Optimal allocation of resources in female sex worker targeted HIV prevention interventions: model insights from Avahan in South India

Abstract

Background: The Avahan programme has provided HIV prevention activities, including condom promotion, to female sex workers (FSWs) in southern India since 2004. Evidence suggests Avahan averted 202,000 HIV infections over 4 years. For replicating this intervention elsewhere, it is essential to understand how the intervention's impact could have been optimised for different budget levels.

Methods: Behavioural data were used to determine how condom use varied for FSWs with different levels of intervention intensity. Cost data from 64 Avahan districts quantified how district-level costs related to intervention scale and intensity. A deterministic model for HIV transmission amongst FSWs and clients projected the impact and cost of intervention strategies for different scale and intensity, and identified the optimal strategies that maximise impact for different budget levels.

Results: As budget levels increase, the optimal intervention strategy is to first increase intervention intensity which achieves little impact, then scale-up coverage to high levels for large increases in impact, and lastly increase intensity further for small additional gains. The cost-effectiveness of these optimal strategies generally improves with increasing resources, while straying from these strategies can triple costs for the same impact. Projections suggest Avahan was close to being optimal, and moderate budget reductions (≥ 20%) would have reduced impact considerably (>40%).

Discussion: Our analysis suggests that tailoring the design of HIV prevention programmes for FSWs can improve impact, and that a certain level of resources are required to achieve demonstrable impact. These insights are critical for optimising the use of limited resources for preventing HIV.

Figure 1. Condom use among FSWs in last sex act before and after the intervention (a), and how condom use amongst reached FSWs in 2008 relates to intervention intensity (average annual number of condoms distributed per reached FSWs in 2008) (b).

IBBA data from rounds 1 and 2 amongst FSWs were used to derive these functions. %CCU is the consistency of condom use in the last commercial sex act, CI denotes confidence interval, FSW denotes female sex worker, and IBBA denotes integrated behavioural and biological surveys.

Figure 2. The relationship between total intervention impact and annual cost (in $100,000 per year) over 4 years (2004–2007) and intervention scale (Annual number of FSWs reached by the intervention over 4 years - #FSW) (a and c) or intervention intensity (annual number of condoms distributed per reached FSW over 4 years - #CD) (b and d).

Impact is projected by the model (described in main text and Appendix S1) and total cost is estimated by the cost function TC = scale^ae ^10.18+b*^intensity (a = 0.256, b = 0.00071 as per equation (1) of Appendix S3).

Figure 3. Model-projected intervention impact (a) or total annual cost over 4 years (2004–2007) (b) for different scale (average number of FSWs reached by the intervention each year over 4 years) and intensity (average annual number of condoms distributed per reached FSW over 4 years or #CD).

For each contour, the impact (number of HIV infections averted over 4 years) (a) or cost (b) remain the same for different combinations of scale and intensity. The red triangles in the figures refer to intervention examples discussed in the text that all avert 300 HIV infections over 4 years.

Figure 4. The cost-effectiveness of different modelled intervention combinations that avert 300, 600 or 900 HIV infections between 2004 and 2007 for different levels of scale (average annual number of FSWs reached by the intervention over 4 years) and intensity (average annual number of condoms distributed per reached FSW over 4 years).

The optimal intervention for averting 300, 600 and 900 HIV infections is the minima of each curve (the red numbered triangles on each curve; with respective cost-effectiveness of $2136, $1279 and $947 per HIV infection averted) when respective scale is 636, 1278 and 1926 FSWs reached per year and average intensity is #CD = 216. The red triangles in the figures refer to these optimal (scale, cost-effectiveness) (in (a)) and (intensity, cost-effectiveness) (in (b)) combinations.

Figure 5. The relationship between incremental annual cost and impact over 4 years (a) or scale and intensity (b) for the optimal intervention strategies that maximise impact for different budget levels.

In (a) we projected the maximum intervention impact, as the number of HIV infections averted, for different budget levels between 2004 and 2007. In (b) we show the optimal combination of scale (average number of FSWs reached each year between 2004 and 2007) and intensity (average number of condoms distributed per FSWs per year: #CD) to attain this maximum intervention impact, with contours of constant annual cost over 4 years also shown for reference. The cross in (a) signifies the estimated cost and impact of Avahan in the representative district, whereas in (b) it signifies the average scale and intensity of Avahan in the representative district.

Figure 6. Illustration of the relationship between yearly budget levels (averaged over 4 years of intervention period (2004–2007)) and both the average incremental cost-effectiveness ratio (ICER = cost/impact; dark-grey curve) and the marginal cost-effectiveness ratio (MCER = Δcost/Δimpact; light-grey curve) for each additional HIV infection averted when the optimal intervention combination is adhered to in the typical/representative district.