Columbia University Analysis for the
Scaling Up Nutrition (SUN) Secretariat
Simulating Potential of
Nutrition Sensitive Investments
Slide Deck to Accompany the Technical Report
January 2014
The Lives Saved Tool (LiST) Visualizer
Five outcome areas, intermediary to stunting, served as key starting point for our study
Lancet, 2013
FAMILY
PLANNING
Source: http://list.cherg.org/
MATERNAL
NUTRITION
Theoretical Framework
Nutrition Sensitive
Sector
Literature Review
And
Modeling
Education
Social
Protection
Agriculture
Predictors
Exclusive
Breastfeeding
Health
Environment
and Water
Outcome Areas
1.Selected outcomes
delivered through
nutrition sensitive
channels for
literature search.
2.Used data available
at the national level
to model
associations of
outcomes with
contextual factors.
3.Used data available
at the national level
and in the literature
review to model
associations of
interventions with
outcomes.
[EBF for 0-6 months}
Complementary
Feeding
[minimum acceptable
diet]
Maternal
Nutrition
[low birth weight and
dietary patterns]
Family Planning
[Contraceptive use]
Diarrhea
incidence
[diarrhea rate]
Strong or Weak
Associations
for decision
making
Public Health Model: Theoretical Framework
Nutrition
Sensitive
Sector
Contextual
Factors
GNI per
capita
Adult literacy
rate
Interventions/
Delivery Channels
Peer counseling
Commercial packets
Facility based education
Maternal education
Health professional training
Supplement provision
Adolescent
birth rate
Public
Health
Female labor
participation
Sec. School
Enrollment
Maternity
leave
Community Health Workers
School promotion
Media campaign
Outcome
Areas
Exclusive
Breastfeeding
[EBF for 0-6 months}
Complementary
Feeding
[minimum acceptable
diet]
Family Planning
[Contraceptive use]
Done by LIST:
Iron/ Folic Acid Supplementation
Multiple Micronutrient Supp
Calcium Supplementation
Balanced Energy Protein Supp
Predictors
Maternal
Nutrition
[Low-birth weight]
Strong or
Weak
Associations
for decision
making
Environment Model: Theoretical Framework
Nutrition
Sensitive
Sector
Environment
and Water
Contextual
Factors
Interventions/
Delivery Channels
Promotion of access to water
and sanitation
% Rural
Population
Adult Literacy
Rates
Vaccination
Rates
Safety of complementary
foods
Outcome
Areas
Reduced
diarrhea
[diarrhea rate]
Predictors
Strong or
Weak
Associations
for decision
making
Agriculture Model: Theoretical framework
Nutrition
Sensitive
Sector
Agriculture
Contextual factors Agricultural Investments
Economic setting and
agricultural role in
society
• GNI per capita
• GINI index
• % rural population
• % Ag value added
• % Ag employment
Health setting
• Life expectancy
• # physicians per
1000
Gender
• Girls/boys ratio in
secondary school
Education
• Literacy
Inputs
• Ag biodiversity
• Fertilizer
• Land available for agriculture
• Water available for
agriculture
• Mechanization
Rural Infrastructure
• Irrigation
• Crop storage facilities
• Road infrastructure
• Port infrastructure
• Mobile network
Institutions and governance
• Access to finance
• Policy and legal framework
• Accountability and
transparancy
• Allocation of pub resources
Markets
• Ag imports and exports
• Ag import tarifs
Research
• Ag R&D
Outcome areas
Predictors
Maternal nutrition
[low birth weight;
dietary patterns:
% energy from non
staples, calories per
capita, micronutrient
availability]
Complementary
feeding
(minimum acceptable
diet)
Strong or
Weak
Associations
for decision
making
Public Health and Environment: Methodology
We investigate the estimated effect of interventions in these two sectors on the following key outcomes:
exclusive breastfeeding, complementary feeding, maternal diet and family planning (public health sector),
and promotion of access to improved water and sanitation (environment and water sector).
Contextual models
we used country level data from The State of World Children (SOWC) 2013 report, and most recent
data available from International Labor Organization, World Bank or other sources . The available data
ranges from period 2006-2010 for adolescent birth rate to year 2011 for vaccination data.
we ran cross country multiple regression analysis with 1000 simulations to investigate the predictors
of the outcome measures. We estimated the 95% confidence interval
Intervention models: Interventions in these two sectors were assessed via meta-analysis, as following:
we conducted an extensive literature review on intervention impacts on the following outcome
measures: exclusive breastfeeding, minimum acceptable diet, contraceptive use, percent of low births
and diarrhea rate.
we selected relevant studies and formatted the results as needed for meta regressions
we estimated the pooled relative risks (RR) and their respective confidence intervals (CI) using a
restricted maximum likelihood estimator (REML) on a random effect model. The pooled estimates were
calculated using the natural logarithms of the RRs and their standard errors from the individual
studies.
we explored sources of heterogeneity using sub-group analysis. The sub-groups (moderators) were
identified based on participant or study characteristics. For moderators that were not systematically
available at the study level, we used country level data that was matched to the country of the study.
Significant moderator were identified using contextual models – described below.
Agriculture model: Methodology
We focused on two intermediary outcome areas for the agriculture model including: maternal nutrition and
complementary feeding. For both outcome areas, we developed a national-level contextual model that allows
us to identify associations rather than causal relationships between agricultural and nutrition variables.
Quantitative model:
National level data from seven publicly available databases were collated and organized to populate the
model integrating agriculture (FAO, IFAD, EIU), socio-economic (World Bank), human nutrition and health
(WHO NILS, WHO IYCF) data using the country as the unit
we first identified model indicators for maternal nutrition and complementary feeding that are
significantly related to stunting, while controlling for income level
Starting from these results we fit the agricultural factors into multivariate regressions against each of
these indicators while taking into account a set of contextual factors
Additional literature review:
Starting from the results of the quantitative model, we revisited the literature to identify specific
programs/ interventions and delivery channels that contribute to the implementation of the agricultural
investments that were identified as significantly related to nutrition specific indicators.
Public Health Model: Results
Contextual model
Adolesc
ent
birth
rate
GNI per
capita
0.28***
[0.07,0.71]
-0.10
[-0.21,0.03]
8.60***
[0.022,3295]
0.01***
[0.0001, 1.84]
-0.56**
[-0.90,-0.19]
Exclusive
Breastfeeding
Complementary
Feeding
Maternal
Nutrition
[EBF for 0-6 months}
[Minimum acceptable diet]
[Low-birth weight]
0.24*
[0.004, 0.48]
-6.13** [-10.79,1.60]
Female to
male labor
participati
on *
maternity
leave
Adult
literacy:
females
as a % of
males
African -17.90*
[-32.16, -3.39]
Asian -14.41*
[-28.89, 1.09]
Secondary
school
enrollment
female/mal
e ratio
%
urban
-64.31***
Asian 19.10*, [-95.5,-35.0]
[0.64, 37.66]
Mixed
24.59*
Female
[4.98, 44.50] to male
labor
participa
tion
Ethnicity
Latino/Hispanic
12.46* ,95% CI
[1.48, 23.75]
-0.18*
[0.02,0.33]
0.47***
[0.26,0.69]
Family Planning
[Contraceptive use]
0.21**
[0.09,0.34]
Access
to
improved
rural
sanitatio
n
Public Health Model: Results
Intervention model EBF
Exclusive
Breastfeeding
[EBF for 0-6 months}
RR 2.46***
RR 1.33***
RR 1.55***
Facility
based
education
Non
provision of
commercial
packets
Peer
counseling
Contextual factors influencing the intervention effect
Duratio
n of BF
Adult
literacy
rate
% rural
populat
ion
Materni
ty leave
Female
labor
particip
ation
Adolesc
ent
birth
rate
For interpretation: Relative Risk (RR) = 1 indicates that the outcome in intervention and control groups are equally likely to occur; RR<1 outcome in intervention is
less likely to occur compared with control; RR>1 outcome in intervention is more likely to occur compared with RR>control. E.g. RR 0.6 is usually interpreted in the
following way (exp for RR=0.6). (1-0.6)*100=40%, the outcome in intervention is 40% less likely to occur. If RR is 1.5, the outcome is 1.5 times more likely to occur
in the intervention compared with control (or 50% increased risk.
Results of meta-regressions for the effect of peer counseling on exclusive
breastfeeding in randomized controlled trials and quasi-experimental studies
Covariate
Duration of breastfeeding (study)
3 months or less
4 to 6 months
Adult literacy rate, female as % of
male (country)
<80
>=80
Rural population, % (country)
<=30
>30
Maternity leave, # of weeks
(country)
<=12
13 to 20
>20
Female labor participation rate (%)
(country)
<40
>=40
Adolescent birth rate (%)
<=60
>60
No. of
observations* RR (95% CI)
p-value
36
26
1.84 (1.44 to 2.35)
3.82 (2.80 to 5.22)
<0.0001
<0.0001
13
36
3.09 (1.94 to 4.92)
2.61 (1.96 to 3.48)
<0.0001
<0.0001
41
21
2.24 (1.72 to 2.92)
2.89 (2.04 to 4.10)
<0.0001
<0.0001
19
34
9
2.79 (1.93 to 4.02)
2.63 (2.00 to 3.47)
1.38 (0.80 to 2.38)
<0.0001
<0.0001
0.2462
16
46
3.99 (2.63 to 6.03)
2.11 (1.69 to 2.64)
<0.0001
<0.0001
53
9
2.61 (2.07 to 3.29)
1.81 (1.06 to 3.07)
<0.0001
0.0286
Results of meta-regressions for the effect of facility based education on exclusive
breastfeeding in randomized controlled trials and quasi-experimental studies
No.
of
RR (95% CI)
observations*
p-value
10
1.60 (1.28 to 2.01)
<0.0001
4 to 6 months
9
Adult literacy rate, female as % of
male (country)
<80
4
1.56 (1.20 to 2.04)
0.001
1.98 (1.33 to 2.95)
0.0008
>=80
7
2.09 (1.58 to 2.77)
<0.0001
<=30
13
1.33 (1.11 to 1.59)
0.0016
>30
7
2.05 (1.60 to 2.63)
<0.0001
<=12
10
1.27 (1.05 to 1.54)
0.0136
13 to 20
9
2.07 (1.66 to 2.58)
<0.0001
>20
1
Female labor participation rate (%)
(country)
<40
5
1.12 (0.63 to 2.00)
0.6974
2.24 (1.63 to 3.08)
<0.0001
>=40
15
1.39 (1.16 to 1.65)
0.0002
<=60
14
1.43 (1.19 to 1.71)
<0.0001
>60
6
1.89 (1.42 to 2.51)
<0.0001
Covariate
Duration of breastfeeding (study)
3 months or less
Rural population, % (country)
Maternity leave, # of weeks (country)
Adolescent birth rate (%)
Public Health Model: Results
Intervention model Family planning
Family Planning
[Contraceptive use]
RR 1.15*
School
promotion,
media
campaign,
communitybased education
Environment model: Results
Contextual model
Diarrhea
treatment
[% treatment with ORS]
0.31*
[0.07, 0.56]
Adult
literacy
rate:
females
as % of
males
0.34*
[0.03, 0.68]
Vaccinati
on rate
Environment Model: Results
Intervention model
Diarrhea
incidence
[Diarrhea rate]
RR 0.76***
RR 0.71**
Hand
washing
Water
treatment
Contextual factors controlled for
Adult
literacy
% rural
populat
ion
Vaccina
tion
rate
GNI per
capita
Agriculture Model
Identification of model indicators for nutrition-related intermediary outcomes
Stunting
Adj R2 0.84
Dietary patterns
(proxy for maternal nutrition)
% Energy
from non
Calories per
staples in
capita
supply
(-6.86***)
(-4.75***) Fe
availability
from
animalproducts
(-4.15*)
Adj R2 0.73
Low-birth weight
(proxy for maternal nutrition)
% Low-birth
weight
(2.82***)
Adj R2 0.43
% energy from non staples in national
food supply significantly related to %
low birth weight (-1.91**)
CONTEXT- SPECIFICITY
Log GNI per capita
Adj R2 0.32
Complementary feeding
% Minimum
acceptable
diet
(-6.51***)
Adj R2 0.63
Fe availability from animal based
products significantly related to %
minimum acceptable diet (9.93*)
Agriculture Model
Agricultural investments related to dietary patterns
Dietary patterns
(proxy for maternal nutrition)
% Energy
from non
staples in
supply
Calories per
capita
-0.34 **
6.88*** to -1.81
dependent on
income level
% Energy
from non
staples in
production
Fe
availability
from
animalproducts
-0.39 ***
0.14**
0.48**
Access to
finance for
farmers
# tractors
per land
unit
Fertilizer
use per
land unit
0.47**
0.15**
% land for
agriculture
Contextual factors influencing outcomes
Log GNI
per capita
Road
infrastruct
ure
Exports as
% of GDP
Ag R&D as
% of GDP
-0.48 **
Ag import
tariffs
Agriculture Model
Supply diversity as a function of production diversity
The relationship between food supply
diversity and food production diversity
depends on the income level of a
country. For low-income countries the
diversity of agricultural goods
produced by a country is a strong
predictor for food supply diversity; for
middle- and high-income countries
national income and trade are better
predictors.
Agriculture Model
Agricultural investments related to complementary feeding
Complementary feeding
% Minimum
acceptable
diet
9.02*
9.93*
Fe
availability
from
animalproducts
-4.85*
0.47**
% Energy
from non
staples in
production
% land for
agriculture
Ag R&D as
% of GDP
Contextual factors influencing outcomes
Log GNI
per capita
Exports as
% of GDP
-0.48 **
Ag import
tariffs
Summary Table of Model Results
Intervention
Outcome
Impact
Peer counselling
EBF
The likelihood of EBF is 2.46 higher for mothers who
received peer counseling than for mothers who
weren’t counseled (95% CI: 1.99 to 3.04, p<0.001).
Facility based promotion
EBF
The likelihood of EBF is 1.55 times higher for
mothers receiving the facility based intervention than
for mothers who didn’t (95% CI: 1.31 to 1.84,
p<0.001).
Commercial packets of infant
formula
EBF
The likelihood of EBF for mothers who were not given
commercial packets is 1.34 higher than for mothers
who received the packets (95% CI: 1.12 to 1.59,
p=0.0011).
Combined health interventions
(minus mass media campaigns
which was assessed
qualitatively)
EBF
The results show that the likelihood of exclusive
breastfeeding for mothers that received the public
health interventions is 2.02 higher than mothers
who did not (95% CI: 1.74 to 2.34, p<0.001)
Summary Table of Model Results
Intervention
Outcome
Impact
Family Planning Promotion
Contraceptive Use
Contraceptive use for participants who were exposed
to school promotions, media campaigns and
community based education is 1.16 higher than for
the control groups (95% CI: 1.01 to 1.35, p<0.0425).
Hand washing
Diarrhea rates
The likelihood of diarrhea for those who were
exposed to hand washing interventions is 24% less
likely than for those in the control group (RR= 0.76%
CI: 0.62 to 0.93, p=0.0074)
Water treatment
Diarrhea rates
The likelihood of diarrhea those who were exposed to
water treatment intervention is 29% less likely than
for those in control group (RR= 0.7073% CI: 0.56 to
0.90, p=0.0043).
Summary Table of Model Results
Investment
Examples specific
interventions
Outcome
Association
Agricultural
diversification
Promotion of animal-based
products, homegardens,
irrigation, legume intercropping,
agro-forestry
Dietary patterns
Complementary
feeding
Increased food supply diversity in low-income
countries
Potential trade-off with calories available per
capita
Increased % of children meeting minimum
acceptable diet
Agricultural
intensification
Increased fertilizer use per land
unit (e.g. subsidy program)
Increased number of tractors
per land unit
Dietary patterns
Increased amount of calories available per
capita
Potential trade-off with food supply diversity
Agricultural
extensification
Increased % land for agriculture
Dietary patterns
Complementary
feeding
Increased amount of calories available per
capita
Potential trade-off with % children meeting
minimum acceptable diet
Rural development
Increased access to finance for
farmers (e.g. microcredits)
Road infrastructure
Dietary patterns
Increased food supply diversity
Trade policies/
strategies
Ag import tariffs
Export crops
Dietary patterns
Potential trade-off with food supply diversity
and micronutrient availability
Ag R&D
e.g. biofortification, livestock
health programs
Dietary patterns
Complementary
feeding
Increased micronutrient availability
Implications
•
•
•
•
•
•
•
With limited evidence, the evidence at hand suggests that public health,
environment and agriculture investments could support nutrition specific
interventions that address undernutrition.
A country’s contextual factors (relating to income, social, education and governance)
are important to consider in their impact on nutrition outcomes with nutrition
sensitive approaches.
Examining agriculture through large scale investments, rather than nutrition
interventions, can provide insight for MoA on impact for nutrition, indirectly.
Alternative delivery channels for public health and environment, through marketing,
commercialization, food safety, and social protection, are less clear in their evidence
of impact.
Evidence published in the literature remains scant and varied for nutrition sensitive
interventions, and more implementation science should be published.
Using a quantitative statistical simulation model can only go so far as with the
current literature and data. This has resulted in some interesting insights but
unfortunately it is not a tool that is user friendly for countries looking to scale up
nutrition.
Costing tools and perhaps game tools could provide an entry point for decision
making in which this quantitative modeling could be used as a first step resource.
Schematic Summary of Findings
SOCIAL PROTECTION
Conditional cash transfers
Exclusive
breastfeedin
g
PUBLIC HEALTH
Peer counseling
Facility-based education
Non provision of
commercial packages
Maternal counseling
Health worker training
School promotion
Media campaigns
Community-based
education
Complement
ary feeding
(LiST results)
Maternal
Nutrition
Diarrhea
incidence
Family
planning
ENVIRONMENT AND WATER
Water treatment, Handwashing
Food safety measures
CONTEXT- SPECIFICITY
Income, Education, Urbanization, Geographic Location, Employment Policy,
AGRICULTURE
Agriculture
diversification
- Animal products
- Homegardens
- Legumes
- Agroforestry
- Small scale irrigation
Access to Finance
Fertilizer use
Agriculture research and
development
- Biofortification
Rural infrastructure
Women empowerment