By the end of 2005 India was estimated to have approximately 5.2 million people living with HIV, representing almost 13% of the global HIV burden.¹ With more than a billion people distributed across 31 states and 593 districts, the enormous size and dispersion of India’s population presents a substantial challenge for scaling up an effective national prevention programme. Analysis of data from India’s sentinel surveillance system has shown that the HIV epidemic in India is highly heterogeneous, with wide variations between regions and states.^2–4 In particular, outside the small northeastern districts the epidemic is largely concentrated in four large southern states: Maharashtra, Andhra Pradesh, Tamil Nadu and Karnataka.^2,3 Most HIV-1 transmission in India is heterosexual,^2–5 and empirical and theoretical research has indicated that a substantial proportion of this transmission involves sexual networks that include female sex workers.^6,7 Accordingly, a major focus of the second phase of India’s National AIDS Control Programme has been scaling up targeted preventive interventions for female sex workers and their clients, and most of these interventions have been focused in the southern states where the prevalence is high.⁸ This regional focus is appropriate because behavioural surveys indicate that men in the southern states report a higher number of partners in the past year than those in the northern states,⁹ and social mapping in urban areas suggests that the population of female sex workers in south India is higher than in many of the northern states.²

The observed variations at the regional or state level in India mask important small area variations which exist at the district and subdistrict levels. Within the four southern states the HIV prevalence among antenatal women varies substantially by district, with the prevalence greater than 2.5% in several districts that are in two clusters, one along the coast of Andhra Pradesh and the other along the border of Maharashtra and Karnataka.^2,10 Another significant epidemiological observation is that in the southern states, and particularly in the districts with high prevalence, the HIV prevalence among antenatal clinic attenders is as at least as high among rural residents as among urban dwellers.¹⁰ As little is known about the sexual structure in India, especially in rural areas, the reasons for the extent of the rural epidemic and the apparent small area variability in transmission are not well understood.

In this paper we present an analysis of the demographic features and sexual structure of Bagalkot, a predominantly rural district in the northern region of the south Indian state of Karnataka with an estimated HIV prevalence among adults of 2.6%.

METHODS

Study setting

Bagalkot is a largely agricultural district with a population of approximately 1.65 million over an area of 6600 km². The district is prone to drought, with less than 40% of the agricultural land area being well irrigated. It has 6 talukas, which are subdistrict administrative jurisdictions, 10 towns and cities with a population of at least 10 000, and 629 villages. An important feature of Bagalkot is the presence in this region of northern Karnataka of the traditional devadasi system of sex work. The devadasi tradition dates back several centuries and involves the dedication of young girls through marriage to different gods. Girls were traditionally required to carry out various temple duties, including the provision of sexual services to priests. Over time, the practice has become more commercialised, but sex work associated with the tradition is socially accepted and culturally sanctioned.^11–13

Our group has delivered district-wide HIV-related prevention programmes and services since early 2003, with funding from the Canadian International Development Agency (CIDA). The programmes and services are extended to all of the 629 villages and the 10 towns and cities. In the villages, the project has trained and supported a network of “link workers” (an average of one male and one female for every five villages) who provide HIV preventive services including education and condoms. They also provide linkages and referrals for management of sexually transmitted infection and HIV counselling, and care and support services that are supported by the project and the government. As a part of the project the link workers and their supervisors conducted social mapping and a household census to develop a detailed profile of each village. The project has also supported Chaitanya, a collective of female sex workers, to deliver programmes for reducing risk and vulnerability of female sex workers across the district through peer outreach and education and the provision of STI services and various social supports.

Demographic data

Data on population distribution and literacy are derived from the 2001 census of India. The mean village population was computed by dividing the total rural population by the number of villages. Information on land use was provided by the Bagalkot district administration. Information on the occupations of rural men and women was gathered through a systematic door-to-door census conducted in all 629 villages. To do this, link workers and their supervisors were trained in the use of a brief census form, which was filled out through an interview, usually with the head of the household and/or spouse. The process included a listing of all adults who normally resided in the home, their age and their current occupation.

Sex worker enumeration and distribution

The number of female sex workers was estimated for all 629 villages and each of the 10 urban towns and cities in Bagalkot. Rural enumeration was done systematically by link workers and female sex worker peer educators. In each village key informants and female sex workers initially provided information on the numbers currently living and working in the village, and the numbers who were from the village but were currently living and working elsewhere. As the project was implementing programmes in each village over a prolonged period of time, this information was validated and regularly updated by the project staff, including the female sex workers who worked as peer educators in the villages. The data for this analysis is based on the enumeration figures for August 2006.

Geographically referenced social mapping and subsequent site-wise validation were used to estimate the numbers of female sex workers in urban areas. The social mapping proceeded in two phases. First, key informants (minimum of 70 in each urban zone), including local autorickshaw drivers, police officers, shop owners, doctors and social workers, were systematically interviewed by a field research team to identify locations and more specific sites where female sex workers solicit and entertain clients. This process was continued until the information about locations became redundant. Second, the identified sites were visited and primary key informants (ie, female sex workers, madams, pimps) were interviewed to validate the site and to estimate the number of female sex workers operating there. Each urban site with female sex workers was included in the intervention programme, with a peer educator assigned to provide outreach to those working in each site. Thereafter, the peer educators conducted a further site-wise estimation of the number of female sex workers working there. For both the rural and urban enumeration, information is collected and maintained on the number of devadasi female sex workers.

The relative size of the female sex worker population for the rural and urban areas of each taluka was estimated for every 1000 adult males and females (aged 15–49 years). The relative concentration of female sex workers in each taluka was analysed in two ways. First, villages were classified according to the estimated number of female sex workers present: <5, 5–9, 10–19 and 20+. Second, the proportion of female sex workers who were in urban zones and in villages classified according these cluster sizes was estimated.

The rural distribution of female sex workers was also analysed using Lorenz curves and the Gini coefficient. These measures have been used to compare inequalities in the distribution of income and health outcomes.^14,15 In our study we used two approaches. First, we plotted Lorenz curves and computed Gini coefficients for the cumulative distribution of female sex workers in relation to the cumulative number of villages, ordered by the number of female sex workers within a village regardless of village population size. This approach is relevant for programme planning because it indicates the extent to which programmes can achieve high coverage of female sex workers by focusing on villages within a taluka with the largest absolute female sex worker populations. Second, we plotted Lorenz curves for the cumulative distribution of female sex workers in relation to the cumulative size of the adult female population, ordered by village according to relative size of the female sex worker population (ie, female sex workers per capita). Distributional inequality in this analysis indicates the extent to which female sex workers are clustered in villages with the highest per capita number of female sex workers within the taluka. In both analyses, if female sex workers are equally distributed among villages then the Lorenz curve is diagonal. The magnitude of the deviation from the equality diagonal line indicates the extent to which the total population of female sex workers is clustered by village type. The Gini coefficient is a numerical summary of this deviation equal to twice the total area between the equality diagonal and the Lorenz curve. A higher Gini coefficient therefore indicates greater inequality in the village-wise distribution of female sex workers. As the Gini coefficient is independent of the absolute magnitude of the measures, it permits comparison of the clustering of female sex workers in geographic areas (talukas) with differing sizes of the female sex worker population.

Sexual behaviour

Sexual behaviour was assessed using “polling booth surveys”. This approach was developed to reduce social desirability biases by providing confidentiality to respondents, and is similar to the informal confidential voting interview described by Gregson et al.^16,17 Briefly, each polling booth survey is conducted in groups of 15–25 individuals (divided demographically—married and unmarried, male and female, etc.). Each individual is sequestered in a private “polling booth” set up in a large room or hall. Each participant has numbered cards of different colours: one colour for “Yes” and another for “No”. A series of numbered questions is read aloud by a surveyor and for each question the respondents are asked to “vote” either “Yes” or “No” by putting the appropriate numbered and coloured card into a closed voting box. All the cards in all the voting boxes are then combined and the total number of “Yes” and “No” answers for each question are tallied.

HIV prevalence

Estimates of the HIV prevalence in three of the talukas are based on a community-based sample survey of males and females aged 15–49 years, which was conducted as part of the baseline assessment for the Bagalkot demonstration project.¹⁸ Using a probability proportional to estimated population size (PPES) method of sampling, 10 villages in Bagalkot were randomly selected; 20 urban blocks from six towns were also selected, using a systematic random sampling procedure. A complete census of households and individuals in selected villages and urban blocks was carried out. Through this census, a list of all individuals aged 15–49 years was prepared for both the rural and urban areas. Using sex, age and marital status as stratification variables, random samples were selected from each of the rural and urban lists. Following informed written consent, information was collected on socio-demographic characteristics and sexual behaviour of the participants. Serum samples were tested for HIV-specific antibodies using an enzyme-linked immunosorbent assay (ELISA), the Detect HIV 1/2 system (BioChem ImmunoSystems, Montreal, Canada). Reactive specimens were confirmed using a second ELISA test, Genedia HIV 1/2 ELISA 3.0 (Green Cross Life Science Corporation, South Korea). Samples were considered positive when both ELISA tests were positive.

Subjects were enrolled between April and September 2003. The institutional review boards of the University of Manitoba, Winnipeg, Canada and St John’s Medical College, Bangalore, India approved the study. Of the 6703 individuals selected for the study, 4949 agreed to participate (73.8%).

RESULTS

Demography

Bagalkot district has a population of approximately 1 652 000 with 71% of its inhabitants living in rural areas. The population of the six talukas ranges from approximately 144 000 to 389 000. Table 1 presents the basic demography of these talukas and illustrates the substantial differences between them. The proportion living in rural areas (defined as communities with fewer than 10 000 population) varies from 61% to 89%. Talukas A, B and C have fewer and larger villages than the other talukas, and these talukas are much more irrigated than the other three talukas. Interestingly, literacy levels among both males and females are higher in the less irrigated talukas. Consistent with their high levels of agricultural irrigation, talukas A, B and C have relatively higher proportions of rural males and females engaged in occupations of cultivation and agricultural labour, whereas the other three, and especially talukas E and F have more rural males and females engaged in other labour occupations.

Distribution of female sex workers

Female sex workers are widely distributed in all six talukas, but the relative size and concentration of this population differs between them (table 2). It is estimated that there are almost 7300 in the district, or approximately 17.1/1000 adult males and 17.8/1000 adult females (aged 15–49 years). Almost 75% of sex workers in the district are devadasis, with a higher proportion in talukas A, B and C. District-wide, the relative size of the female sex worker population is almost threefold higher in rural areas than in urban areas. There is also considerable variation between the talukas in the relative size of the female sex worker population, with the number of female sex workers/1000 adult males ranging from 11.7 (taluka E) to 38.2 (taluka B) in rural areas, from 0.3 (taluka D) to 14.0 (taluka A) in urban areas, and from 9.6 (taluka E) to 30.5 (taluka B) overall. An estimated 186 female sex workers are currently working outside the district for every 1000 still working in the district.

Based on fixed cluster sizes, there are large differences in the extent of clustering of rural female sex workers between the talukas (table 2 and fig 1). Overall, 15% of villages have clusters of at least 20 female sex workers, but this ranges from less than 1% of villages (taluka F) to 47% of villages in taluka B. Overall, 54% of rural female sex workers are found in clusters of 20 or more; this proportion ranges from approximately 3% (taluka F) to almost 80% (taluka B). More than 80% of female sex workers in taluka A are concentrated in urban areas or village clusters of greater than 20, whereas less than 15% are so clustered in taluka F.

A different picture emerges when the analysis of distribution of rural female sex workers is based on the Lorenz curve and the Gini coefficient (figs 2 and 3). Specifically, when looking strictly at village-level clustering all talukas show a similar and relatively high level of distributional inequality with Gini coefficients ranging from 0.45 (taluka F) to 0.56 (taluka A) (fig 2). This suggests that the amount of clustering seen in the first analysis was a function of the scale used to classify female sex worker clusters, with talukas A, B and C having larger clusters but not a higher degree of distributional inequality in the rural female sex worker population. When the analysis accounts for the size of villages, there is still substantial distributional inequality in all of the talukas with Gini coefficients ranging from 0.38 (taluka C) to 0.58 (taluka F). This indicates that there is substantial clustering of female sex workers in villages with higher per capita female sex worker populations and larger absolute village populations. The higher Gini coefficients for talukas E and F suggest that there is greater female sex worker clustering in villages with a high per capita female sex worker population.

Male sexual behaviour

Overall, 12.3% of rural men in the district reported ever visiting a female sex worker, with a higher proportion of married men reporting this behaviour than unmarried men. The proportion of men reporting previous sexual relations with a female sex worker was highest in taluka B (18.0%), which was significantly higher than all of the other talukas except taluka A. Similarly, taluka B had the highest proportion of men reporting non-marital sexual relations with women other than female sex workers, and this was significantly higher than all of the other talukas expect taluka E.

DISCUSSION

Our analysis of the sexual structure and HIV epidemiology in Bagalkot district has important implications for India’s HIV prevention strategy. The HIV epidemic in rural areas will probably not be contained by an urban-centric prevention strategy. Already, there is evidence from many of the districts with high prevalence in south India that HIV prevalence is now higher in rural areas than in urban centres.¹⁰ This suggests sexual networks in rural areas can sustain and amplify transmission independent from connections to urban sexual networks. In Bagalkot also we found a substantially higher HIV prevalence in rural dwellers, and importantly, we found that there was a substantially higher concentration of female sex workers in rural areas than in urban areas. A notable finding was that rural HIV prevalence was particularly high (6.0%) in the taluka with the highest per capita number of female sex workers and where a higher proportion of men reported female sex worker and other non-marital sexual partners. It is possible that we have overestimated the rural–urban differences in the relative size of the female sex worker population. Village-based female sex workers in that region are often visible and well known to the local community, and therefore they could be enumerated more easily than the urban ones. However, our experience in implementing programmes and services for female sex workers in urban areas across 16 districts has indicated that a mature programme such as the one in Bagalkot district usually has accurate estimates of urban female sex workers, particularly those who are working regularly. Recently we have also conducted rapid female sex worker enumeration exercises in 10 other districts of Karnataka and have found that overall the relative size of the female sex worker population is at least as high in rural areas as in the urban areas.

We are not aware of other data on rural sex work in India. However, our estimates for Bagalkot district are considerably higher than published estimates from urban centres in Maharashtra, which is the state directly north of Karnataka. A mapping study in the state capital and three urban towns estimated that female sex workers represented 0.2–0.4% of the adult female population,¹⁹ which is less than a third of our estimate for Bagalkot district overall. Differences could be partly due to mapping and enumeration methodology; our mapping of female sex workers in Karnataka’s capital city (population 6.1 million) resulted in an estimate of approximately 14/1000 adult women, which is lower than our estimates for Bagalkot district but considerably higher than estimates from the Maharashtra mapping study. A recent review paper estimates of the female sex worker population in different regions of the world reported national-level estimates in Asia ranging from 0.2% to 2.6%,²⁰ which would place Bagalkot district at the upper end of the range. In other world regions, estimates of urban female sex worker populations range from 0.4% to 4.3% in sub-Saharan Africa, from 0.1% to 1.5% in ex-Russian Federation, from 0.4% to 1.4% in east Europe, from 0.1% to 1.4% in west Europe and from 0.2% to 7.4% in Latin America.²⁰ Estimates from rural areas were only available from one country (Kenya) but the population proportion was not reported.²⁰ The high number of female sex workers in rural areas highlights the need for urgent implementation of focused prevention programmes and services for them because they currently have limited access to information, education, condoms and STI services.

Our data on sexual behaviours are not sufficiently detailed to draw definitive conclusions about the contribution of female sex work networks to the overall transmission dynamics and the small area variations in HIV prevalence in rural India. However, research from other populations suggests that the relative size of the female sex worker population in Bagalkot is sufficient to explain a high proportion of the HIV transmission. Analyses of the HIV epidemic in Cotonou, Benin, indicate that with a comparable female sex worker population (1.2% of adult women) and a similar HIV prevalence, a high proportion of HIV transmission was related to sexual networks including female sex workers.^21,22 Our results showing that HIV prevalence is markedly higher in the talukas with higher female sex worker populations is suggestive but not conclusive that much of the variability in HIV prevalence is due to variability in the size of the female sex work population.

Our study further emphasises the importance of implementing specific HIV prevention programmes for migrating female sex workers in this region. We found that many female sex workers from the rural areas of this district are currently working elsewhere. Previously, we have reported that almost 20% of devadasi female sex workers currently in Karnataka have worked outside the state, and almost 5% have worked in another district in the state.¹¹ These figures almost certainly underestimate the total volume of sex worker migration from this area because it is not uncommon for young devadasi sex workers to move directly from villages to distant brothels immediately after sex work initiation.¹² Most of these younger female sex workers would not be enumerated in our mapping process. Together, these data show the strong linkages between commercial sexual networks in these rural areas and those outside the state. The most common destinations for these migrant female sex workers are urban centres in Maharashtra, where many work in brothels from a young age.^11,12 Therefore, the local HIV transmission dynamics in Bagalkot and other districts in this region are probably greatly influenced by the effectiveness of HIV prevention programmes in these other urban centres where client volume and HIV prevalence among sex workers has been shown to be high.^19,23,24 Moreover, migrating female sex workers are likely to be more vulnerable than non-migrants. Those migrating out of the state find themselves in an unfamiliar urban environment where they do not know the local language and the sex work environment is often much different. Most female sex workers in the rural areas of northern Karnataka work in their own homes,¹¹ whereas a high proportion of those working in cities in Maharashtra work in brothels.¹⁹ Previously we found that female sex workers in Karnataka who worked in home environments reported considerably lower levels of client violence and police harassment than did those working in brothels, lodges or public places.¹¹ Therefore, programmes that focus specifically on reducing the vulnerability of migrant female sex workers from these areas are necessary.

We found substantial small area variation in the relative size of the female sex worker population. Although more research is required to better understand the reasons for this variability, it is our view that some of this variability is related to the overall population distribution and economic activity. The three talukas with the highest female sex worker populations (A, B and C) are also characterised by more clustering of their rural population with fewer and larger villages. These three talukas are also more affluent than the other three because they are highly irrigated and support a more robust rural agricultural economy. We speculate that both of these factors are determinants of higher female sex worker populations. Larger villages provide a more conducive environment for sex work because they attract a greater number and variety of men from the surrounding areas, and they provide more anonymity for the sex client and the sex worker. More economic activity also leads to greater purchasing power for men, which might increase demand for sex work. The variability in distribution of female sex workers has important implications for scaling up HIV prevention programmes in rural areas. Our analysis shows that a high proportion of rural female sex workers are concentrated in a small proportion of the villages. Interestingly, this clustering pattern was similar in all of the talukas, so the main difference between them was the absolute number and size of the clusters. Strategically, this pattern shows that efficient outreach can be achieved by focusing on the larger rural clusters. In this regard, our experience and findings have practical implications for scaling up focused prevention programmes in rural areas. Mapping of female sex worker populations in rural areas is feasible and affordable. Using 15 two-member mapping teams (one field researcher and one female sex worker) we were able to map a district of approximately 630 villages within 15 days, and have subsequently mapped 11 additional districts within three months. The female sex worker mapping data is crucial for prioritising rural areas for interventions for this population and for designing efficient outreach strategies. Based on our findings of clustering in rural areas, we conclude that in similar circumstances priority can be given to those areas with high absolute and relative concentrations of female sex workers. For example, in Bagalkot district efficient resource allocation would be achieved by focusing predominantly on talukas A, B, C and D, and within each taluka orienting outreach and services to reach the largest clusters of female sex workers. If such rural clustering is not observed, other intervention strategies would need to be considered, perhaps including the use of allied village-level health workers and local female sex worker volunteers to support risk reduction in small village-based clusters.

More research is required to better understand the sexual structure and sex work dynamics in rural India. We recommend three research priorities. First, it will be important to replicate the mapping and enumeration exercise in other districts in different parts of the country to determine how much regional variability there is. Second, more detailed study is also required into the sexual behaviours and networking patterns in rural populations, and to use those findings to better understand rural HIV transmission dynamics and the extent to which rural epidemics are self-contained. Third, research is needed to evaluate the migration patterns of rural female sex workers, and the effect of that migration on their HIV risk and vulnerability.

Notwithstanding the need for more research, we conclude that the findings of this study provide compelling evidence for the need to scale up HIV prevention programmes and services for female sex workers in rural areas of India. Failure to do so will leave many of them without the information, skills and resources to reduce their risk for HIV, and will result in poor control of the large HIV epidemic in rural India.