Detailed data on sexual behaviour were collected in confidential interviews with 6661 men who have sex with men and 6648 women who sell sex from 40 sampled geographic sites covering urban and rural locations in 13 districts of the Indian state of Andhra Pradesh. This was done from July 2003 to April 2004 as part of the Frontiers Prevention Project baseline study, which was approved by the Ethics Committees of the Administrative Staff College of India, Mexico's National Institute of Public Health, the International HIV/AIDS Alliance, and by the Indian Health Ministry's Screening Committee, Indian Council of Medical Research, New Delhi, India.
Details of the sampling and data collection procedures for men who have sex with men and women who sell sex are described elsewhere [3, 4]. Forty geographic sites in 13 districts of the Telangana and Rayalseema regions of Andhra Pradesh state were identified where men who have sex with men and women who sell sex were considered to be present in reasonably large numbers and access to them seemed feasible through peer facilitators and non-governmental organisations having links with them. Each geographic site consisted of one or more close-by cities/towns/villages. The total number of cities, towns or villages included in the 40 geographic sites were 62 for men who have sex with men and 72 for women who sell sex, which included a range of rural and urban categories of various sizes [3, 4].
The data collection instruments were developed by a multidisciplinary team through review of worldwide literature, focus group discussions and in-depth interviews with men who have sex with men and women who sell sex for the local context in Andhra Pradesh, and pre-pilot studies. The instruments were translated into the local language Telugu, translation checked through back-translation into English, and refinement done where necessary. Extensive training of the interviewers was done by a variety of survey experts, as well as men who have sex with men and women who sell sex, in order to address the technical, ethical and cultural issues.
At each study location, peer facilitators helped contact and recruit respondents. Standardised procedures were established and followed for contacting and interviewing respondents. Written informed consent was obtained from each respondent. One-to-one interviews were done confidentially and the identity of respondents was not recorded. The methods used were similar for men who have sex with men and women who sell sex. Relevant to this report, the number of sex partners, number of times sex done in a unit time, type of sex done, condom use, whether men who have sex with men also sold sex to men, and if men sold sex to men was this done frequently or infrequently, were documented.
The probability of HIV infection was estimated using a previously published formula [5]: Pr = 1-{P [1-R(1-FE)]N+(1-P)}M where Pr is the probability of HIV infection in uninfected, P is the average HIV prevalence among sex partners of the group for which the probability is being estimated, R is the risk of HIV transmission per act of unprotected sex, F is the fraction of sex acts in which condom is used, E is the effectiveness of condoms, N is the average number of sex acts per partner, and M is the average number of sex partners. In order to estimate the probability of new HIV infection in an entire group including those uninfected as well as infected by HIV, Pr was multiplied by (1 – I), where I is the proportion in the susceptible group that is already infected with HIV. Probabilities were calculated separately for acquiring and transmitting HIV. For the calculation of acquiring HIV, the application of the formula is self-explanatory. For the calculation of transmitting HIV, the group to which transmission was assessed became the acquirer in the formula, with the value of variables in the formula being those that were applicable to sex between these two groups. For example, while calculating transmission by men who sell sex to men of HIV to men who do not sell sex, the value for P was that for men who sell sex to men and the value for the other variables were those that applied to sex between these two groups.
The average HIV prevalence was 16% among men who have sex with men and 16% among women who sell sex in the sentinel surveillance of 2004 in Andhra Pradesh [6]. As the sample in our study was recruited through peers and non-governmental organisations, it was felt that the HIV risk characteristics were likely to be similar in our sample and the sample recruited through non-governmental organisations in the sentinel surveillance. Therefore, HIV prevalence from the sentinel surveillance was used for our sample. Among men who have sex with men, HIV prevalence was assumed to be higher for those who sell sex to men (24%) than in those who do not sell sex to men (12%). Among men who sell sex to men, HIV prevalence was assumed to be higher for those who sell sex frequently (25%) as compared with those who sell sex infrequently (22%). When sex partners of men who have sex with men formed a mix of those who sold sex and those who did not, HIV prevalence intermediate to those for these groups was used for calculations based on the proportions – of the men who have sex with men who do not sell sex 17.3% paid for sex and of the men who sell sex and also do insertive anal sex 30.3% paid for sex in our sample. HIV prevalence among clients of women who sell sex was assumed to be half that in women who sell sex (8%). HIV prevalence among women sex partners of men who have sex with men assumed to be 4% – about twice the estimate for general women in the antenatal sentinel surveillance of 2004 in Andhra Pradesh [6].
The risk of HIV transmission per unprotected sex act was adapted from previous literature [7–9]. As probabilities for acquiring HIV per unprotected sex act are estimated to be higher in developing countries due to co-existing sexually transmitted infections and other factors [9], among the reported ranges relatively higher values were assumed for our sample: 0.015 for receptive anal sex, 0.001 for insertive anal sex, 0.0015 for receptive vaginal sex, 0.0007 for insertive vaginal sex, and 0.0004 for receptive oral sex.
Estimation of the effectiveness of condoms was based on a Cochrane systematic review, which reported that the effectiveness of condom in reducing HIV transmission during vaginal sex was 0.80 [10]. Since condom use is less effective for anal sex than for vaginal sex, especially without lubricant use as in our setting [3], we assumed a lower condom effectiveness of 0.70 for anal sex. On the other hand, since condom use is likely to be more effective for oral sex than for vaginal sex, we assumed a higher condom effectiveness of 0.90 for oral sex. It is to be noted that this effectiveness is in the real life situation, as compared with condom efficacy in an ideal situation which would be higher.
The following range of values were considered plausible for the variables in the formula: 20% lower and higher than the point estimates for P, R and I used in the base calculations; the low and high values of the 95% confidence intervals for the estimates of F, N and M in our data; E for anal sex 0.65–0.75, for vaginal sex 0.75–0.85 and for oral sex 0.85–0.95. Due to limited availability of published data from India, the plausible ranges assumed by us were based on what seemed reasonable to us. A 20% variation on either side of the P, R and I estimates and 0.05 variation on either side of risk per unprotected anal, vaginal or oral sex seemed reasonable plausible ranges. For F, N and M, variables for which we had data from our field study, the 95% confidence intervals were considered as the plausible range. Sensitivity analysis for the probabilities of acquiring and transmitting HIV was performed based on the Monte Carlo simulation principle by doing 100,000 iterations with the @Risk software (Palisade Corporation, Newfield, New York, USA), using random values between these plausible ranges to obtain the range and 95% confidence intervals for the estimates for the probabilities of acquiring and transmitting HIV. The @Risk software was also used to assess the sensitivity of the probability ratios between men who sell sex and women who sell sex to the different variables in the formula.