Answers are grounded in this paper and cite their sources.
Early theoretical and empirical studies of the sexual transmission of HIV focused on numbers of sexual partners as a key variable [1,2]. For individuals this dominates the risk of acquiring a sexually transmitted disease (STD) such as HIV, but within a population the incidence and prevalence of an STD is the consequence of the pattern of contacts of the group rather than just individuals. As our understanding has improved as an additional set of characteristics of the network of sexual partnerships has become the subject of both theoretical and empirical study [3–5]. In this issue, Morris and Kretzschmar [6] add to a growing debate about the relative importance of the different network variables. Extending the early insight of Watts and May [7], that the static as well as the dynamic pattern of sexual partnerships influences the early stages of HIV spread, they use a stochastic simulation of HIV spread through a network of sexual partners to explore the importance of ‘concurrency’. They illustrate how their method of increasing concurrency leads to larger groups of people connected at any one time through chains of sexual partnerships, even though the average rate of sexual partner change is held constant. In their results, concurrency correlates with the rate of spread of HIV. Their mechanism to increase concurrency is to increase the probability that someone forms a new partnership if they are in a partnership already. In Fig. 1 we illustrate the kind of change this produces. In the network shown in Fig. 1b there is greater ‘concurrency’ than in the network in Fig. 1a, but also a greater variance in the number of partnerships and more assortative mixing.Fig. 1: . Two illustrative networks of 15 individuals (circles represent individuals and connecting lines their sexual partnerships). In graphs (a) and (b) the numbers of sexual partnerships are the same, but they are more concentrated in (b) where more concurrency is generated by a component of seven people. The bar charts show the distribution of numbers of sexual partners for the two networks. The component in graph (b) is the result of some people having more partners and others fewer partners than in graph (a). If infection were to enter person 1, then how far it could spread through the component would depend on when the contacts with person 2, and so on, were made.In a modelling exercise, or within a real population, it is not possible to alter concurrency without changing some other aspect of sexual partnerships. For example, an increase in concurrency without an increase in the variance in number of partners would only be possible with an increase in the duration of sexual partnerships. With many behavioural variables inextricably linked how can we identify and measure the most significant variables? Using a simulation model similar to that of Morris and Kretzschmar [6], but representing gonorrhoea transmission, Ghani and colleagues [8] have compared a range of different statistics describing a series of simulated networks. In a logistic regression analysis on simulation results they explored the best predictors of whether infection became established and, if it did, the best predictors of its endemic prevalence. Establishment was best predicted by the proportion of partnerships where both partners had other partners, and by the size of groups of people with many sex partners who were directly connected to others with many sex partners (termed ‘high activity coherence’). The former allows chains of infection to be maintained and the latter act as a core group increasing the number of cases. The endemic prevalence was best predicted by the high activity coherence and by the number of links that this high activity group had with the rest of the population. The two together provide a core group where an HIV epidemic could start, and then the subsequent contacts that allow it to become widespread. An example of this scenario has been seen in Thailand with a core group of sex workers and clients subsequently spreading HIV infection to longer-term sexual partners [9]. Such findings are in agreement with those of Morris and Kretzschmar that the growth of an STD epidemic is faster with more concurrency. However, Ghani and colleagues [8] found that their predictions were not as good if they recorded those in partnerships at any one time rather than over a longer period of 3 months. Although the static network is influential, the build up in connections in the sexual partner network over time also has a role. The relative contribution of the static and dynamic network will depend upon how infectious people are and for how long they remain infectious. For short-duration, highly transmissible infections the relative importance of the static network is likely to be greater. This is an issue that could be explored in future modelling work. Just as there can be differences in the type and context of sexual partnerships, one can easily envisage different types and contexts of concurrent sexual relationships. For example, the epidemiological impact of a large component would be very different if people came together at one time, in one place, than if they repeatedly moved backwards and forwards between their partners. Just as the timing and sequence of sexual partnerships has an influence, so can the timing and sequence of sexual acts. Infection is more likely to spread if both the infectiousness and the sexual acts with concurrent partners persist. In the case of HIV, infectiousness is long-lived, although repeated sexual acts would still be required for the full effect of concurrency. For example, in Fig. 1b, if person 1 were infected, infection could only reach persons 3 and 4 after person 2 had become infected, and only if they had sex with person 2 after the period between becoming infected and being infectious. Such details clearly have an influence on a local level, but how they influence HIV epidemics nationally is not so clear. The conclusion of Morris and Kretzschmar [6] is that sexual partner network statistics have more explanatory power than individual attributes. Such insights are important. They emphasize that to understand geographical differences in the scale of HIV epidemics we require a more sophisticated understanding of partnership characteristics than those afforded by simple counts of the number of partnerships and their distribution. Advances in sexual behaviour research in the last decade have considerably improved our understanding of variability in sexual lifestyles between both individuals and populations. Studies in developed countries are remarkable more for their similarities than their differences in the distribution of number of sexual partners. Internationally, surveys show considerable variability in the frequency of contact with prostitutes, casual partnerships and rates of formation of partnerships that are concurrent with marriage or cohabiting relationships [10]. The results of network analysis suggest that we should pay more attention to these variations. Morris and Kretzschmar [6] rightly emphasize the need to include measures of concurrency in sexual behaviour surveys. If this is to be done in future work, then the best way to collect such data requires further study, as does the definition of the most important parameters. Data from the British National Survey of Sexual Attitudes and Lifestyles showed considerable demographic variability in patterns of concurrency [2]. The empirical experience of this group suggests that detailed data on concurrent partnerships are more difficult to collect than counts alone and are subject to high rates of non-completion. The greatest difficulties are likely to arise when networks are measured through individuals' reports of their partners activities. The in-depth study of both social and sexual networks in Colorado Springs [11] is an alternative approach. Such studies, along with model simulations, may help put the analysis of the reports of individuals into perspective. There are now growing developments in the study of sexual networks. This reflects an iterative process whereby developments in sexual behaviour research and epidemic modelling proceed hand-in-hand to improve our understanding of epidemic dynamics. This understanding may help to guide public health policy. For example, Rothenberg and Narramore [12] describe how a network perspective is being used in Nashville, Tennessee, where intervention specialists are attempting to identify the social networks where syphilis may spread in order to be proactive rather than reactive in syphilis control. Network analysis may begin to guide policy on the more efficient use of resources, the targeting of interventions, and the type of health promotion messages that may best achieve control of HIV epidemics.
No takes yet. Share an insight, caveat, or question.
Geoff P. Garnett (1997) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: