A Nutrition-Sensitive Agroecology Intervention in Rural Tanzania Increases Children's Dietary Diversity and Household Food Security But Does Not Change Child Anthropometry: Results from a Cluster-Randomized Trial

Abstract

Background: There are urgent calls for the transformation of agriculture and food systems to address human and planetary health issues. Nutrition-sensitive agriculture and agroecology promise interconnected solutions to these challenges, but evidence of their impact has been limited.

Objectives: In a cluster-randomized trial (NCT02761876), we examined whether a nutrition-sensitive agroecology intervention in rural Tanzania could improve children's dietary diversity. Secondary outcomes were food insecurity and child anthropometry. We also posited that such an intervention would improve sustainable agricultural practices (e.g., agrobiodiversity, intercropping), women's empowerment (e.g., participation in decision making, time use), and women's well-being (e.g., dietary diversity, depression).

Methods: Food-insecure smallholder farmers with children aged <1 y from 20 villages in Singida, Tanzania, were invited to participate. Villages were paired and publicly randomized; control villages received the intervention after 2 y. One man and 1 woman "mentor farmer" were elected from each intervention village to lead their peers in agroecological learning on topics including legume intensification, nutrition, and women's empowerment. Impact was estimated using longitudinal difference-in-differences fixed-effects regression analyses.

Results: A total of 591 households (intervention: n = 296; control: n = 295) were enrolled; 90.0% were retained to study end. After 2 growing seasons, the intervention improved children's dietary diversity score by 0.57 food groups (out of 7; P < 0.01), and the percentage of children achieving minimum dietary diversity (≥4 food groups) increased by 9.9 percentage points during the postharvest season. The intervention significantly reduced household food insecurity but had no significant impact on child anthropometry. The intervention also improved a range of sustainable agriculture, women's empowerment, and women's well-being outcomes.

Conclusions: The magnitude of the intervention's impacts was similar to or larger than that of other nutrition-sensitive interventions that provided more substantial inputs but were not agroecologically focused. These data suggest the untapped potential for nutrition-sensitive agroecological approaches to achieve human health while promoting sustainable agricultural practices.

Keywords: agrobiodiversity; agroecology; child diet; dietary diversity; food security; nutrition-sensitive agriculture; participatory interventions; smallholder farmers; sub-Saharan Africa; women's empowerment.

FIGURE 1

Intention-to-treat impact of the Singida Nutrition and Agroecology Project intervention on children's diet, the primary study outcome (n = 591). (A) β coefficients and 95% CIs of the intervention impact on children's dietary diversity score (range: 0–7 food groups). At baseline, children ate (mean ± SD) 2.3 ± 1.1 and 1.8 ± 0.9 food groups during the postharvest and growing seasons, respectively. (B) β coefficients and 95% CIs of the impact on the proportion of children meeting minimum dietary diversity. At baseline, 14.5% and 4.2% of the children met the minimum acceptable diet during the postharvest and growing seasons, respectively. β coefficients were derived using linear regression for continuous outcomes and linear probability regression for binary outcomes, controlling for child age and household-level fixed effects; 95% CIs were calculated accounting for clustering at the village level using the wild bootstrap method. Full regression results are presented in Supplemental Table 2. The arrows indicate the direction of favorable outcome. pp, percentage points.

FIGURE 2

Intention-to-treat impact of the Singida Nutrition and Agroecology Project intervention on food security, a secondary study outcome (n = 591). (A) β coefficients and 95% CIs of the intervention impact on the Household Food Insecurity Access Scale score. At baseline, the Household Food Insecurity Access Scale score (mean ± SD) was 6.8 ± 5.7 and 12.4 ± 6.4 at the postharvest and growing seasons, respectively. (B) β coefficients and 95% CIs of the impact on the proportion of households experiencing moderate or severe food insecurity. At baseline, 71.4% and 86.8% of households experienced moderate or severe food insecurity during the postharvest and growing seasons, respectively. β coefficients were derived using linear regression for continuous outcomes and linear probability regression for binary outcomes, controlling for household-level fixed effects; 95% CIs were calculated accounting for clustering at the village level using the wild bootstrap method. Full regression results are presented in Supplemental Table 3. The arrows indicate the direction of favorable outcome. pp, percentage points.

FIGURE 3

Intention-to-treat impact of the Singida Nutrition and Agroecology Project intervention on child anthropometry, a secondary study outcome. (A and C) β coefficients and 95% CIs of the intervention impact on HAZ and WHZ, respectively. At baseline, children's HAZ (mean ± SD) was –1.4 ± 1.1, children's WHZ at postharvest seasons was –0.62 ± 1.04, and children's WHZ at growing seasons was –0.46 ± 1.01. (B and D) β coefficients and 95% CIs of the impact on the proportion of children who were stunted (B) and wasted (D). At baseline, 26.9% of children were stunted, 10.4% were wasted during the postharvest seasons, and 6.1% were wasted during the growing seasons. β coefficients were derived using linear regression for continuous outcomes and linear probability regression for binary outcomes, controlling for child age and household-level fixed effects; 95% CIs were calculated accounting for clustering at the village level using the wild bootstrap method. Full regression results are presented in Supplemental Tables 4a and 4b. The arrows indicate the direction of favorable outcome. HAZ, height-for-age z score; pp, percentage points; WHZ, weight-for-height z score.

FIGURE 4

Intention-to-treat impact estimates of the Singida Nutrition and Agroecology Project intervention on sustainable agriculture practices (n = 591). At baseline, the crop species richness (mean ± SD) was 1.8 ± 0.9 crops, 5.7% of households practiced intercropping, the number of sustainable practices to improve soil health was 0.39 ± 0.53, and the number of sustainable practices to improve pest management was 0.32 ± 0.53. β coefficients were derived using linear regression for continuous outcomes and linear probability regression for binary outcomes, controlling for household-level fixed effects; 95% CIs were calculated accounting for clustering at the village level using the wild bootstrap method. Full regression results are presented in Supplemental Table 5. The arrows indicate the direction of favorable outcome. pp, percentage points.

FIGURE 5

Intention-to-treat impact estimates of the Singida Nutrition and Agroecology Project intervention on women's empowerment (n = 591). At baseline, women's income allocation and agriculture decision-making power (mean ± SD) was 0.37 ± 0.26 and 0.36 ± 0.22, respectively; husbands helped with 2.1 ± 1.8 household chores in the past month; women spent 10.3 ± 3.2 h on household work in the previous 24 h; and the empowerment score according to A-WEAI was 0.46 ± 0.17. β coefficients were derived using linear regression for continuous outcomes and linear probability regression for binary outcomes, controlling for household-level fixed effects; 95% CIs were calculated accounting for clustering at the village level using the wild bootstrap method. Full regression results are presented in Supplemental Table 6. The arrows indicate the direction of favorable outcome. A-WEAI, Abbreviated Women's Empowerment in Agriculture Index.

FIGURE 6

Intention-to-treat impact estimates of the Singida Nutrition and Agroecology Project intervention on women's well-being (n = 591). At baseline, 14.7% women achieved minimum dietary diversity, 78.4% reported adequate social support, and 41.2% reported probable depression. β coefficients were derived using linear regression for continuous outcomes and linear probability regression for binary outcomes, controlling for household-level fixed effects; 95% CIs were calculated accounting for clustering at the village level using the wild bootstrap method. Full regression results are presented in Supplemental Table 7. The arrows indicate the direction of favorable outcome. pp, percentage points.