HIV prevention programmes for female sex workers in Andhra Pradesh, India: outputs, cost and efficiency.

BACKGROUND
Female sex workers and their clients play a prominent role in the HIV epidemic in India. Systematic data on the outputs, cost and efficiency for HIV prevention programmes for female sex workers in India are not readily available to understand programme functioning and guide efficient use of resources.


METHODS
Detailed output and cost data for the 2002-2003 fiscal year were obtained using standardised methods at 15 HIV prevention programmes for female sex worker in the state of Andhra Pradesh in southern India. The services provided and their relation to the total and unit economic costs were analysed using regression techniques. The trends for the number of sex workers provided services by the programmes since inception up to fiscal year 2004-2005 were examined.


RESULTS
The 15 programmes provided services to 33941 sex workers in fiscal year 2002-2003 (range 803-6379, median 1970). Of the total number of contacts with sex workers, 41.6% were by peer educators and 58.4% by other programme staff. The number of sex worker contacts in a year by peer educators varied 74-fold across programmes as compared with a 2.7-fold variation in sex worker contacts by other programme staff. The annual economic cost of providing services to a sex worker varied 6-fold between programmes from Indian Rupees (INR) 221.8 (4.58 US dollars) to INR 1369 (28.29 US dollars) with a median of INR 660.9 (13.66 US dollars) and mean of INR 517.8 (10.70 US dollars). Personnel salaries made up 34.7% of the total cost, and recurrent goods made up 38.4% of which 82.1% was for condoms. The cost per sex worker provided services had a significant inverse relation with the number of sex workers provided services by a programme (p < 0.001, R2 = 0.75; power function). There was no correlation between the full time equivalents of programme staff and the number of sex workers provided services by the programmes, but there was a modest inverse correlation between the number of sex workers served and the average time spent with each sex worker in the year adjusted for the full-time equivalents of programme staff (p = 0.011, R2 = 0.40; exponential function). The average number of sex workers provided services annually by the first batch of 7 programmes started in early 1999 plateaued after the fourth fiscal year to 3500, whereas the 8 second-batch programmes started in late 2000 reached an average of 2000 sex workers in 2004-2005 with an increasing trend up to this fourth fiscal year.


CONCLUSION
The HIV prevention efforts in this Indian state would benefit from standardisation of the highly variable services provided by peer educators, who form an important part of the sex worker programmes. The cost per sex worker served decreases with increasing number of sex workers served annually, but this has to be weighed against an associated modest trend of decrease in time spent with each sex worker in some programmes.


Background
The number of people infected with HIV in India is estimated to be one of the highest in the world [1,2]. The state of Andhra Pradesh in southern India has a population of 80 million, and the sentinel surveillance suggests that it has one of the highest estimated burden of HIV among the Indian states [3]. It has been suggested that female sex workers and their clients play a very prominent role in the HIV epidemic in India, and that effective interventions for them could help substantially curb this epidemic [4]. In the sentinel surveillance of female sex workers at seven locations in Andhra Pradesh in 2004, the HIV prevalence ranged from 8% to 41% with a mean of 16% [5].
HIV prevention programmes for female sex workers form a major component of the HIV prevention efforts in Andhra Pradesh. With recent substantial funding from the Gates Foundation for HIV prevention [6], the number of intervention programmes for female sex workers is increasing rapidly at present in Andhra Pradesh. However, there are few data assessing the outputs, cost and efficiency of the various strategies to control HIV in India [7,8]. Such data are needed for informed planning and efficient utilisation of resources available for HIV control. As part of a study to assess the cost-efficiency of various HIV prevention strategies in Andhra Pradesh, we report data on the outputs, cost and efficiency of HIV prevention programmes for female sex workers.

Methods
This study forms part of a multi-country effort to study cost and efficiency of HIV prevention in India, Mexico, Russia, South Africa and Uganda by the Prevent AIDS Network for Cost-Effectiveness Analysis (PANCEA) [9]. The methods for the overall multi-country study are described in detail elsewhere [10]. The methods relevant for this report follow.

Selection of programmes
At the time of starting data collection for this study in mid-2003, 15 HIV prevention programmes for female sex workers were being run by non-governmental organisations in Andhra Pradesh with the support of the Andhra Pradesh State AIDS Control Society. As the HIV epidemic in Andhra Pradesh was initially estimated to be more prevalent in the eastern Coastal region of the state with nine districts, 13 of the 15 programmes were in this region. The other two programmes were in the northern Telangana region of the state that has the state capital and nine other districts, whereas the smaller southern Rayalseema region with four districts had none of these programmes. For this study, we included all the 15 programmes, of which 7 had started functioning in February 1999 and 8 in December 2000.

Data collection procedures
The initial versions of the data collection instruments from the global PANCEA study were reviewed and refined to suit the context of Andhra Pradesh. The data collection team, consisting of six investigators with background in economics or finance, was involved with the refinement of the instruments and received extensive training to ensure a standardised approach to data collection. A pilot study was done to make final refinements in the data collection format and approach.
Detailed data were initially collected for the April 2002 -March 2003 fiscal year at the 15 sex worker programmes, which included a history of the evolution of the programme, and output and cost data by month. Formal consent of the senior-most person responsible for each programme, generally the director of non-governmental organisation, was obtained to collect data. The programme project director, project manager, accounts officer, outreach coordinator, outreach workers and counsellor were interviewed and the available written records reviewed to obtain the programme data. Each visit started with an interview containing structured open-ended questions on the history of the programme, and operational or community factors that may have affected the programme. Data collection at a programme by three investigators lasted one week. Data were recorded in the field on laptop computers in MS Excel and MS Word files, which after review were entered into an MS Access database. The number of sex workers provided services annually by each programme since inception up to the 2004-2005 fiscal year was documented in another round of data collection that finished in April 2005.

Output data
Detailed data were obtained from the written records at the sex worker programmes regarding the services provided by month. The Andhra Pradesh State AIDS Control Society categorises the services delivered by sex worker intervention programmes into four major components: behaviour change communication, sexually transmitted infections care, condom promotion, and creating an enabling environment [11]. Behaviour change communication comprises several types of sessions by outreach workers and peer educators to teach and encourage the sex worker individually and in groups to follow safe sex practices. Sexually transmitted infection care includes the programme staff taking or referring sex workers and occasionally their partners for treatment of sexually transmitted infections (the cost of this is covered by the programme), HIV counselling and referrals to voluntary counselling and testing centres, help in utilising the services of HIV care and support centres, and organising training programmes and sensitisation meets for sexual health service providers such as registered medical practitioners. Condom promotion includes free condom distribution and sale of condoms under social marketing. Creating an enabling environment for sex workers includes meeting with a variety of external stakeholders by organising sensitisation meets with the general public, advocacy meets with the police, media and policy makers, establishing linkages with relevant governmental and non-government organisations, and also assisting sex workers to organise and form community based organisations and with some of their non-sexual health needs. The average time spent by the programme staff with a sex worker for each of the different types of contact was estimated based on information given by programme staff.

Cost data
The cost of the sex worker programme was divided into five categories: salaries, recurrent goods, recurrent services, office space rentals and capital goods. These cost data were collected for each month, as far as possible. Economic cost was computed, i.e. the true resource cost incurred rather than just the financial cost, as described below.
Salary cost was recorded for all personnel contributing to the work of sex worker programme, which included the project director, project manager, accounts officer, outreach coordinator, outreach workers, counsellor and attender. In addition, the compensation paid by the programme to sex worker peer educators in cash or kind was included in salary costs. Personnel compensation was noted from the official records of the programme. There were no non-salary personnel costs.
The major component of recurrent goods was male condoms. The majority of condoms distributed to the sex workers at these programmes are of a particular brand and are provided free by the Andhra Pradesh State AIDS Control Society. The market price of this brand of condoms is subsidised by 70% by the government. We considered the economic cost of these condoms as what it would have been without the subsidy. Some condoms of other brands are also sold to sex workers under the condom social marketing effort at the cost at which they are procured. Since these condoms are not subsidised, the market price of these condoms was used for economic cost calculations. The other recurrent goods utilised by the sex worker programme included medications for sexually transmitted infections, behaviour change communication materials, stationery and condom outlet boxes. The cost of the other recurrent goods was noted from the official records of the programmes, except for behaviour change communication materials about which information was obtained from the Andhra Pradesh State AIDS Control Society as it supplies these materials to the sex worker programmes.
Recurrent services included local travel, organising various workshops and meets, training for staff and peer educators, computer rentals and maintenance, consultation fees for sexually transmitted infections (investigations were usually not done, diagnosis and treatment were based on symptoms and signs), educational tour for staff, organising special events such as the World AIDS Day and candle light movement, cleaning and building maintenance, telephone, out of station travel, printing, photocopying, electricity, postage and courier, gas and some miscellaneous items. The cost of recurrent services was noted from official records of the programme. The cost for staff training was calculated by including travel fare, per diem, trainer fees, training materials, and training facility cost, information about which was obtained not only from the programme, which incurred some of these costs, but also from the Andhra Pradesh State AIDS Control Society, which incurred many of these costs.
The office of all these programmes run by non-government organisations were located in rented buildings. The details of the monthly rent paid were obtained from the programme records.
Capital goods used for the work of the sex worker programmes included computer and accessories, office furniture, electrical fixtures, telephone, white/black board, scooter, television, bicycle, public address system, video recorder, type writer, stove, refrigerator, audio system, inverter, projector, fax machine, lamination machine, sewing machine, spiral binding machine, water filter, camera, cash box and wall clock. If information about the cost of the capital goods was not available from the sex worker programme or its parent organisation, the market price was determined from retail sellers of these goods. Three quotations for these goods were obtained from the market for the 2002-2003 fiscal year and the average of these taken as the cost. The life of the capital goods was assumed to be five years, and therefore, one-fifth of the cost was allocated to the 2002-2003 fiscal year if the good was used for the full year. If a good was used only for part of the year, the cost was pro-rated for that use. If a capital good, for example a projector or a public address system, was also being used by the parent organisation for work other than that of the sex worker programme, a cost was allocated to the sex worker programme according to the proportion use reported by programme staff.
The average exchange rate of Indian Rupees (INR) 48.40 to a US$ for the 2002-2003 fiscal year was used for INR to US$ conversions [12].

Quality control
Quality control measures included a thorough pilot study before commencing formal data collection, comprehensive training of a qualified data collection team including their conceptual understanding of all data issues, full back-up and justification for any data recorded, supervision of data collection at each programme by the project coordinator, thorough review by the study team of the data obtained at each programme, and contacting programmes again to obtain information about data issues that needed clarification after the review.

Data analysis
Data were analysed using SPSS statistical software. The total number of sex workers provided services by each programme, different types of services, services provided per full time equivalent of programme staff, and services provided per sex worker, were analysed. The average economic cost per sex worker served in that fiscal year was taken as the main measure of cost-efficiency, and its relation assessed with the total number of sex workers served. The cost per contact with sex workers was also assessed. Incremental costs for each of the four major types of services provided by the programmes were assessed using a multiple regression model. Statistical significance was defined as p < 0.05.
The relation between the number of sex workers provided services and the intensity of services in the form of time spent with each sex worker was assessed. Comparison was done between the services provided by programmes that were started in the first phase in February 1999 (first batch) and those started later in December 2000 (second batch).

Results
A total of 33941 female sex workers were provided services by the 15 programmes during the 2002-2003 fiscal year. The managers for these programmes estimated that there were a total of 42900 female sex workers in the areas covered by these programmes in this fiscal year. The number of sex workers provided services annually by each programme ranged from 803 to 6379, with a median of 1970 and mean of 2263 (Table 1). Based on estimates of the types of female sex workers served by each programme, of the total sex workers served by all the programmes considered together 57% were street-based, 27% brothel-based and 16% home-based. There were a total of 245367 contacts with sex workers. The average number of contacts of programme staff with each sex worker in a year ranged 6.5-fold from 2.6 to 16.9 with a median of 8 for the different programmes (Table 1). There were a total of 49142 contacts with external stakeholders. The average number of contacts of programme staff other than peer educators with the external stakeholders in a year ranged 15-fold from 98 to 1463 with a median of 215 (Table 1).
Of the total number of contacts with sex workers by the 15 programmes, 58.4% were by staff other than peer educators and 41.6% by peer educators. The full time equivalents of peer educators with the different programmes had a wider range of 3.3 to 18.5 (median 9.5) as compared with the range of 6.9 to 11.7 (median 8.5) for the programme staff other than peer educators (Table 2). In addition, the number of sex worker contacts in a year per full time equivalent of peer educators ranged 74-fold from 36 to 2626 (median 259) whereas those per full time equivalent of programme staff other than peer educators ranged 2.7-fold from 568 to 1520 (median 958) ( Table 2). There was no significant correlation between the number of full time equivalents of peer educators in a programme and the number of sex worker contacts per peer educator (p = 0.66, R 2 = 0.01; linear function), or between full time equivalents of programme staff other than peer educators in a programme and the number of sex worker contacts per staff (p = 0.26, R 2 = 0.09; linear function), or between sex worker contacts per full time equivalents of programme staff other than peer educators in a programme and sex worker contacts per full time equivalents of peer educators (p = 0.20, R 2 = 0.12; linear function). The major portion of the contacts with sex workers by programme staff was related to behaviour change communication (93.7%) ( Table 3). The number of such contacts per sex worker in a year ranged from 2 to 16.6 (median 7.4) for the different programmes. The number of sexually transmitted infection care related contacts in a year ranged from 0.15 to 0.84 (median 0.39) per sex worker provided services by the different programmes (Table 3). Of these contacts related to sexually transmitted infections, 77.9% were for referral of sex workers for treatment of sexually transmitted infections, 14.9% for HIV counselling, referral to voluntary counselling and testing centres, and referral to care and support centres, 5.5% for training of sexual health service providers, and 1.7% for referral of partners of sex workers for treatment of sexually transmitted infections. The number of condoms distributed by the different programmes in a year per sex worker provided services ranged from 46 to 432 with a median of 168 (Table 3). Of all the condoms distributed, 95% were given free and 5% sold as part of the condom social marketing effort.
The number of contacts by the programmes related to creating an enabling environment for sex workers ranged from 0.15 to 5.64 (median 1.18) per sex worker provided services (Table 3). Of these contacts, 77% were with the community at large in the form of mass or special events, 12.1% were with other external stakeholders in the form of advocacy meets and with agencies to enhance linkages, and 10.9% were with sex workers regarding formation of community based organisations and facilitation of nonsexual health needs. There was no significant correlation    Table 4). The average cost of each sex worker contact at all the 15 programmes combined was INR 71.6 (US$ 1.48). Both, the cost per sex worker provided services and the cost per sex worker contact for the different programmes, were significantly inversely associated with scale. The best fit for this relation was obtained with the power function (Figures 1 and 2), with which scale explained 75% of the variability in cost per sex worker provided services and 73% of the variability in cost per sex worker contact by programme staff. There was also a statistically significant direct linear relation between the cost of providing services to each sex worker and the cost of each sex worker contact at a programme, and this explained 35% of the variability (p = 0.02, R 2 = 0.35).
There was no significant correlation between the full time equivalents of programme staff and the number of sex workers provided services by the programmes (p = 0.42, R 2 = 0.05; linear function). Not only was there no inverse relation between the number of sex workers provided services and the contacts per sex worker by the programmes, there was a modest positive linear relation (p = 0.02, R 2 = 0.34). However, there was also a borderline inverse correlation between the number of sex workers served and the average time spent with each sex worker by the programme staff during the year (p = 0.08, R 2 = 0.22; exponential function) (Figure 3), which became statistically significant after adjusting for the number of full time equivalents of staff in each programme (p = 0.01, R 2 = 0.40) (Figure 4).
The 15 programmes had started in two phases, 7 in February 1999 (first batch) and 8 in December 2000 (second batch), resulting in a difference of a 1.75 years in their duration. The first batch programmes were started in places estimated to have the highest numbers of sex workers in the state, followed by the second batch programmes at places also estimated to have sizeable number of sex workers. All 7 first batch programmes (numbers 8, 10, 5, 9, 7, 1, 12 in Table 1) Table  4).
Since scale was a major determinant of efficiency, we studied the trends of the average number of sex workers provided services annually by the first and second batch programmes since inception ( Figure 5). The fiscal year 1999-2000 was mostly spent in planning and establishing the first batch of sex worker programmes in Andhra Pradesh. The average number of sex workers served by each of the first batch programmes increased over the first four fiscal years and then settled around 3500 annually. The second batch of programmes, established at places estimated to have the next highest number of sex workers where Ĉ is the total economic cost in INR, P is the number of behaviour change communication contacts with sex workers, Q is the number of condoms distributed, R is the number of sexually transmitted infections related contacts, and S is the number of contacts to enhance enabling environment. This model explained 88% of the variability in the economic cost (R 2 = 0.88, standard error = 125362) and the fit was significant at p < 0.001. In this model, the constant was significant at p = 0.001 (F = 27.38; degrees of freedom: 4 for regression, 10 for residual, 14 total). Two of the predictor variables were statistically significant (number of behaviour change communication contacts with sex workers and number of condoms distributed) whereas the other two predictor variables were not (number of sexually transmitted infections related contacts and number of contacts to enhance enabling environment at) ( Applying these incremental costs to the total services provided by the 15 programmes (Table 3)

Discussion
In this study of 15 HIV prevention programmes for female sex workers in the Indian state of Andhra Pradesh, the most important determinant of cost per sex worker provided services was scale. As the number of sex workers provided services increased, the cost per sex worker trend towards a modest decrease in the average time spent with each sex worker as the total number of sex workers served increased. The distribution of this relationship ( Figure 4) implies that at least for some low-to-medium scale programmes it should be possible to increase the efficiency without compromising time spent with each sex worker. Interestingly, there was also a modest increase in the number of contacts with each sex worker as the total number of sex workers served increased. However, no standardised estimates are available about how many contacts and how much time with each sex worker would be optimal for providing effective HIV prevention services in India. Our data imply that although the efficiency increases with the scale of the programme, further development of sex worker programmes should also include assessment of the optimal number of contacts and time spent with sex workers for the various types of services provided in order to make these programmes not only efficient but also effective. This would have to include variables such as geographical coverage of the programme and transport availability, relative mix of fresh and old sex workers covered by the programme, and relative mix of street-based, home-based and brothel-based sex workers covered.  There was a very high 74-fold variation between the programmes for the number of sex worker contacts in a year per peer educator, whereas the variation for the number of sex worker contacts per programme staff other than peer educators was much smaller (2.7-fold). If attempts were made to standardise the output of peer educators, the overall output and efficiency of sex worker programmes in this Indian state would improve. It should be noted though that our study did not assess effectiveness, which would also have to receive attention in attempts to increase efficiency.
The contact by programme staff with sex workers for behavioural change communication forms a major portion of the work of the programme ( The estimates of outputs, cost and efficiency of HIV prevention services in India are generally only scantily available [7,8]. The outputs, cost and efficiency estimates, relationship of efficiency with scale, and the unit incremental costs for each of the major activities of the sex worker programmes, presented in this paper could be useful for planning sex worker programmes and estimating the resources needed by them in Andhra Pradesh and other states in India. This is particularly relevant as there has been a major recent increase in the number of sex worker programmes in Andhra Pradesh and a few other Indian states with substantial funding from the Gates Foundation [6]. The data presented in this paper also highlight that the role of peer educators needs to be Changes in the average number of sex workers provided services annually by each of the first and second batch pro-grammes since inception Figure 5 Changes in the average number of sex workers provided services annually by each of the first and second batch programmes since inception. This study is part of the multi-country PANCEA study [9,10]. The outputs and cost-efficiency estimates of HIV prevention programmes for sex workers from India can be compared with similar estimates from other countries in the PANCEA study using similar methodology. Policy-relevant comparisons of these estimates would also be possible with other HIV prevention strategies in Andhra Pradesh that we have undertaken with similar methodology. For example, we have recently reported cost-efficiency estimates for HIV voluntary counselling and testing in Andhra Pradesh [14]. Standardised comparisons of cost-efficiency estimates for HIV prevention services are needed across countries and across the different types of prevention in the background of the considerable attention recently to estimating the resources needed for HIV control, which would enable more firm local and global estimates for planning.

Conclusion
The annual economic cost of providing services to a sex worker varied 6-fold between the 15 HIV prevention programmes for female sex workers that were being run in the fiscal year 2002-2003 by non-governmental organisations in the southern Indian state of Andhra Pradesh. The cost per sex worker served decreased with increasing number of sex workers served annually by a programme, and this was associated with a modest trend of decrease in time spent with each sex worker in some programmes. However, it should be possible for at least some low-tomedium scale programmes to increase scale and efficiency without compromising time spent with each sex worker. The services provided by peer educators, who form an important part of the sex worker programmes, were highly variable, standardisation of which would be useful. The data of this study imply that although the efficiency increases with the scale of the programme, further development of sex worker programmes should also include assessment of the optimal number of contacts and time spent with sex workers for the various types of services provided in order to make these programmes not only efficient but also effective.