Received: 20 March 2018 Revised: 31 July 2018 Accepted: 14 August 2018
DOI: 10.1111/mcn.12677bs_bs_banner
S U P P L EMEN T AR T I C L EAn agriculture–nutrition intervention improved children's diet
and growth in a randomized trial in Ghana
Grace S. Marquis1 | Esi K. Colecraft2 | Roland Kanlisi3 | Bridget A. Aidam4 |
Afua Atuobi‐Yeboah2 | Comfort Pinto3 | Richmond Aryeetey51School of Human Nutrition, McGill
University, Montreal, Quebec, Canada
2Department of Nutrition and Food Science,
University of Ghana, Legon, Ghana
3Heifer International, Accra, Ghana
4Evidence and Learning Unit, World Vision
International, Washington, DC, USA
5School of Public Health, University of Ghana,
Legon, Ghana
Correspondence
Grace S. Marquis, School of Human Nutrition,
McGill University, 21,111 Lakeshore Rd., Ste.
Anne de Bellevue, Montreal, QC H9X 3V9,
Canada.
Email: grace.marquis@mcgill.ca
Funding information
McGill University; Heifer International; World
Vision Canada; Global Affairs Canada, Grant/
Award Number: S065653Matern Child Nutr. 2018;14(S3):e12677.
https://doi.org/10.1111/mcn.12677Abstract
Stunting in Ghana is associated with rural communities, poverty, and low education;
integrated agricultural interventions can address the problem. This cluster randomized
controlled trial tested the effect of a 12‐month intervention (inputs and training for
poultry farming and home gardening, and nutrition and health education) on child diet
and nutritional status. Sixteen clusters were identified and randomly assigned to
intervention or control; communities within clusters were randomly chosen, and all inter-
ested, eligiblemother–child pairswere enrolled (intervention: 8 clusters, 19 communities,
and 287 households; control: 8 clusters, 20 communities, and 213 households). Inten-
tion‐to‐treat analyses were used to estimate the effect of the intervention on endline
minimum diet diversity (≥4 food groups), consumption of eggs, and length‐for‐age
(LAZ)/height‐for‐age (HAZ), weight‐for‐age (WAZ), and weight‐for‐length (WLZ)/
weight‐for‐height (WHZ) z‐scores; standard errors were corrected for clustering.
Children were 10.5 ± 5.2 months (range: 0–32) at baseline and 29.8 ± 5.4 months (range:
13–48) at endline. Compared with children in the control group, children in the
intervention group met minimum diet diversity (adjusted odds ratio = 1.65, 95% CI
[1.02, 2.69]) and a higher LAZ/HAZ (β = 0.22, 95% CI [0.09, 0.34]) and WAZ (β = 0.15,
95% CI [0.00, 0.30]). Sensitivity analyses with random‐effects and mixed‐effects
models and as‐treated analysis were consistent with the findings. There was no
group difference in WLZ/WHZ. Integrated interventions that increase access to
high‐quality foods and nutrition education improve child nutrition.
KEYWORDS
agriculture, dietary diversity, length‐for‐age, nutrition education, poultry, weight‐for‐age1 | INTRODUCTION
Nutrition indicators among Ghanaian children under 5 years of age
have improved over the past decade at the national level; the preva-
lence of stunting declined by one third to reach 19% in 2014 (Ghana
Statistical Service, Ghana Health Service, & ICF International, 2015;
Ghana Statistical Service, Ghana Health Service, & ICF Macro, 2009).
However, a large disparity in rates across subpopulations persists. The
prevalence of stunting was 63% higher in rural compared with urban
areas in the latest national survey (Ghana Statistical Service et al.,wileyonlinelibrary.com2015). There was about a threefold variation in rates across regions
and wealth quintiles and a sevenfold difference between children of
mothers with no education and those with secondary education. Child
diet also showed variation. Whereas 28.1% of all 6–23 months met the
criteria for minimum diet diversity (≥4 food groups), only 16.9% did in
the Eastern Region (Ghana Statistical Service et al., 2015). Targeted
efforts are needed to address these problems to diminish the physical,
cognitive, and social consequences of poor child nutrition.
An agricultural‐based intervention has the potential to diminish
nutrition disparities as it typically targets a rural population affected/journal/mcn © 2018 John Wiley & Sons Ltd 1 of 10
2 of 10 MARQUIS ET AL.
bs_bs_banner
Key messages
• Stunting remains an issue in rural Ghana, and most
young children in rural Upper Manya Krobo District,
Ghana, did not meet the recommendation for minimal
dietary diversity.
• Integrated support for agricultural production of
nutrient‐dense foods and poultry, combined with
nutrition and health training, improves diet and growth
of young children in rural Ghana.
• Intersector collaborations to implement and sustain
integrated agriculture–nutrition programmes that reach
vulnerable rural populations are warranted.by poverty and low educational opportunities. There are several path-
ways by which such interventions may work. Meeker and Haddad
(2013) suggested that agricultural practices influence nutrition via (a)
food availability, (b) income for food purchases, (c) food prices, (d)
women's social status and empowerment, (e) women's time, and (f)
women's health. Surprisingly, there is limited evidence that agricultural
interventions lead to improved child nutritional status. Over the past
14 years, a series of reviews of published research on agriculture inter-
ventions (e.g., Berti, Krasevec, & Fitzgerald, 2004; Girard, Self,
McAuliffe, & Olude, 2012; Masset, Haddad, Cornelius, & Isaza‐Castro,
2011; Pandey, Dev, & Jayachandran, 2016; Ruel, Quisumbing, &
Balagamala, 2018) have reported mixed results for nutrition outcomes,
often reflecting weak study designs. The wide range of agriculture
interventions and diverse indicators of nutritional status add to the
challenge of linking agriculture and nutrition (Webb & Kennedy,
2014). The greatest effect has been noted with agricultural interven-
tions that directly addressed specific nutritional deficiencies (e.g.,
orange‐fleshed sweet potatoes improved vitamin A status; Hotz
et al., 2012; Low et al., 2007).
The multiple pathways by which agriculture may affect nutrition
suggest that an integrated approach is necessary. Educational activ-
ities can complement agricultural initiatives by helping participants
(a) use agricultural outputs to improve their dietary practices, (b)
mitigate health risks associated with agricultural activities, and (c)
manage agriculture‐related time demands that compete with child
caregiving. Two recently published studies demonstrated the value
of an integrated approach. Olney, Pedehombga, Ruel, and Dillon
(2015) reported on a 2‐year cluster randomized trial in Burkina
Faso that integrated home garden production and education on
child feeding. Anaemia in young infants (difference‐in‐differ-
ence = −14.6 percentage points; P = 0.03) and wasting in infants
(difference‐in‐difference = −8.8 percentage points; P = 0.08)
improved. A 16‐month quasiexperimental project in Ghana inte-
grated microcredit, entrepreneurial training for small businesses,
and nutrition education, with an emphasis on animal source foods
(Marquis et al., 2015). Participation was associated with an increase
in preschool children's height‐for‐age (HAZ; +0.19 z; P < 0.05) and
weight‐for‐age (WAZ; +0.28 z; P = 0.01).
Recent studies support a focus on animal source foods for agri-
culture interventions. Semba et al. (2016) found that essential
amino acids were lower in the diets of stunted compared with
nonstunted Malawian children, suggesting that the lack of high‐
quality protein in the diet may be limiting linear growth. Iannotti
et al. (2017) reported that a randomized controlled trial in Ecuador,
which provided children with one egg per day for 6 months,
reduced the prevalence of stunting by 47% and underweight by
74%. These results demonstrate that agriculture interventions need
to assure a pathway that leads to high‐quality diets for young
children.
Interventions that include poultry farming have several
advantages for child nutrition. First, healthy poultry produce a
large number of eggs that are low‐cost, small packages of high‐
quality macronutrient and micronutrient that can be kept without
refrigeration for an extended duration (Brown, 2003). Second,
poultry farming and the money gained from egg sales are oftenunder women's control and can be used to make market pur-
chases to diversify and enrich the home diet (Homiah, Sakyi‐Daw-
son, Mensah Bonsu, & Marquis, 2012). Finally, as eggs are fragile
and often break, they are likely to be used regularly in home
meals.
Given the evidence that (a) children's diets were poor and stunting
was prevalent in rural Ghana and (b) integrated and targeted interven-
tions work, we proposed an integrated agriculture–nutrition interven-
tion to improve children's diets through increased home production of
nutrient‐rich foods and improved child‐feeding knowledge, income,
and empowerment that would encourage purchases of nutrient‐rich
food from markets. This analysis tests whether the integrated
intervention improved indicators of young Ghanaian children's diet
and growth.2 | METHODS
2.1 | Study design and site
This study was a cluster randomized controlled community trial carried
out within the context of a 5‐year capacity‐building and research pro-
gramme (Nutrition Links [NL]) in the Upper Manya Krobo District
(UMKD) of the Eastern Region of Ghana.
The NL programme provided training on nutrition, gender and
diversity, data management and analysis, and evidence‐based decision
making to government and private sector service providers in the
health, education, agriculture, governance, and finance sectors of the
district. The NL team stratified the six UMKD subdistricts by popula-
tion size and randomly selected three subdistricts to serve as the
study site for this trial.
In the three selected subdistricts, we completed a census of com-
munities (n = 86) with GPS location of all households. Three additional
communities were included in the study site (total n = 89) because
they received services from the Ghana Health Service subdistrict per-
sonnel even though they were slightly outside the subdistricts' politi-
cal boundaries. Based on census data generated, the 89 communities
were then organized geographically into 16 clusters. Our aim was to
have at least 14 households with infants/young children in each
MARQUIS ET AL. 3 of 10
bs_bs_bannercluster, that is, the minimum target for group activities for the inter-
vention. The clusters consisted of either one distinct community or
multiple adjacent small communities (range of 2–10). Within each
cluster, we randomly chose communities until we reached a minimum
of our target number of eligible households per cluster. A total of 39
communities were selected (range: 1–6 communities/cluster) as the
study area.2.2 | Randomization and allocation
The 16 clusters were randomly assigned to treatment group
(sequential, using random numbers). The eight intervention clusters
had 19 communities (range: 1–6), and the eight control clusters
had 20 communities (range: 1–4; Figure 1). Given the nature ofFIGURE 1 Flow of participants through the agriculture–nutrition clusterthe intervention, it was not possible to mask the treatment assign-
ment; therefore, the project maintained separate field staff for the
implementation of the intervention and survey data collection. The
clusters were geographically distant enough from each other to
avoid direct contamination—that is, no control community partici-
pants received inputs or took part in educational activities planned
for intervention participants.2.3 | Participant selection and enrolment
Given limited human and financial resources, enrolment and inter-
vention implementation were carried out in two phases. In Phase
1 (2014–2015), all women with infants (0–12 months) who lived
in the selected communities and who planned to remain in therandomized controlled trial in Upper Manya Krobo District, Ghana
4 of 10 MARQUIS ET AL.
bs_bs_bannercommunity for the duration of the project were invited to enrol in
the trial. In Phase 2 (2016–2017), the age range was expanded to
target young children <18 months to include the planned sample
size. For both phases, additional eligibility criteria for the interven-
tion participants included the timely preparation of (a) a chicken
coop that met project specifications and (b) a fenced home garden
plot. Although the trial was directed to women, the project staff
encouraged the woman's household and community to support
the activities.
All eligible households (n = 277) in the selected communities of
the eight intervention clusters were invited to enrol in Phase 1 in 2014
(Figure 1). After the end of the first phase and the completion of the
12 months of trial activities, we identified newly eligible households
(n = 95) from the same communities and invited them to enrol in
Phase 2 in 2016. Two of the intervention clusters had no newly eligi-
ble households, so only six intervention clusters were active in Phase
2. A total of 34 eligible households were not enrolled, and 51 were
enrolled, but baseline data were lost due to a malfunction of the elec-
tronic tablets. We considered it untenable to enrol participants a sec-
ond time from control cluster communities that had received no
benefit. Thus, the order of including the control clusters was randomly
assigned. To mimic the intervention enrolment, five control clusters
were used in Phase 1 (135 eligible households) and three control clus-
ters (114 eligible households) in Phase 2. Among the control clusters,
36 households were not enrolled.
Ethics approval for the trial was obtained from the institutional
review boards of McGill University (# 822‐0514) and the Noguchi
Memorial Institute for Medical Research at the University of Ghana
(#060/13‐14). All participants provided written informed consent for
themselves and their children. The trial was registered at
Clinicaltrials.gov (NCT01985243).2.4 | Intervention
The 12‐month intervention was an integrated package of agricultural
inputs and training as well as education in nutrition, health care, and
child stimulation for participants. Beekeeping was introduced for
interested households only in Phase 1 for honey harvesting after the
end of the trial. The relevant intervention components are described
below in more detail.
1. Poultry for egg production. Participants received 4 days of inten-
sive training from livestock extension and veterinary officers on a
wide variety of topics to build their knowledge and skills in poul-
try farming. These included 2 days on coop construction using
local materials and 2 days on feeding and caring of poultry, use
of poultry manure, and handling and marketing of eggs. The inter-
vention used Heifer's Passing on the Gift® (POG) community
development approach where repayment of the cost of inputs
provides funds for inputs for new participants. During Phase 1,
each participant received 40 Swiss Brown chickens at point of
lay. The POG funds supported the purchase of 30 chickens for
each Phase 2 participant. To compensate for the lower number
of chickens provided so that women in Phase 2 would have sim-
ilar income to Phase 1, the POG repayment requirement wasreduced by 50%. Initial feed for 1 month and vaccinations were
provided to all participants at no cost; access to purchase feed
after the first month was facilitated because there was no feed
distributor in the district at the time of the trial. Weekly technical
assistance on poultry production and poultry health management
was available in the community throughout the year, provided by
the project staff, sometimes accompanied by district agricultural
extension officers. To assist women with their poultry‐based
small business, the project facilitated egg sales for women who
could not access markets.
2. Home gardens. Project agricultural staff trained participants at
the University of Ghana's Nutrition Research and Training Centre
and in the communities on vegetable gardening, providing infor-
mation on site selection, fencing, seedbed preparation, compost
preparation and use, and organic weed, insect, and pest control.
Households with limited space were encouraged to prepare their
garden in available household receptacles (container gardening).
Participants received planting materials (e.g., one sachet of seeds
and 5–10 kg of vines) for nutrient‐rich vegetables such as
kontomire (Cocoyam leaves, Colocasia esculenta), tomatoes, and
orange‐fleshed sweet potato. Weekly technical assistance was
available in the community throughout the year.
3. Group education. Weekly group education sessions were carried
out using a curriculum of 12 lessons that was repeated during
the year. The lessons emphasized young child diet and health,
with special emphasis on diet diversity and consumption of eggs,
green leafy vegetables, and orange‐fleshed sweet potatoes. The
preparation of the nutrition education activity was delayed and
therefore was provided only during the final 5 months of Phase
1; Phase 2 participants received lessons during all 12 months.
Eight lessons on psychosocial stimulation of young children were
added during Phase 2. These additional lessons focused on child
play and parent–child communication.
4. Community‐wide education. The intervention communities
received training that was accessible to all residents. The training
included (a) food demonstration sessions that emphasized the
consumption of vegetables promoted for home gardens and eggs,
(b) mother‐to‐mother support groups that encouraged optimal
child‐feeding practices, (c) enhanced community‐based growth
monitoring and promotion, and (d) community‐wide discussions
on gender and diversity. Training in the community was provided
by the project as well as through collaborations with district
government staff.2.5 | Data collection
Household and maternal sociodemographic data (e.g., maternal ethnic-
ity and education) had been collected through the NL district‐wide
baseline survey (November 2013–June 2014) and were incorporated
into the data set for this analysis. The intervention‐specific data were
collected using electronic tablets through baseline and endline surveys
completed during the months before and after each phase of the trial.
Household data included characteristics such as family compo-
sition and demographics, household assets, water and sanitation
MARQUIS ET AL. 5 of 10
bs_bs_bannerfacilities, agricultural practices including raising of poultry, use of
district services, and food insecurity. Household food insecurity
was measured with the 15‐item Latin American and Caribbean
Food Security Scale (Food and Agriculture Organization, 2012).
Maternal‐ and child‐specific information included diet, anthropomet-
ric measurements, haemoglobin concentration, health behaviours,
and symptoms of physical and mental (mother only) health. Weight
was measured to the nearest 100 g with a digital scale (Tanita Cor-
poration of America, Inc., Arlington Heights, IL, USA) and length/
height to the nearest 0.1 cm with a stadiometer (Shorr Productions,
Olney, MD, USA). All measurements were done using standard pro-
cedures, and weight and length/height measurements were taken in
duplicate. A third measurement was taken if the discrepancy was
above the World Health Organization (WHO) cut‐off for acceptable
difference in repeated measurements (WHO Multicentre Growth
Reference Study Group, 2006).2.6 | Statistical analysis
The sample size was calculated with an α = 0.05, power = 0.80, effect
size d = 0.35, and variance inflation factor = 1.79, resulting in 227 house-
holds/group. Assuming a loss‐to‐follow‐up of 10%, the sample size
estimate was 250 per treatment group or a total of 500 mother–child
pairs. The data were analysed with STATA version 13 (StataCorp, 2013).
The primary outcome measures of interest were endline diet qual-
ity (minimum dietary diversity [≥4 out of 7 food groups] and con-
sumption of eggs during the previous day) and endline nutritional
status (WAZ, length‐for‐age [LAZ]/height‐for‐age [HAZ], and weight‐
for‐length [WLZ]/weight‐for‐height [WHZ]). A nonquantitative list of
foods consumed yesterday was used to identify children's intake of
seven food groups: grains, roots, and tubers; legumes and nuts; dairy
products (milk, yogurt, and cheese); flesh foods (meat, fish, poultry,
and organ meats); eggs; vitamin A‐rich fruits and vegetables; and other
fruits and vegetables (WHO, 2008). The minimum diet diversity score
of children was coded as a dichotomous variable (<4 food groups [not
minimally diverse] or ≥4 food groups [minimally diverse]). Weight and
length/height data were transformed into standardized deviation
scores using the WHO age and sex growth references (WHO
Multicentre Growth Reference Study Group, 2006).
The wealth variable was derived from a principal components
analysis of 13 household asset variables (floor material, wall material,
cooking fuel, electricity, and ownership of a telephone, radio, televi-
sion, video player, DVD/CD player, refrigerator, sewing machine,
motorcycle, and car). Wealth scores were extracted from the first
component and categorized by tertiles (low, medium, and high). A
food security score was constructed with the 15 questionnaire items
(Food and Agriculture Organization, 2012). Households were catego-
rized by the number of affirmative answers: food secure (0), mildly
food insecure (1–5), moderately food insecure (6–10), and severely
food insecure (11–15).
Unadjusted bivariate analyses were performed to test the rela-
tionship between outcomes and possible covariates using indepen-
dent Student's t test for continuous variables and Pearson's
goodness‐of‐fit chi‐square for categorical variables. Factors were
included initially in the multivariable models if baseline groupcomparisons had a P value < 0.20 or if factors were considered to
be important to child diet or growth based on previous research.
We completed an intention‐to‐treat (ITT) analysis first. We esti-
mated the size of the effect of the intervention on continuous outcomes
(WAZ, LAZ/HAZ, and WLZ/WHZ) and dichotomous outcomes (mini-
mum diet diversity and consumed eggs) using linear regression models
with cluster‐robust standard errors based on the Eicker–Huber–White
robust approach as implemented in the “cluster()” option to the “regress”
and “logit” commands in STATA (Cameron & Miller, 2011). For all
outcomes, we conducted an initial model without covariates and then
a second model that included phase of enrolment and covariates for
the child (baseline age, sex, baseline value of the outcome, and time
elapsed between measurements), mother (education, marital status, and
ethnicity), and household (food security, wealth, and raised poultry previ-
ously). Endline diet diversity was also included initially in the models for
anthropometric outcomes. Backward elimination stepwise covariate
selection procedure was used to select covariates with a P value of
<0.10 (testing across categories) to adjust for the final models. No
interaction terms with intervention were significant and therefore
were not included in the final models. Adjusted odds ratios (aOR) or
beta coefficients from the models are reported. Statistical significance
was set at P < 0.05 unless otherwise indicated.
To assess the robustness of the findings, we used different statis-
tical models and indicators of the outcomes of interest in sensitivity
analyses (Thabane et al., 2013). We have included two additional sta-
tistical approaches run for each of the final ITT models: (a) a random‐
effects model (using “xtreg” and “xtlogit”) and (b) a mixed‐effects
model (using “mixed” and “melogit”). In addition, we ran an as‐treated
analysis that replaced “intervention” with “received inputs” as an indi-
cator of level of participation in the trial. Among the 287 participants
who were enrolled in intervention clusters, 233 received the poultry
and garden inputs (144 in Phase 1, 89 in Phase 2). The primary reason
for not receiving the inputs was because participants had not prepared
the coop and garden. The three models (linear regression with cluster‐
robust standard errors, random effects, and mixed effects) were used
for the as‐treated analyses.3 | RESULTS
A total of 500 women and their infants were enrolled in the trial and
completed the baseline; 287 lived in the eight intervention clusters,
and 213 lived in the eight control clusters (Figure 1). The rate for
enrolment with baseline completion was lower among the interven-
tion compared with the control clusters, partly due to a malfunction
of the electronic data collection system (77.2% vs. 83.5%; P < 0.01).
The first phase of the trial enrolled 316 mother–infant pairs, and the
second phase enrolled 184 women with their infants. There were no
enrolment phase differences in baseline values for the infant anthro-
pometric indices (data not shown). Baseline dietary outcome values
(egg consumed and minimum diet diversity) were not compared, as
phase was associated with child age and diet changed with age. In
Phase 2, children were about 3 months older (12.4 ± 6.3 vs.
9.4 ± 3.9 months; P < 0.01); mothers were more educated (46.7%
vs. 28.9% had completed secondary education or above; P < 0.001)
6 of 10 MARQUIS ET AL.
bs_bs_banner
TABLE 1 Baseline characteristics of participants of an agriculture–
nutrition intervention in rural Ghana, by treatment groupa
Intervention Control P
Characteristic n = 287a n = 213a valueb
Child
Age, months 10.52 ± 5.17 10.43 ± 5.07 0.85
Length‐for‐age, z‐score −0.88 ± 1.27 −0.78 ± 1.30 0.39
Weight‐for‐age, z‐score −0.78 ± 1.12 −0.68 ± 1.27 0.34
Weight‐for‐length, z‐score −0.37 ± 1.08 −0.31 ± 1.24 0.61
Female 143 (49.8) 97 (45.5) 0.34
Consumed eggs in previous 56 (25.3) 35 (21.5) 0.38
24 hrc
Minimal diverse dietd 67 (30.9) 54 (33.8) 0.55
Maternal
Marital status 0.94
Not married/cohabitation 48 (21.8) 46 (22.1)
Married/cohabiting 172 (78.2) 162 (77.9)
Education level completed 0.17
None 54 (24.5) 40 (19.2)
Primary 100 (45.5) 89 (42.8)
Secondary or higher 66 (30.0) 79 (38.0)
Ethnicitye 0.80
Krobo 217 (76.4) 161 (77.4)
Others 67 (23.6) 47 (22.6)
Householdand were in a marriage/union (86.7% vs. 74.7%; P < 0.01), and more
households reported being food secure (50.5% vs. 40.3%; P < 0.04)
compared with Phase 1. Phase was tested in all models and retained
if significant.
There were no baseline treatment group differences in child,
maternal, or household characteristics (Table 1). There tended to be
a group difference in the time interval between baseline and endline
anthropometric measurements (intervention: 19.7 ± 3.2 months vs.
control: 19.1 ± 4.1 months; P = 0.07). Over half of the households
(56%, n = 276) reported experiencing some level of food insecurity
at baseline. Among the children who were over 6 months of age at
baseline, only about one quarter (23.7%, n = 91) consumed eggs on
the previous day and one third (32.1%, n = 121) had a minimally
diverse diet. The mean baseline values for LAZ (−0.84 ± 1.28 z) WAZ
(−0.74 ± 1.18 z), and WLZ (−0.34 ± 1.15 z) demonstrated poor growth
status during infancy.
The rate of study attrition was 14.4%. Total loss‐to‐follow‐up
cases were due to refusal (n = 2), participant moved outside study area
(n = 49), and maternal or child death (n = 5). The remaining cases could
not be found (n = 16). There were no significant differences in child,
maternal, or household characteristics (see list of variables in Table 1)
between those participants who were lost to follow up and those
who completed the study (data not shown). There was no difference
in attrition rate by treatment group (13.9% intervention vs. 15.0%
control; P = 0.73).
Wealth tertilef 0.93
Low 92 (33.0) 70 (33.8)
Middle 95 (34.0) 67 (32.4)
High 92 (33.0) 70 (33.8)
Food securityg 0.87
Food secure 123 (43.3) 95 (45.2)
Mild food insecurity 79 (27.8) 54 (25.7)
Moderate food insecurity 48 (16.9) 39 (18.6)
Severe food insecurity 34 (12.0) 22 (10.5)
Raised poultry in past 140 (48.8) 114 (53.5) 0.29
12 months
Note. Data shown are mean ± standard deviation or n (%).
aTotal n = 428–500 for all but “egg consumed” and “minimal diverse diet”
(intervention n = 220–287; control n = 207–213). Includes all participants
with baseline data for these variables. bIndependent Student's t test for con-
tinuous variables and Pearson's goodness‐of‐fit chi‐square test for categorical
variables. cIncludes only children≥6months (n = 384). dMinimal diet diversity:
includes only children≥6 months (n = 377);≥4 of the following food groups:
grains, roots, and tubers; legumes and nuts; dairy products; flesh foods; eggs;
vitamin A‐rich fruits and vegetables; and other fruits and vegetables (World
Health Organization, 2008). eKrobo: the local ethnic group; others: Akan,
Ewe, Ga, among others. fWealth: tertiles for the first component of a principal
components analysis using 13 household assets: floor material, wall material,
cooking fuel, electricity, and ownership of a telephone, radio, television, video
player, DVD/CD player, refrigerator, sewing machine, motorcycle, and car.
gFood security: classification based on the 15‐item Food Insecurity Experi-
ence Scale (Food and Agriculture Organization, 2012).3.1 | Dietary outcomes
The availability of eggs during the project implementation was high
among the intervention households, with a production of
110.7 ± 50.6 eggs per week (sold, given as a gift, consumed, or lost
to breakage). At endline, the unadjusted prevalence of consuming eggs
in the previous 24 hr was higher in the intervention than control group
(31.5% vs. 22.6%, respectively; P < 0.05). Children who consumed
eggs at baseline were more than twice as likely to consume them at
endline (aOR = 2.25, 95% CI [1.38, 3.66]; Table 2). The aOR for con-
suming an egg over the previous 24 hr did not differ by treatment
group (aOR = 1.35, 95% CI [0.83, 2.20]). The effect of the intervention
on consuming eggs was almost identical in the sensitivity analyses that
used random‐effects and mixed‐effects models (data not shown). The
as‐treated analysis, however, demonstrated a tendency for a higher
odds of consuming eggs among those who “received inputs” compared
with those who did not (aOR = 1.59, 95% CI [0.98, 2.59]).
The endline prevalence of having minimum diet diversity was
higher in the intervention than control group (80.2% vs. 69.5%;
P = 0.02); the unadjusted odds ratio for intervention (but accounting
for clusters), however, was not significant (Table 2). Adjusting for
covariates, children in the intervention group had a 65% higher odds
of having minimum diet diversity at endline compared with children
in the control group (OR = 1.65, 95% CI [1.02, 2.69]). The sensitivity
analyses that used the random‐effects and mixed‐effects models dem-
onstrated a similar estimate for the odds ratio but a slightly weaker
relationship (aOR = 1.65, 95% CI [0.93, 2.94] for both models). The
as‐treated model gave a slightly lower odds ratio that also tended to
be significant (aOR = 1.51, 95% CI [0.94, 2.42]).3.2 | Nutritional status outcomes
During the project period, overall stunting increased (14.0% to 24.3%;
P < 0.001) andwasting decreased (6.3% to 2.9%; P < 0.05); underweight
did not change (11.9% to 12.2%; P = 0.89). The intervention had a pos-
itive direct effect on linear growth. The intervention group LAZ/HAZ
MARQUIS ET AL. 7 of 10
bs_bs_banner
TABLE 2 Logistic regression models for the effect of an agriculture–
nutrition intervention on the diet of Ghanaian rural children, unad-
justed and adjusted for covariatesa
Minimal diet Egg consumption
diversitya,b in last 24 hr
Model 1 Model 2 Model 1 Model 2c
Group
assignment
Control
(reference)
Intervention 1.78 (0.70) 1.65 (0.41)* 1.57 (0.43)† 1.35 (0.33)
Baseline value of 2.25 (0.66)**
outcome
Second phase of 0.21 (0.03)
enrolment ***
Maternal
education
None
(reference)
Primary 1.53 (0.63)
Secondary or 2.68 (0.99)
above **
Marriage status
Married
(reference)
Not married/ 0.31 (0.09)
cohabitation ***
Wealthd
Low
(reference)
Middle 1.02 (0.28)
High 1.53 (0.39)
Constant 2.28 (0.64) 3.09 (0.09) 0.29 (0.06) 0.26 (0.04)
** ** *** ***
Pseudo R2 0.01 0.13 0.008 0.03
Sample n 425 354 425 327
Note. Values are odds ratios (standard errors).
aThis is an intention‐to‐treat analysis with logistic regression models with
standard errors adjusted, accounting for clustering. For Model 2, models
initially included phase of enrolment and covariates for child (baseline
age, sex, baseline value of the outcome, and time elapsed between mea-
surements), mother (education, marital status, and ethnicity), and house-
hold (food security, wealth, and raised poultry prior to project). Backward
elimination stepwise covariate selection procedure was used; the models
retained covariates with a P value of <0.10 for the overall significance
for the variable (not individual categories). No interaction terms with inter-
vention were significant. bMinimal diet diversity: includes only children
≥6 months (n = 377); ≥4 of the following food groups: grains, roots, and
tubers; legumes and nuts; dairy products; flesh foods; eggs; vitamin A‐rich
fruits and vegetables; and other fruits and vegetables (WHO, 2008).
cIncludes only children who were at least 6 months of age at baseline as
the baseline value was retained in the model. dWealth: tertiles for the first
component of a principal components analysis using 13 household assets:
floor material, wall material, cooking fuel, electricity, and ownership of a
telephone, radio, television, video player, DVD/CD player, refrigerator,
sewing machine, motorcycle, and car.
†P < 0.10. *P < 0.05. **P < 0.01. ***P < 0.001.declined less than that of the control group over the trial period (unad-
justed change: −0.38 ± 1.0 z‐score vs. −0.64 ± 0.86 z‐score;
P = 0 < 0.01). In the unadjusted LAZ/HAZ model, the intervention beta
coefficient was not significant (Table 3). After adjusting for baselineanthropometric status and other covariates, children in the interven-
tion group at endline had a higher LAZ/HAZ (β = 0.22, 95% CI
[0.09, 0.34]) than children in the control group. The sensitivity analy-
ses showed similar results for LAZ/HAZ (random‐effects model
β = 0.21, 95% CI [0.09, 0.34]; mixed‐effects model β = 0.22, 95%
CI [0.07, 0.36]; as‐treated model β = 0.25, 95% CI [0.10, 0.41]).
The intervention estimate from the unadjusted model WAZ was
not significant; however, the adjusted model estimate reflected a
higher WAZ for the intervention group compared with the control
group (β = 0.15, 95% CI [0.00, 0.30]; Table 3). The effect of the inter-
vention on WAZ was identical in the sensitivity analyses that used
random‐effects and mixed‐effects models (data not shown). The as‐
treated analysis gave a similar result (β = 0.17, 95% CI [0.03, 0.31]).
There was no treatment group difference in WLZ/WHZ in the ITT
analysis (Table 3) or in any of the sensitivity analyses. Similarly, the ITT
and sensitivity analyses did not reveal any treatment group differences
for stunting, underweight, and wasting outcomes (models not shown).4 | DISCUSSION
Agricultural interventions have the potential to improve child growth;
however, the scarcity of well‐designed studies has limited researchers'
ability to examine causal relationships (Pandey et al., 2016). To the
best of our knowledge, our study is the first randomized controlled
community trial of an integrated agriculture–nutrition intervention to
demonstrate a measureable effect on both LAZ/HAZ and WAZ in
young children.
The intervention mitigated the decline in linear growth that
occurs in late infancy and toddlerhood in Ghanaian communities.
The 2014 mean national LAZ z‐score was −0.5 for 9–11 months and
−1.3 for 24–35 months (for HAZ), a −0.8 z‐score difference across
about the same age range as our participants (Ghana Statistical Service
et al., 2015). The LAZ/HAZ decline in our intervention group
(Δ = −0.38 z‐scores) was <50% of the national cross‐sectional differ-
ence. Our adjusted 1‐year difference in LAZ/HAZ would be consid-
ered a small intervention effect (see Cohen, 1977); for a 20‐month‐
old girl, for example, 0.22 z would represent about 0.75 cm. The
results are consistent with the findings (+0.19 HAZ) from our previous
agricultural intervention work carried out in three different regions of
Ghana (Marquis et al., 2015). In both studies, opportunities existed for
rural women to engage in income generation activities and improve
child caregiving practices without providing any food or supplements
directly. The consistency of results suggests that future comparable
interventions may expect about a 0.20 z‐score improvement in LAZ/
HAZ in young children over 1 year. If the intervention had continued
longer, the results may have mirrored the 2.45‐cm difference associ-
ated with intake of a high protein energy supplement in the Guatemala
3‐year trial (Habicht, Martorell, & Rivera, 1995). Given the reported
long‐term benefits of that intervention (e.g., cognitive development,
Stein et al., 2008; and economic productivity, Hoddinott, Maluccio,
Behrman, Flores, & Martorell, 2008), the small length gains seen here
may reflect future benefits for these Ghanaian children.
The treatment difference in LAZ/HAZ as well as WAZ in this pro-
ject was smaller than that reported by the Lulun project (Iannotti
8 of 10 MARQUIS ET AL.
bs_bs_banner
TABLE 3 Regression models for the effect of an agriculture–nutrition intervention on anthropometric indices (z‐scores) of Ghanaian rural chil-
dren, unadjusted and adjusted for covariatesa
Length‐for‐age/Height‐for‐age z‐score Weight‐for‐age Weight‐for‐length/Weight‐for‐height
Model 1 Model 2 Model 1 Model 2 Model 1 Model 2
Group assignment
Control (reference)
Intervention 0.14 (0.11) 0.22 (0.06)** 0.09 (0.12) 0.15 (0.07)* 0.04 (0.13) 0.07 (0.08)
Baseline value of outcome 0.59 (0.03)*** 0.61 (0.03)*** 0.55 (0.04)***
Baseline age, month 0.02 (0.01)* 0.03 (0.01)* 0.05 (0.01)**
Sex
Male (reference)
Female −0.19 (0.04)** −0.21 (0.06)**
Maternal ethnicityb
Krobo (reference)
Non‐Krobo −0.18 (0.05)** −0.15 (0.06)*
Household food securityc
Food secure (reference)
Mild insecurity −0.02 (0.09)
Moderate insecurity −0.01 (0.09)
Severe insecurity 0.12 (0.14)
Reared chickens at baseline 0.14 (0.08)†
Constant −1.41 (0.06)*** −1.16 (0.11)*** −0.99 (0.08)*** −0.83 (0.15) *** −0.36 (0.10)** −0.64 (0.15)**
R2 0.005 0.48 0.002 0.51 0.0005 0.44
Sample n 415 408 416 411 415 408
Note. Values are beta coefficients (standard errors).
aThis is an intention‐to‐treat analysis with logistic regression models with standard errors adjusted, accounting for clustering. For Model 2, models initially
included phase of enrolment and covariates for child (baseline age, sex, baseline value of the outcome, and time elapsed between measurements), mother
(education, marital status, and ethnicity), and household (food security, wealth, and raised poultry prior to project). Backward elimination stepwise covariate
selection procedure was used; the models retained covariates with a P value of <0.10 for the overall significance for the variable (not individual categories).
No interaction terms with intervention were significant. bKrobo: the local ethnic group; others: Akan, Ewe, Ga, among others. cFood security: classification
based on the 15‐item Food Insecurity Experience Scale (Food and Agriculture Organization, 2012).
†P < 0.10. *P < 0.05. **P < 0.01. ***P < 0.001.et al., 2017). The randomized controlled trial in Ecuador saw a large
increase in LAZ (0.63, 95% CI [0.38, 0.88]) and WAZ (0.61, 95% CI
[0.45, 0.77]). In contrast to the Lulun project where eggs were given to
participants at no cost, the women in our study made the decision about
how to use their eggs each day (sell, give away, or consume). Egg
income was used to meet many of their needs—from purchasing market
foods to paying for health and educational expenses. Thus, our children
did not have the same level of dietary exposure to eggs as the Lulun
intervention children, which may explain part of the difference in results.
The path by which the intervention affected growth indicators is
likely to be multidimensional. The present analysis gives support to
improved diet diversity as one path. This may have happened because
of (a) increased home production, (b) increased income for purchasing
market foods, and (c) increased child‐feeding knowledge. We did not
see a group difference in egg consumption in our ITT analysis. How-
ever, the as‐treated analysis suggested that those who received the
intervention inputs tended to be more likely to consume eggs. In addi-
tion, the outcome data reflected the egg intake after the end of the
project; group differences may have existed during the trial. Further
examination of the dietary data will provide a more in‐depth picture
of how change in specific parts of the diet may be one of the agricul-
ture–nutrition pathways.Another pathway that is likely to have contributed to improving
LAZ/HAZ is women's empowerment. Weekly meetings with educators
and technical staff would be expected to increase women's knowledge
and skills in their income generation activities and caregiving behav-
iours. Nutrition education alone can improve child nutrition if access
to food is not limiting. Improved nutrition education in Peruvian health
services resulted in a 0.272 (0.099 to 0.445) LAZ difference at
18 months of age (P = 0.002; Penny et al., 2006). Peruvian households
were poor, but they had the capacity to act on the nutrition messages
and purchase foods in local markets. Our smaller effect on LAZ/HAZ
may be, in part, due to the limitations that households faced in carrying
out our recommendations. Low wealth ranking and moderate‐to‐
severe food insecurity were reported by about one third of households,
and access to markets was limited for some. Thus, our results are rele-
vant for communities similar to the UMKD; larger effects might be
expected where poverty is less acute.
The strengths of the study included the implementation of a cluster
randomized controlled trial design, the selection of clusters sufficiently
separated to prevent treatment contamination, and the use of two
unique teams of field staff for data collection and for the implementa-
tion of the trial. There were a number of weaknesses. First, there may
have been some selection bias due to enrolment procedures. At
MARQUIS ET AL. 9 of 10
bs_bs_bannerenrolment, all eligible women were informed of the project require-
ments including preparation of a chicken coop and a garden plot. Not
all participants completed the requirement in time; those who did not
received no project inputs. This additional requirement may have led
to a group bias in willingness to participate. However, we did not detect
any baseline differences by treatment group, and the sensitivity analy-
sis results were consistent with the ITT analysis, suggesting that selec-
tion bias was not large enough to affect the results.
Second, due to financial limitations, the project was carried out in
two phases with slightly different inputs available and cluster inclu-
sion. The baseline anthropometric characteristics were not different
by phase, and diet diversity was the only model that retained phase
as a covariate. As the second phase children were slightly older, the
indicator may be reflecting child age more than differences between
the clusters or the years of enrolment.
In summary, this study demonstrated that integrated agricultural
interventions that increase access to high‐quality foods, women's
income‐generating activities, and women's nutrition knowledge can
improve child dietary diversity, LAZ/HAZ, and WAZ. All of the project
activities can be integrated into the mainstream activities of local
district institutions. Financial support for small businesses can be
addressed with microcredit programmes by the rural banks. Depart-
ments of agriculture and health can meet the educational and exten-
sion service needs of the population. Local governments can
facilitate women's access to markets. Support is needed for implemen-
tation research to develop the methods for successfully expanding
integrated agriculture–nutrition activities into sustainable programmes
for vulnerable populations throughout rural Ghana.
ACKNOWLEDGEMENTS
The authors thank Boateng Bannerman and Theresa Thompson‐Colón
for data management and Oduro Oppong‐Nkrumah for statistical
assistance.
CONFLICTS OF INTEREST
The authors declare that they have no conflicts of interest.
CONTRIBUTIONS
GSM was responsible for the data analysis and wrote the manuscript.
GSM, EKC, RK, BA, and RA were involved in the study design and
implementation and interpretation of results. CP and AA‐Y were
involved in implementation of the intervention and analysis interpreta-
tion. All authors contributed to and approved the final manuscript.
ORCID
Grace S. Marquis http://orcid.org/0000-0002-2349-4897REFERENCES
Berti, P. R., Krasevec, J., & Fitzgerald, S. (2004). A review of the effective-
ness of agriculture interventions in improving nutrition outcomes.
Public Health Nutrition, 7, 599–609.
Brown, D. L. (2003). Solutions exist for constraints to household produc-
tion and retention of animal food products. Journal of Nutrition, 133,
4042S–4047S.Cameron, A. C., & Miller, D. L. (2011). Robust inference with clustered
data. In A. Ullah, & D. E. A. Giles (Eds.), Handbook of empirical economics
and finance (pp. 1–28). Boca Raton: CRC Press.
Cohen, J. (1977). Statistical power analysis for the behavioral sciences. New
York: Academic Press.
Food and Agriculture Organization. (2012). Escala Latinoamericana y
Caribeña de seguridad alimentaria (ELCSA): Manual de uso y
aplicaciones. Santiago, Chile, FAO Regional Office, Latin America.
(www.rlc.fao.org/es/publicaciones/elcsa/).
Ghana Statistical Service, Ghana Health Service, & ICF International
(2015). Ghana Demographic and Health Survey 2014. Rockville, Mary-
land, USA: GSS, GHS, and ICF International.
Ghana Statistical Service, Ghana Health Service, & ICF Macro (2009).
Ghana Demographic and Health Survey 2008. Accra, Ghana: GSS, GHS,
and ICF Macro.
Girard AW, Self JL, McAuliffe C, Olude O. (2012). The effects of household
food production strategies on the health and nutrition outcomes of
women and young children: A systematic review. US National Library
of Medicine, National Institutes of Health. Paediatric and Perinatal Epi-
demiology, 26(Suppl 1), 205–222.
Habicht, J., Martorell, R., & Rivera, J. A. (1995). Nutritional impact of sup-
plementation in the INCAP longitudinal study: Analytic strategies and
inferences. Journal of Nutrition, 125, 1042S–1050S.
Hoddinott, J., Maluccio, J., Behrman, J., Flores, R., & Martorell, R. (2008).
Effect of a nutrition intervention during early childhood on economic
productivity in Guatemalan adults. Lancet, 371, 411–416.
Homiah, P. A., Sakyi‐Dawson, O., Mensah Bonsu, A., & Marquis, G. S.
(2012). Effects of microenterprise development on caregivers' incomes
and consumption of animal source foods. African Journal of Food, Agri-
culture, and Nutrition, 12, 5725–5745.
Hotz, C., Loech, C., de Brauw, A., Eozenou, P., Gilligan, D., Moursi, M.,
… Meenakshi, J. V. (2012). A large‐scale intervention to introduce
orange sweet potato in rural Mozambique increases vitamin A
intakes among children and women. British Journal of Nutrition,
108, 163–176.
Iannotti, L. L., Lutter, C. K., Stewart, C. P., Gallegos Riofrío, C. A., Malo, C.,
Reinhart, G., … Waters, W. F. (2017). Eggs in early complementary
feeding and child growth: A randomized controlled trial. Pediatrics,
140, e20163459. https://doi.org/10.1542/peds.2016‐3459
Low, J. W., Arimond, M., Osman, N., Cunguara, B., Zano, F., & Tschirley, D.
(2007). A food‐based approach introducing orange‐fleshed sweet pota-
toes increased vitamin A intake and serum retinol concentrations in
young children in rural Mozambique. Journal of Nutrition, 137,
1320–1327.
Marquis, G. S., Colecraft, E. K., Sakyi‐Dawson, O., Lartey, A., Ahunu, B. K.,
Birks, K. A., … Huff‐Lonergan, E. (2015). An integrated microcredit,
entrepreneurial training, and nutrition education intervention is associ-
ated with better growth among preschool‐aged children in rural Ghana.
Journal of Nutrition, 145, 335–343.
Masset, E., Haddad, L., Cornelius, A., & Isaza‐Castro, J. (2011). A systematic
review of agricultural interventions that aim to improve nutritional status
of children. London: EPPI‐Centre, Social Science Research Unit, Institute
of Education, University of London. www.r4d.dfid.gov.uk/PDF/Out-
puts/SystematicReviews/Masset_etal_agriculture_and_nutrition.pdf
Meeker J, Haddad L. (2013). A state of the art review of agriculture‐nutri-
tion linkages. Agridiet position paper 2013. (https://opendocs.ids.ac.
uk/opendocs/bitstream/handle/123456789/3035/
AgiDiet%20Global%20Review%20FINAL.pdf?sequence=1).
Olney, D. K., Pedehombga, A., Ruel, M. T., & Dillon, A. (2015). A 2‐year
integrated agriculture and nutrition and health behavior change com-
munication program targeted to women in Burkina Faso reduces
anemia, wasting, and diarrhea in children 3–12.9 months of age at
baseline: A cluster‐randomized controlled trial. Journal of Nutrition,
145, 1317–1324.
10 of 10 MARQUIS ET AL.
bs_bs_bannerPandey, V. L., Dev, S. M., & Jayachandran, U. (2016). Impact of agricultural
interventions on the nutritional status in South Asia: A review. Food
Policy, 62, 28–40.
Penny, M. E., Creed‐Kanashiro, H. M., Robert, R. C., Narro, M. R., Caufield,
L. E., & Black, R. E. (2006). Effectiveness of an educational intervention
delivered through the health services to improve nutrition in young
children: A cluster‐randomised controlled trial. Lancet, 365,
1863–1872.
Ruel, M. T., Quisumbing, A. R., & Balagamala, M. (2018). Nutrition‐sensitive
agriculture: What have we learned so far? Global Food Security, 17,
128–153.
Semba, R. D., Shardell, M., Sakr Ashour, F. A., Moaddel, R., Trehand, I.,
Maleta, K. M., … Manary, M. J. (2016). Child stunting is associated
with low circulating essential amino acids. eBioMedicine, 6,
246–252.
StataCorp (2013). StataCorp. 2013). Stata statistical software: Release 13.
College Station, TX: StataCorp LP.
Stein, A. D., Wang, M., DiGirolamo, A., Grajeda, R., Ramakrishnana, U.,
Ramirez‐Zea, M., … Martorell, R. (2008). Nutritional supplementation
in early childhood, schooling, and intellectual functioning in adulthood:
A prospective study in Guatemala. Archives of Pediatrics and Adolescent
Medicine, 162, 612–618.Thabane, L., Mbuagbaw, L., Zhang, S., Samaan, Z., Marcucci, M., Ye, C.,
… Goldsmith, C. H. (2013). A tutorial on sensitivity analyses in clin-
ical trials: The what, why, when and how. BMC Medical Research
Methodology, 13, 92.
Webb, P., & Kennedy, E. (2014). Impacts of agriculture on nutrition: Nature of
the evidence and research gaps. Food and Nutrition Bulletin, 35, 126–132.
WHO Multicenter Growth Reference Study Group (2006). WHO child
growth standards: Length/height‐for‐age, weight‐for‐age, weight‐for‐
length, weight‐for‐height and body mass index‐for‐age: Methods and
development. Geneva, Switzerland: World Health Organization.
World Health Organization. (2008). Indicators for assessing infant and
young child feeding practices: Conclusions of a consensus meeting held
6–8 November 2007 in Washington D.C., USA. (http://apps.who.int/
iris/bitstream/10665/43895/1/9789241596664_eng.pdf)
How to cite this article: Marquis GS, Colecraft EK, Kanlisi R,
et al. An agriculture–nutrition intervention improved children's
diet and growth in a randomized trial in Ghana. Matern Child
Nutr. 2018;14(S3):e12677. https://doi.org/10.1111/mcn.12677