Assessing effectiveness and cost-effectiveness of concurrency reduction for HIV prevention

Abstract

We estimated the effectiveness and cost-effectiveness of changes in concurrent sexual partnerships in reducing the spread of HIV in sub-Saharan Africa. Using data from Swaziland, Tanzania, Uganda and Zambia, we estimated country-specific concurrency behaviour from sexual behaviour survey data on the number of partners in the past 12 months, and we developed a network model to compare the impact of three behaviour changes on the HIV epidemic: (1) changes in concurrent partnership patterns to strict monogamy; (2) partnership reduction among those with the greatest number of partners; and (3) partnership reduction among all individuals. We estimated the number of new HIV infections over 10 years and the cost per infection averted. Given our assumptions and model structure, we find that reducing concurrency among high-risk individuals averts the most infections and increasing monogamy the least (11.7% versus 8.7% reduction in new infections, on average, for a 10% reduction in concurrent partnerships). A campaign that costs US$1 per person annually is likely cost-saving if it reduces concurrency by 9% on average, given our baseline estimates of concurrency. In sensitivity analysis, the rank ordering of behaviour change scenarios was unaffected by potential over-estimation of concurrency, though the number of infections averted decreased and the cost per HIV infection averted increased. Concurrency reduction programmes may be effective and cost-effective in reducing HIV incidence in sub-Saharan Africa if they can achieve even modest impacts at similar costs to past mass media campaigns in the region. Reduced concurrency among high-risk individuals appears to be most effective in reducing HIV incidence, but concurrency reduction in other risk groups may yield nearly as much benefit.

Keywords

HIV/AIDS sub-Saharan Africa concurrent partnerships sexual partnership network cost-effectiveness

INTRODUCTION

HIV prevalence in parts of sub-Saharan Africa is orders of magnitude higher than in Europe or the USA.¹ This disparity may be due in part to the existence of concurrent sexual partnerships, which have been documented at high rates in parts of sub-Saharan Africa.^1–5 Concurrent partnerships allow for more efficient transmission of HIV by increasing the number of individuals connected together through a network of sexual partnerships. In a monogamous relationship, the virus can be transmitted between partners, but cannot escape the partnership, whereas in a concurrent sexual network the virus can rapidly spread.⁶ Previous work has suggested that for the same number of lifetime partners, having concurrent partners is associated with a more rapid spread of sexually transmitted infections (STIs) than having one partner at a time.^6–8 This is of particular concern with HIV, since transmission of the virus is most efficient during the early phase of the infection, when viral load is high and the ability to diagnose individuals is limited.⁹

Individuals engage in multiple concurrent partnerships for a variety of reasons (e.g. sexual variety, increased social status and material gain^10–14) and to different degrees (some individuals may only have two partners simultaneously, while others may have many concurrent partners). Strategies to reduce concurrent partnerships may achieve different levels of success depending on the target population and behaviour change. For example, a campaign that promotes faithfulness in married or co-habitating partnerships may have a different effect on the spread of HIV than a campaign that reduces partnerships among highly promiscuous individuals.

Reducing concurrent sexual partnerships is increasingly an important goal for HIV prevention in highly endemic areas.^15–17 A recent review of HIV prevention activities argued that partner reduction programmes have had a powerful impact on containing the spread of HIV, and should be more assertively promoted.¹⁷ Several countries in Africa have targeted concurrent sexual partnerships for HIV prevention. In the late 1980s, Uganda launched a ‘Zero Grazing’ campaign which cost an estimated US$0.23 per person in the population annually and promoted faithfulness as a means of reducing concurrent sexual partnerships.^18–21 The introduction of this campaign in Uganda was associated with a significant decrease in HIV prevalence in the early 1990s, while other countries in the region were still seeing HIV prevalence rise.^20,21 While the ‘Zero Grazing’ campaign may not have been solely responsible for the reduction in HIV prevalence, several analyses have concluded that other factors such as condom use or HIV mortality cannot explain the large reductions in prevalence seen in Uganda compared with its neighbours.^20–24 More recently, observational studies in Zimbabwe found a reduction in the number of individuals reporting multiple sexual partners, including concurrent partners, from the late 1990s to the 2000s^25–27 Mathematical models suggest that this change in sexual behaviour contributed to the observed coincident decline in HIV prevalence, which cannot be explained by the natural history of the epidemic alone.^23,26,27 These examples suggest that behaviour change can have a major impact on the course of the HIV epidemic.

Many African countries now focus on concurrent partnerships in their HIV prevention programmes. For example, Zambia encourages faithfulness to a single partner as part of its sexual risk reduction programmes.^12,28 Botswana recently embarked on a comprehensive nationwide campaign to reduce concurrency,²⁹ targeting the underlying motivations for engaging in multiple concurrent partnerships, such as a desire for sexual variety or material gain.^30,31 However, the extent to which these programmes will induce changes in concurrent partnership behaviour is uncertain.

Previous behaviour change campaigns in the region have seen reductions in number of sexual partners or increases in monogamous relationships in direct response to mass media campaigns in a variety of settings.³² Studies of various mass media campaigns in sub-Saharan Africa found changes in sexual risk behaviour among individuals within the broadcast area as compared with those outside of the broadcast area: in Zimbabwe, monogamous behaviour among young adults increased by 18% in response to a reproductive health multimedia campaign targeting adolescents;³³ and in Tanzania, there was a 17% reduction in the number of sexual partners reported by men in the previous year in response to an educational radio drama highlighting the consequences of risky sexual behaviour.³⁴

We sought to evaluate the relative effectiveness and cost-effectiveness of concurrency reduction campaigns based on the level and type of behaviour change achieved. We used country-specific data on demographics, HIV prevalence, and sexual behaviour from sub-Saharan Africa. Because the true level of concurrency in these populations is uncertain, we varied the level of concurrency in sensitivity analysis.

METHODS

Overview

We developed a stochastic network model to simulate the course of the HIV epidemic in one-month intervals over a 10-year time horizon in a population of approximately 9000 adults aged 15–49 years (details in Appendix 1, available online only at: http://www.ijsa.rsmjournals.com/cgi/content/full/22/10/558/DC1). The model tracks sexual partnerships, entry into the population, HIV transmission and disease progression, HIV treatment, deaths from HIV and other causes, life years experienced in the population, and HIV incidence and prevalence. We used data on sexual partnership patterns and the current state of the HIV epidemic from four sub-Saharan African countries: Swaziland, Tanzania, Uganda and Zambia (Table 1). We examined four countries in order to determine the impact of concurrency reduction in different demographic and epidemiological settings. We chose these countries because sufficient nationally representative data were available that were generated using a consistent collection methodology across countries, on their demographic structure, HIV epidemic and sexual behaviour (see Appendix 1). We used the model to evaluate several types of behaviour change that might be achieved by a concurrency reduction campaign and the effects that such behaviour change might have on the spread of HIV. We also determined the cost-effectiveness of concurrency reduction programmes as a function of programme cost and the level and type of behaviour change achieved.

Table 1

Model inputs and sources

Variable	Value	95% confidence interval	Source
Population characteristics
Population size	∼9000*
% married			MEASURE DHS^36–39
Swaziland	33%	±0.36%
Tanzania	52%	±0.42%
Uganda	58%	±0.33%
Zambia	58%	±0.68%
Sexual partnerships parameters
Degree distribution – Swaziland (M, F)			MEASURE DHS³⁶
0 partners	39%, 30%	±0.72%, ±0.59%
1 partner	47%, 68%	±0.76%, ±0.60%
2 partners	13%, 2%	±0.33%, ±0.04%
3 partners	0.5%, 0%	±0.02%, ±0.00%
4 partners	0.5%, 0%	±0.02%, ±0.00%
Degree distribution – Tanzania (M, F)			MEASURE DHS³⁷
0 partners	29%, 24%	±0.53%, ±0.44%
1 partner	51%, 64%	±0.65%, ±0.54%
2 partners	16%, 9%	±0.35%, ±0.20%
3 partners	2%, 2%	±0.06%, ±0.03%
4 partners	2%, 1%	±0.05%, ±0.03%
Degree distribution – Uganda (M, F)			MEASURE DHS³⁸
0 partners	28%, 27%	±0.42%, ±0.37%
1 partner	52%, 71%	±0.52%, ±0.38%
2 partners	17%, 2%	±0.30%, ±0.03%
3 partners	2%, 0.2%	±0.05%, ±0.00%
4 partners	1%, 0%	±0.02%, ±0.00%
Degree distribution – Zambia (M, F)			MEASURE DHS³⁹
0 partners	19%, 25%	±0.66%, ±0.42%
1 partner	61%, 73%	±1.01%, ±0.44%
2 partners	15%, 2%	±0.55%, ±0.04%
3 partners	3%, 0%	±0.10%, ±0.00%
4 partners	2%, 0%	±0.09%, ±0.00%
Duration of partnerships (years)			Morris⁸, NACA²⁹
Spousal	10	4.0–19.9^†
Non-spousal	1.5	0.8–2.4
HIV disease parameters
HIV transmission probabilities (per act)			Wawer⁴⁵, Quinn⁷³
Acute phase	0.0082	0.0039–0.0150
Chronic phase	0.0007	0.0006–0.0011
With treatment	0.0001	0.00008–0.00014
Initial HIV prevalence			MEASURE DHS^36–39
Swaziland	26.2%	±0.40%
Tanzania	6.7%	±0.11%
Uganda	5.8%	±0.08%
Zambia	15.3%	±0.26%
CD4 count at HIV treatment initiation	200 cell/μL	N/A	WHO⁵⁴
HIV treatment coverage (of eligible patients)			UNAIDS¹
Swaziland	42%	36–50^‡
Tanzania	31%	26–38%
Uganda	33%	27–40%
Zambia	46%	40–56%
HIV additional mortality			Badri⁴⁹
<200 cells/μL	0.0222	0.0190–0.0271
200–350 cells/μL	0.0066	0.0042–0.0100
>350 cells/μL	0.0042	0.0025–0.0058

*We initialized the network model with a population of 4000 men. We formed partnerships according to the degree distribution for each country.^34–37 The number of women in the network was then determined by these partnerships. For the degree distributions we used, this led to approximately 4900 women in the population (the number varied slightly by country), and thus a population size of approximately 9000 individuals

^†Range found in literature

^‡Low and high estimates

Population characteristics

We stratified the model population by age and gender to match the population demographics of the four study countries.³⁵ At baseline, our infected population was age- and gender-matched to the HIV prevalence in each country, determined from demographic and health survey data.^36–39

Individuals enter the population at age 15 years. We used population pyramids at baseline to determine the number of males and females ageing into the population each month.³⁵ Individuals exit the population at age 50 years (unless involved in a sexual partnership) or due to death. We used age- and gender-specific death rates, adjusted for HIV mortality, to estimate the risk of death among uninfected individuals.^40,41 HIV infected individuals experience an additional mortality risk, determined by their disease state.

Sexual partnership network

We constructed a network model with gender-specific concurrent sexual partnership degree distributions estimated from population surveillance data for the four study countries (Table 1).^36–39 These surveys collected data on the number of sexual partners reported by men and women in the past 12 months, but did not differentiate between concurrent and serial partnerships. We used the number of sexual partners reported in the past 12 months as a surrogate measure for the number of concurrent partnerships. In sensitivity analysis, we considered reduced numbers of concurrent partnerships.

The degree distribution describes the fraction of individuals in the population who have a given number of partnerships. Less than 0.1% of the population in each country reported more than four concurrent sexual partnerships, so we assumed that individuals have at most four sexual partnerships at any time. On average, men report having more sexual partners than women.^36–39 We balanced the reported partnership distribution at baseline by adjusting the female-to-male ratio appropriately. Over the 10-year time horizon, we assumed a fixed degree distribution; that is, the fraction of individuals who are involved in one, two or more concurrent partnerships is stable from month to month.

We modelled two types of partnerships: spousal and non-spousal. The number of spousal partnerships was determined by the proportion of married adults in the general population. On average, spousal partnerships last longer than non-spousal relationships (Table 1). The model captures the formation and dissolution of partnerships each month. Each partnership has a constant monthly probability of dissolution. Spousal partnerships have a lower monthly probability of dissolution than non-spousal partnerships. Partnership dissolution also occurs in the event of the death of a partner. We chose partnership formation probabilities to balance partnership dissolution rates while maintaining the reported gender-specific partnership distributions (see Appendix 1 and reference⁴²)

HIV transmission and progression

The model includes both acute and chronic HIV infection. Acute infection is characterized by high infectivity for a period of three months.^43,44 The subsequent chronic infection phase is characterized by a slow decline in CD4 counts and a lower infectivity.^45–47

Changes in CD4 counts influence an infected individual's risk of HIV-related mortality and of developing an AIDS-defining illness (ADI). We modified a previous model of HIV disease progression to estimate HIV-related mortality as a function of CD4 counts.⁴⁸ As CD4 counts drop, an individual's risk of developing an ADI and HIV-related mortality increases.^49,50 Individuals become eligible for treatment when their CD4 counts fall below 200 cells/μL or when they experience an ADI. Country-specific coverage of antiretroviral therapy (ART) determines the fraction of eligible individuals who ultimately receive treatment.¹ When an infected individual receives effective ART, CD4 counts rise as a function of the CD4 count nadir and time on treatment.^51–53 However, if treatment fails, CD4 counts again decline. We modelled treatment options based on World Health Organization (WHO) guidelines, and treatment failure rates based on clinical experience in southern Africa.^54,55

The probability of acquiring HIV infection depends on the HIV disease state of the infected partner and the level of sexual activity. The risk of infection per sexual act is substantially higher during acute infection than during chronic infection with and without effective treatment (Table 1).⁴⁷ We estimated the level of sexual activity by using the number of sexual acts per month described in a cohort of discordant couples, and then adjusted these values to calibrate the model.⁴⁷

Model calibration

We calibrated our model to projected HIV prevalence trends from the UNAIDS Spectrum Package.^1,56 We chose HIV prevalence as a calibration measure because of the availability of longitudinal HIV prevalence data that have been obtained with a consistent methodology over time. In the four study countries, HIV prevalence is estimated to be stable or in slight decline.¹ To match projected country-specific HIV prevalence trends, we adjusted the monthly risk of HIV infection per partnership (Table 2). These monthly infection risks differ by country and implicitly reflect potential differences in sexual behaviour across countries, such as differences in coital frequency, condom use and prevalence of male circumcision.

Table 2

Model calibration parameters and measures

	Swaziland	Tanzania	Uganda	Zambia
Calibration parameter
HIV monthly transmission probabilities (per partnership)
Acute phase	0.34	0.18	0.25	0.25
Chronic phase	0.029	0.015	0.021	0.022
With treatment	0.0041	0.0022	0.0030	0.0031
Calibration measures*
Change in HIV prevalence (%)
Data, 2004–2007¹ (low and high estimates indicated)	−0.4% (−2.5%, 1.7%)	−0.3% (−1.0%, 0.5%)	−0.9% (−2.1%, −0.1%)	0.1% (−2.1%, 2.2%)
Model prediction, 2004–2007 (95% CIs)	−0.9% (−1.8%, −0.2%)	−0.1% (−0.7%, 0.4%)	−0.2% (−0.7%, 0.2%)	−1.0% (−1.6%, −0.3%)
Number of partners
Data^34–37
Married (%) (95% CIs)	33% (32%, 34%)	52% (51%, 53%)	58% (57%, 59%)	58% (56%, 60%)
More than 1 concurrent partner (95% CIs)	14% (13%, 15%)	20% (19%, 21%)	20% (19%, 21%)	20% (18%, 22%)
Model prediction
Married (%) (95% CIs)	31% (30%, 32%)	52% (51%, 53%)	58% (58%, 59%)	57% (57%, 58%)
More than 1 concurrent partner (95% CIs)	13% (12%, 13%)	20% (19%, 20%)	21% (20%, 21%)	19% (19%, 20%)

CI = confidence interval

*The monthly probabilities of HIV transmission were adjusted to match country-specific HIV prevalence trends. Other measures, such as number of married and unmarried partners, are also well-matched

Behaviour change scenarios

A campaign to reduce concurrent partnerships may achieve different changes in sexual behaviour, depending on the campaign message and target populations. We considered the following types of behaviour change that might be achieved:

Increased monogamy: individuals with more than one partnership change to having at most one partnership at any one time. This may be achieved by promoting faithfulness to a single partner;

High-risk partnership reduction: individuals with the highest number of concurrent partnerships reduce their number of partners, but do not necessarily become monogamous. This may be achieved by a campaign with a partnership reduction message targeting individuals with (or behaviours leading to) a high number of concurrent partnerships;

Untargeted partnership reduction: individuals with more than one partnership reduce their number of sexual partners. This may be achieved by a broad partnership reduction campaign targeting all individuals with concurrent partnerships.

We compared the relative effectiveness of concurrency reduction under these behaviour change scenarios to answer two important questions: which type of behaviour change achieves the greatest benefits and how much better is the most effective behaviour change compared with other (and perhaps more realistic) behaviour changes? For example, it may be more difficult to change the behaviour of high-risk individuals. We sought to determine how much benefit is gained if high-risk individuals are targeted and, therefore, how important it is to find effective behaviour change messages for such individuals.

All behaviour change scenarios were compared against a baseline projection of the HIV epidemic in the absence of any changes in concurrent sexual partnerships (status quo). Because the effectiveness of concurrent partnership reduction campaigns is unknown, we varied the programme effectiveness for each of the scenarios from 0% to 50%. We measured effectiveness as the percentage of non-monogamous individuals who become monogamous under the increased monogamy scenario. We compared the results of the other scenarios by removing the same number of partnerships. For example, 10% effectiveness for each of the scenarios implies that the number of concurrent partnerships removed is equal to 10% of the number of partnerships removed under increased monogamy. If an increased monogamy programme that is 10% effective involves a reduction of 200 partnerships, then 10% effectiveness in each of the other scenarios involves 200 fewer partnerships.

Analyses

For each of the three behaviour change scenarios, we estimated the number of HIV infections averted over a 10-year time horizon as a function of programme effectiveness. We simulated each behaviour change scenario 500 times to obtain stable estimates. We express a reduction in the number of new HIV infections as a percentage reduction compared with the status quo.

We also examined the cost-effectiveness of concurrency reduction using cost estimates of behaviour change campaigns implemented in Uganda and Botswana. Uganda's ‘Zero Grazing’ campaign was estimated to cost approximately US$0.23 per person annually,¹⁸ while Botswana's campaign targeting multiple concurrent partnerships has a budget of over US$1 per person annually (T Kasper, personal communication, 8 August 2009). We estimated the cost per HIV infection averted for the increased monogamy scenario as a function of behaviour change and annual campaign costs per person.

RESULTS

Status quo

The status quo was associated with an estimated average drop in HIV prevalence of 1.0% (0.3–1.9%) in the four study countries over the next 10 years, consistent with UNAIDS estimates of the prevalence trends.⁵⁶ We estimated an average HIV incidence of 2.3% (1.1–4.3%) in the first year and 1.5% (0.7–2.9%) after 10 years, consistent with recent estimates.^57,58

New HIV infections

Each modelled behaviour change scenario resulted in fewer HIV infections compared with the status quo (Figure 1). Across all four study countries and for any level of behaviour change, the high-risk partnership reduction scenario resulted in the greatest reduction in new infections. For example, a programme targeting high-risk individuals with 10% effectiveness led to an average decrease in new infections of 11.7% (7.9–14.2%) over 10 years in the four study countries, as compared with average decreases of 8.7% (6.6–10.0%) and 8.9% (6.8–10.0%) in new infections under the increased monogamy and untargeted partnership reduction scenarios, respectively. The reduction in new infections for each country is shown in Figure 1 for a range of programme effectiveness from 0% to 50%.

Figure 1

The effect of behaviour change (percentage reduction in concurrent partnerships) on reducing new HIV infections. The figure shows the estimated percentage reduction in the number of new HIV infections (compared with the Status Quo) for a range of campaign effectiveness from 0% to 50% under the different behaviour change scenarios (Increased Monogamy, High-Risk Partnership Reduction and Untargeted Partnership Reduction) for the four study countries, Swaziland (a), Tanzania (b), Uganda (c) and Zambia (d). The 95% confidence intervals are indicated for each sample point

Cost-effectiveness

We estimated the cost per HIV infection averted for a single scenario (increased monogamy) in each study country as a function of behaviour change achieved and annual programme cost (Figure 2). Using an estimate of US$1200–3800 for the lifetime medical cost of caring for an HIV-infected individual in Africa,^59–62 our analysis suggests that concurrency reduction is cost-saving under a broad range of cost and effectiveness assumptions. For example, according to model predictions, a programme that costs US$0.25 per person annually (similar to Uganda's ‘Zero Grazing’ campaign)¹⁸ would need a programme effectiveness of less than 5% in any of the study countries to be cost-saving, while a programme that costs US$1 per person annually (similar to Botswana's campaign) would need to be 9% (<5–15%) effective on average. A programme that costs US$2 per person annually would need to be 20% (10–35%) effective in order to be cost-saving, and less than 5% (<5–12%) effective to cost less than US$5000 per infection averted.

Figure 2

Cost per HIV infection averted for the increased monogamy scenario. This figure shows the cost per HIV infection averted for the increased monogamy scenario, as a function of annual programme costs (measured as $US/person/year) and programme effectiveness (measured as the percentage reduction in concurrency). Costs and infections averted are discounted to the present at 3% per year. Results are shown for each of the four study countries, Swaziland (a), Tanzania (b), Uganda (c) and Zambia (d). The 95% confidence intervals are indicated for each sample point

Sensitivity analysis

Our primary analysis includes sensitivity analysis on a range of concurrency, demographic and epidemic patterns in sub-Saharan Africa, as well as the impacts of different levels of behaviour change up to a programme effectiveness of 50%. In sensitivity analysis, we considered the case where concurrency is completely eliminated (100% behaviour change). In this case, the model predicts an average decrease in new infections of 54.8%.

We relied upon sexual behaviour surveys to estimate concurrency behaviour in the study countries. However, different survey methodologies can lead to different estimates of numbers of sexual partners.⁶³ The sexual behaviour survey data used in this analysis provided information on the number of sexual partners in the past 12 months reported by men and women but did not differentiate between concurrent and serial partnerships. These surveys may therefore overestimate concurrency behaviour. Over-estimating the number of concurrent partnerships in the study countries may overstate the benefits of concurrency reduction. We therefore conducted a sensitivity analysis in which we assumed that these surveys over-estimated concurrency by 50%. With levels of concurrency that are half as high as we estimated in the base case, we found that fewer infections are averted under the three behaviour change scenarios as compared with the status quo (e.g. 30% versus 45% reduction in new infections in Tanzania for a 50% reduction in concurrency) with a greater cost per HIV infection averted (e.g. US$2400 versus US$1400 per HIV infection averted in Tanzania for a programme that is 10% effective and costs US$1 per person annually). However, the rank ordering of the different types of behaviour change was unaffected (see Figure 6, available online only at: http://www.ijsa.rsmjournals.com/cgi/content/full/22/10/558/DC1).

In sexual behaviour survey data, men and women typically report different numbers of partners. In a heterosexual population, the total number of partnerships engaged in by men should equal the total number of partnerships engaged in by women. To resolve this discrepancy, in our base-case analyses we adjusted the male-to-female ratio appropriately and left the reported gender-specific partnership distributions unchanged. For two of the study countries, this ratio was quite different than that estimated from demographic data (1:1.3 versus 1:1 in Uganda and 1:1.4 versus 1:1 in Zambia; Table 3, available online only at: http://www.ijsa.rsmjournals.com/cgi/content/full/22/10/558/DC1). In sensitivity analysis, we explored the effects of fixing the male-to-female ratio to 1:1 and then averaging the male- and female-specific partnership distributions and applying this average distribution to both men and women. With an average partnership distribution, slightly fewer infections are averted for each of the three behaviour change scenarios as compared with the status quo (e.g. 30% versus 35% reduction in new infections in Zambia for a 50% reduction in concurrency). However, the rank ordering of the different types of behaviour change was again unchanged (Figure 7, available online only at: http://www.ijsa.rsmjournals.com/cgi/content/full/22/10/558/DC1).

We also varied a number of other parameters (e.g. rate of condom use in non-spousal partnerships, duration of non-spousal partnerships, level of access to HIV treatment and the CD4 count threshold for HIV treatment eligibility). Changes in the duration of non-spousal partnerships, level of access to HIV treatment and HIV treatment eligibility criteria had little effect on the number of infections averted through concurrency reduction. Reducing the risk of HIV transmission in non-spousal partnerships (e.g. increasing condom use) resulted in fewer infections averted for a given level of concurrency reduction; however, over all parameter ranges considered in sensitivity analysis, the rank ordering of the different types of behaviour change in averting HIV infections was unchanged (Table 4, available online only at http://www.ijsa.rsmjournals.com/cgi/content/full/22/10/558/DC1).

DISCUSSION

Using a stochastic network model based on data from four representative sub-Saharan African countries with a large HIV burden, we provide the first estimate of the epidemiological benefits and the cost-effectiveness of changes in sexual behaviour that may arise from concurrent partnership reduction campaigns. Our analysis suggests that reducing concurrent partnerships among high-risk individuals would result in the greatest reduction in new HIV infections compared with the status quo. However, all three of the modelled behaviour change scenarios achieved relatively similar reductions in new HIV infections. In contrast to previous work, which argues that concurrency reduction without targeting the most connected individuals will fail,⁶⁴ our results suggest that the message and target population matter less than the extent of partnership reduction.

Concurrent sexual partnerships increase the rate of HIV transmission by creating a connected network, and high-risk individuals are the most important links in such a network. This explains our finding that reducing the number of partnerships among the highest risk individuals has the greatest impact on HIV transmission. However, reaching the most promiscuous individuals and changing their behaviours is often difficult. Our analyses suggest that other forms of concurrency reduction provide relatively similar benefits, without the need to specifically target groups that may be hard to reach. These findings lend support to previous suggestions that reaching all members of a sexual network, not just the most promiscuous, is of great importance in preventing HIV.⁶⁵

Campaigns explicitly targeting concurrent partnership reduction are relatively new and their effectiveness has yet to be measured. However, our analyses suggest that if campaigns can achieve at least modest reductions in concurrency at similar costs as other mass media campaigns, they are likely to be cost-effective and even cost-saving. For example, given our baseline estimates of concurrency, our model predicts that a campaign that costs US$1 per person annually need only reduce concurrent partnerships by between less than 5–15% in the study countries to be cost-saving, when comparing the cost per HIV infection averted to the lifetime medical costs of an HIV infection. For a campaign that is 10% effective in reducing concurrent partnerships and that costs US$1 per person annually, we estimate that the cost per HIV infection averted ranges from US$582 to US$1583 in the study countries considered. This places concurrency reduction on par with other established HIV prevention interventions in Africa: the costs per HIV infection averted for male circumcision in South Africa and prevention of mother-to-child transmission are US$181 and US$20–2198, respectively.^61,66–68

We estimate that the maximal potential reduction in new infections due to concurrency reduction (i.e. if there are no concurrent partnerships) is about 55%, corresponding to 0.2, 2.0, 1.5 and 1.4 million infections averted in Swaziland, Tanzania, Uganda and Zambia, respectively, over 10 years. While this is a substantial reduction in new infections, it highlights the importance of implementing complementary HIV prevention approaches. Even complete elimination of multiple concurrent sexual partnerships cannot stop HIV transmission.

Our analysis focused on the benefits of concurrency reduction for HIV prevention. However, programmes that reduce concurrency may also generate additional benefits, such as reduction in the incidence of other STIs. Targeting concurrent sexual partnerships may also have social benefits such as reducing intergenerational sex, a phenomenon that has likely contributed to the feminization of the HIV epidemic in Africa, and reducing domestic violence, which is commonly associated with having multiple sexual partners.^11,69,70

The results presented here are contingent on the prevalence of concurrency in each of the study countries. Our estimates of concurrency behaviour relied on limited data. We assumed that sexual behaviour survey data on the number of sexual partners reported in the past 12 months could be used as a surrogate for the concurrency behaviour in the population. This assumption may over-estimate the true concurrency behaviour. At the same time, sexual behaviour surveys may underestimate true behaviour patterns. While discussions are ongoing about how best to measure concurrency using survey methods,⁷¹ sexual behaviour surveys have yet to systematically incorporate these recommendations. In an effort to quantify the importance of this limitation on our results, we varied the prevalence of concurrency in sensitivity analysis to account for potential over-estimation. We found that the rank ordering of the behaviour change scenarios was unchanged, though the expected number of infections averted decreased as the prevalence of concurrency was reduced and the cost per infection averted increased. This highlights the need for better estimates of baseline sexual behaviour, including the prevalence of concurrent partnerships, when considering investments in campaigns to change sexual behaviours.

Our analysis has several additional limitations. First, we assessed the relative effectiveness of different types of behaviour change, but do not comment on the feasibility of such behaviour changes. These considerations may vary by country. Second, our prevalence estimates depend on HIV mortality, which may change as treatment coverage changes in sub-Saharan Africa. Finally, our analysis was limited by the data informing the network structure in our model. We based the sexual partnership network structure for each of the four study countries on country-specific survey data. These data only provided us with partnership distributions, and we assumed non-preferential mixing (lacking data informing us otherwise), though populations typically do exhibit some kind of assortative mixing.⁷² Mixing patterns induced by cross-generational partnerships, in particular, could prove extremely important in the context of concurrent partnership reduction campaigns. Future work characterizing these relationships would greatly help in evaluating the impact of concurrency reduction on the course of the epidemic and could inform an extension of the current model incorporating preferential mixing patterns.

The message(s) that a concurrency reduction campaign chooses to convey have important implications. The target population may change, the cultural background may affect the message's effectiveness and the cost of reaching the target population may vary. Our analysis suggests that the beneficial effect of concurrency reduction is only minimally dependent on the target population and message, and greatly dependent on the achieved reduction in concurrency.

Moreover, our analysis suggests that concurrency reduction campaigns can be cost-effective, or even cost-saving, under a broad range of plausible cost and effectiveness assumptions, and even if the level of concurrency is half as high as we estimated in our base case. This lends further support to the potential importance of these campaigns as a means of controlling HIV in sub-Saharan Africa.

Footnotes

ACKNOWLEDGEMENTS

This research was funded by the National Institute on Drug Abuse, grant number R01-DA15612. E A Enns is supported by a National Defense Science and Engineering Graduate Fellowship, a National Science Foundation Graduate Fellowship and a Rambus Stanford Graduate Fellowship. T K Igeme was supported by a Stanford Undergraduate Research Fellowship. E Bendavid is supported by the National Institute of Allergy and Infectious Diseases (K01-AI084582). We appreciate the help of Tobias Kasper in providing important input on Botswana's campaign to reduce concurrent sexual partnerships. The funding agencies had no part in the study design, the collection of data, the analysis, or the conclusions. No medical writers or editors were used for this manuscript.

References

UNAIDS. Report on the Global AIDS Epidemic. Geneva: Joint United Nations Programme on HIV/AIDS, 2008

Morris

, Epstein

, Wawer

. Timing is everything: international variations in historical sexual partnership concurrency and HIV prevalence. PLos One 2010;5:e14092

Lagarde

, Auvert

, Caraël

, Concurrent sexual partnerships and HIV prevalence in five urban communities of sub-Saharan Africa. AIDS 2001;15:877

Carter

, Kraft

, Koppenhaver

, ‘A bull cannot be contained in a single kraal’: concurrent sexual partnerships in Botswana. AIDS Behav 2007;11:822–30

Morris

, Wawer

, Podhisita

. The Thailand and Ugandan sexual network studies. In: Morris

, ed. Network Epidemiology: A Handbook for Survey Design and Data Collection. Oxford: Oxford University Press, 2004

Morris

, Kretzschmar

. Concurrent partnerships and the spread of HIV. AIDS 1997;11:641–8

Kretzschmar

, Morris

. Measures of concurrency in networks and the spread of infectious disease. Math Biosci 1996;133:165–95

Morris

, Kretzschmar

. A microsimulation study of the effect of concurrent partnerships on the spread of HIV in Uganda. Math Popul Stud 2000;8:109–33

Pilcher

, Tien

, Eron

Jr , Brief but efficient: acute HIV infection and the sexual transmission of HIV. J Infect Dis 2004;189:1785–92

10.

Gourvenac

, Taruberekera

, Mochaka

, Kasper

. Multiple Concurrent Partnerships Among Men and Women Aged 15–34 in Botswana. Gaborone, Botswana: PSI-Botswana, 2007

11.

Jana

, Nkambule

, Tumbo

. One Love: Multiple and Concurrent Sexual Partnerships in Southern Africa. Johannesburg, South Africa: Soul City Institute for Health & Development Communications, 2008

12.

Serlemitsos

. Mass media campaigns on multiple concurrent sexual partnerships (MCP) in Zambia. In: Conference on Addressing Multiple and Concurrent Partnerships in Southern Africa. Developing Guidance for Bold Action. Gaborone, Botswana, 2009

13.

Mah

, Halperin

. Concurrent sexual partnerships and the HIV epidemic in Africa: evidence to move forward. AIDS Behav 2008;14:11–16

14.

Shelton

. Why multiple sex partners? Lancet 2009;374:367–69

15.

Wilson

, Halperin

. ‘Know your epidemic, know your response’: a useful approach, if we get it right. Lancet 2008;372:423–26

16.

Halperin

, Epstein

. Concurrent sexual partnerships help to explain Africa's high HIV prevalence: implications for prevention. Lancet 2004;364:4–6

17.

Potts

, Halperin

, Kirby

, Public health. Reassessing HIV prevention. Science 2008;320:749–50

18.

Low-Beer

. This is a routinely avoidable disease. Financial Times, 28 November 2003

19.

Ciantia

, Orach

, Pariyo

HIV prevention conundrum: did the Pope have a case?

J Med Pers 2009;7:63–69

20.

Green

, Nantulya

, Stoneburner

, Stover

. What Happened in Uganda? Declining HIV Prevalence, Behavior Change, and the National Response. Washington, DC: USAID, 2002

21.

Green

, Halperin

, Nantulya

, Hogle

. Uganda's HIV prevention success: the role of sexual behavior change and the national response. AIDS Behav 2006;10:335–46

22.

Shelton

, Halperin

, Nantulya

, Partner reduction is crucial for balanced ‘ABC’ approach to HIV prevention. BMJ 2004;328:891–93

23.

Hallett

, Aberle-Grasse

, Bello

, Declines in HIV prevalence can be associated with changing sexual behaviour in Uganda, urban Kenya, Zimbabwe, and urban Haiti. Sex Transm Infect 2006;82(Suppl 1):1–8

24.

Stoneburner

, Low-Beer

. Population-level HIV declines and behavioral risk avoidance in Uganda. Science 2004;304:714–18

25.

Halperin

, Mugurungi

, Hallett

A surprising prevention success: why did the HIV epidemic decline in Zimbabwe?

PLoS Med 2011;8:e1000414

26.

Gregson

, Garnett

, Nyamukapa

, HIV decline associated with behavior change in eastern Zimbabwe. Science 2006;311:664–66

27.

Gregson

, Gonese

, Hallett

, HIV decline in Zimbabwe due to reductions in risky sex? Evidence from a comprehensive epidemiological review. Int J Epidemiol 2010;39:1–13

28.

HIV/AIDS/STI/TB Council of Zambia. National HIV/AIDS Communication Strategy. See http://zambia.jhuccp.org/resources/nac_pubs/NationalHIVAIDSCommunicationStrategy.pdf (last checked 1 April 2009)

29.

National AIDS Coordinating Agency. National Campaign Plan: Multiple Concurrent Partnerships. See http://www.psiwash.org/resources/pubs/National%20MCP%20strategy.pdf (last checked 1 March 2009)

30.

National AIDS Coordinating Agency. Botswana Campaign on MCP. In: UNAIDS/Soul City Regional Meeting on Multiple Concurrent Partnerships . Johannesburg, South Africa, 2008

31.

Matlhare

. An overview of National MCP Campaign Botswana. In: Conference on Addressing Multiple and Concurrent Partnerships in Southern Africa. Developing Guidance for Bold Action . Gaborone, Botswana, 2009

32.

Bertrand

, O'Reilly

, Denison

, Systematic review of the effectiveness of mass communication programs to change HIV/AIDS-related behaviors in developing countries. Health Educ Res 2006;21:567–97

33.

Kim

, Kols

, Nyakauru

, Promoting sexual responsibility among young people in Zimbabwe. Int Fam Plan Perspect 2001;27:11–19

34.

Vaughan

, Rogers

, Singhal

, Swalehe

. Entertainment-education and HIV/AIDS prevention: a field experiment in Tanzania. J Health Comm 2000;5(Suppl):81–100

35.

US Census Bureau. International Data Base. See http://www.census.gov/ipc/www/idb/tables.html (last checked 29 January 2009)

36.

MEASURE DHS. Swaziland Demographic and Health Survey. See http://www.measuredhs.com/countries/country_main.cfm?ctry_id=220&c=Swaziland (last checked 1 January 2009)

37.

MEASURE DHS. Tanzania Demographic and Health Survey . See http://www.measuredhs.com/countries/country_main.cfm?ctry_id=39&c=Tanzania (last checked 1 January 2009)

38.

MEASURE DHS. Uganda Demographic and Health Survey. See http://www.measuredhs.com/countries/country_main.cfm?ctry_id=44&c=Uganda (last checked 1 January 2009)

39.

MEASURE DHS. Zambia Demographic and Health Survey. See http://www.measuredhs.com/countries/country_main.cfm?ctry_id=47&c=Zambia (last checked 1 January 2009)

40.

World Health Organization. Life Tables for WHO Member States. See http://www.who.int/whosis/database/life_tables/life_tables.cfm (last checked 1 November 2008)

41.

Marston

, Todd

, Glynn

, Estimating ‘net’ HIV-related mortality and the importance of background mortality rates. AIDS 2007;21(Suppl 6):S65–71

42.

Enns

, Brandeau

. Inferring network structure and dynamics in network-based disease simulation. Health Care Manag Sci 2011;14:174–88

43.

Vergis

, Mellors

. Natural history of HIV-1 infection. Infect Dis Clin North Am 2000;14:809–25

44.

Hollingsworth

, Anderson

, Fraser

. HIV-1 transmission, by stage of infection. J Infect Dis 2008;198:687–93

45.

Wawer

, Gray

, Sewankambo

, Rates of HIV-1 transmission per coital act, by stage of HIV-1 infection, in Rakai, Uganda. J Infect Dis 2005;191:1403–09

46.

McCormick

, Walensky

, Lipsitch

, The effect of antiretroviral therapy on secondary transmission of HIV among men who have sex with men. Clin Infect Dis 2007;44:1115–22

47.

Gray

, Wawer

, Brookmeyer

, Probability of HIV-1 transmission per coital act in monogamous, heterosexual, HIV-1-discordant couples in Rakai, Uganda. Lancet 2001;357:1149–53

48.

Bendavid

, Young

, Katzenstein

, Cost-effectiveness of HIV monitoring strategies in resource-limited settings – a southern African analysis. Arch Intern Med 2008;168:1910–18

49.

Badri

, Lawn

, Wood

. Short-term risk of AIDS or death in people infected with HIV-1 before antiretroviral therapy in South Africa: a longitudinal study. Lancet 2006;368:1254–59

50.

Holmes

, Wood

, Badri

, CD4 decline and incidence of opportunistic infections in Cape Town, South Africa: implications for prophylaxis and treatment. J Acquir Immune Defic Syndr 2006;42:464–69

51.

Lawn

, Myer

, Bekker

, Wood

. CD4 cell count recovery among HIV-infected patients with very advanced immunodeficiency commencing antiretroviral treatment in sub-Saharan Africa. BMC Infect Dis 2006;6:59

52.

Kaufmann

, Perrin

, Pantaleo

, CD4 T-lymphocyte recovery in individuals with advanced HIV-1 infection receiving potent antiretroviral therapy for 4 years: the Swiss HIV Cohort Study. Arch Intern Med 2003;163:2187–95

53.

Kaufmann

, Furrer

, Ledergerber

, Characteristics, determinants, and clinical relevance of CD4 T cell recovery to <500 cells/microL in HIV type 1-infected individuals receiving potent antiretroviral therapy. Clin Infect Dis 2005;41:361–72

54.

World Health Organization. Antiretroviral Therapy For HIV Infection in Adults And Adolescents: Recommendations for a Public Health Approach. Geneva: WHO, 2006

55.

Orrell

, Harling

, Lawn

, Conservation of first-line antiretroviral treatment regimen where therapeutic options are limited. Antivir Ther 2007;12:83–88

56.

Stover

, Walker

, Grassly

, Marston

. Projecting the demographic impact of AIDS and the number of people in need of treatment: updates to the Spectrum projection package. Sex Transm Infect 2006;82(Suppl 3):ii45–iii50

57.

Matovu

JKB

, Gray

, Kiwanuka

, Repeat voluntary HIV counseling and testing (VCT), sexual risk behavior and HIV incidence in Rakai, Uganda. AIDS Behav 2007;11:71–78

58.

Shelton

, Halperin

, Wilson

. Has global HIV incidence peaked? Lancet 2006;367:1120–22

59.

Stover

, Bertozzi

, Gutierrez

J-P

, The global impact of scaling up HIV/AIDS prevention programs in low- and middle-income countries. Science 2006;311:1474–76

60.

Cleary

, Boulle

, McIntyre

, Coetzee

. Cost-Effectiveness of Antiretroviral Treatment for HIV-Positive Adults in a South African Township. Cape Town: Health Systems Trust, 2004

61.

Kahn

, Marseille

, Auvert

. Cost-effectiveness of male circumcision for HIV prevention in a South African setting. PLoS Med 2006;3:2349–58

62.

Yazdanpanah

, Losina

, Anglaret

, Clinical impact and cost-effectiveness of co-trimoxazole prophylaxis in patients with HIV/AIDS in Cote d'Ivoire: a trial-based analysis. AIDS 2005;19:1299–308

63.

LePont

, Pech

, Boelle

, A new scale for measuring dynamic patterns of sexual partnership and concurrency: application to three French Caribbean regions. Sex Transm Dis 2003;30:6–9

64.

Liljeros

, Edling

, Amaral

LAN

, The web of human sexual contacts. Nature 2001;411:907–08

65.

Jones

, Handcock

. Social networks: sexual contacts and epidemic thresholds. Nature 2003;423:605–06

66.

Sweat

, Gregorich

, Sangiwa

, Cost-effectiveness of voluntary HIV-1 counselling and testing in reducing sexual transmission of HIV-1 in Kenya and Tanzania. Lancet 2000;356:113–21

67.

Soderlund

, Zwi

, Kinghorn

, Gray

. Prevention of vertical transmission of HIV: analysis of cost effectiveness of options available in South Africa. BMJ 1999;318:1650–56

68.

Wilkinson

, Floyd

, Gilks

. Antiretroviral drugs as a public health intervention for pregnant HIV-infected women in rural South Africa: an issue of cost-effectiveness and capacity. AIDS 1998;12:1675–82

69.

Hallett

, Gregson

, Lewis

JJC

, Behaviour change in generalised HIV epidemics: impact of reducing cross-generational sex and delaying age at sexual debut. BMJ 2007;83(Suppl 1):50–54

70.

Andersson

, Ho-Foster

, Mitchell

, Risk factors for domestic physical violence: national cross-sectional household surveys in eight southern African countries. BMC Womens Health 2007;7:11

71.

UNAIDS Reference Group on Estimates, Modelling and Projections. Consultation on Concurrent Sexual Partnerships. Nairobi: UNAIDS, 2009

72.

Garnett

, Hughes

, Anderson

, Sexual mixing patterns of patients attending sexually transmitted diseases clinics. Sex Transm Dis 1996;23:248–57

73.

Quinn

, Wawer

, Sewankambo

, Viral load and heterosexual transmission of human immunodeficiency virus type 1. Rakai Project Study Group. N Engl J Med 2000;342:921–29