Methods and protocol of a mixed method quasi-experiment to evaluate the effects of a structural economic and food security intervention on HIV vulnerability in rural Malawi: The SAGE4Health Study

HIV/AIDS, poverty, and food insecurity contribute substantially to morbidity and mortality in sub-Saharan Africa (Masanjala 2007; Tsai et al. 2011). The Republic of Malawi, in southeastern Africa, bears one of the heaviest HIV disease burdens globally, ranking ninth worldwide for prevalence of HIV/AIDS among adults (ages 15–49) (Central Intelligence Agency CIA 2010). The 2010 Malawi Demographic and Health Survey (MDHS) (National Statistical Office NSO and ICF Macro 2011) reported an adult HIV prevalence rate of 11%, with a disproportionately higher rate for women than for men (13% versus 8%; (National Statistical Office NSO and ICF Macro 2011). Poverty is endemic in Malawi; more than half of its estimated 15 million people live on less than a dollar a day (Masanjala 2007; Integrated Household Survey IHS3 2010-2011). Food insecurity, defined as having uncertain or limited access to nutritionally adequate food, or being unable to procure food in socially acceptable ways (Anderson 1990; Olson 1999), is an aggravated problem in Malawi. The latest Integrated Household Survey (IHS3, 2010–2011) showed that half of Malawian households (49%) faced food shortages in the year preceding the survey. Taken together, Malawi seems to be caught in a vicious cycle wherein poverty and food insecurity increase people’s vulnerability to HIV infection, while HIV infection reduces peoples’ ability to earn money and produce food (Aarehag et al. 2006; Bryceson et al. 2004; de Waal and Whiteside 2003).

To better understand the context of HIV in Malawi, and to determine potential responses, it is important to consider HIV within an ecosocial framework (Krieger 1994; Loevinsohn and Gillespie 2003). Moving beyond the conventional focus on proximal factors contributing to HIV vulnerability, like individual risk behaviors, it is essential that interventions address poverty and food insecurity (Anema et al. 2009; Miller et al. 2011) as interrelated distal factors in the HIV pandemic, especially in countries like Malawi. Poverty has been consistently recognized as a risk factor for food insecurity (Maes et al. 2010; Nelson et al. 1998; Normén et al. 2005). Focusing on poverty, the Intervention with Microfinance for AIDS and Gender Equity (IMAGE) study, which combined a microfinance program with a gender and HIV training curriculum, was associated with decreased levels of self-reported intimate partner violence and unprotected sex (Kim et al. 2007; Pronyk et al. 2006).

Food insecurity, in turn, is associated with decreased adherence to HIV antiretroviral therapy (Weiser et al. 2009a, b), incomplete virologic suppression (Wang et al. 2011; Weiser et al. 2009a), declines in physical health status, (Weiser et al. 2009a, 2010), worse immunologic status (Kalichman et al. 2010; Weiser et al. 2009b), increased incidence of serious illness (Weiser et al. 2010) increased mortality (Anema et al. 2013) and decreased survival (Anema et al. 2013). In particular, a cross-sectional study in Brazil found that severe food insecurity with hunger is associated with reduced condom use and with increased occurrence of symptoms that may indicate sexually transmitted disease among sexually active women (Tsai et al. 2012). These findings indicate that HIV prevention strategies in food insecure areas should include interventions that target food insecurity.

Increasing critique has targeted the limitations of proximally focused HIV prevention interventions (Weinhardt et al. 1999; Johnson et al. 2003) and pointed out the need for development and assessment of complex, multilayered structural interventions (Gupta et al. 2008) that address root causes and causal pathways linking social, economic, political, and environmental factors to HIV vulnerability in specific contexts (Gupta et al. 2008; Hargreaves 2013). There are significant gaps in knowledge, however, about the development, implementation, and evaluation of structural interventions. First, while integration of food security interventions into HIV/AIDS prevention programs may be essential to curtail the HIV/AIDS pandemic and improve health and quality of life among those infected in resource-poor settings, (Weiser et al. 2011) the literature has offered little guidance to international policy makers, such as the World Food Programme (World Food Programme 2003). To our knowledge, there have been no published intervention studies examining the impact of economic status and food security on HIV outcomes in Malawi. Second, complex multilevel structural interventions are expensive. Typically, non-governmental organizations (NGOs) or government agencies implement them. The cost and complexity of study designs that would adequately evaluate real-world structural interventions do not align well with the typical NIH-funded randomized control trial (RCT) model; this presumably could explain the dearth of research. Third, major challenges remain in evaluating the impact of structural interventions. Few NGO interventions are evaluated rigorously to rule out alternative explanations of effects. Perhaps most importantly, few NGO program evaluations involve a control group. Further, most structural intervention assessments are limited to either structural variables on which they directly intervene (such as social norms that condone intimate-partner violence or microcredit program use rates) or key HIV health outcomes only (Gupta et al. 2008). These research gaps in the development, implementation, and evaluation of structural interventions limit their wider dissemination and scale-up in resource-poor countries, where services are desperately needed.

This manuscript describes the rationale, design, and methodology of a five-year study, called SAGE4Health, examining a structural, multilevel intervention carried out by a large NGO, CARE International, in rural Malawi. This longitudinal study represents one of the first attempts to understand the mechanisms and processes through which changes in food security and economic outcomes (i.e., income, household assets, livelihood options) can impact HIV vulnerability (i.e., HIV risk behaviors, malnutrition, HIV infections). It also represents one of the first NIH-funded studies based on an academic-economic development NGO partnership (Delisle et al. 2005; Weinhardt et al. 2009). This type of partnership leverages strengths of NGOs (i.e., their ability to respond quickly to crises and their capacity for large-scale, sustainable development work) and HIV/AIDS researchers’ rigorous study designs and evaluations. In addition to examining pathways linking distal ecosocial factors to HIV vulnerability, this study will provide important information for understanding the impact of multilevel structural interventions on HIV with the potential for sustainable long-term public health benefits. Finally, this collaboration provides a unique opportunity to conduct a detailed study of a multilevel intervention on a scale unlikely to be supported entirely by NIH research funding; in effect, we use the NIH and NGO program funding to enhance both contributions.

The SAGE4Health research team consists of an interdisciplinary NGO-University partnership between CARE USA, CARE Malawi, the University of Wisconsin-Milwaukee (UWM) Joseph J. Zilber School of Public Health, the UWM College of Nursing’s Center for Global Health Equity, the Medical College of Wisconsin’s Center for AIDS Intervention Research, and the University of Malawi. The SAGE4Health study name reflects the key goals of the CARE’s SAFE Program: Savings, Agriculture, Governance and Empowerment. This partnership allowed the academic institutions and NGO to each contribute areas of expertise and learn from one another, rather than creating a costly situation whereby academia attempts to reinvent and implement an intervention in isolation. During the project, the academic collaborators provided the field staff, who had much experience in NGO-based field data collection but less experience in scientific research, with extensive in-person training in human subject research ethics, interviewing skills, data management, data analysis, scientific literature reviews, electronic data collection, participant tracking, oral presentation skills, and dissemination of results.

The study is being conducted in the rural areas of the Kasungu District of central Malawi. Among Malawian adults aged 15–49, approximately 11% live with HIV (National Statistical Office (NSO) and ICF Macro (2011)). Malawi’s population is young, rural, and poor National Statistical Office (NSO) and ICF Macro (2011) and 74% live below international poverty line of US$1.25 per day (UNICEF 2013). The Malawi economy is dominated by the agriculture sector, which employs 80% of the population, accounts for 42% of national GDP, supplies 81% of foreign exchange earnings and contributes significantly to national and household food security (Mataya 2003). Aside from agriculture, Malawi’s economy is also highly influenced by foreign aid. Based on the World Bank Africa Development Indicators 2011 Report, foreign aid accounted for 16.3% of Malawi’s GDP in 2009 (http://data.worldbank.org/sites/default/files/adi_2011-web.pdf). Given that Malawi’s economy receives substantial amounts of assistance, there is great interest from both donors and the Malawian government to understand the types of interventions that effectively create sustainable change in both the health and economic sectors.

Below we delineate investigations using quantitative samples one and two, and the qualitative study, sample three. We discuss the assessment, power calculation (for quantitative studies only), and the data collection and analysis plan separately.

Quantitative studies

Primary outcome assessment for quantitative investigation one and investigation two

Quantitative studies that involved samples one and two use the same survey instrument consisting of 23 modules (Table 2). We selected those measurement modules because they correspond to constructs of a social-ecological framework (Krieger 1994; Loevinsohn and Gillespie 2003) of HIV vulnerability that aligns with the multilevel nature of the SAFE intervention (Figure 3). The scales assess major structural variables, including economic status and food security, HIV vulnerability outcomes, such as self-reported HIV testing, infections, and individuals’ HIV/AIDS risk behavior and risk perceptions. In addition, gender empowerment and gender-based violence are measured. Also included is a verbal autopsy section to record and determine the cause of death of family members. Anthropometric measurements are taken to quantitatively assess the nutritional status of study participants and children under 5 in the household.

1.	Respondent characteristics
2.	Household sociodemographic characteristics
3.	Household economic status/livelihood strategies
4.	Housing and assets
5.	Use of assets for emergencies
6.	Income and expenses
7.	Household dietary diversity
8.	Household food security
9.	Household’s poverty perceptions
10.	Access to services
11.	Sustainable agriculture practices
12.	Personal health
13.	Illness occurrence and healthcare seeking
14.	Childbirth experiences
15.	Family planning
16.	Chronic illnesses
17.	Self-reported STD infections
18.	HIV/AIDS perception of risks, stigma and testing
19.	Male circumcision
20.	HIV risk activities
21.	Gender power
22.	Community cohesion
23.	Anthropometric measurements

Malawian field scientists administered the surveys face-to-face in Chichewa, the language spoken in the study area. Interview data were collected using Samsung Galaxy tablets utilizing the survey research platform Open Data Kit (Anokwa et al. 2009). The password-protected data were uploaded daily from the tablets to a desktop computer.

Major outcomes: Structural factors

Economic outcomes

Instruments measuring economic factors were adapted from the Malawi Diffusion and Ideational Change Project (MDICP) led by Watkins et al. (Anglewicz et al. 2009; Chin 2010) and previous CARE surveys. We focus selectively on important indicators of standard of living, consumption, production and income. Standard of living is measured by assets, income and consumption of basic commodities. Household production is measured by information on the quantities of major agricultural staples produced and the revenues of family enterprises in the past year. We also asked respondents about their relative wealth compared with other family members (specifically, siblings), and compared with other community members.

Food security, agricultural practices and nutrition outcomes

Food security is assessed with six questions from the preliminary CARE survey. These include questions about the months in the past year when the respondent’s household had enough food, projections for sufficient food crop production for the coming year and the household’s strategies when there was not enough food. Agriculture and natural resources are assessed with questions from CARE’s preliminary survey. For each of 23 sustainable agriculture practices (e.g., crop diversification, terracing, contour ridging, mulching), we ask respondents (1) if they know how to do it, (2) if they are applying it, and (3) if they are receiving benefit from it.

Nutrition is assessed with questions about the numbers of meals that household members ate in the past 24 hours, and 13 categories of foods consumed in the past 24 hours, with a yes/no response format. Anthropometric indicators including height, sitting height and weight are also collected for all respondents. Each measure is taken twice. If the difference is too large (e.g., > 1 cm for height), the measurement is repeated and an average of the two closest measures is used. BMI (Body Mass Index = weight/height²) and height-for-age (related to stunting) and weight-for-height (related to wasting) for children under the age of 5 are calculated, using the WHO/NCHS/CDC reference standards (de Onis et al. 2009, de Onis et al. 2006).

Major outcomes: HIV outcomes, HIV risk behavior and vulnerability variables

Subjective HIV testing, subjective HIV status and infection risk

Subjective HIV status and subjective HIV infection risks are elicited by having the respondents state the likelihood (from 0 = no risk to 10 = absolutely infected) that they (i) perceive themselves to be already infected, (ii) become infected in the next 5 years and (iii) ever become infected. These measures were developed for the MDICP (Anglewicz et al. 2009). HIV risk behavior (Weinhardt et al. 1999, 1998) measures (i.e., unprotected and protected sexual behaviors and sexually transmitted diseases) and a concise battery of measures derived from the Information-Motivation Behavior Skills Model (Fisher et al. 2002), as well as other relevant individual-level psychological factors, also were collected. HIV testing was assessed via self-report. Each participant was asked if they had ever been tested for HIV and, if so, what the result was.

In linking academic research to a “real world,” NGO-led structural intervention, this study design may yield many benefits. First, the cost sharing of the implementation and the research between the NGO and the academic institution allowed for a fully-funded project that may not have been possible had only one funding stream been available. In an environment of constricting research funding, it is prudent to develop such mutually beneficial partnerships. Furthermore, the NGO-academia relationship extended beyond a grant application into capacity building. The NGO field staff developed expertise in research methods, including study protocols, human research ethics trainings, institutional review board (IRB) procedures, sampling methods, qualitative interview skills, data management techniques and computer-assisted quantitative interview skills; the academic staff gained insight into the logistics of on-the-ground program implementation, including international IRB processes, multi-agency grants management, logistics planning in challenging environments, NGO operations and interdisciplinary communication.

Due to the rapid pace with which NGOs undertake programs, compared to the relatively slow pace of federally funded academic research and result dissemination, this partnership has the potential to lead to rapid use of results. Given the program ownership by the NGO, the organization and its funders have a vested interest in understanding the effectiveness, or lack thereof, of its interventions. As the NGO and academic institution work in partnership, results are fed back to the implementers on a more real-time basis. The NGO, thus, has the opportunity to adjust future programming based on this study’s findings.

Limitations of this study include the use of self-reported outcomes and the quasi-experimental design. There is no random assignment to condition at the individual or community level. As such, the intervention and control groups may systematically differ in factors other than intervention exposure. As shown in Table 4, however, despite the nonequivalent-control group design, the intervention and control groups were very similar at baseline on key outcome variables. Another limitation of our design is that we cannot measure the separate effects of each intervention component. However, we feel the benefits of understanding the effects of the program as a whole, as designed, outweigh the benefit of a randomized treatment dismantling design. Bonell et al. (2006), Victora et al. (2004) and others have proposed designs such as ours to evaluate complex interventions in field studies.

There are many perspectives on whether structural interventions and public health should be evaluated using the RCT. Victora et al. argue that blind acceptance of RCT designs may not provide the entire picture of an intervention’s effects (Victora et al. 2004). They note, “Current trends toward acceptance of RCTs as the gold standard source of evidence may limit the knowledge base needed to make sound decisions about public health priorities, and policies” (Victora et al. 2004). External validity and generalizability are key to evaluations of the type presented here, as the partner NGO has the power to utilize findings to make programming changes, as well as replicate the intervention to other areas. RCTs are not the ideal study design when equipoise does not exist, where the intervention is already underway, and where tailoring of the intervention is not possible — for example, it must be consistent across an entire country as in some educational programs (Bonell et al. 2011). Though it would have been possible to construct an RCT of the SAFE intervention, it would not have fit into the established NGO mode of operation.

In addition to examining pathways linking environmental factors to HIV vulnerability, this research will yield important information for understanding the impact of multilevel environmental/structural interventions on HIV with the potential for sustainable long-term public health benefits. Finally, this collaboration represents a unique opportunity to conduct a rigorous effectiveness study of a multilevel intervention being implemented on a scale that is unlikely to be supported entirely by NIH research funding; in effect, we are leveraging the NIH and NGO program funding to enhance the contributions of both.