Cost effectiveness of HIV and sexual reproductive health interventions targeting sex workers: a systematic review

Search strategy

A literature search was conducted on Medline, Web of Science, Econlit and the NHS Economic Evaluation Database. The keywords used were divided into three groups: “cost-effectiveness”, “sexual reproductive health intervention” and “sex worker”. The full keywords used are shown in the Additional file 1, these were used in combination with each other.

Inclusion and exclusion criteria

We included studies that were published from January 1995 to May 2018 in English, with a full economic evaluation, i.e. studies performed comparative analysis of costs and outcomes of at least two interventions. Full economic evaluation analyses include cost minimization analysis (CMA), cost-effectiveness analysis (CEA), cost-utility analysis (CUA) or cost–benefit analysis (CBA) [18]. We included the studies that specifically focused on sex workers or presented separate cost-effectiveness results for sex workers. Interventions included were all types of SRH interventions including preventative, curative or education or information provision that were targeted towards improving the SRH of sex workers. The full exclusion and inclusion criteria are presented in detail in Additional file 2.

Data extraction and quality assessment

Relevant papers were selected in two steps: in first step, titles and abstracts of papers were screened and in second step entire papers were screened according to the inclusion criteria. The papers were screened independently by GR and HHB and disagreements were resolved by discussion.

The interventions included in this review were classified into four main categories: behavioral, biomedical, structural and mixed interventions. The behavioral or behavior change interventions aim to reduce the risk of HIV/STD infection through modification of sexual and substance use-related behaviors [21, 22]. These interventions seek to delay onset of sexual intercourse, decrease the number of sexual partners, reduce incidence of unprotected sex, reduce sharing of needles and syringes, and reduce or eliminate substance use. These interventions generally focus on counseling individuals, couples, peer groups or networks, institutions, and entire communities through different means such as peer groups, workshops, social networks, social marketing or mass media [21, 22]. Biomedical interventions use chemical and physical technology to prevent infection or decrease infectiousness, through targeting biological and physiological processes that are responsible for HIV acquisition and transmission. These include interventions such as male and female condoms, treatment of STIs, pre- or post-exposure prophylaxis, ART, male circumcision, microbicides, and vaccines [21, 23]. Structural interventions, also known as social, environmental, ecological, or upstream interventions, aim to change the underlying determinants of risk, vulnerability or disease. These interventions, which are varied in nature, include legal changes, microfinance, vouchers, women empowerment, income-generating activities, etc. [21, 24]. Mixed or integrated interventions are those interventions that consisted a combination of biomedical, structural, or behavioral strategies.

The quality of reporting the economic evaluation evidence was assessed using the Consolidated Health Economic Evaluation Reporting Standards (CHEERS) checklist [25]. Twenty-four items of the checklist were scored using ‘yes’ (met the criteria in full), ‘partially met’, ‘no’ (not met), and ‘not applicable’. GR and HHB independently assessed the papers using the checklist and any discrepancies were resolved by discussion.

Interpreting cost-effectiveness of the interventions

The cost-effectiveness of the interventions was judged using the WHO recommendation on cost-effectiveness threshold stating that an intervention is highly cost-effective if cost-effectiveness ratio (cost per DALY averted) is less than the GDP per capita, is cost-effective if it is between one and three times the GDP per capita and it is not cost-effective if it is more than triple the GDP per capita [26]. In addition to the WHO recommendation, the cost-effectiveness results were judged using alternative thresholds, stating that an intervention, in low and middle income countries, is cost-effective if the cost-effectiveness ratio is < 50% of GDP per capita [27]. The outcomes reported in the studies, mainly in terms of cost per disability-adjusted-life-years (DALYs) averted, cost per quality-adjusted-life-years (QALYs) gained or cost per HIV infection (or STI) averted, were used in a cross-study comparison. For the studies that only reported cost per infection averted, the cost per infection averted was converted to cost per DALYs averted, using a conversion factor used by previous studies [16, 28, 29] (i.e. it was assumed that cost per DALY averted was equal to cost per infection averted divided by 20). To facilitate the cross-study comparison, all cost-effectiveness ratios were inflated to 2016 values using the consumer price indices for the country where the study conducted and then converted to 2016 international dollar (INT$) using the 2016 Purchasing power parity (PPP) conversion factor for each country [30].

Search results

The literature search identified a total of 6627 papers (Fig. 1). Initial title and abstract scanning excluded partial economic evaluations (i.e. cost description, cost of illness studies) and identified studies that provided data for sex workers. Through this process 27 potential papers were identified and analyzed and eight were excluded either due to lack of cost data, appropriate outcome or separate sex worker results. The reference lists for the remaining 19 studies [31–49] were scanned for other possible studies.

General and economic features of the papers

The majority (74%, N = 14) of 19 included studies were based in middle-income countries with the exception of three low-income and two high-income settings. Most of the studies were conducted in Asia (48%, N = 10) following by sub-Saharan Africa (25%, N = 5).

A majority of the studies (58%, N = 11) were CUA, using DALYs (n = 8) or quality-adjusted life years (QALYs, n = 3) as the outcome measures. The remaining were CEA, using cases of infection prevented as the outcome. Overall, 15 out of the 19 (79%) studies employed modeling, four were observational studies, and there was no randomized control trial design. The perspectives were mostly provider (42%, N = 8) followed by government/public sector perspective (21%, N = 4), and five studies didn’t specify their perspective.

Six studies (32%) did not specify time horizon, 5 (26%) studies had a time horizon of over 10 years, 4 (21%) studies had a time horizon of 1–10 years and 4 (21%) had a time horizon of < 1 year. There was one behavioral intervention, which consisted of voluntary counseling and testing, 4 (21%) biomedical interventions, including STI test and treatment, HIV vaccination, introducing female condom, and 1 (5%) structural interventions, including provision of vouchers for sex workers. Furthermore, there were 13 (68%) mixed interventions, which consisted of either all three types of interventions or a combination of two. Majority of (n = 8, 62%) of mixed intervention included a combination of behavioral and biomedical strategies. Detailed features of the selected studies are reported in Additional file 3 and summarized in Table 1.

Quality of reporting the evidence

The findings of the assessment of reporting quality using the CHEERS checklist are presented in Additional file 4 and summarized in Table 2 and Fig. 2. The findings showed that on average, the compliance with each item on the CHEERS checklist was 62% (ranging from 8 to 100%). The most frequently not reported items were item 9 ‘discount rate’ (47% compliant), item 11b ‘measurement of effectiveness in synthesis-based estimates’ (10% compliant) and item 12 ‘measurement and valuation of preference based outcome’ (33% compliant). Furthermore, item 15 ‘choice of model’ (46% compliant) and item 17 ‘describing analytical models’ (8% compliant) were major areas of weakness for the included studies.

Item	Item no.	Yes	No	Partially	NA
Title and abstract
Title	1	17	1	1	0
Abstract	2	10	0	9	0
Introduction
Background and objectives	3a	15	1	3	0
	3b	17	0	2	0
Methods
Target population and subgroups	4	10	5	4	0
Setting and location	5	15	2	2	0
Study perspective	6	12	5	2	0
Comparators	7	12	1	6	0
Time horizon	8	10	6	3	0
Discount rate	9	7	5	3	4
Choice of health outcomes	10	10	0	9	0
Measurement of effectiveness	11a	4	2	3	10
	11b	1	3	6	9
Measurement and valuation of preference based outcomes	12	1	1	1	16
Estimating resources and costs	13a	4	1	1	13
	13b	9	1	3	6
Currency, price date, and conversion	14	11	5	3	0
Choice of model	15	6	4	3	6
Assumptions	16	11	2	0	6
Analytical methods	17	1	12	0	6
Results
Study parameters	18	15	2	2	0
Incremental costs and outcomes	19	13	3	3	0
Characterising uncertainty	20a	3	0	1	15
	20b	13	0	2	4
Characterising heterogeneity	21	8	5	1	5
Discussion
Study findings, limitations, generalisability, and current knowledge	22	12	0	7	0
Other
Source of funding	23	19	0	0	0
Conflicts of interest	24	10	9	0	0

Cost-effectiveness by intervention type, income level and geographical region

There was only one behavioral intervention (included promoting voluntary counseling and testing and condom use components), implemented in West-Java [44] which costs INT$33 per DALY averted (or INT$889 per HIV infection averted).

Four studies (21%) evaluated cost-effectiveness of biomedical interventions. In this category, a facility-based STI case management study in a province in Colombia by Carrara et al. [34] showed that the cost per STI treated was INT$658. Contrastingly, Wilson et al. [48] demonstrated that the current scheme in Australia for sex workers’ STI checkups costs the government INT$4,078,956 per HIV infection averted, which, is not cost effective.

Only one study (5%) implemented structural interventions, which was a competitive voucher scheme in Managua, Nicaragua to increase STI testing and treatment uptake in high-risk groups (such as female sex workers and their clients, transvestites etc.) [32]. ICER for this intervention was INT$364 per STI cured.

The ICER among the remaining 13 (48%) mixed interventions ranged from INT$51 per DALY averted (or INT$869 per HIV infection averted) in a district in India, for adding a structural component (i.e. community empowerment and involvement in programme management and services, violence reduction, and addressing legal policies and police practices) to existing behavioral and biomedical strategies [45], to INT$3017 per DALY averted (INT$71,747 per HIV infection averted) in Santo Domingo district in Dominican Republic, for an intervention package consisted of community mobilization activities (such as educational workshops and materials), promotional media (such as posters and stickers to reinforce the message that the sex establishments were ‘‘100% condom’’ settings), and interpersonal communication (such as informational booths, and interactive theater presentations, training sex workers as health educators/counselors and peer facilitators in STI clinics) [43].

With the exception of the Wilson et al. study [48] in Australia, in all reviewed settings, the ICERs of all other interventions is < 20% of the GDP per capita, proving to be highly cost-effective interventions (Fig. 3). The ICERs of biomedical interventions were consistently less than 2% of GDP per capita. Similarly, ICERs of the mixed interventions are on average 3% of GDP per capita (ranging from 0.2 to 20%).

ICER of SRH interventions in low-income countries ranged from INT$4 to INT$33 per DALYs averted, compared to INT$18–INT$3017 in middle income countries (Fig. 4). ICERs also varied in different geographical areas, the lowest in sub-Saharan Africa (on average, INT$98, ranging from INT$4 to INT$310) and the highest in Latin America (on average, INT$939, ranging from INT$18 to INT$3017) (Fig. 4).

Cost-effectiveness by intervention structure and platform of delivery

Majority (n = 13) of the interventions evaluated in this review were stand-alone interventions and only five interventions were integrated into the existing programs. On average, ICER for the integrated interventions was INT$79 (ranging from INT$18 to INT$264), while ICER for stand-alone intervention was INT$469 (ranging from INT$4 to INT$3017) (Fig. 5).

Majority of the interventions were either delivered as outreach (n = 7) or at the clinic or health care facility (n = 6) and only five interventions were delivered using both platforms. Examining the ICER by platform of delivery indicated that on average mixed platform interventions had lower ICER, with INT$83 (ranging from INT$25 to INT$173), compared to clinic-based (with average ICER of INT$169, ranging from INT$18 to INT$301) and outreach (with average ICER of INT$634, ranging from INT$4 to INT$3017) interventions (Fig. 5).

Drivers of cost effectiveness

The drivers of cost-effectiveness were evaluated based upon the results from one-way/univariate sensitivity analyses. Most (89%, N = 17) studies conducted sensitivity analyses including one-way/univariate (63%, n = 12), multivariate (42%, N = 8) and probabilistic (32%, N =  6 analyses. The remaining 2 (11%) studies did not perform or specify any sensitivity analysis. Additional file 5 summarizes the results of the sensitivity analyses conducted in each study and the main drivers of cost effectiveness reported by the authors.

Studies showed [33, 40, 43] that the cost-effectiveness of HIV and SRH interventions for sex workers was most sensitive to HIV incidence and prevalence amongst sex workers. For example, Burgos et al. [33] showed that when incidence of HIV amongst sex workers increased to 4% the cost per QALY decreased to 122$ per QALY gained whilst, when it fell to 0.3% the cost per QALY substantially increased (US$1202 per QALY gained). The number of partners per sex worker was also a recurring factor causing significant changes in cost-effectiveness results. Marseille et al. [40] reported that the numbers of partners per sex worker (or female condom user in this context) altered the cost-effectiveness results, indicating that more clients per year resulted in the female condom to be more cost saving. You et al. [49] demonstrated their results were sensitive to the number of clients per day and contact rate between female sex workers with their regular partners.

Commodity costs were another determinant factor in the cost-effectiveness in some studies. One study [47] reported that the cost of rapid diagnostic test was the largest determinant of the ICER. Changes in personnel costs and laboratory test prices caused significant changes in the cost-effectiveness outcomes in Borghi et al. [32] study. Another driver of cost effectiveness results, which was mentioned in several studies, was the effectiveness of the intervention services provided such as the impact of voluntary counselling and testing (VCT) on condom use [44] or the actual effectiveness of female condoms in preventing STI transmissions [40].

Sweat et al. [43] found that cost-effectiveness of HIV and SRH interventions are sensitive to the choice of discount rate and marginally sensitive to STD prevalence in sex workers. However, Vassall et al. [45] stated that variations in discount rate did not increase the incremental cost per DALY averted above the cost-effective thresholds suggested by the WHO. In summary, the most important drivers to rendering a SRH intervention cost-effective amongst sex workers is the high incidence and prevalence of HIV and other STIs, the average number of partners each sex worker has per year, and lastly, the commodity costs of the intervention.

Potential limitations

This review has several potential limitations that should be considered when interpreting its findings. Firstly, it is not possible to differentiate which intervention types are more cost-effective than others, since the studies reviewed did not take into account interactions between different types of interventions. Moreover, the cost-effectiveness of each intervention is not directly compared, so it is difficult to determine the optimal intervention. Further analyses, including modelling, are needed to comparatively assess the cost-effectiveness of each intervention in the same population to determine which intervention is most cost-effective. In addition, comparisons of the results between different studies and countries are problematic since the interventions evaluated are not homogenous, and cost-effectiveness results are influenced by a variety of factors such as epidemiological characteristics, coverage, unit prices or supply costs and the technical efficiency for implementation.

There are also few potential limitation in regards to the literature review process that need to be considered. This study only included published studies implying that the results might have skewed by publication bias. Furthermore, as in any review study, it is difficult to rule out selection bias or disagreement between the criteria of the reviewers. To minimize this bias, we used pre-defined inclusion criteria and discussion of disagreement between the investigators throughout the review process. Lastly, for cross study comparison, we used a conversion factor of 20 to convert cost per infection averted to cost per DALYs averted for the studies that cost per DALY averted was not available (n = 8). This conversion factor, has been widely used in the literature [16, 28, 29] and originated from the 1996 Global Burden of Disease [54] study when the HIV epidemic did not necessarily have the same characteristics as the present time. Research to create a contemporary conversion factor to convert non-DALY estimates to cost per DALY ratios is ongoing [55].