AIDS and Behavior

, Volume 16, Issue 7, pp 1746–1752

Is Concurrency Driving HIV Transmission in Sub-Saharan African Sexual Networks? The Significance of Sexual Partnership Typology

Authors

    • Julius Centre for Health Sciences and Primary CareUniversity Medical Centre Utrecht
    • Centre for Infectious Disease ControlRIVM
  • Michel Caraël
    • Department of Social SciencesFree University of Brussels
Commentary

DOI: 10.1007/s10461-012-0254-6

Cite this article as:
Kretzschmar, M. & Caraël, M. AIDS Behav (2012) 16: 1746. doi:10.1007/s10461-012-0254-6

Abstract

Recently, there has been debate about the role of concurrent partnerships in driving the transmission of HIV, particularly in Southern Africa, where HIV prevalence is up to 25 % in many heterosexual populations and where evidence from sexual behavior surveys also suggests high levels of male concurrency. While mathematical modeling studies have shown that concurrency has the potential to enhance the speed at which HIV spreads in a population, empirical studies up to now have failed to provide conclusive evidence supportive of these effects. Here we discuss some reasons for the apparent discrepancy between theoretical and empirical studies. We propose that studying the impact of concurrency on HIV transmission should be differentiated by taking more insight from social and behavioral studies on sexual partnerships into account. We also suggest that a more rigorous definition is needed for when a factor is considered a driving force for HIV epidemic spread. We illustrate this with a modeling example.

Keywords

Concurrent partnershipsHIV transmissionSub-Saharan AfricaSexual networksMathematical modelsBasic reproduction number

Resumen

Reciéntemente se ha debatido el rol que las parejas concurrentes tienen en impulsar la transmisión de VIH, particularmente en Sudáfrica, donde la prevalencia en muchas poblaciones heterosexuales es de hasta 25 % y la evidencia de comportamiento sexual también sugiere altos niveles de concurrencia en hombres. Mientras que estudios con modelos matemáticos muestran que la concurrencia tiene el potencial de aumentar la velocidad de propagación de VIH en la población, hasta ahora estudios empíricos no han logrado mostrar evidencia concluyente sobre estos efectos. Aquí discutimos algunas razones de la aparente discrepancia entre los estudios teóricos y empíricos. Proponemos que estudios sobre el impacto de la concurrencia en la transmisión de VIH se deberían diferenciar considerando mayor entendimiento proveniente de estudios sociales y conductuales sobre parejas sexuales. También proponemos que se necesita una definición más rigurosa para cuando un factor se considera fuerza impulsora de propagación de una epidemia de VIH. Empleamos un modelo para ilustrar un ejemplo.

Introduction

The ongoing debate about whether partnership concurrency is a driving force of HIV transmission in sub-Saharan Africa (SSA), in particular in the hyper epidemic regions of Southern Africa, unveils the intrinsic difficulties of traditional epidemiological methods to deal with network phenomena. Similarly, it demonstrates the difficulties to adequately anchor mathematical modeling approaches in sexual network structure as observed in real societies. The concurrency debate provides an excellent example of the role mathematical modeling can and should play in infectious disease epidemiology, namely to inspire new ways of thinking about risks, transmission and possibly intervention. However, empirical evidence of the effectiveness of interventions has to be obtained from field studies. Here we highlight some issues that need to be addressed in order to clarify the role of partnership concurrency in HIV transmission dynamics.

Why has it proven so difficult to empirically test the seemingly simply hypothesis of concurrency being a main driver of HIV transmission? One reason lies in the gap between the simplicity of the hypothesis as an abstract modeling result and the complexity of factors and forces that determine real sexual network structure. The strength and at the same time the weakness of mathematical models are their ability to abstract from reality and in that way study the effects of specific mechanisms on epidemic spread in isolation. However, identifying the effects of these mechanisms on real behaviors and transmission dynamics is difficult, because many factors interact to shape the observed epidemiological patterns and it is often not possible to isolate the effects of one. This is especially true in dynamic sexual networks that link individuals through different types of partnerships (e.g. marriages and extramarital short term partnerships) into larger connected components. It is inherent in networks that effects on the population level may not be directly derived from a sum of individual behaviors, but may be an emerging phenomenon visible only on the population level [1]. Other reasons for the difficulty in linking HIV transmission to concurrency in empirical studies are problems of defining quantities that can readily be measured in the field, and difficulty in measuring HIV incidence directly.

The concurrency hypothesis is conceptually simple as was the model used in [2]. This model took only idealized features of sexual behavior into account by assuming that all sexual relationships have the same dissolution rate, equal rates of sexual intercourse during a relationship, and tendency to form concurrent relationships. The conclusions drawn about the impact of concurrency are strong and are convincing because the model was stripped of all complexity and succeeded in isolating a causal relationship between concurrency and HIV transmission dynamics. However, by isolating this mechanism from others that influence transmission dynamics, it is not possible to quantify how important its contribution is in relation to those other mechanisms that were not included in the model. Furthermore, isolation of such a mechanism is not easily possible in empirical studies.

As Lurie and Rosenthal have convincingly argued [3, 4] the role of concurrency in driving the HIV epidemic in this region with hyper epidemics has not been proven empirically. The reasons are multiple: the definition of concurrency is not standardized and therefore different things are measured in different studies. The impact of concurrency on transmission dynamics is likely to vary at different stages of the HIV epidemics. Other critical co-factors of the sexual transmission of HIV such as male circumcision, condom use, and other STIs may have masked other transmission factors associated with sexual partnerships. Recording sexual network and sexual behavior data is fraught with many difficulties such as non-response, refusal, social desirability and memory biases. Finally, concurrency is usually investigated as an isolated phenomenon without relationship with other behavioral factors and cultural embedding of sexual relationships. Sexuality in all societies is part of the culture with institutions and norms [5]. A social institution such as for example polygyny is a structure that has many other implications [6] such as early first marriage for girls, late marriage for men, short time before remarriage for widows or separated women, sexual post-partum abstinence for breastfeeding mothers [7], and emotional distance between spouses [8]. All of these impact the structure and dynamics of the sexual network and cannot be viewed in isolation. For example, in Rakai, Uganda, long-term concurrent polygamous relationships reduce HIV risk, perhaps due to closed sexual networks [9]. A recent ecological study in a range of SSA countries and regions showed that polygyny is associated with lower HIV prevalence [10, 11].

In the following, we discuss some reasons for the difficulties of quantifying the possible contribution of concurrent partnerships to the spread of HIV, the need for a classification of types of concurrent partnerships and a possible definition for when a factor is a “driving force” for an epidemic.

Individual Level Versus Population Level

Concurrency is necessarily first defined as a behavioral characteristic of an individual person, namely as having more than one partnerships simultaneously (whatever the duration of overlap). This behavior per se does not put the individual at a higher risk of acquiring HIV, because it does not matter from whom infection is acquired. However, concurrent partnerships put an individual’s partners at increased risk once this person becomes infected. Therefore, concurrency is a risk behavior that increases the risk for persons other than the one who is practicing the behavior [12]. To determine who is at increased risk, we have to investigate the local networks of infected persons. This is in contrast with traditional epidemiological methods, which basically view individuals sampled from a population as independent entities, such that statistical methods can be applied assuming independence of measurements. Those methods cannot deal with a situation where risk behavior and risk are not attached to the same sampling unit. These ideas have been taken up in a recent study investigating the impact of concurrent partnerships on HIV transmission in South Africa [13], where the authors relate HIV acquisition risk in women to some sexual behavior characteristics of the local male population they are living in. In that study, no impact of concurrency on HIV acquisition risk was found.

Next, concurrency is also a population level phenomenon. Some studies have used the term “prevalence of concurrency” meaning the fraction of persons involved in more than one partnership at a given time or in a given time period. The index k used in the simulations in [2] measures population level of concurrency and therefore results in [2] show that prevalence of concurrency is related to HIV transmission. There is a crucial assumption underlying these modeling results, which is that all other network properties are random. In other words, concurrent partnerships are distributed at random over the population without differences between men and women, or any other structural limitations. But it was also shown that including assortative mixing or asymmetry between men and women will alter the impact of concurrency on HIV transmission [12, 14]. In short, assortative mixing by number of partners increases the size of connected network components and their density and therefore enhances the speed of transmission through the network. Disassortative mixing also results in larger network components, but the connected components tend to be more heterogeneous in terms of numbers of partners, so the epidemic spread is not as fast as in the assortative mixing network. Asymmetry between men and women slows down transmission because it leads to heterogeneity in the network and prevents the existence of larger densely connected structures.

The question of how concurrency on individual and population level is related remains to be answered. In [15, 16] we introduced the concept of the line graph in an attempt to provide a technical framework for doing that. This rather abstract mathematical concept in essence is a shift of focus from the individual towards the partnership as a unit of analysis. With this step, local and global measurements and dynamics are linked to each other. The essence of the network paradigm is that global phenomena cannot be predicted from local variables only. This implies that we need to use the partnership as the unit of investigation and analysis if we want to be able to identify the signature of concurrent partnerships—if any—on the HIV epidemic.

Typology of Concurrent Partnerships

How can those abstract concepts be translated into definitions that can be used in empirical studies of real sexual networks? In consultations of the UNAIDS reference group for modeling and estimation, various ways of measuring levels of concurrency in populations were discussed and their limitations and difficulties were assessed [17]. In those measurements, one tries to quantify concurrency in terms of prevalence of concurrency, duration of overlaps and numbers of concurrent partnerships within a given time period[18]. Those measures do not take into account different social roles that the partnerships may play within society [19]. Indeed, different types of partnerships may be associated with different risk behaviors and social norms, and thus have varying impact on disease transmission, even if quantitative measures such as duration of overlap are equal. Therefore, populations with differences in types of concurrent partnerships rather than their quantitative statistics may also differ in HIV transmission dynamics. These relationships can only be understood, if types of concurrent relationships are studied in terms of associated risk behavior and their quantitative and qualitative implications on HIV transmission. Association of concurrency and risk behavior can differ between individual and population levels [20].

The social and cultural background of a society defines various types of concurrent partnerships with different motives including reproduction, benefits and expectations for partners [21]. The types of partnerships usually involve different ages, social and marital status among the partners between whom concurrent partnerships take place [22, 23]. Different social roles may lead to correlation between various types of behavior, for example certain types of partnerships may be correlated with high or low sexual risk behavior, because those behaviors follow from the same set of norms. To better understand what role different types of partnerships play for HIV transmission, we need to develop a classification of types of concurrent partnerships based on these societal determinants and their sexual behavior. This typology is starting to emerge from anthropological studies on sexual relations conducted in Southern Africa and will provide the framework for a more complex definition, not only based on variables that can be quantified, such as duration, gaps and overlap or partnerships [24]. In the context of a US study, Gorbach et al. [25] distinguished between different forms of concurrency depending on whether an individual has a primary, steady partner or not. For individuals with a primary partner, concurrent partnerships may occur when both partners agree with it, as a reaction to behavior of the primary partner, when separated for a longer time period from the primary partner, shortly before ending the relationship with the primary partner, or to compensate for frustration in the relationship with the main partner. Individuals without primary steady partnership may have concurrent partnerships without stronger commitment to any of the partners and go through different forms of experimental behavior. All of these different forms of concurrency have different implications for the ensuing network structure and hence on disease transmission.

In SSA, besides being in a first marriage, men and women can engage in other long term relationships such as being a co-wife (women) or having other wives or a “small house” or “deuxième bureau” (men) [26] (Fig. 1). The latter are long term relationships resulting in second households and families that are not based on official marriage. There are also various types of short term sexual partnerships, either involving some kind of economical exchange, or as a consequence of unavailability of the male or female partner due to spatial separation [27]. For the different types of concurrent partnerships, information is then needed about how often they occur by age and sex, their durations, what determines their formation and separation, mixing by age and other demographic characteristics, and sexual behavior within these partnerships. Finally, to study HIV transmission, we also need data about HIV concordance and discordance in different types of partnerships. However, as mentioned above obtaining this kind of descriptive data in different cultures is fraught with many difficulties.
https://static-content.springer.com/image/art%3A10.1007%2Fs10461-012-0254-6/MediaObjects/10461_2012_254_Fig1_HTML.gif
Fig. 1

Types of sexual partnerships/encounters that may lead to concurrency with varying duration of overlap. Co-wives are married to a man in official marriages, while a “small house” is a second family without official marriage. Women may engage in long term sexual relationships to obtain economic support, possibly needed for raising a child born from that relationship. Transactional sex is sex traded for material goods such as luxury articles. In relationships, where partners are separated for longer periods of time, because one partner works at a long distance from his/her home village, other sex partners are sought for intermittent time periods

A question arises concerning the time window of concurrency: what is considered concurrent at the time of interview? Methodologically the most correct is to consider only ongoing partnerships at the time of interview. From the practical point of view this might be more difficult than defining some time window during which partnerships were overlapping (e.g. in the last 6 months). In the latter case, statistical methods have to be developed to translate instantaneous rates into accumulated fractions and vice versa. To do this, knowledge about the temporal homogeneity of the underlying process is needed. Some methods for translating data from sexual behavior surveys into rates of partnership formation and dissolution have been presented in [24]. Methodological issues here are how to deal with sampling biases (e.g. partnerships with long duration have a higher probability of being ongoing than short term partnerships) and how to define whether a partnership is ongoing or not. In the extreme, a one off partnership is never ongoing, but can only be sampled retrospectively. So we need methods that collect both retrospective and instantaneous information in a consistent framework.

Data collected in the Likoma network study [28], but also earlier analysis of the time evolution of sexual networks in the Colorado Springs study [29], shows that network structure quickly increases in density when links are accumulated over time. However, the larger the time window is across which links are accumulated, the looser is the association between network structure and disease transmission, simply because many paths in the accumulated network did not occur ordered in time. Also, clustering of contacts in time has been shown to greatly influence transmission dynamics in sexual networks [30]. Therefore, what we need to know about a network is not its accumulated structure, but its structure at a given point in time (cross sectional) and its time evolution. In particular, do certain structural network properties like the mean degree or degree variance remain approximately constant in time or do they change in some way—maybe as a consequence of disease transmission, demographic changes or behavioral changes. To some extent changes in network structure over time can be deduced from changes in numbers of partners and partnership duration as measured by sexual behavior surveys. We need to develop methods to measure the dynamic behavior of sexual networks with time. Only then can the impact of intervention on network structure and disease transmission be properly monitored.

Concurrency as a Driving Force of HIV Transmission

When asking the question whether concurrency is the driving force of sexual HIV transmission in SSA and more particularly in Southern Africa, one first has to clearly define what is meant by “driving force”. When is a factor a driving force? One way to define this is by using the concept of the basic reproduction number R0, which is the number of secondary cases that one infected individual produces in a susceptible population during his/her entire infectious period [31]. If R0 > 1, an infection can persist and spread in a population, if R0 < 1 it will die out. Obviously, R0 > 1 for HIV in SSA populations. R0 is influenced by many factors, but roughly one can always say that it depends on the number of contacts an individual has, the probability of transmission per contact, and on the duration of the infectious period. For HIV, we have to consider that the infectious period consists of various phases of lower and higher infectivity, so transmission probability per contact changes during the course of the infection. The numbers of secondary transmissions per infected cases are influenced by the level of concurrency, because in monogamous relationships, once an individual has infected his/her partner, there are no others that he/she can infect until the partnership has dissolved and a new one formed. With concurrency, that restriction is circumvented; similarly, if partnerships are sufficiently short, partnership duration is less influential, because sexual contacts with many different individuals are possible within a short time interval. One can now see that it will matter when during a relationship concurrent relationships occur, how long they last, and whether they occur within a web of other concurrent relationships or as solitary events in a largely monogamous environment. Under which circumstances can concurrent partnerships increase a reproduction number from below one to above one? We can distinguish between biological factors pertinent to disease progression and behavioral factors that influence network dynamics and structure.

In the early 1990’s it was recognized that the viral load of HIV and therefore its infectivity could be high shortly after infection during the so-called primary phase, while infectivity is lower during the long chronic phase of infection. In recent years this variability in infectiousness has been corroborated in studies of the viral load [32] and transmission studies between discordant couples [33]. Early on it was recognized that concurrent partnerships could lead to high rates of HIV transmission during primary infection and therefore pave the way for explosive spread in the beginning phase of an epidemic [34]. Recently, molecular epidemiology studies have shown that in some populations clusters of HIV infected persons occur who infected each other during primary infection [35, 36]. Estimates for the proportion of transmissions during primary infection range between 15 and 40 % [3739]. It is plausible that concurrent partnerships can provide the substrate for fast spread of HIV with transmission occurring mainly during primary infection, which would enhance epidemiological differences between populations with serial monogamy and overlapping partnerships. Here also the concept of a gap between partnerships plays a role, as was discussed in the context of STI transmission by Chen et al. [40]. If the gap between two partnerships is longer than the primary infectious period, transmission rates will be lowered.

Behavioral factors influence the structural properties of the network and together with concurrency may increase (or decrease) the potential of transmission. Long term concurrent partnerships result in larger connected network components, assortative mixing results in network clustering, and higher risk behavior may result in higher transmission rates per partnership. Some answers can be obtained from a deterministic model—originally designed to describe HIV transmission amongst men who have sex with men in Amsterdam—that takes one off partnerships occurring concurrently to steady long lasting partnerships into account [41, 42]. HIV infection was modeled as a two-stage process of infection with a short period of primary infection with high infectivity and a long lasting chronic infection with low infectivity. Assuming that the process of partnership formation and separation is at equilibrium (i.e. the numbers of steady partnerships are constant over time), we investigated the impact on HIV prevalence of having one off contacts either while being single or while being in a steady relationship. Figure 2 shows the endemic HIV prevalence as a function of the proportion of one off partnerships that are formed as concurrent to ongoing steady partnerships. We see that when an increasing proportion of all one off partnerships are formed by those who are already participating in a steady relationship, the basic reproduction number crosses the threshold above which spread of HIV is possible. With further increase in the proportion of one off partnerships formed by those in steady partnerships, HIV prevalence in endemic equilibrium increases. In this model, to sustain a high prevalence of HIV in the population, high levels of concurrent partnerships are required.
https://static-content.springer.com/image/art%3A10.1007%2Fs10461-012-0254-6/MediaObjects/10461_2012_254_Fig2_HTML.gif
Fig. 2

The prevalence of chronic HIV infections in endemic equilibrium as a function of the proportion p of all casual partnerships, which are formed by individuals in a steady partnership. The overall rate of formation of casual partnerships is kept constant, only the tendency of individuals who are already in a steady partnership to also participate in casual partnerships changes with increasing p. Parameter values are taken as in [41]. At p = 0.34 the basic reproduction number increases to above 1 and HIV can establish itself in an endemic steady state with increasing prevalence as more casual partnerships are with individuals who already have a steady partner

The model has recently been extended to describe strictly polygynous populations, where it can be shown that the basic reproduction number in the polygynous population is always larger than in a serial monogamous population provided all other parameters remain unchanged [43]. This result, however, does not necessarily mean that endemic prevalence will be higher in polygynous versus serially monogamous populations. A further step towards explicitly modeling concurrency in dynamic networks was taken in [44], where a more general framework was developed for studying the dynamics of sexual networks. In ongoing work, this framework will be exploited to analyse impact of concurrency in a wider range of dynamic networks.

Mathematical modeling cannot answer the question of whether concurrent partnerships are driving HIV transmission in Southern Africa, a region with hyper epidemics; that question can only be answered by empirical studies. However, mathematical modeling can play an important role in increasing the understanding of how concurrent partnerships shape the sexual network, how concurrency is interrelated with other network structural properties, and what might be the parameters and quantities that should be measured in the field in order to identify the impact of concurrent partnerships on HIV transmission dynamics. Modeling can be applied in different ways to reach those goals. Firstly, strategic models can be defined whose aim it is to provide a “proof of principle” for a given hypothesis. This means that models with a simple structure are defined that aim to show the effects of a specific mechanism in isolation by keeping other model parameters and variables constant. Such models will in many aspects not be realistic, but they will be able to clearly demonstrate that the mechanism under consideration has a particular impact on HIV transmission; however, they cannot directly be used to design prevention strategies. Another modeling strategy consists in defining models that reflect many aspects of transmission dynamics in a more realistic way, such that they can be reasonably fitted to available HIV incidence, prevalence and/or behavioral data. Aim of such a modeling strategy is to plausibly relate the observed transmission dynamics to the underlying model structure. This also would provide a tool for investigating alternative scenarios and in that way give some information about the contribution of concurrent partnerships (and other factors) on HIV transmission dynamics. Both modeling strategies are useful and should be followed in order to give guidance to epidemiologists and behavioral scientists for future field studies. Having established the theoretical potential of the impact of concurrent partnerships on HIV transmission dynamics, we now should move on to including more detail of determinants of sexual behavior into our models. We need to understand how behaviors that shape network structure are correlated with other behaviors that modify the risks of acquiring infection (such as condom use, periods of abstinence, re-marriage within a family).

Conclusions

Although theoretical work has shown that concurrent partnerships may play or may have played an influential role in the transmission dynamics of HIV in SSA, we argue that their role as a “driving force” of HIV transmission has not yet been established empirically or sufficiently understood from the theoretical point of view. Many alternative hypotheses are available for what the role of concurrent partnerships is in shaping HIV transmission dynamics and why it is difficult to support these hypotheses with empirical studies [45]. There is at present still too wide a gap between knowledge of sexual behavior and partnerships in their societal and cultural context in Southern Africa and what that implies for sexual network structure and HIV transmission. Much work still has to be done to close that gap and that can only be done by constructive dialogue between those who do empirical studies and those who use mathematical models for their analysis. All efforts to curb further transmission of HIV are needed, but at present, we are not in a position to argue convincingly that the reduction of levels of concurrency in SSA will have a major impact on the rates of transmission of HIV. As long as that is not proven, we have to focus our efforts on intervention strategies with sound evidence base while ensuring that the prevention message is clear [46]. It remains an important challenge for both epidemiologists and mathematical modelers to clarify whether and how concurrent partnerships drive HIV transmission, and how different types of concurrent partnerships may contribute to HIV transmission dynamics.

Copyright information

© Springer Science+Business Media, LLC 2012