D, et al.. The incidence of HIV-1 associated with injections and transfusions in a prospective cohort, Rakai, Uganda. AIDS 2004; 18:342–344.
20. Quigley M, Morgan D, Malamba S, et al.. Case-control study of risk factors for incident HIV infection in rural Uganda. J Acquir Immun Defic Syndr 2000; 23:418–425.
21. Anderson R, May R. Social heterogeneity and sexually transmitted diseases. Infectious Diseases of Humans. In: Dynamics and Control. Oxford, England: Oxford University Press, 1991:228–303.
22. Hyman J, Stanley E. Using mathematical models to understand the AIDS epidemic. Math Biosc 1988; 90.
23. Todd J, Grosskurth H, Changalucha J, et al.. Risk factors influencing HIV infection incidence in a rural African population: A nested case-control study. J Infect Dis 2006; 193:458–466.
24. Boily M-C, Alary M, Baggaley R. Neglected issues and hypotheses regarding the impact of sexual concurrency on HIV and sexually transmitted infections. AIDS Behav 2011; 1–8.
25. Kretzschmar M, White RG, Carael M. Concurrency is more complex than it seems [Editorial]. AIDS 2010; 24:313–315.
26. Seeley J, Wagner U, Mulemwa J, et al.. The development of a community-based HIV/AIDS counselling service in a rural area in Uganda. AIDS Care 1991; 3:207–217.
27. Mulder DW, Nunn AJ, Kamali A, et al.. Two-year HIV-1-associated mortality in a Ugandan rural population. Lancet 1994; 343:1021–1023.
28. Mulder DW, Nunn AJ, Wagner HU, et al.. HIV-1 incidence and HIV-1-associated mortality in a rural Ugandan population cohort. AIDS 1994; 8:87–92.
29. Shafer LA, Biraro S, Nakiyingi-Miiro J, et al.. HIV prevalence and incidence are no longer falling in southwest Uganda: Evidence from a rural population cohort 1989–2005. AIDS 2008; 22:1641–1649.
30. Maher D, Waswa L, Karabarinde A, et al.. Concurrent sexual partnerships and associated factors: A cross-sectional population-based survey in a rural community in Africa with a generalised HIV epidemic. BMC Public Health 2011; 11:651.
31. Harrison A, Cleland J, Frohlich J. Young people's sexual partnerships in KwaZulu-Natal, South Africa: Patterns, contextual influences, and HIV risk. Stud Fam Plann 2008; 39:295–308.
32. Gourvenec D, Taruberekera N, Mochaka O, et al.. Multiple concurrent partnerships among men and women aged 15–34 in Botswana—Baseline study, Gaborone, Botswana, 2007.