The effectiveness of the intervention varied across
The effectiveness of the intervention varied across districts, with the proportion of infections averted inversely related to pre-intervention consistent condom use (appendix). Understandably, for districts in which previous interventions had already led to high consistent condom use among high-risk groups before Avahan, only small, incremental effects could be achieved. Districts with low HIV prevalence and low consistent condom use at baseline, such as those in Tamil Nadu, had more infections averted than other districts, because the epidemic had more potential to grow. Our results differ in several important ways from those of the previous assessment of Avahan by Ng and colleagues, which estimated that 100 200 HIV infections were averted by Avahan between 2004 and 2008. The earlier analysis compared HIV prevalence trends in women attending antenatal care clinics between Avahan and non-Avahan districts, with the assumption that intervention coverage was higher in Avahan districts. By contrast, our method compared the HIV epidemic trends in each district with what might have occurred in the absence of Avahan or any other intense, core-group intervention. Ng and colleagues\' analysis probably underestimated the effectiveness of Avahan, since the Indian Government, through the National AIDS Control Organisation (NACO), implemented high-coverage, targeted interventions in many non-Avahan districts, so non-Avahan districts have been exposed to interventions and cannot always be used as a valid counterfactual. For example, in Tamil Nadu, Ng and colleagues reported no evidence for the effectiveness of Avahan, probably because of the long history of interventions in many of the non-Avahan districts in that state. By contrast, because Avahan was usually the first and only intervention in the districts of Tamil Nadu in which it buy GSK503 operated , our analysis estimated that 25% of HIV infections averted across all Avahan districts were in that state. Although Banandur and colleagues used a similar method to Ng and colleagues, their estimate of 87 000 HIV infections averted in Karnataka state between 2004 and 2008 is fairly close to our estimate of 60 300 over 4 years (data not shown). The two approaches to the assessment of Avahan differ in other ways. The previous analyses took into account only the eventual effect on the general population, without investigating the causal pathway through which the intervention achieved its effects. By contrast, our analysis first assessed effectiveness in the high-risk groups that were the focus of the Avahan programme, and then projected how pepsinogen effect propagated to the general population, thereby taking into account the targeted nature of the intervention. It thus addresses some of the issues related to causation (panel). Assessment of the effectiveness of HIV preventive interventions is crucial for determining which strategies should be prioritised. This study sought to determine whether there is evidence that Avahan reduced the transmission of HIV among high-risk groups and the general population. It represents the first preplanned, integrated use of mathematical modelling and data collection for assessment of a real-life, large-scale HIV intervention programme, and its success suggests that our assessment design could be a viable alternative to randomised controlled trials. The mathematical model used was developed specifically to assess Avahan, with a structure reflecting important sources of heterogeneity in IBBA data. The model was refined in consultation with epidemiologists and other expert non-modellers, and through exploratory modelling work. The IBBA surveys used a detailed sampling frame derived from careful mapping of venues, and were designed for this assessment, providing previously unavailable information on HIV prevalence and risk behaviour of high-risk individuals across a large number of districts. Combined with systematically gathered programme data and size-mapping estimates, these survey data allowed for detailed and robust mathematical modelling projections of the effectiveness of Avahan in many different settings, while accounting for uncertainty in estimates. We used an assessment design established at the beginning of the study to minimise assessor biases. The effectiveness estimates were deliberately chosen to be conservative, with the assumption of the counterfactual scenario that condom use would have continued increasing at pre-Avahan rates in the absence of Avahan. Although the results for all 69 districts are based on an extrapolation from a linear regression model of IBBA districts, and therefore have more uncertainty, we present results with different degrees of strength of evidence to quantify how uncertain our results are.