The district specific estimates of

The district-specific estimates of effectiveness (figure 1) include infections averted in all population subgroups over the first 4 years, even for districts without surveys of men who have sex with men and the general population, and the state-level and overall proportion of infections averted across all modelled districts. Table 2 shows the overall infections averted over 4 and 10 years, by increasing degree of uncertainty. In 18 IBBA districts, the modelled HIV prevalence among female sex workers fell during the first 4 years under the intervention condom SLx-2119 manufacturer by 11% to 52% dependent on the district, whereas it increased in four districts. Over 10 years, the projected median HIV prevalence fell in all districts to less than 14% among female sex workers, with only six districts having a median prevalence higher than 5% in this subpopulation (district-level prevalence-trend graphs are provided in the appendix). By comparison, under the control condom hypothesis HIV prevalence among female sex workers remained high, with seven districts having a prevalence higher than 20% after 10 years. Figure 2 shows the mean prevalence trends weighted by the size of the female sex worker population, grouped by strength of evidence. After the first 4 years, the HIV incidence ratio between the intervention and matched counterfactual varied by between 0·08 and 0·92 (median 0·30) across modelled districts. For the intervention condom hypothesis, incidence in low-risk women fell during the first 4 years in 19 districts, and after 10 years it had fallen by at least 70% in all districts, with larger reductions in incidence among female sex workers (appendix). When only the subset of variables available across all Avahan districts was used, 62% of the variability in the number of HIV infections averted over the first 4 years across IBBA districts, as measured by the R2 value of the linear regression model, was accounted for by: the number of female sex workers in a district (positively associated, accounted for 24% of the variability); whether Avahan was the main intervention provider for female sex workers in the first year in that district (positively associated, 26%); and being a district in Maharashtra state (negatively associated, 12%). When this regression model, described in the appendix, was used to extrapolate effectiveness estimates to all non-IBBA districts, the overall effectiveness of Avahan for all 69 districts was estimated to be 202 000 HIV infections averted over the first 4 years (table 2), with 37% of infections averted in Andhra Pradesh, 30% in Karnataka, 8% in Maharashtra, and 25% in Tamil Nadu. Over 10 years, the number of infections averted increased to 606 000 across all districts. The two regression models for effectiveness over 4 and 10 years had the same independent variables, although they were built independently. Figure 3 shows how the number of infections averted varied geographically across all Avahan districts.
Discussion For districts in which hypothesis testing did not show evidence of increased consistent condom use due to Avahan, this finding was probably because consistent condom use was generally quite high before Avahan (figure 2). Our attribution of effectiveness to Avahan is supported by other analyses. First, research by Bradley and colleagues showed that condom distribution increased substantially in Karnataka after 2004, mainly because of Avahan, and suggested that the proportion of sex acts between female sex workers and their clients that were protected by a condom increased from 16–24% in 2004 to 81–89% in 2008. Second, statistical analysis of the reconstructed trend data for consistent condom use by female sex workers suggests that use increased faster after the beginning of Avahan than before in ten of 18 districts. Finally, survey data suggest a dose-response relation between condom use between female sex workers and their clients and both time since first contact with Avahan staff and the number of condom demonstrations seen.