Spatial clustering of food insecurity and its association with depression: a geospatial analysis of nationally representative South African data, 2008–2015

While food insecurity is a persistent public health challenge, its long-term association with depression at a national level is unknown. We investigated the spatial heterogeneity of food insecurity and its association with depression in South Africa (SA), using nationally-representative panel data from the South African National Income Dynamics Study (years 2008–2015). Geographical clusters (“hotpots”) of food insecurity were identified using Kulldorff spatial scan statistic in SaTScan. Regression models were fitted to assess association between residing in food insecure hotspot communities and depression. Surprisingly, we found food insecurity hotspots (p < 0.001) in high-suitability agricultural crop and livestock production areas with reliable rainfall and fertile soils. At baseline (N = 15,630), we found greater likelihood of depression in individuals residing in food insecure hotspot communities [adjusted relative risk (aRR) = 1.13, 95% CI:1.01–1.27] using a generalized linear regression model. When the panel analysis was limited to 8,801 participants who were depression free at baseline, residing in a food insecure hotspot community was significantly associated with higher subsequent incidence of depression (aRR = 1.11, 95% CI:1.01–1.22) using a generalized estimating equation regression model. The association persisted even after controlling for multiple socioeconomic factors and household food insecurity. We identified spatial heterogeneity of food insecurity at a national scale in SA, with a demonstrated greater risk of incident depression in hotspots. More importantly, our finding points to the “Food Security Paradox”, food insecurity in areas with high food-producing potential. There is a need for place-based policy interventions that target communities vulnerable to food insecurity, to reduce the burden of depression.

Despite two decades of political advancement since liberation from the apartheid regime and the advent of democracy in 1994, South Africa continues to face intractable poverty 1 , income inequality 2 , and unemployment 3 , that results in many citizens struggling to meet their basic household needs. Consistent with recent global trends 4 , there has also been a disconcerting decline in South African food security, a concept defined internationally as having enough food at all times for an active, healthy life 5 . Since 2011, there has been a growing number of individuals living below the food poverty line (25.2% in 2015 versus 21.4% in 2011 1 ), where approximately a quarter of households are exposed to food insecurity in South Africa (which ranged from 21.5% to 23.9% between the years 2010-2015 6 ).
Depression, our study outcome of this current investigation, is also a serious health condition that affects 9.8% of South Africans in their lifetime 7 . As a leading cause of disability globally 8 , depression is understood to have the highest negative impact on productivity among all medical conditions 9,10 . Evidence based on financial modelling strongly underpins the assertion that untreated depression can be considered both a health and development challenge for many low-and middle-income countries 11 . In South Africa, workplace-related economic loss from depression is estimated to be 4.9% of gross domestic product (GDP) 12 , a major impediment to economic development for a developing nation, which is further compounded by an additional GDP loss of 5% due to high rates of malnutrition 13 .
Food insecurity and depression are dual public health challenges that are often closely linked. Several systematic reviews confirm a link between food insecurity and mental health 14,15 , with likely mechanisms including both biological processes related to nutritional deficiencies and the psychological stress of having insufficient food [16][17][18] . Consumption of several nutrients, including vitamin B12 and antioxidants, appear important to prevent depression 19,20 . As a self-reported construct, food insecurity at household-or individual-level also encapsulates an important psychological aspect (i.e. anxiety about restricted food supplies) that is central to the experience of living with food insecurity 21 .
Achieving food security, as part of the United Nation's Sustainable Development Goal (SDG) 2.1 22 , and in terms of Section 27(1)(b) of the South African constitution 23 , as well as preventing depression onset (SDG 3.4 24 ), remain monumental challenges in the light of budgetary limitations to address multiple development priorities. This dilemma warrants targeted strategies to first identify food-vulnerable geographical locations, so as to inform prioritized interventions for the most affected areas (which may have mental health implications, yet to be examined). With some exceptions from Nigeria 25 and Ethiopia 26 , there are few recent national-level studies from sub-Saharan African (SSA) countries 27 on the spatial variability of household food insecurity. In this current study, we will use of novel spatial statistical methods to first identify geographical clusters ("hotpots") of food insecurity, and assess its association with depression, based on panel data from the South African National Income Dynamics Study (SA-NIDS), a unique nationally-representative sample of South Africans, with geographical coordinates for each household. As alluded to previously, the association between food insecurity and depression is well-established at the individual-level [28][29][30] . Second, we aimed to identify risk of depression onset between individuals living in/outside food insecure hotspot communities over and above the individual effect of household-level food insecurity.

Methods
Data from the South African National Income Dynamics Study (SA-NIDS) wave 1 (year 2008), wave 2 (2010), wave 3 (2012), and wave 4 (2015) were utilized. As the first panel survey of a nationally representative sample of households in South Africa, the SA-NIDS provides unique insights into population trends in living conditions and well-being that is rarely observed in SSA. The SA-NIDS employs a stratified, two-stage cluster sample design to attain a nationally representative sample of households. In the first stage, 400 of the 3,000 Primary Sampling Units (PSUs) from Statistics South Africa's Master Sample were selected for inclusion and proportionally allocated, based on the 53 district councils in South Africa. In the second stage, clusters of dwelling units were systematically drawn within each PSU, with two clusters of 12 dwelling units being selected from each. All consenting adult resident household members (≥ age 15) at the selected dwelling units were administered the SA-NIDS questionnaire, its details being described in a published methodological report 31 . For study participants between the ages of 15 and 17, the SA-NIDS obtained written informed consent from a parent and or legal guardian. In minor cases where there were no parents or legal guardians, written informed consent was obtained from caregivers, consistent with South African National Department of Health Guidelines for ethics in health research 32 .
Our study constructed an incident cohort based on similar methods utilized in previous studies 33,34 , to better isolate the effect of exposure to food insecurity (i.e. residing in food insecurity hotspot communities) at baseline on subsequent depression onset; thus, reducing the likelihood of reverse causation. The purpose of constructing the incident cohort was to ensure that the observed study participants were free of depression initially; and then to track the risk of depression onset over time between individuals exposed and unexposed to food insecurity. Household food insecurity information was only available for wave 1 (baseline) in the SA-NIDS. Therefore, the incident cohort of our current study consisted of wave 1 participants who screened negative for depression in wave 1 (baseline) and for whom there was a further depression rating score available in any of waves 2-4 (thus participants who screened positive for depression in wave 1 were excluded). We right censored the data, either at the earliest observation at which a participant subsequently screened positive, or at the last observation if the participant did not subsequently screen positive for depression. The SA-NIDS study, approved by the Ethics Committee of the University of Cape Town, obtained written informed consent for all study participants. Our use of SA-NIDS data was approved by the University of KwaZulu-Natal Biomedical Research Ethics Committee (BE 111/14).

Measures
Depression. Depression was the main study outcome. Information on depression, based on the ten-item abridged version of the Center for Epidemiologic Studies Depression Scale (CES-D), were obtained from the SA-NIDS Adult questionnaire. The CES-D is a commonly-used psychometric valid/reliable instrument 35,36 that captures self-reported depression-associated symptoms during the past week. Each of the items has four possible responses in a Likert format, ranging from 0 = rarely/none of the time (less than 1 day) to 3 = almost/all of the time (5-7 days). Depression symptomatology is based on a composite score of the 10 items (Cronbach's α = 0.75), with a total score of ≥ 10 being classified as a cutoff to signify significant depressive symptoms, consistent with a previous study 37 . Importantly, in this study we use this cutoff to define 'depression' as a presentation characterized by significant depressive symptomatology, although this cannot be considered equivalent to a clinical diagnosis of major depressive disorder.
Household food insecurity. The information on household food insecurity was obtained from wave 1 (2008) of the SA-NIDS Household questionnaire and captured the adequacy of household food needs over the last month. The measure was based on a three-point Likert scale on the adequacy of the food needs, ranging from 1 = less than adequate for household's needs, to 3 = more than adequate. We generated a final household food insecurity measure (where food adequate/more than adequate = 0 and less than adequate = 1) to be used for the regression analysis (and for generating geographical clustering described immediately below).
Geographical clusters ("hotpots") of food insecurity. Geographical clusters ("hotpots") of food insecurity were identified using Kulldorff spatial scan statistic 38 implemented in SaTScan software 39,40 . The space permutation model was utilized to identify spatial clusters of households with food insecurity, which were unlikely to have arisen by chance, by testing whether they were significantly adjacent in space. The scan statistical analysis imposed a circular window with varying radii continuously for each global positioning system (GPS) coordinate of the SA-NIDs households located throughout South Africa. The analysis identified a number of distinct potential clusters of affected households with the statistical significance of each being tested using a likelihood ratio test. After a food insecurity hotspot was identified (p < 0.05), its strength within compared with outside the hotspot was estimated using relative risk (RR). Individuals were subsequently either classified as exposed to a food insecurity hotspot (i.e. residing in a household located within a hotspot) or unexposed to a food insecurity hotspot (i.e. residing in a household located outside the hotspots). The SA-NIDS household GPS coordinates were accessed (with permission) from the DataFirst's Secure Data Centre at the University of Cape Town.

Statistical analysis.
First, a descriptive analysis of the sociodemographic details of the incident cohort was conducted. Second, geographical clusters of food insecurity (i.e. hotspots) were identified (based on the method described above) and the socio-demographic correlates of these 'hotspot populations' assessed. Socio-demographic disparities between the hotspot exposed and unexposed were tested using the second-order correction method for survey design 41 and subsequently converted into F statistics. Third, we investigated the baseline association between residing in food insecure hotspot communities and the likelihood of depression by fitting a generalized linear model 42 (GLM) based on the prevalence cohort (labelled Model 1). The prevalence cohort, as opposed to incident cohort, included all adults (i.e. 15,630 adults) regardless of depression status at baseline. This analysis was conducted to supplement the results from the above-mentioned second analysis, and to quantify the 'immediate effect' of residing in food insecurity hotspot on depression at baseline. Lastly, for the primary focus of our study, we investigated the long-term risk of depression onset due to exposure to hotspots over time based on the incident cohort who were depression-free at baseline. For this final analysis, given the repeated measurements of the SA-NID data structure, we fitted two generalized estimating equation 43 (GEE) regression models. The first (labelled Model 2a) is a model based on variables included in Model 1. The second (labelled Model 2b) is our full model with household food insecurity variable nested within Model 2a. The significance of the nested model that added household food insecurity was tested using Wald test. All regression models were adjusted for sociodemographic variables (e.g. gender, race, age, educational attainment, employment status, income, and urban/rural). Given the nature of the complex survey design in the SA-NIDS, all the analyses involving proportion and regressions were adjusted by post-stratification weight to allow our results to better represent the South African population. The construction of post-stratification weight by SA-NIDS is documented in the published report 44 . All methods were performed in accordance with the relevant guidelines and regulations.

Results
Sociodemographic characteristics. Our  Geographical clusters of food insecurity community and its association with baseline depression. Spatio-temporal scan statistics analysis identified three significant clusters of food insecurity (p < 0.05) in South Africa ( Fig. 1 and Table 2

Geographical clusters of food insecurity community and its association on incident depression.
The results of the adjusted regression analysis ( www.nature.com/scientificreports/  Table 2. Spatial clustering of food insecurity was derived from the SA-NIDS household GPS coordinates accessed (with permission) from the DataFirst's Secure Data Centre using SaTScan. The map was created using ArcGIS software by Esri version 10.3. Table 2. Description of the spatio-clusters of food insecurity in South Africa as depicted in Fig. 1

Discussion
Our study investigated the spatial heterogeneity of food insecurity, and its association with incident depression in South Africa, and yielded two significant findings. First, we found a significantly greater incident depression for those residing in a hotspot that had high levels of food insecurity over and above the individual effect of household food insecurity based on incident cohort. This finding points to the likelihood that depression can be associated with either household food insecurity or residing in hotspot communities or both. Second, we found significant differences in the study participant profile, pointing to social vulnerabilities (e.g. [low] household income, employment, and educational attainment) of individuals residing in food insecure hotspot communities. While alleviating depression is often assumed to require psychotherapeutic and pharmacotherapeutic approaches 45 in low-resource setting, addressing the underlying social causes (e.g. poverty and hunger www.nature.com/scientificreports/ in the community) may well be called for. Our findings highlight the need to go beyond biomedical approach, and address broader social determinants of depression in many under-resourced communities in South Africa.
To the best of our knowledge, this is the first study in South Africa that has identified the spatial variability of food insecurity at a national scale, with greater risk of incident depression among individuals residing in food insecure hotspot communities. Our findings on the spatial patterns of "hotpots" provide evidence supporting the "Food Security Paradox"; i.e. food insecurity in areas with a high food producing potential 46 . This paradox poses nuanced challenges in devising place-based policy interventions tailored to vulnerable communities. South Africa in general, being regarded as a food secure nation with the means to produce enough staple foods for all individuals 47 , is nonetheless plagued with widespread chronic household food insecurity; this "Food Security Paradox" phenomenon being particularly evident in KwaZulu-Natal Province (KZN) and its border area with Table 4. Regression model assessing the relationship between food insecurity (both hotspot and household) and depression. The "coloured" is term used by Statistics South Africa 62 , a South African race label that includes children/descendants from black-white, black-Asian, black-colored, and white-Asian unions 63 . HS high school, aRR adjusted relative risk, SE standard error, CI confidence interval. ├ The regression model adjusted based on post-stratification weight (from final observation of the individual panel) to reflect more recent population estimates produced by Statistics South Africa. www.nature.com/scientificreports/ the Eastern Cape Province. KZN's fertile soils are suitable for commercial and small-scale/subsistence farming, being the best-watered province 48 and home to many large agri-business firms 49 . The province has the country's highest proportion of agricultural households (23.0%) 50 , with the hotspot communities being located within municipalities with the highest percentage of subsistence-based livelihoods (South African census 51 ).
In the 1970s, viewing hunger as a social rather than a technical problem, Moore and colleagues in Food first: Beyond the myth of scarcity 52 , documented how the Sahel zone of North Africa exported food to Europe/North America and wealthy African clientele, the amount exceeding that of provided by international food relief in the midst of the famine. The "Food Security Paradox" is also a South African legacy, with apartheid policies replacing adequate and self-sufficient subsistence farming systems with commercial farming, where black South Africans were labor pools, reliant on cash from employment as a basis for attaining food security 47 .
Weaver and Hadley suggest three pathways from food insecurity to mental health problems such as depression namely: nutritional deficiencies with neurobiological consequences; stress generated by "uncertainty in the household ecology"; and perhaps most pertinent to the current South African context, the negative psychological consequences of relative social comparisons of wellbeing within and between communities 14 . Food has significant social functions 53 , and food insecurity may "amplify or magnify relative differences in wellbeing", being a "particularly honest signal" of relative wealth/income and wellbeing 14 , both between households within and between different communities. There is substantial evidence showing the negative mental health effects of social comparisons in relation to household income inequality 54 . In the context of contemporary South Africa, a country characterised by one of the highest indexes of income inequality globally, households and communities experiencing relative food insecurity (in comparison with their neighbouring households and communities) are likely to experience feelings of inadequacy, shame and 'social defeat'-all highly correlated with depressive symptoms and disorders in particular 54 .
The question remains regarding what equity policies can address food insecurity that is also closely linked to improved mental health outcomes for socioeconomically vulnerable populations. While market-based policies for improving food security are necessary, there is also a case to be made for well-planned government policies and interventions, particularly when food security is viewed as a public good 55 . Access to food is also a right guaranteed under Section 27(1)(b) of the Constitution of the Republic of South Africa 23 . There is also a case for reducing negative market externality, given that agriculture was the major driver of habitat loss in KZN 56 . Several government options are worth noting 57 , as they have cross-cutting food security and human well-being implications, such as mental health. Acknowledging that it is plausible that government interventions may give rise to unintended consequences that threaten food security, it is also important to explore the provision and restoration of land/property rights of (previously) disenfranchised individuals for subsistence farming. This seems obvious if ending hunger is to be achieved through the production of sufficient and nutritious amounts of food where it is most needed, and not just through social protection mechanisms such as cash transfers/social grants. Currently, the focus of the South African government policy is not on supporting subsistence farmers, but on transitioning them to semi-commercial farming 58 . A land/property rights-based approach to food security may be less dependent on market forces that determine agricultural commodity and individual purchasing power; and may relate more to the restoration of dignity and psychological liberation [given that black South Africans were forcibly removed from their ancestral (fertile) land for whites as part of the Natives Land Act #27 of 1913 59 ]. While the discussion of implementation choices in the provision and restoration of land/property rights, with or without compensation, is outside the purview of this current study, it is important to contextualize this research within the socioeconomic and political debate around food, land, and social justice in contemporary South Africa. Finally, we urge prioritized interventions focused on food security and its mental health impacts for individuals residing in hotspot communities in South Africa, and in particular throughout KwaZulu-Natal province and its bordering area with the Eastern Cape. These are the communities that the data suggest are most affected by the "Food Security Paradox" and, therefore, most in need of comprehensive and integrated interventions that deal, not only with food insecurity per se, but also the consequent negative impacts on human wellbeing, as well as the human rights of dignity and equality.
The major limitation of this investigation is the lack of clinical data that would allow for a diagnosis of depressive disorder; and we have therefore utilized a measure of significant depressive symptomatology as a 'depression' outcome. Second, food insecurity status was based on self-report, with limited longitudinal measures. Although we assumed the spatial-temporality of food insecure hotspots, we argued, as indicated in the Introduction section, that household food insecurity, although high, remains consistent over time. Notwithstanding these limitations, for the first time, we identified spatial variability of food insecurity at a national scale, with greater risk of incident depression among individuals residing in food insecure hotspot communities, linked directly to a decline in human wellbeing detectable at the population level. The strength of our investigation rests on spatially analyzing GPS data to identify food insecurity patterns, and linking this to the onset of depression at a national-scale; this, to our knowledge, not having been reported previously in studies from SSA. Furthermore, by constructing an incident cohort, we are better able to establish directionality and a causal path from exposure to food insecurity hotspot communities to onset of depression.
Our study highlights cross-cutting SDGs challenges (i.e. 2.1 and 3.4) and comes at a critical juncture when South Africa is grappling with highly contentious land reform issues to address chronic racial inequality 60 -an issue that is further complicated by climate change 61 . As we acknowledge the important role of food insecurity in mental health outcomes, hunger in the community is also an emotional and 'distributive justice' social issue that is linked to the basic human dignity of South Africans previously dispossessed of their land. Addressing symptoms or manifested mental health challenges will require recognition of this deep-rooted (economic) land injustice that undermined human dignity. Our study underscores the need for place-based policy and prioritized interventions that target communities vulnerable to food insecurity to prevent depression and its associated damaging impact on social development in South Africa.