Efficacy of HIV interventions among factory workers in low- and middle-income countries: a systematic review

In this paper, studies about factories (also known as manufacturing plants) or people working in factories were included. The Mesh terms “workplace/industry/acquired immunodeficiency syndrome” and the key word “worker” were combined using the Boolean operator and with the following key words: (enterprise, firm, company, workshop, floor shop, machine shop, mill, factory, manufactory) and (worker, workman, workingman, employee). Key words in parentheses were connected to operators. The search strategy was implemented in the databases PubMed, PsycINFO, Scopus and EMBASE using a date range of January 1, 1990 through December 31, 2018. Grey literature online (e.g., AIDS Conference, International AIDS Society Conference), was also searched, but no relevant articles were found. Details of the search strategies are presented in Additional file 2: Table S1.

Two authors of this study (DC and GL) each independently searched for relevant articles. Titles, abstracts, full texts and reference lists of all identified reports were reviewed in duplicate by the two authors, and extracted articles were double-checked. Disagreements were resolved by discussion among the three authors (DC, GL and HZ). Reference lists from related main studies and review articles were also checked for additional relevant reports.

The inclusion criteria were: (1) The scope was formulated using the population, intervention, comparison, outcomes, and study design (PICOS) [22, 23] format, which are listed in Table 1; (2) Studies were conducted in LMICs according to the World Bank (the World Bank classifies countries by income. Countries can be categorized into LMICs and High-Income Countries (HICs). The LMIC status of a country where an included study was conducted was judged according to the World Bank classification of country by income in 2017, the most up-to-date data when we conducted our review) [24]; (3) Studies report on a specific intervention time span; (4) Articles were written in English. In order to ensure that we were not overlooking relevant studies, we had no restrictions on intervention methods.

Studies were excluded based on the following criteria: (1) Participants were not factory workers; (2) No intervention; (3) Article published before 1990; (4) Studies were observational, and did not describe an intervention’s efficacy on reducing HIV infection, changing HIV risk behaviors and attitudes, and decreasing HIV stigma (e.g., a cross-sectional study); (5) Systematic review, literature review, case series; (6) Article published in languages other than English.

The following data were extracted from publications: year of publication, first author, country in which the study took place, study design, sample size, length of follow-up, intervention method, and intervention outcomes. In addition, for studies that did not provide a χ² value, we calculated the χ² value using R version 3.6.0, including P value, if the necessary figures were provided in the paper.

In order to assess the quality of the included articles, we used the Quality Assessment Tool for Quantitative studies from the Effective Public Health Practice Project (EPHPP) [25, 26]. This tool has been widely used in literature to evaluate randomized control trials of HIV research [27, 28], and was recommended by the Cochrane Library in the area of Health Promotion and Public Health [29, 30]. Quality assessment included six components: selection bias, study design, confounders, blinding methods, data collection method, and withdrawals and drop-outs. The scores of each component are based on the documents [25, 26]. A study will receive a “strong” overall rating when none of the individual components has been rated as “weak”. If a study cannot get more than one “weak” score on any single component, it will receive a “medium” overall rating. A study of at least two “weak” ratings for individual components will be given a “weak” overall rating. Two authors (DC and GL) independently conducted quality assessment of included articles. If the quality evaluation results were different, the authors (DC and GL) would recheck the original article, and finally disagreements were resolved by discussion with the corresponding author (HZ)”.

Figure 1 shows the procedure of study inclusion. We identified 4856 articles using the specified search criteria (PubMed: n = 732; EMBASE: n = 1612; Scopus: n = 2277; PsycINFO: n = 235). 940 duplicated papers were removed. Based on the inclusion criteria, 3867 papers were excluded. 49 papers remained for full text review, and 13 papers met inclusion criteria.

Table 2 provides details regarding the 13 studies included in the review. Studies included in our review were published between 1996 and 2018. All countries included in our review either remained as low/mid-income or progressed from low to mid-income, and none of these countries progressed to a higher income status, according to the World Bank [24]. Eight took place in Africa (three in Zimbabwe [19, 31, 32], two in South Africa [20, 33], two in Ethiopia [34, 35] and one in Tanzania [36]), while the remaining five were performed in Asia (three in Thailand [18, 37, 38] and two in China [39, 40]). Two were RCTs [31, 37] and eleven were cohort studies [18‐20, 32‐36, 38‐40]. All studies reported a statistically significant effect on one or more outcomes, which were reported as follows: eight reported HIV risk behaviors [18, 34‐40], four reported HIV/AIDS knowledge [18, 37, 38, 40], four reported HCT uptake [19, 20, 32, 33], one reported HIV public stigma [20], one reported HIV/AIDS attitude [18] and one reported HIV incidence [31]. Risk behaviors included presence of multiple sex partners, commercial sex, recreational drugs or alcohol use before sex, and condom use, etc. According to the UNAIDS [41], the definition of HIV public stigma was a process of devaluation of people either living with or associated with HIV infection, such as, those who blamed foreigners/migrant workers/prostitutes for spreading HIV/AIDS.

The types and definitions of HIV intervention among factory workers in LMICs are summarized in Table 3. Further details on the risk of bias are reported in Additional file 3: Table S2. In selection bias, seven articles [18, 20, 32, 37‐40] were rated “strong” because those study participants were factory workers, and more than 80% of the selected individuals agreed to participate. Six articles [19, 31, 33‐36] were rated “medium” because only 60–79% of the selected individuals agreed to participate. In study design, two studies [31, 37] were randomized controlled trials (RCTs), so they were rated “strong”, and eleven studies [18‐20, 32‐36, 38‐40] were cohort studies, so they were rated “medium”. In confounders, two of the thirteen studies [35, 40] were rated as “strong”, five as “medium” [18, 34, 36, 38, 39] and the remaining six [19, 20, 31‐33, 37] as “weak”. For two RCTs [31, 37], the authors did not report whether confounding factors were balanced at baseline, so they were rated “weak”. In blinding methods, twelve [18, 19, 31‐40] were rated “medium” or “weak” because the evaluator or participant knew the task of the study group. In data collection methods, the research data in most studies [18, 20, 32‐40] were obtained from the survey and proved to be effective, so they were rated as “strong”, but there were two studies [19, 31] that did not evaluate the quality of the acquisition method (such as validity and reliability), so they were rated as “weak”. Finally, withdrawals and drop-outs were not related to the six studies [20, 33‐35, 38, 40], as they did not include subsequent evaluations. Overall, five cohort studies were assessed as strong quality [18, 34, 35, 39, 40], four as moderate quality [20, 32, 36, 38], and two as weak quality [19, 33]. Two RCTs only show that RCT method was used, but did not describe how to control mixing and how to achieve blind method, so their final evaluation results were weak [31, 37].