Mathematical models of HIV transmission are useful for exploring the relationship between risk behaviour — including that of vulnerable populations such as sex workers — and risk of infection in the wider population. Many models have been published illustrating the importance, for instance, of numbers of sex partners, rates of needle sharing, transmission probabilities, heterogeneities in their values and the mixing between different groups (5, 27, 28). The difficulty of accurately measuring these key variables means that such models are better suited to generating qualitative rather than quantitative predictions. Such qualitative predictions can be compared with observed surveillance data indicating whether gross changes in the epidemiology of HIV are under way. For example, following an epidemic of HIV among injecting drug users, a model based on behavioural data from urban areas of the Russian Federation predicts significant sexual transmission among drug users and also to others through both commercial and noncommercial sex (Fig. 1). Although models are unable to predict the final extent of this sexual transmission, they highlight the importance of preventing sexual transmission among injecting drug users as well as among their non-injecting partners.
Comparison of qualitative model predictions with sentinel surveillance HIV prevalence data and, where available, behavioural data makes clear where changes in patterns of exposure — and therefore prevention — must be occurring. This can also highlight significant gaps in surveillance systems. For example, simple models of HIV transmission among sex workers or injecting drug users estimate significant HIV prevalence among individuals who have stopped sex work or injecting drug use and are therefore no longer considered vulnerable. In the case of female sex workers, their infections may be detected at antenatal clinic surveillance sites if they become pregnant, but men who cease injecting drug use will not be picked up by most surveillance systems and may represent an overlooked source of prevalent and incident HIV infections.
Mathematical models are also essential to estimating the likely impact of behavioural change resulting from interventions on the incidence of HIV. This may be particularly important when transferring a successful intervention from one location to another where the epidemiological context may differ (29). In the past, models of HIV transmission have been used to highlight the importance of good intervention coverage and the efficiency gained by targeting vulnerable populations (30). Ongoing behavioural surveillance as part of a risk assessment framework can help track trends in behaviour and identify effective and ineffective programmes. The integrated analysis of behavioural and biological surveillance data can provide a compelling argument for the success of national programmes (31).