From 1990 to 2014, the number and proportion of undernourished people increased in nine countries in sub-Saharan Africa( 1 ). This increase occurred despite a global focus over the last two decades on the first Millennium Development Goal calling for reductions in poverty and hunger. In 2015, the Sustainable Development Goals were created where Goal 2 was to ‘end hunger, achieve food security and improve nutrition, and promote sustainable agriculture’ by 2030( 2 ). Food insecurity is ‘defined by the lack of secure access to sufficient amounts of safe and nutritious food for normal growth and development and an active and healthy life’( Reference Anderson 3 ). Food insecurity experienced as an adult can lead to serious physical health consequences( Reference Stuff, Casey and Szeto 4 – Reference Hadley and Crooks 9 ) and has been shown to correlate with psychological distress in some populations( Reference Siefert, Heflin and Corcoran 5 , Reference Hadley and Crooks 9 – Reference Moraes, Marques and Reichenheim 17 ). In sub-Saharan Africa, past research has linked food insecurity to psychological distress among vulnerable populations including HIV-positive patients, pregnant and postpartum women, refugees and immigrants, and other samples of note( Reference Hadley and Patil 18 – Reference Maes, Hadley and Tesfaye 25 ). In addition, a longitudinal study in Zambia found that food insecurity affected mental health among a group of adults in eighty-one households across sixteen villages, and did so even more during the dry season( Reference Cole and Tembo 26 ).
Conceptual framework
First, the relationship between food insecurity and mental health may differ between men and women across different societies depending on local gender norms. In cultures where women traditionally are responsible for food preparation such as in sub-Saharan Africa( Reference Quisumbing, Brown and Feldstein 27 ), the relationship between food insecurity and depression may be stronger for women, which previous research suggests( Reference Carter, Kruse and Blakely 10 , Reference Tsai, Bangsberg and Frongillo 21 , Reference Gupta, Dandu and Packel 28 , Reference Wu and Schimmele 29 ). However, men might still feel responsible for making sure that enough money is available to purchase (or grow) sufficient food. Thus, it is important to assess the relationship between food insecurity and mental health separately for men and women to improve the design of interventions. In addition, the extent to which an independent and direct association exists between food insecurity and mental health after adjusting for socio-economic status should be assessed for both men and women as past work has found economic status to be a predictor of food insecurity( Reference Misselhorn 30 – Reference De Cock, D’Haese and Vink 33 ) as well as a predictor of mental health( Reference Kinyanda, Woodburn and Tugumisirize 34 , Reference Lund, Breen and Flisher 35 ).
Second, the social support and stress-buffering literature( Reference Cobb 36 – Reference Cohen and Wills 38 ) would suggest that social support may moderate the relationship between food insecurity and psychological distress. Although a few studies have found that perceptions of social support buffer the association between food insecurity and psychological distress( Reference Tsai, Bangsberg and Frongillo 21 , Reference Tsai, Tomlinson and Comulada 23 , Reference Wu and Schimmele 29 , Reference Kollannoor-Samuel, Wagner and Damio 39 , Reference Natamba, Mehta and Mou 40 ), at least one study has shown no evidence of effect modification( Reference Kapulsky, Tang and Forrester 41 ). Moreover, no studies have examined how the actual arrangement and composition of social ties around an individual (i.e. the social network) may inhibit (or enhance) the food security–depression relationship. On the one hand, social networks represent the structure through which support may flow and which may also themselves generate support, particularly in places lacking formal sources of assistance( Reference Berkman, Glass and Brissette 42 , Reference Tsai and Papachristos 43 ). People experiencing food insecurity but who have many social ties, who are highly embedded within their community network or who have a tightly connected personal network may feel supported or be able to access resources if needed and exhibit fewer depression symptoms. In contrast, people experiencing substantial food insecurity who have few social ties, who are on the social periphery of a community network or who have a loosely connected personal network may feel less instrumentally or emotionally supported, experience more stress and exhibit more depression symptoms. Similarly, for people with personal contacts who are mostly poor, experiencing food insecurity may lead to more depression symptoms due to perceiving that their personal network would likely not be able to provide material help if asked, as compared with people who are also experiencing food insecurity but who have fewer poor contacts.
On the other hand, social network factors may also serve to heighten negative feelings associated with food insecurity such as shame and embarrassment( Reference Weaver and Hadley 13 , Reference Nanama and Frongillo 44 , Reference Oliva, de Mendonca and Sant’Anna 45 ). For example, a food-insecure individual with many personal ties or who is highly centralized in a social network may begin to feel acute stress as gossip about his or her situation could spread to more people. Similarly, feelings of shame might be greater among food-insecure people with a personal network whom they assume to be less understanding of food insecurity, such as networks with a lower composition of visibly poor people. In addition, food-insecure individuals with a network comprising a small proportion of food-insecure contacts may internalize a negative sense of perceived ‘otherness’ as compared with food-insecure people with proportionally more food-insecure contacts. (This latter form of modification may only happen, however, if contacts do not attempt to hide their food insecurity status.)
Thus, social network work factors may modify the food insecurity–depression relationship; for example, by strengthening or weakening perceptions of support or actual support, or enhancing or reducing feelings of shame. These perceptions and feelings may, in turn, directly increase or reduce symptoms of depression. Alternatively, they may indirectly do so; for example, by reducing the likelihood that food-insecure individuals will request support from their networks. No prior research has assessed whether social network characteristics moderate the association between food insecurity and mental health within the general population. Moreover, sex differences could exist in how social network factors act as modifiers given sex-based effect modification results from past studies when using measures of perceived social support( Reference Hadley, Lindstrom and Tessema 46 – Reference Tsai, Bangsberg and Emenyonu 48 ). For example, food-insecure men might feel embarrassed by their situation among wealthy friends and thus develop further symptoms of depression. In contrast, food-insecure women might not if they are more used to the situation or it is more socially acceptable to talk about being very food insecure among their friends regardless of wealth status.
The current study
To address these gaps in the literature, we undertook a cross-sectional, population-based study in rural Uganda. In this context, the local economy is largely based on subsistence agriculture, and food and water insecurity are common( Reference Tsai, Bangsberg and Emenyonu 48 , Reference Tsai, Kakuhikire and Mushavi 49 ). Transportation is a difficult economic barrier and reinforces geographic isolation( Reference Tuller, Bangsberg and Senkungu 50 , Reference Siedner, Lankowski and Tsai 51 ). Few households have access to electricity, unequal gender norms are strong( Reference Tsai, Kakuhikire and Perkins 52 ) and fertility rates are high( 53 ). Health-care infrastructure is limited and agricultural assistance and training are rare( 53 ).
Our study had two aims. First, we assessed the direct relationship between food insecurity and depression symptoms severity among adult men and women, separately, in rural Uganda after adjusting for several potential confounders such as socio-economic status. Second, we assessed whether social network position, structure and composition (e.g. number of ties, social location within a community network, the structural arrangement of personal ties around an individual, the wealth composition and food security composition of an individual’s immediate contacts) moderated the association between food insecurity and depression for men and women separately.
Methods
Study population
The study targeted all adults (aged 18 years or older) whose main household was located within one parish containing eight villages in rural south-western Uganda, a very low-resource setting about 260 km south-west of the capital city, Kampala. The study team was comprised of several well-trained research assistants from the surrounding district who spoke the local language. They used a census enumeration to continuously search for all 1939 potential participants across the 716 households listed as belonging to the parish. By the end of the data collection period, the research assistants had found and interviewed 1669 eligible people. Among the remaining 270 people, sixteen refused, sixty-two could not be contacted (because the person was away from the parish during every attempted contact), 166 became ineligible as their primary residence had shifted outside the parish, eleven were consistently too incapacitated/sick to participate and fifteen had died. Thus, after excluding the people who were ineligible, incapacitated or who had died, the response rate was 1669/1747=96 %. There was little variation in response rate across villages.
Procedures
Interview materials were translated, back-translated and pilot-tested in an iterative process to ensure accuracy, consistent word choice and linguistic equivalence. Between October 2011 and August 2012, trained local research assistants conducted one-on-one structured interviews (lasting about an hour) with eligible participants, typically at a participant’s place of residence. All participants provided written informed consent, either with a signature or, if there were cultural literacy reasons why a signature was not appropriate, a thumbprint.
Depression symptom severity
We used a slightly modified version of the fifteen-item Hopkins Symptom Checklist for Depression (HSCL-D)( Reference Derogatis, Lipman and Rickels 54 ), by including a sixteenth item (‘Feeling like I don’t care about my health’) that prior studies had included for use in the Ugandan context( Reference Tsai, Bangsberg and Frongillo 21 , Reference Bolton and Ndogoni 55 – Reference Ashaba, Kakuhikire and Vorechovska 57 ). Participants were asked how often in the last 7 d they had experienced each of the sixteen symptoms using a 4-point scale representing ‘not at all’ to ‘extremely often’ (coded 0 to 3). Cronbach’s α was 0·84. The total depression symptom severity score was the mean of all items (no missing permitted), with higher scores representing more symptoms. Some 4·6 % of participants had a missing depression symptom severity score. For use in describing the population and in a sensitivity analysis, participants were classified as symptomatic of having ‘probable depression’ if they had a score of 1·75 or greater, which is a commonly used threshold( Reference Tsai, Bangsberg and Frongillo 21 , Reference Derogatis, Lipman and Rickels 54 , Reference Winokur, Winokur and Rickels 58 ).
Food insecurity
Individual perception of household food insecurity was measured using the nine-item Household Food Insecurity Access Scale (HFIAS or simply food insecurity)( Reference Coates, Swindale and Bilinsky 59 ), which a previous study had adapted for use in Uganda( Reference Tsai, Bangsberg and Frongillo 21 ). Participants were asked how often in the past 30 d they had experienced different food insecurity-related situations using a 4-point scale representing ‘never’ to ‘often’ (coded 0 to 3). Cronbach’s α was 0·85. A summary score was created (no missing was permitted). Higher scores represented greater food insecurity (maximum=27). Some 1·9 % of participants had a missing food insecurity score. Using a validated scoring algorithm, the scores on the raw scale were used to assign respondents to categories of food insecurity severity: ‘none’ (food secure), ‘mildly food insecure’, ‘moderately food insecure’ and ‘severely food insecure’( Reference Coates, Swindale and Bilinsky 59 ).
Social network position, structure and composition
We first employed ‘name generator’ questions to elicit study participants’ social ties( Reference Burt 60 ). The canonical example of a name generator frequently used in US-based surveys is the question embedded in the US General Social Survey: ‘From time to time, most people discuss important matters with other people. Looking back over the last six months – who are the people with whom you discussed matters important to you?’( Reference Burt 60 ). Consistent with the literature( Reference Perkins, Subramanian and Christakis 61 ), however, we adapted five name generator questions for the local context. All participants were asked to name up to six adult parish residents in response to each of the name generator questions. The questions inquired about with whom or to whom, in the past 12 months, the participant (i) spent free time, (ii) discussed financial topics, (iii) discussed health matters, (iv) went to for emotional support and (v) shared, exchanged, received or gave food, in that order. Names could be repeated for each network type. Previous research has shown that by utilizing information from across multiple name generators, estimates of network characteristics are more accurate( Reference Marin and Hampton 62 ). Thus, we collapsed information across the five network types and created several variables to describe an individual’s position within the community social network, the structure of his or her personal network, and the composition of his or her personal network.
We started by calculating: (i) the number of people whom an individual nominated not including duplicates (i.e. ‘out degree’); (ii) the number of people whom an individual nominated who also nominated the individual (i.e. ‘reciprocal degree’); and (iii) the total number of out-going plus in-coming nominations not including duplicates (i.e. ‘total degree’), which is also referred to as personal network size. We then calculated the density of a participant’s personal network by dividing the total number of ties among a participant’s contacts (ignoring nomination direction) by the total number of ties that could have possibly existed among a participant’s contacts. In addition, we calculated a variable representing a participant’s embeddedness within his or her village social network. That is, we calculated how close a participant was to all other people within his or her village network by taking the inverse of the sum of the shortest distance across same-village ties between a participant and every other person in his or her village (i.e. ‘closeness’). We then created equally distributed quintile categories representing an individual’s closeness, which could range from very peripheral (lowest quintile) to very central (highest quintile). These variables are typically included in social network analysis and represent traditional measures of an individual’s centrality in a network and network structure( Reference Wasserman and Faust 63 , Reference Sabidussi 64 ). Finally, to measure personal network composition, we calculated the percentage of a participant’s personal network that was poor (see next paragraph for wealth definition) and the percentage of the participant’s personal network that reported moderate or severe food insecurity. We then categorized these composition percentages of 0 to 100 into quintiles.
Other explanatory variables
Participants also reported age, tribe, marital status, educational attainment, alcohol consumption frequency and HIV status. Age was categorized as: (i) less than 30 years old; (ii) 40–49 years; (iii) 50–59 years; (iv) 60–69 years; and (v) 70 years or older. Participants were either part of the Banyankore tribe or another tribe. Marital status was split as married or other. Education was categorized as primary schooling or less v. secondary schooling or more; and alcohol consumption as two or more times per week v. less often. Participants reported being HIV positive or negative/unknown. In addition, two household-level measures were included: (i) whether there had been a death in the household in the past 12 months; and (ii) household wealth. The latter was measured via a household asset index, by conducting a principal components analysis on twenty-six separate variables representing household assets and housing characteristics. We retained the first principal component to define the asset index( Reference Filmer and Pritchett 65 ) and categorized anyone in a household in the first two quintiles of the asset index as ‘poor’. We also created a variable indicating whether the interview was conducted in the rainy season, as prior research has shown food insecurity to be greater during the dry season in Uganda( Reference Patterson, Berrang-Ford and Lwasa 66 , Reference Mayanja, Rubaire-Akiiki and Greiner 67 ) and across sub-Saharan Africa more generally( Reference Haile 68 ). In addition, dummy variables were created to represent village of residence. These explanatory variables were utilized in our analyses to adjust for potential confounding as prior research has linked them to food security and mental health outcomes in similar populations( Reference Carter, Kruse and Blakely 10 , Reference Tsai, Bangsberg and Frongillo 21 , Reference Cole and Tembo 26 , Reference Anema, Vogenthaler and Frongillo 69 – Reference Whitaker, Phillips and Orzol 71 ).
Statistical analyses
After removing participants with missing responses on any of the study variables, the final analytic population consisted of 1499 people (677 men and 822 women). The distribution of men and women across sociodemographic characteristics was calculated. To test for simple differences between men and women in terms of depression symptom severity, having symptoms indicative of ‘probable depression’ and, separately, being food insecure, we used Pearson’s χ 2 test for categorical variables and the non-parametric equality-of-medians test for the continuous variable. To assess the continuous association between food insecurity and depression symptom severity among adults in rural Uganda, we first fitted a linear regression model stratified by sex with random effects at the household level and a series of fixed effects for explanatory variables (Model 1). Our main independent variable was food insecurity although we also adjusted for age, tribe, marital status, HIV status, alcohol consumption, education, household wealth, having had a death in the household in the past year, village and season in which the interview was conducted. We also fitted a fully adjusted, sex-stratified logistic regression model (where having ‘probable depression’ was the outcome) to check whether those estimates were similar to estimates produced by the linear model. Finally, to test for effect modification by sex, we fitted a supplemental linear regression model where we pooled data for men and women and included an interaction effect between food insecurity and sex.
To address our second study aim and look for evidence of effect modification by social network characteristics, we ran a series of sex-stratified analyses using a second model that included the main effect for a network variable and the associated interaction effect between that network variable and food insecurity in addition to the variables in Model 1. (When Model 2 included the personal network structure variable or a network composition variable, we also adjusted for the main effect of total personal network size.) All statistical tests were two-sided and significance was considered at P<0·05 for all tests.
Results
Overall, the mean level of depression symptom severity was greater among women than among men (1·52 v. 1·28, P<0·001), as was the prevalence of symptoms that were indicative of having ‘probable depression’ (women: n 205, 25 %; men: n 63, 9 %; P<0·001; Table 1). Likewise, 76 % of women and 67 % of men reported any degree of food insecurity (P<0·001), with 24 % of women and 16 % of men reporting severe food insecurity (P<0·001). While the mean number of nominated contacts was a little over five contacts for both men and women, men had a mean total personal network size of 9·1 (sd 6·0) contacts and women had a mean total personal network size of 8·6 (sd 4·7) contacts. Men’s mean closeness score was 0·23 and women’s mean closeness score was 0·24, both with sd of 0·02. About one-third of an individual’s personal network contacts (33 % for men and 30 % for women) were categorized as poor, on average (with sd of 24 and 23 %, respectively). Finally, 48 % of men’s personal network contacts and 56 % of women’s personal network contacts, on average, reported moderate or severe food insecurity (with sd of 23 % for both men and women).
Table 1 Descriptive characteristics of men and women aged 18 years or older across eight villages in rural south-western Uganda
According to sex-stratified regression analyses, the results from Model 1 showed that severe food insecurity among women was associated with a 0·41-point increase in depression symptom severity (95 % CI 0·31, 0·50) compared with women reporting no food insecurity, adjusting for several other factors (Table 2). Separately, severe food insecurity among men was associated with a 0·31-point increase in depression symptom severity (95 % CI 0·24, 0·39) compared with men reporting no food insecurity. Similarly, moderate food insecurity was associated with greater depression symptom severity for both men and women although the estimate was three to four times less than the estimate for severe food insecurity. In relative terms, the magnitude of the association of severe (and moderate) food insecurity with depression symptom severity was greater than that of all the other sociodemographic variables. The logistic regression analyses using the binary outcome of having symptoms indicative of ‘probable depression’ demonstrated similar results; women and men reporting severe food insecurity were about 5·2 times (95 % CI 2·9, 9·4) and 4·6 times (95 % CI 1·9, 11·0) more likely, respectively, to have symptoms indicative of ‘probable depression’ than adults reporting no food insecurity. In addition, women who experienced moderate food insecurity were 2·0 times more likely to have symptoms indicative of ‘probable depression’ (95 % CI 1·1, 3·5). The estimate for the parallel group of men was similar, but not significant.
Table 2 Multilevel linear regression estimates for food insecurity and descriptive characteristics predicting depression symptom severity among men and women aged 18 years or older in eight villages in rural south-western Uganda
Estimates were obtained using a two-level, random-intercepts linear regression model which accounted for clustering at the household level and included dummy variables for rainy season and villages.
*P<0·05, **P<0·01, ***P<0·001.
Supplemental regression results indicated a statistically significant interaction effect between sex and severe food insecurity (b=0·14, 95 % CI 0·02, 0·25, P=0·03); women with severe food insecurity were predicted to have 0·28-point increase in depression symptom severity compared with men with severe food insecurity. There were no differences in depression symptom severity between women and men at lower levels of food insecurity.
Regression analyses testing for evidence of moderator effects found no statistically significant interactions between any of the social network variables and food insecurity for women. (Moreover, there were no main effect associations of social network characteristics with depression symptom severity for women.) For men, however, regression results from Model 2 indicated a statistically significant interaction between food insecurity and closeness quintile (i.e. an individual’s social location within the village network; F=3·41, P=0·019; Table 3). Thus, when stratifying Model 2 by categories of closeness, among men in the most peripheral category of closeness, there was no statistically significant association of any level of food insecurity with depression symptom severity. In contrast, among men in the most central category of closeness, severe food insecurity had a statistically significant association with depression symptom severity (b=0·53, 95 % CI 0·28, 0·77, P=0·004), which was more than 1·5 times larger in magnitude than the overall estimate of the association between food insecurity and depression symptom severity for men. Similarly, among men with the greatest social closeness in their village network, moderate food insecurity also had a statistically significant association with depression symptom severity (b=0·20, 95 % CI 0·02, 0·37, P=0·03), which was more than twice the size of the overall estimate in magnitude.
Table 3 Multilevel linear regression estimates of interaction effects between food insecurity and two social network characteristics, separately, in predicting depression symptom severity among men aged 18 years or older in eight villages in rural south-western Uganda
Estimates were obtained using a two-level, random-intercepts linear regression model which accounted for clustering at the household level and adjusted for all covariates included in Model 1 in Table 2, as well as for total degree.
*P<0·05, **P<0·01, ***P<0·001.
In addition, a statistically significant interaction effect was also estimated in Model 2 between food insecurity and the poverty composition of men’s personal networks (e.g. quintiles of percentage poor; F=2·71, P=0·047). Thus, when stratifying Model 2 by network poverty composition, among men with few (0–20 %) poor contacts in their personal network, severe food insecurity had a statistically significant association with depression symptom severity (b=0·42, 95 % CI 0·28, 0·57, P<0·001). This estimate was about one-third more in magnitude than the overall estimate of the relationship between severe food insecurity and depression symptom severity for men. In contrast, among men with 40 % or more poor contacts in their network, severe food insecurity had a smaller statistically significant association with depression symptom severity (b=0·19, 95 % CI 0·04, 0·35), which was about one-third less in magnitude than the overall estimate. No other social network variables exhibited interaction effects or main effect associations for men.
Discussion
More than two-thirds of adults in the current rural, population-based study in Uganda reported some level of food insecurity. The study demonstrates that within a general population of adults in a rural setting in Uganda, severe food insecurity (and to some extent moderate food insecurity) was a primary predictor of depression symptom severity for both men and women. These results are similar to findings from population-based studies in New Zealand, the USA and Canada( Reference Carter, Kruse and Blakely 10 , Reference Wu and Schimmele 29 , Reference Okechukwu, El Ayadi and Tamers 72 ), and to a nationally representative study on US adults who fell 130 % below the federal poverty level( Reference Leung, Epel and Willett 73 ). The relationship between food insecurity and depression symptom severity for both men and women in the present study remained robust after accounting for other factors including socio-economic status and social network characteristics. Therefore, the potential for food insecurity to lead to significant mental health consequences among the general population in Uganda is of critical public health concern.
In addition, the study found novel evidence that, for men, two social network characteristics (i.e. social centrality in one’s village network and the percentage of contacts in one’s personal network who are poor) may moderate the way their perceived household food insecurity is related to their depression symptoms severity. In contrast, social networks did not seem to matter in the food insecurity and depression dynamic for women. These results differ from a study finding that instrumental social support moderated the association between food insecurity and depression symptom severity for women on antiretroviral therapy in Uganda, but not for men( Reference Tsai, Bangsberg and Frongillo 21 ). The new results also differ from another study finding that instrumental social support moderated the relationship between food insufficiency and depression symptom severity among pregnant women in peri-urban South Africa( Reference Tsai, Tomlinson and Comulada 23 ). It is possible that severe food insecurity is experienced so regularly or profoundly by most women in rural Uganda that they typically experience less shame and stigma compared with men who may not experience severe food insecurity in such a persistent or profound way. Therefore, the mechanisms linking food insecurity and depression may differ, in part, between women in general and men in general in rural Uganda. For example, perhaps the link between food insecurity and depression has a greater biological and/or generalized stress basis for woman and a more social basis for men. If so, then social network factors could be less likely to affect the food insecurity–depression relationship for Ugandan women and more so for men. (Separately, the mechanisms linking food insecurity and depression may also differ between the general population of adults and specific marginalized populations given results from prior work.)
Future studies should explore whether perceived food insecurity-related stigma and shame could be possible mechanisms through which social networks moderate the role of food insecurity on men’s mental health. Perhaps an inability to provide food for oneself or family is (perceived to be) judged very harshly by peers for men in patriarchal settings( Reference Silberschmidt 74 – Reference Wilson 76 ) and thus it leads to feelings of shame. If so, then perhaps severely food-insecure men in the centre of their village social network would feel even more embarrassed and stressed about their situation (due to the greater possibility that more people would find out about their severe food insecurity) than would severely food-insecure men on the periphery of their village social network. This additional stress may turn into increased depression symptom severity for these food-insecure men who are highly embedded within their village social network. Similarly, men may make assumptions about whether their contacts are experiencing food insecurity based on their contacts’ visible assets, which may not be a good indicator if men tend to hide food insecurity. If food-insecure men with few poor contacts assume a significant proportion of their network to either not experience food insecurity and/or not understand their situation, then such assumptions could create a heightened sense of ‘otherness’ and shame, and ultimately lead to greater depression symptoms as compared with food-insecure men with many poor contacts who may not feel like an outsider or as judged.
Although future research is needed to test those hypotheses, the lack of moderation on the food insecurity–depression dynamic by the prevalence of moderate to severe food insecurity among men’s personal networks supports this theory about the potential role of food-related stigma. If men generally try to hide their food insecurity status, then men may not be aware of others’ food insecurity in the same way they are aware of others’ tangible assets. Thus, feeling the need to hide one’s food insecurity in the presence of wealthy people could increase shame and, therefore, depression symptoms, among severely food-insecure men with less poor networks, as well as reduce their likelihood to ask for help, as compared with severely food-insecure men with poor networks. Evidence from a study on men and women in rural Zambia of a negative relationship between relative deprivation and BMI would seem to support this idea of shame and men not wanting to reach out to others if they feel relatively worse off( Reference Cole 77 ).
Taken together, these results suggest that food insecurity interventions may help alleviate depression symptoms (in addition to food insecurity) and that they might be more effective (and produce fewer unintended consequences) if they are designed with different populations in mind. For example, among the overall population of food-insecure women, general food insecurity interventions may reduce women’s depression symptom severity (as well as food insecurity). The same interventions may be less effective, however, in addressing food insecurity and depression for severely food-insecure men and may be contingent on their social network context (e.g. the wealth of their personal network or their embeddedness within their community network). Yet, if the interventions targeting men are designed to reduce the potential for public embarrassment (and thus shame and stigma), then such interventions might be effective among all severely food-insecure men. Finally, the present study does not suggest that food insecurity interventions should be designed to manipulate social network factors, such as network composition. Rather, the present results suggest that the social network context should be taken into account when designing food insecurity interventions that target men. In-depth assessment of the direct relationship between social networks and food insecurity, or between social networks and depression, may be important areas for future study.
Interpretation of the study’s findings is subject to several important limitations. First, we lacked data on the extent to which study participants met formal diagnostic criteria for major depressive disorder. Sub-syndromal symptoms are commonly experienced during the course of mood disorders, however, and are associated with significant psychosocial impairment( Reference Judd, Akiskal and Maser 78 , Reference Judd, Akiskal and Zeller 79 ). In addition, our food insecurity measure did not differentiate between household food insecurity, child food insecurity and other specific aspects of nutritional insecurity. Future studies could validate the study’s results by using another measure( Reference Leyna, Mmbaga and Mnyika 80 ). Second, we lacked data to adjust for any physical health outcomes and other potential confounders that may be linked to both food insecurity and depression symptoms. Third, the data are self-reported and therefore are subject to the challenges inherent to all studies based on self-reported data. Collecting objective data such as stress-related biomarkers, anthropometric data and micronutrient levels may help clarify the physical and psychosocial health-related consequences of food insecurity( Reference Pike and Williams 81 ).
Finally, the cross-sectional design precludes our ability to make causal claims. It is possible that the direction of influence could run from depression symptom severity to food insecurity or in both directions as a longitudinal study of rural families in the USA found( Reference Huddleston-Casas, Charnigo and Simmons 82 ). However, our findings are consistent with the results from four recent longitudinal studies that employed a fixed-effects study design (of which three were conducted with sub-populations in sub-Saharan Africa)( Reference Tsai, Bangsberg and Frongillo 21 , Reference Tsai, Tomlinson and Comulada 23 , Reference Cole and Tembo 26 , Reference Palar, Kushel and Frongillo 83 ). Adjusting for unobserved time-invariant confounding, they found that food insecurity led to depression symptoms. In addition, a study from the USA recently demonstrated that participation in a nutrition assistance programme reduced psychological distress( Reference Oddo and Mabli 84 ), and a recent quasi-experimental study from Bangladesh found that food insecurity was the primary mediator between poverty and distress as well as having a direct effect( Reference Jalal, Frongillo and Warren 85 ). Future research employing a study design to determine causal direction between food insecurity and mental health outcomes, and including measures of perceived food insecurity-related stigma, social support and overall time spent experiencing food insecurity, may uncover potential mechanisms linking social network characteristics, food insecurity and depression, particularly for men.
Conclusions
The current study provides clear evidence of a robust link between moderate to severe food insecurity and depression symptoms for both men and women in the general population in rural Uganda. This finding is particularly concerning given that the vast majority of people in Uganda live in rural areas and many rely on subsistence farming( 86 , 87 ), that Uganda is one of the few countries where the proportion of undernourished people actually increased from 1990–92 to 2012–14( 1 ), that the majority of adults in the present study reported some level of food insecurity, and that depression is a leading cause of disability( Reference Marcus, Taghi Yasamy and van Ommeren 88 ). Thus, population-wide nutrition interventions aimed at improving food insecurity in Uganda might actually address at least two major health and development issues for the general population. Future research is needed, however, to elucidate the role of social network composition on the food insecurity–mental health dynamic for men and why evidence of effect modification does not appear for women. The effectiveness of food insecurity interventions might be greater if men’s social context and social interpretation of food insecurity were acknowledged in the design of such interventions and opportunities for public embarrassment reduced.
Acknowledgements
Acknowledgements: The authors thank the Emikago Study team members for their assistance with data collection and study administration; and Niels Rosenquist, Sae Takada, Peggy Bartek, Anna Baylor, Pamela Mbabazi, Nozmo F.B. Mukiibi and Roberts Muriisa for their assistance with study administration and infrastructure development. Financial support: This study was supported by Friends of a Healthy Uganda and by a Roybal Center grant through the US National Institutes of Health (NIH; grant number P30AG034420). The authors also acknowledge salary support from the NIH (grant number K23MH096620). The funders had no role in the design, analysis or writing of this article. Conflicts of interest: None. Authorship: J.M.P. conceived of and designed the study. J.M.P., V.N.N. and B.K. participated in the data collection. J.M.P. wrote the first draft. J.M.P., V.N.N., B.K., A.C.T., S.V.S., D.R.B. and N.A.C. provided critical revisions. A.C.T., D.R.B. and N.A.C. provided study oversight. Ethics of human subject participation: This study was conducted according to the guidelines laid down in the Declaration of Helsinki and all procedures involving human subjects/patients were approved by the Committee on the Use of Human Subjects in Research, Harvard University; the Partners Human Research Committee, Massachusetts General Hospital; and the Institutional Review Committee, Mbarara University of Science and Technology. We also received clearance for the study from the Uganda National Council for Science and Technology and from the Research Secretariat in the Office of the President. Written informed consent was obtained from all subjects/patients.
