Depression is a leading cause of disability worldwide(1). In 2017, depressive disorders ranked as the third leading cause of disability globally and led to over 40 million years lived with disability lost in low- and middle-income countries(Reference James, Abate and Abate2). Furthermore, the burden of depressive disorders continues to rise globally(Reference Patel, Saxena and Lund3,4) . The economic consequences of mental disorders are severe, and the global economy is estimated to lose over US$16 trillion between 2010 and 2030, driven partly by early onset of mental health conditions and subsequent productivity loss across the life course(Reference Patel, Saxena and Lund3,Reference Bloom, Cafiero and Jané-Llopis5) . Depression is associated with poor quality of life, cognitive impairment, negative physical health outcomes such as cardio- and cerebrovascular diseases and higher levels of mortality(Reference Patel, Saxena and Lund3,Reference Charlson, Baxter and Dua6,Reference Lépine and Briley7) ; it is also a point of concern for individuals with substance abuse disorders and dementia(Reference Patel, Saxena and Lund3). Moreover, parental depression can also impede the capacity to provide quality childcare(Reference McLearn, Minkovitz and Strobino8–Reference Butler, Young and Tuthill10), therefore casting negative downstream effects to children(Reference Rahman, Patel and Maselko11–Reference Madlala and Kassier14). As such, the need to reduce mortality, morbidity and disability from mental disorders was identified as a specific target within the 2030 Sustainable Development Goals (SDGs) #3(Reference Patel, Saxena and Lund3). Additionally, the WHO has emphasised the importance of strengthening the prevention and treatment of mental health in their Mental Health Action Plan (2013–2020)(15).
A complicated and dynamic web of risk factors interact to cause depression. In addition to biological and developmental factors, such as serotonin and dopamine metabolism(Reference Syvälahti16), socio-environmental influences have an integral and modifiable role in depression(Reference Patel, Saxena and Lund3). Specifically, these determinants include political and environmental factors (e.g. climate change, clean water and violence); social and cultural factors (e.g. education and social support); demographic factors (e.g. age, sex and ethnicity) and economic factors (e.g. food security, employment and assets)(Reference Patel, Saxena and Lund3,Reference Lund, Brooke-Sumner and Baingana17) . The influence and interaction of economic and demographic factors are demonstrated by the fact that the prevalence of depression is consistently higher in women than in men(1,Reference Weissman, Bland and Canino18,Reference Whiteford, Degenhardt and Rehm19) , especially in low- and middle-income countries(1,Reference Pereira, Andrew and Pednekar20) .
Food security, defined as adequate access to the quality and quantity of food needed for an active and healthy life(21), plays a prominent role in mental health(21). The link between food security and mental health has been found to be bidirectional(Reference Huddleston-Casas, Charnigo and Simmons22) and is posited to operate through biological(Reference Wachs23) and psychosocial pathways(Reference Weaver and Hadley24,Reference Hadley and Patil25) . Biological pathways include inadequate access to nutritious foods leading to nutritional deficiencies associated with poor mental health status(Reference Wachs23,Reference Jebena, Lindstrom and Belachew26,Reference Sparling, Nesbitt and Henschke27) and physical effects associated with food insecurity, such as stomach aches and headaches, which can impact mental health status(Reference Jebena, Lindstrom and Belachew26,Reference Heflin, Siefert and Williams28) . Psychosocial pathways include inadequate access to sufficient preferred food creating stress(Reference Weaver and Hadley24,Reference Hadley and Patil25,Reference Cole and Tembo29–Reference Hadley and Crooks32) and experiences of stigma from not being able to fulfil social expectations of providing food for the household(Reference Piperata, Schmeer and Rodrigues33).
Empirical findings demonstrate the association between food security and women’s depression. Cross-sectional observational evidence from Weaver and Hadley’s systematic review(Reference Weaver and Hadley24), Pourmotabbed et al.’s systematic review(Reference Pourmotabbed, Moradi and Babaei34) and Tribble et al.’s meta-analysis(Reference Tribble, Maxfield and Hadley35) demonstrate this relationship. Longitudinal studies from multiple countries, including Tanzania(Reference Hadley and Patil25), India(Reference Patel, Rodrigues and DeSouza36) and Uganda(Reference Palar, Wagner and Ghosh-Dastidar37), strengthen this evidence since they all found that food-insecure women were more likely to experience depressive symptoms. Multiple studies, including Tribble et al. (unpublished results) and Weaver and Hadley(Reference Weaver and Hadley24), have called for the need to further explore the directionality between food security and depression in order to establish causality. The study by Huddleston-Casas et al.(Reference Huddleston-Casas, Charnigo and Simmons22) is the only work, to our knowledge, in which a causal link between food security and depression is stated. This conclusion was reached using data from three annual cross-sectional surveys in a high-income country (the US), analysed using structural equation modelling (sem)(Reference Pearl38). As such, there is still a need for stronger, interventional evidence of a causal relationship between food security and depressive symptoms, especially in low-income countries.
Even without strong evidence that food insecurity causes depression, there is still a strong push to address food security as a depression reduction strategy. For example, in their review on global mental health literature in the context of SDG targets, the Lancet Commission on global mental health emphasised the importance of addressing social determinants of mental health by meeting various SDG(Reference Patel, Saxena and Lund3) in addition to addressing the stigma, cost and availability barriers of psychological therapies(Reference Patel, Saxena and Lund3,Reference Singla, Kohrt and Murray39,Reference van Ginneken, Tharyan and Lewin40) . The Commission even specifically recommends policymakers to ‘reduce the prevalence of depression through improved food security’(Reference Patel, Saxena and Lund3). This approach to depression reduction may be important because although there are known and effective treatments for mental disorders, a significant proportion of people affected by depression in low- and middle-income countries never receive such treatments(Reference Araya, Zitko and Markkula41). The Commission’s recommendation was also echoed by the WHO Mental Health Action Plan for 2013–2020(15) and global health academics, such as Tsai(Reference Tsai, Bangsberg and Frongillo30) and Hadley(Reference Hadley, Tegegn and Tessema42).
Nutrition-sensitive agriculture interventions, i.e. agricultural interventions aimed to improve underlying determinants of nutrition(Reference Ruel and Alderman43,Reference Ruel, Quisumbing and Balagamwala44) are one example of interventions to address food insecurity. Agroecology, a holistic approach to growing food using ecological methods, such as crop diversification and compost, and concurrently addresses the health, social and economic inequities of the food system, is another(Reference Gliessman45). Both approaches are expected to improve food security by improving the diversity of household agricultural production, increasing household resilience in times of climatic shock and improving women’s nutritional knowledge, input to and control over household and agricultural decisions(Reference Ruel and Alderman43,Reference Rosenberg, Maluccio and Harris46) . There is evidence in support of these mechanisms. For example, a systematic review of nutrition-sensitive agriculture interventions found that they have consistently improved the dietary diversity of their participants(Reference Ruel, Quisumbing and Balagamwala44). In Zambia, a nutrition-sensitive agriculture study improved food access, one facet of food security, over 4 years of interventions(Reference Rosenberg, Maluccio and Harris46). A participatory nutrition-sensitive agroecology intervention in Malawi, which incorporated lessons on gender equity, nutrition and ecological approaches to agriculture, increased food security after 2 years(Reference Kangmennaang, Bezner Kerr and Lupafya47).
Despite the plausibility of nutrition-sensitive agriculture and agroecology (in particular, because of its emphasis on equity) to improve mental health, there is a dearth of empirical evidence to support this relationship. In fact, we believe our recent finding that a participatory nutrition-sensitive agroecology intervention reduced the prevalence of probable depression is the first agricultural intervention to report positive mental health impacts(Reference Santoso, Bezner Kerr and Kassim48). Specifically, we found that Singida Nutrition and Agroecology Project (SNAP-Tz) reduced the prevalence of probable depression among Tanzanian women farmers by 11·4 percentage points. However, the role of food security in this impact remains unclear.
We, therefore, investigated the plausible but untested potential for changes in food security to drive the decrease in the prevalence of probable depression among women smallholder farmers in SNAP-Tz. We first assessed covariate associations of probable depression at the baseline and then conducted mediation analyses modelling probable depression after 3 years, controlling for these covariates.
Methods
Study design & settings
The current study took place in the Singida rural district of Tanzania’s semi-arid central region. In Tanzania, depressive disorders have increased by 35 % between 2007 and 2017 and are ranked as the third leading cause of disability(49). While these rates may be increasing due in part to the more frequent measurement of mental health outcomes, the prevalence suggests that depressive disorders are a significant issue. Smallholder farming is the primary source of livelihood in Singida; households cultivate an average of 2·15 ha(50). Food insecurity is also a persistent issue for the majority of smallholder households in rural Tanzania(Reference Leyna, Mmbaga and Mnyika51,Reference Wandel, Holmboe-Ottesen and Manu52) . In 2012, 49 % of households in the Singida region had poor household dietary diversity, an indicator of food insecurity(50).
Intervention
The Singida Nutrition and Agroecology Project (SNAP-Tz; NCT02761876) was a cluster-randomised effectiveness trial that investigated the effects of a participatory, nutrition-sensitive, agroecological intervention on improving child’s diet through improvements in food security, sustainable agriculture, gender equity and women’s well-being( Reference Santoso, Bezner Kerr and Kassim48 ,Reference Bezner Kerr, Young and Young53,Reference Santoso54) . In-depth details on study design are reported elsewhere(Reference Bezner Kerr, Young and Young53,Reference Santoso54) . Briefly, the intervention consisted of a male and female ‘mentor farmer’ leading their village peers in participatory learning about sustainable farming, legume intensification, nutrition and women’s empowerment.
The study enrolled 598 households: 25–30 households from each of twenty villages, with ten villages randomised to receive the intervention and the other ten to receive the intervention at the end of the study. Village selection criteria included their village leadership’s willingness to participate in the study, having more than 200 children under 5 years, and not participating in other interventions. Twenty villages were ultimately included and paired based on the number of months of food security, predominant soil type and proximity to health clinics. Household eligibility criteria included (1) being food insecure as defined by the community; (2) having a child under one year old in Jan 2016; (3) having access to land and planning to farm in the coming year; (4) intending to reside in that village for the next 3 years and (5) being interested in experimenting with new farming techniques. From amongst these households, two ‘mentor farmers’ (one man, one woman) were elected by participating households in each village to facilitate participatory learning on nutrition-sensitive agriculture.
The twenty mentor farmers from the ten intervention villages were trained on nutrition-sensitive agroecology during a farmer-to-farmer learning exchange with Malawian farmers(Reference Kangmennaang, Bezner Kerr and Lupafya47) and the Farming for Change curriculum(Reference Bezner Kerr, Young and Young53). The curriculum encouraged participatory learning methods, e.g. experiential-based learning and theatre, to educate smallholder farmers with limited literacy on topics of agroecology, climate change, nutrition, gender and social equity. The mentor farmers then facilitated learning exchanges on curriculum topics during monthly community meetings and regular household visits. Additionally, each participating household received a mix of legume seeds (e.g. cowpea, pigeon pea, groundnut and soya), adequate to plant 0·1 ha at the beginning of the farming season during the first 2 years of the study.
For the current analysis, we only included married women (n 548) from the study because the relationship of food insecurity and depression would likely greatly differ from single and widowed women(Reference Kornblith, Green and Casey55,Reference Islam56) .
Data collection
Four annual household surveys were conducted between 2016 and 2019 using questionnaires administered by local enumerators at the participant’s residence or public village meeting place. The data collection team consisted of twenty local enumerators, and each survey took about 1 hour to administer. Survey pre-testing was performed to ensure participant comprehension and accurate outcome measurement within the questionnaire.
Key outcomes
Depressive symptoms were evaluated using the Center for Epidemiologic Studies Depression Scale (range: 0–60)(Reference Radloff57). The Center for Epidemiologic Studies Depression Scale is composed of 20 items that query the frequency (0 = rarely or never, 1 = sometimes and 2 = often, 3 = most of the time) with which participants have experienced various depressive symptoms, such as sadness and trouble sleeping in the past week. Probable depression was modelled as a binary outcome (Center for Epidemiologic Studies Depression Scale > 17), and this cut-off was validated among a similar population in East Africa(Reference Natamba, Achan and Arbach58).
Food security was measured using the Household Food Insecurity Access Scale (range: 0–27); higher values indicate greater insecurity(Reference Coates, Swindale and Bilinsky59). Women and men were jointly asked to report their household access to food in the prior month. Household Food Insecurity Access Scale has been used broadly to assess food security in rural Sub-Saharan Africa(Reference Jones, Ngure and Pelto60), including in Tanzania(Reference Knueppel, Demment and Kaiser61) and Malawi(Reference Kangmennaang, Bezner Kerr and Lupafya47). Food insecurity was modelled as a continuous variable throughout the analyses.
Covariates
A variety of sociodemographic information was collected across all time points (2016–2019). Covariates assessed for confounding include marital status (monogamous or polygamous), farming as main occupation, ethnic group (Nyaturu or other ethnic groups), religion (Muslim or other religion), years of education, years lived in village before 2016 and dependency ratio, calculated as the number of children (≤ 14) and elders (> 65) divided by number of household members between the ages of 15 and 64(62). The indicator for household wealth was tertiles of an index derived from a principal component analysis of self-reported household asset ownership of thirty-three items in January 2016.
Indicators of social support and gender equity were also assessed as potential confounders of the relationship between food security and depressive symptoms since they have previously been found to be associated with food security and depressive symptoms(Reference Patel, Rodrigues and DeSouza36,Reference Nasreen, Kabir and Forsell63–Reference Rudkoski71) . Social support was indicated using an adapted version of Duke’s Perceived Social Support Scale(Reference Antelman, Fawzi and Kaaya72) (range: 0–40), where women were asked to what extent they liked the amount of help they received in ten different circumstances, such as when they are sick or during household work. Those with a mean social support score ≥ 3 were considered as having low social support, based on Antelman and colleagues’ previous use of the scale in urban Tanzania(Reference Antelman, Fawzi and Kaaya72).
Gender equity indicators include measures of women’s experience with and attitude towards domestic violence, the burden of household work and decision-making power. Domestic violence experience was measured by asking if participants had experiences with any emotional, financial, sexual or physical violence with any family members living inside or outside of the household over the past year (dichotomous). Attitude towards domestic violence was then measured by asking participants if physical violence was justified in seven scenarios (range: 0–7)(73). Two additional indicators of women’s burden of household work were included. One was men’s involvement with seven household chores commonly perceived as ‘women’s work’, such as fetching water, within the past month. The mean number of these activities was then calculated (range: 0–1)(Reference Santoso, Bezner Kerr and Hoddinott74). The other was the number of leisure hours women had in the previous 24 h(Reference Malapit, Pinkstaff and Sproule75). Finally, women were asked if and to what extent they had decision-making power within agricultural and income allocation activities; these items came from the Women’s Empowerment in Agriculture Index questions(Reference Malapit, Pinkstaff and Sproule75). Responses were scored no/little = 0, some = 0·5 and final say = 1, and the mean of the seven responses was calculated (range: 0–1). We used linear splines to split income allocation decision-making scores between groups (0–0·4 v. 0·41–1) because exploratory analysis indicated that the association between probable depression and income allocation decision-making power differs between the two groups (online supplementary material, Supplemental Figure S1). This difference in the nature of the association between the two groups is consistent with previous work that reported that women having more say in decision making without adequate resources was associated with dissatisfaction and social stress(Reference Hutton76).
Data analysis
Evaluating predictors of probable depression
We first described baseline characteristics between study arms using t-tests and Pearson χ 2 tests as appropriate. Standard errors were adjusted for village-level clustering in all cases.
To assess associations of covariates with probable depression at baseline, we calculated risk ratios (RR) for all covariates, including demographics, gender equity, social support and physical health variables, using log-binomial regression models (Table 1). We then used a Poisson approximation to a log-binomial multivariable regression model due to convergence issues, including all significant variables from the bivariate risk ratio estimates (Table 1). Finally, backwards stepwise model selection(Reference Hosmer, Lemeshow and Sturdivant77) was used until all variables remaining in the model were significant (P < 0·05). We chose to keep maternal social support in the final model due to epidemiologic reasoning and previous literature demonstrating relationships between social support, food insecurity and depressive symptoms(Reference Tsai, Bangsberg and Frongillo30,Reference Piperata, Schmeer and Rodrigues33,Reference Na, Miller and Ballard78) . Adjusted risk ratios (aRR) were calculated from this parsimonious model (Table 1). Standard errors for all models accounted for clustering at the village level. These analyses were performed using Stata 16(79).
Table 1 The risk of probable depression (CES-D > 17) at baseline of SNAP-Tz (January 2016) in bivariate and multivariate models. Food insecurity, domestic violence experience, men’s involvement with household chores typically done by women and higher income allocation decision-making power were significantly associated with a greater likelihood of probable depression among smallholder farmers in Tanzania in log-binomial multivariable regression (n 548)
* P < 0·01;
** P < 0·05.
† Wealth tertile is based on asset index score, developed using principal component analysis from household’s ownerships of any land, metal roof, electricity, ox plow, solar panels, cell phone, radio, modern beds, mosquito net, books, bicycle and cattle.
‡ Dependency ratio calculated as number of children (< 14 years) and elders (> 65 years) divided by number of adult household members (15–64 years).
§ Cut-off from Antelman et al.(Reference Antelman, Fawzi and Kaaya72).
|| In the past year.
¶ World Bank indicator(73).
†† Modified Women’s Empowerment in Agriculture Index(Reference Malapit, Pinkstaff and Sproule75).
Mediation analysis
To understand food insecurity’s role in the intervention’s impact on women’s depressive symptoms between 2016 and 2019, we carried out mediation analyses using structural equation modelling estimation of total, natural direct and natural indirect effects(Reference Petersen, Sinisi and van der Laan80). In the current analysis, the total effect is an estimate of how much odds of probable depression would change if the control group received the interventions. The natural direct effect is an estimate of the effect of the intervention on odds of probable depression as if the intervention had no impact on food security. The calculation of the natural direct effect contrasts the depression scores of the intervention group with the control group, assuming that food security values are those that participants would have had in absence of the intervention, regardless of their intervention assignment. The natural indirect effect represents the effect of the intervention that is due to the effect of the intervention on food security(Reference Hafeman81) (i.e. the proportion of the intervention effect that is mediated by food security), contrasting the food security values that participants would have had under the intervention v. control if all participants had undergone the intervention.
Probable depression was modelled as a binary outcome (Center for Epidemiologic Studies Depression Scale > 17) and food insecurity as a continuous mediator (assuming a normal kernel). Income allocation decision-making power, men’s involvement with household chores typically done by women, domestic violence experience and social support were a priori identified as time-varying confounders of the mediator–outcome relationship and subsequently controlled for in mediation analyses. To ensure temporality of the mediation analysis, time-varying confounders measured during the prior year were used in models for the subsequent mediator (food insecurity scores) and subsequent outcome (probable depression). For example, the 2016 time-varying confounder data were used in models of food insecurity and probable depression in 2017, and time-varying confounder data from 2017 were used in models of food insecurity and probable depression from 2018. Moreover, as neither food insecurity nor probable depression was significantly different between groups at baseline in 2016 (online supplementary material, Supplemental Table S1), baseline food insecurity nor probable depression status were not considered as potential confounders of the mediator–outcome relationship. Mediation analyses were performed via the ‘mediation’ macro in SAS 9.4(Reference VanderWeele82,83) .
Missing data
Baseline missing values ranged from 0 to 6 % for all variables (online supplementary material, Supplemental Table S2), while the number of missing values for probable depression, food insecurity and covariates ranged from 0 to 13 % during follow-up (from 2017 to 2019) (online supplementary material, Supplemental Table S3). Study attrition differed by participant age, ethnic group and length of time living in the village before the study baseline, so these characteristics were included in the imputation models, along with all confounders, mediators and outcomes discussed above (online supplementary material, Supplemental Table S4). Imputation with chained equations with 20 iterations was used to impute missing values of probable depression, food insecurity and covariate data at each time point(Reference Royston and White84). For imputed values below zero or outside of score ranges, post-estimate rounding was used to adjust values into range. Imputation was performed using Stata 16(79).
Results
Characteristics at baseline
The majority of women enrolled in the study were married monogamously, of the Nyaturu ethnic group, and reported farming as their main occupation (online supplementary material, Supplemental Table S1). On average, they were about 30 years old. At baseline, one-third of the married women in each group had probable depression (Control (C): 32·0 %, Intervention (I): 31·9 %, P difference = 0·97, Table S1). More than three-quarters of participants experienced moderate to severe food insecurity (C: 86·9 %, I: 86·5 %, P difference = 0·63, online supplementary material, Supplemental Table S1).
Baseline covariate associations with probable depression
At baseline, probable depression was associated with food insecurity, any experience of domestic violence, high income-allocation decision-making power and lack of men’s involvement with household chores typically done by women (Table 1). Women who experienced higher food insecurity were at a higher risk of probable depression (aRR = 1·06, 95 % CI: 1·03, 1·08). Measures of gender inequity were also correlated with increased risk of probable depression: married women who experienced domestic violence were at higher risk of probable depression (aRR = 1·47, 95 % CI: 1·15, 1·89, Table 1). Income allocation decision-making power scores higher than 0·4 were associated with an increased risk of probable depression (aRR = 2·90, 95 % CI: 1·79, 4·69), while scores 0–0·4 did not have any significant association with risk of probable depression (Table 1). On the other hand, women who reported men being involved with household chores typically done by women were associated with a decreased risk of probable depression (aRR = 0·60, 95 % CI: 0·40, 0·90). Notably, there were no significant associations between risk of probable depression and social support, dependency ratio, wealth tertiles, occupation, age, marital status, amount of leisure time and education. Sensitivity analyses modelling depression as a continuous variable demonstrated similar results (online supplementary material, Supplemental Table S5).
Mediation analysis
The intervention lowered the odds of probable depression by 43 % (total effect OR = 0·57, 95 % CI: 0·43, 0·70) (Fig. 1a). Differences in food insecurity explained approximately 10 percentage points of the effects of the intervention on odds of probable depression (natural indirect effect OR = 0·90, 95 % CI: 0·83, 0·95). The total effect of the intervention on odds of probable depression was partially attenuated after accounting for differences in food security (natural direct effect OR = 0·63, 95 % CI: 0·47, 0·80) (Fig. 1b). When depression was modelled as a continuous variable, or when income-allocation decision-making power was removed as a confounder, similar results were found (online supplementary material, Supplemental Table S6).
Fig. 1 Diagrams of total effect (Panel A) and natural direct effect (Panel B) estimates for mediation of food insecurity in the nutrition-sensitive agroecology intervention’s impact on odds of probable depression (SNAP-Tz) (n 548). OR with 95 % CI shown correspond to each indicated pathway and ‘X’ represents the muted effect of the intervention on food insecurity in the calculation of the natural direct effect. MI, Men’s involvement with household chores typically done by women; DVE, domestic violence experience; INC, income allocation decision-making power
Discussion
In a nutrition-sensitive agroecology intervention that decreased the prevalence of probable depression amongst women in rural Tanzania, food insecurity played a significant mediating role. Specifically, the intervention lowered the odds of probable depression by 43 % (OR = 0·57, 95 % CI: 0·43, 0·70), and the effect of the intervention on odds of probable depression mediated by food insecurity was approximately 10 percentage points (OR = 0·90, 95 % CI: 0·83, 0·95) (Fig. 1). As such, changes in food insecurity explained approximately 23 % of the intervention’s impact on depression. This is the first evidence from a randomised control trial that lowering food insecurity has a strong effect on reducing women’s depressive symptoms.
This finding is important since it contributes empirical evidence on the previously theorised interdisciplinary, synergistic work between mental health, nutrition and agriculture fields to improve the quality of life for women in low-resource settings through nutrition-sensitive agriculture interventions. It answers the call from proponents of nutrition-sensitive agriculture interventions to target and measure mental health outcomes(Reference Ruel and Alderman43). Simultaneously, it provides evidence in support of the call by those concerned with promoting global mental health for interventions outside of traditional cognitive therapies, such as those discussed by the Lancet Commission on global mental health(Reference Patel, Saxena and Lund3) and others(15,Reference Tsai, Bangsberg and Frongillo30,Reference Hadley, Tegegn and Tessema42) .
In terms of the baseline covariates of risk of probable depression, food insecurity, domestic violence experience, lack of men’s involvement with household chores typically done by women and high-income allocation decision-making power scores were salient (Table 1). These findings largely correspond to existing literature: a multitude of studies have observed significant relationships between depressive symptoms and food insecurity(Reference Weaver and Hadley24,Reference Hadley, Tegegn and Tessema42,Reference Hadley and Patil85,Reference Tsai, Krumdieck and Collins86) , social support(Reference Getinet, Amare and Boru69–Reference Rudkoski71,Reference Prachakul, Grant and Keltner87) and experiences of domestic violence(Reference Patel, Rodrigues and DeSouza36,Reference Nasreen, Kabir and Forsell63–Reference Rogathi, Manongi and Mushi66) amongst women. The finding on income allocation decision-making power, however, is more nuanced. Subgroup analyses revealed positive associations only for women whose income allocation decision-making power scores were above 0·4 (n 237 women, 43 %). Amongst women with lower scores (0–0·4), income allocation decision-making power was not associated with elevated risk of probable depression. One possible explanation for this result is our ethnographic observations that high-income allocation decision-making power might reflect the mental burden that women bear of making decisions about child welfare without having the resources to enact them(Reference Santoso54). For example, in a discussion with participants about the preliminary findings about high rates of probable depression, one woman said ‘husbands put all responsibilities on wives…you may have activities to do and children want some food which you can’t afford, you just wish you could provide…you are depressed because you have a lot to do all alone’(Reference Santoso, Bezner Kerr and Hoddinott74). A similar dynamic was reported amongst Irish women: having more say in decision making without adequate resources was associated with dissatisfaction and social stress(Reference Hutton76).
Our findings on the impacts of the intervention on mental health are consistent with previous qualitative studies on agroecology which document positive impacts on women’s well-being(Reference Bezner Kerr, Hickey and Lupafya88–Reference Calderón, Jerónimo and Praun92). Moreover, the study’s participatory agroecology approach may be important for explaining the substantial impact on depressive symptoms. Hallmarks of an agroecological approach include collective action between farmers, explicit efforts to draw on local knowledge in culturally appropriate ways, attention to improving social support and gender relations, and addressing food production holistically with positive environmental, health, social, and nutrition-related outcomes(Reference Wezel, Herren and Kerr93). All of these aspects could improve women’s and men’s mental health. Nutrition-sensitive agriculture interventions that are not agroecological, i.e. they do not attend to social relations or local knowledge, may not see as large of an impact. Furthermore, participatory women’s groups have been proposed as a low-cost alternative intervention to address mental health, especially in settings with low numbers of specialised workers(Reference Tripathy, Nair and Barnett94,Reference Patel and Kirkwood95) . For example, a participatory intervention in rural India introduced women’s groups discussing care-seeking behaviour and clean birth delivery methods in order to reduce maternal depression and improve birth outcomes(Reference Tripathy, Nair and Barnett94). While the current study found a significant reduction in only moderate depression cases over 3 years, participatory strategies may hold unmeasured potential in impacting women’s depression on a community level compared with a clinical intervention’s individual-level impact.
Future research
We encourage future nutrition-sensitive agricultural or agroecological studies to measure impacts on mental health in men(Reference Santoso, Bezner Kerr and Hoddinott74). In nutrition-sensitive and gender equity interventions, there are many calls for equal involvement from both men and women(Reference Ruel, Quisumbing and Balagamwala44,Reference Santoso, Bezner Kerr and Hoddinott74,Reference Wyatt, Yount and Null96,Reference Barker, Ricardo and Nascimento97) , but men have often not been included(Reference Santoso, Bezner Kerr and Hoddinott74). By not including men in interventions in order to achieve key outcomes such as gender equity and food security and by excluding them from the measurements of the impact of these interventions, the idea that only women are in charge of household responsibilities, including its food supply, is reinforced(Reference Santoso, Bezner Kerr and Hoddinott74,Reference Barker, Ricardo and Nascimento97) . Furthermore, because women bear the mental and physical burden of many household responsibilities(Reference Hadley and Crooks32,Reference Johnston, Stevano and Malapit98) , men’s involvement in household activities could help equalise that burden. Finally, it is plausible that food insecurity could mediate the effect of the intervention on depression in men the same way we observed it to do so amongst women. Insights into the possibly synergistic or cumulative effects between men and women’s mental health, food security improvements and gender relations may generate further insights or recommendations for more effective interventions.
A second consideration for future research is to elucidate the role of other mediators in the relationship between nutrition-sensitive agriculture interventions and depressive symptoms. Indeed, a proportion of the intervention’s impact on depressive symptoms after eliminating the effect of food insecurity was still positive and substantial (natural direct effect OR = 0·63, 95 % CI: 0·47, 0·80, Fig. 1b). It is possible that other pathways such as improved gender equity and social support also contributed to the reduction in depressive symptoms. For example, agroecology’s focus on farmers’ autonomy, meaningful work and social networks for learning could be a critical factor(Reference Timmermann and Felix99,100) ; other studies on agroecology have found significant impacts on social capital and farmers’ autonomy which could have mediating impacts on depressive symptoms(Reference Deaconu, Mercille and Batal101,Reference Kansanga, Luginaah and Bezner Kerr102) .
The plausibility of there being many mediators, such as improved gender equity and social support, in this relationship is high, and often times these exposures are found together. For example, Hernandez and colleagues found that maternal depression mediated the relationship between intimate partner violence and food insecurity(Reference Hernandez, Marshall and Mineo103), demonstrating that there may be more dynamics at play within this complicated web. Furthermore, a study in the UK found that food insecurity within low-socio-economic status women was related to a woman’s domestic violence experience and overall burden of mental health problems (i.e. depression, psychosis spectrum disorder or alcohol/drug-related disorder)(Reference Melchior, Caspi and Howard104).
Similarly, social support could also mediate the relationship between food insecurity and women’s depression. Studies in the USA(Reference Kollannoor-Samuel, Wagner and Damio105) and sub-Saharan Africa(Reference Tsai, Bangsberg and Frongillo30,Reference Na, Miller and Ballard78) found a significant relationship between food insecurity, social support and depressive symptoms. On the other hand, in Nicaragua, Piperata et al. found that spousal support and maternal social support networks were not important modifiers of the link between food insecurity and mental distress. This result was speculated to be due to the fear of gossip and embarrassment about food insecurity preventing social support from being sought(Reference Piperata, Schmeer and Rodrigues33). In contrast, in SNAP-Tz, no significant relationship between social support and risk of probable depression was found at baseline (Table 1). Since women at baseline reported very high levels of social support (C: 82·5 %, I: 76·9 %, online supplementary material, Supplemental Table S1), we believe that a ceiling effect may have masked any associations between changes in social support and depressive symptoms over time.
It is important to note that null associations may be due to the homogeneity of the current study sample. For instance, there was no significant association between baseline probable depression and wealth or years of education (Table 1). Because eligibility was based on food insecurity, the study sample was fairly homogenous. While it may appear that participants have variable wealth levels (online supplementary material, Supplemental Table S1), this variable is illusory as 83 % of the sample was severely food insecure and ownership of most high value resources was rare(Reference Santoso54) (online supplementary material, Supplemental Table S1). Studies with samples with greater variation in socio-economic status are needed to investigate the relationship between wealth and depressive symptoms.
Strengths, limitations
Strengths of the current study include randomisation, large sample size, longitudinal analysis and robust statistical techniques. Limitations include reliance on self-reports, which makes the current analysis vulnerable to social desirability bias. This bias could have resulted in systematic underreporting of experiences with sensitive topics such as food insecurity, depressive symptoms and/or domestic violence experience. We took precautions to address this by avoiding leading questions, using different personnel for enumeration and intervention implementation and assurance of ‘no wrong answers’ throughout the survey. We also measured social desirability bias in July 2018 and found that it was both low and non-differential between study arms(Reference Santoso, Bezner Kerr and Kassim48). Moreover, since these measurements were recorded for each participant at multiple time points, relative changes analysed in the longitudinal mediation analyses likely mitigate this potential bias. Additionally, because income allocation decision-making power has a more complex relationship with probable depression than other gender equity covariates (men’s involvement with household chores typically done by women, experience of domestic violence, attitude towards domestic violence and leisure time, Table 1), this relationship should be looked at more carefully in studies that measure these outcomes. Another possible limitation is the study’s external validity; these analyses only included food insecure, married women with a child < 1 year old at enrolment. It will be useful to know if these relationships are observed in other populations.
Conclusions
These findings highlight that food security mediates roughly a quarter of the impact of a participatory nutrition-sensitive agroecology intervention on women’s depressive symptoms. As such, these results demonstrate that nutrition-sensitive agroecology interventions can have broader impacts than previously demonstrated, i.e. they can go beyond improvements in nutrition to include improving mental health. Indeed, it seems possible that nutrition-sensitive agroecology interventions have the ability to be an accessible method of improving women’s well-being in resource-poor settings, working both through changes in food security and other mediators not tested in the current analysis, such as social support, gender equity and wealth. It is therefore our hope that those working in global mental health consider the role of food insecurity in depression interventions, and that those primarily interested in nutritional outcomes consider the impacts of agricultural interventions beyond nutritional status, to include mental health.
Acknowledgements
Acknowledgements: The authors would like to thank the Singida District Council for their cooperation throughout the study, especially Peter Njau for his role as a community guide, all enumerators for their tremendous contribution during data collection, all mentor farmers for their participation and all research assistants for their help in data cleaning and management. We also acknowledge the Young Research Group’s support in reviewing and editing of the manuscript. Financial support: The Singida Nutrition and Agroecology Project was funded by the Collaborative Crop Research Program of the McKnight Foundation and Atkinson Center for a Sustainable Future of Cornell University. SLY was supported by the National Institutes of Health (K01 MH098902). MVS was supported by the Borlaug Fellows in Food Security Program. HMC was supported by the Institute of Policy Research, Northwestern Office of Undergraduate Research and the John and Martha Mabie Fellowship for Public Health Research. Funders had no role in the design, analysis or writing of this article. Conflict of interest: There are no conflict of interest. Authorship: SLY, RBK and MVS designed the SNAP-Tz study; SLY and RBK were the principal investigators. MVS, SLY, RBK, HM, NK and EM implemented the study. HMC and MVS conceived the idea for the analyses, conducted statistical analyses and wrote the manuscript. LB conducted an analysis of baseline data as the groundwork for this manuscript. LP supported data analysis and interpretation of results. TN assisted in implementation of the study and helped provide a local context of study findings. All authors have critically revised the manuscript and read and approved the final version of the manuscript. Ethics of human subject participation: The current study was conducted according to the guidelines laid down in the Declaration of Helsinki, and all procedures involving research study participants were approved by institutional review boards (I.R.B.) in the U.S and Tanzania (Cornell University I.R.B., reference 1511005983; Northwestern University I.R.B., reference STU00204506; National Institute for Medical Research, reference NIMR/HQ/R.8a/Vol.IX2656). In addition, administrative permission was sought from Singida District Council, and written informed consent was obtained from all subjects.
Supplementary material
For supplementary material accompanying this paper visit https://doi.org/10.1017/S1368980021001014
