Social Science & Medicine 238 (2019) 112374

Contents lists available at ScienceDirect

Social Science & Medicine

journal homepage: www.elsevier.com/locate/socscimed

Assessing associational strength of 23 correlates of child anthropometric T

failure: An econometric analysis of the 2015-2016 National Family Health
Survey, India
Rockli Kima, Sunil Rajpald, William Joeb, Daniel J. Corsic, Rajan Sankard, Alok Kumare,
S.V. Subramaniana,f,∗
a
Harvard Center for Population & Development Studies, Cambridge, MA, USA
b
Institute of Economic Growth (IEG), University of Delhi Enclave, North Campus, Delhi, India
c
Clinical Epidemiology Program, Ottawa Hospital Research Institute, Ottawa, ON, Canada
d
Nutrition, Tata Trusts, India
e
National Institution for Transforming India (NITI), Government of India, New Delhi, India
f
Department of Social and Behavioral Sciences, Harvard T.H. Chan School of Public Health, Boston, MA, USA

ARTICLE INFO ABSTRACT

Keywords: Despite the broad consensus that investments in nutrition-sensitive programmes are required to reduce child
India undernutrition, in practice empirical studies and interventions tend to focus on few nutrition-specific risk factors
Undernutrition in isolation. The 2015–16 National Family Health Survey provides the first opportunity in more than a decade to
Anthropometric failures conduct an up-to-date comprehensive evaluation of the relative importance of various maternal and child health
Stunting
and nutrition (MCHN) factors in respect to child anthropometric failures in India. The primary analysis included
Underweight
Wasting
140,444 children aged 6–59 months with complete data on 20 MCHN factors, and the secondary analysis in-
Social determinants cluded a subset of 25,603 children with additional paternal data. Outcome variables were stunting, underweight
and wasting. We conducted logistic regression models to first evaluate each correlate separately in age- and sex-
adjusted models, and then jointly in a mutually adjusted model. For all anthropometric failures, indicators of
past and present socioeconomic conditions showed the most robust associations. The strongest correlates for
stunting were short maternal stature (OR: 4.39; 95%CI: 4.00, 4.81), lack of maternal education (OR: 1.74;
95%CI: 1.60, 1.89), low maternal BMI (OR: 1.64; 95%CI: 1.54, 1.75), poor household wealth (OR: 1.25; 95%CI:
1.15, 1.35) and poor household air quality (OR: 1.22; 95%CI: 1.16, 1.29). Weaker associations were found for
other correlates, including dietary diversity, vitamin A supplementation and breastfeeding initiation. Paternal
factors were also important predictors of anthropometric failures, but to a lesser degree than maternal factors.
The results remained consistent when stratified by children's age (6–23 vs 24–59 months) and sex (girls vs boys),
and when low birth weight was additionally considered. Our findings indicate the limitation of nutrition-specific
interventions. Breaking multi-generational poverty and improving environmental factors are promising invest-
ments to prevent anthropometric failures in early childhood.

1. Introduction Sustainable Development Goals (Baye, 2017; De Onis et al., 2013;

Ensuring adequate nutrition and optimal growth potential in early
childhood has important implications for cognitive development
(Perkins et al., 2017; Sudfeld et al., 2015) and economic productivity in
adulthood (Horton and Steckel, 2011). India bears more than one third
of the global burden of child stunting (De Onis et al., 2012). Needless to
say, India's progress in child health and nutrition will directly con-
tribute to the worldwide progress in achieving global targets such as the
Haddad et al., 2015). Within India, the National Nutrition Strategy
(NNS), released by the National Institution for Transforming India
(NITI) Aayog in 2017, specifically aims to prevent and reduce pre-
valence of underweight (i.e. low weight-for-age) in children (0–3 years)
by 3 percentage points per annum by 2022 (NITIAayog, 2017). This is
an ambitious goal given the 7 percentage point reduction (from 43% to
36%) in the proportion of underweight among children under 5 years
over the past decade (IIPS, 2017). During the same period of time, India

∗
Corresponding author. Harvard Center for Population & Development Studies, 9 Bow Street, Cambridge, MA, 02138, USA.
E-mail address: svsubram@hsph.harvard.edu (S.V. Subramanian).

https://doi.org/10.1016/j.socscimed.2019.112374
Received 7 January 2019; Received in revised form 12 June 2019; Accepted 17 June 2019
Available online 20 June 2019
0277-9536/ © 2019 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license
(http://creativecommons.org/licenses/BY/4.0/).
R. Kim, et al.
experienced 10 percentage point reduction (from 48% to 38%) in the
proportion of stunting whereas wasting remained steady at 20% (IIPS,
2017). Given the current commitment and political momentum in India
to achieve the NNS goals, there is an urgent need to generate com-
prehensive and up-to-date evidence to inform targeted interventions.
Despite the broad consensus that investments in nutrition-sensitive
programmes to ensure healthy living conditions are required to sub-
stantially reduce undernutrition (Bhutta et al., 2008; R. E. Black et al.,
2008), in practice priorities are given to nutrition-specific interventions
that aim to modify individual level behaviors such as complementary
foods and feeding practices and breastfeeding (Bhutta et al., 2008; R.
Black et al., 2013; Menon et al., 2015; Ruel et al., 2013). However,
when simultaneously considered with other ‘upstream’ or ‘distal’ risk
factors (i.e., socioeconomic conditions and maternal characteristics),
age-appropriate infant and young child feeding practices were not ne-
cessarily the strongest risk factors of anthropometric failures (Corsi
et al., 2016b; Kim et al., 2017; Martorell and Young, 2012). This sug-
gests the limitation of assessing one or two risk factors of child un-
dernutrition in isolation and the need to understand their relative im-
portance conditional on other known correlates (Corsi et al., 2016b;
Kim et al., 2017). The most comprehensive investigation of determi-
nants of anthropometric failures in India (Corsi et al., 2016a, b) and
South Asia (Kim et al., 2017), to our knowledge, were based on older
data from India (2005–06) and did not incorporate information on
antenatal care (ANC) visits, skilled birth attendant (SBA) at delivery,
family planning needs, oral rehydration therapy (ORT) for children
with diarrhea, and care seeking for pneumonia, which are widely ac-
cepted indicators to monitor coverage of maternal and child health
interventions in low- and middle-income countries (Barros and Victora,
2013; Kruk and Freedman, 2008).
The release of the 4th National Family Health Survey (NFHS-4)
(IIPS, 2017) conducted in 2015–16 provides the first opportunity in
more than a decade to update the current understanding of the national
context of child anthropometric failures and their correlates in India.
The objective of this study was to conduct a comprehensive evaluation
of the relative importance of 23 correlates of anthropometric failures
that have been considered separately, but not jointly, in prior literature
and interventions (Bhutta et al., 2008; R. Black et al., 2013; Corsi et al.,
2016b; Kim et al., 2017; Ruel et al., 2013). We also assess the poten-
tially differential influence of these factors on child anthropometric
failures by age and sex in stratified analyses.
2. Methods
2.1. Survey data and study population
For the first time the NFHS-4 included all 36 states and union ter-
ritories and 640 districts in India (IIPS, 2017, 2018). Survey re-
spondents were selected following a stratified two-stage sampling frame
by states and urban and rural areas within each state (IIPS, 2018). The
response rate was 97% for women and 92% for men in the NFHS-4
(IIPS, 2018). A total of 199,314 singleton children aged 6–59 months
alive at the time of survey with non-pregnant mothers were eligible to
be included in our analysis. Of them, 14,399 children who did not have
height/weight measures were excluded. Moreover, 44,471 children
were excluded for missing information on one or more of the maternal
and child health and nutrition (MCHN) correlates listed below. The
final analytic sample included 140,444 children (Fig. 1). A subset of
25,603 children with additional information on paternal characteristics
was used for the secondary analysis because the men's questionnaire
was administered in a representative sample of 15% of households se-
lected for the state module in the NFHS-4.
2.2. Outcomes
Child's weight was measured by trained health investigators using
2
Social Science & Medicine 238 (2019) 112374
digital solar-powered scales along with adjustable Shorr measuring
boards in the NFHS-4 (IIPS, 2018). For older children (above 24
months) standing height was obtained, while for younger children (less
than 24 months) recumbent length was measured (IIPS, 2018). We
constructed the following three anthropometric failure outcomes based
on these height and weight measures. Stunting, which reflects retarded
skeletal growth due to chronic deprivation (UNICEF, 2009), was de-
fined as height-for-age z-scores < -2 standard deviation (SD) below the
WHO child growth reference standards (WHO, 2006). Underweight was
defined as weight-for-age z-scores < -2SD. Wasting, signaling acute
malnutrition (Bhattacharyya, 2006), was defined as weight-for-height
z-scores < -2SD.
2.3. Correlates
Existing studies on determinants of child undernutrition and
UNICEF conceptual framework were reviewed to identify a compre-
hensive set of correlates at multiple levels (Bhutta et al., 2008; R. Black
et al., 2013; Corsi et al., 2016b; Kim et al., 2017; Ruel et al., 2013). In
the NFHS-4, data on most correlates, except for mother's and father's
anthropometry measures, were self-reported. At household level,
wealth index, access to improved drinking water source and sanitation
facilities, air quality, safe disposal of child stools, and use of iodized salt
were considered. We constructed wealth index score using principal
component analysis on household characteristics (main material for
floor, wall, and roof) and assets including electricity, radio, refrigerator,
bicycle, motorcycle/scooter, car/truck, telephone, watch, animal-
drawn cart, bank account, mattress, pressure cooker, chair, bed, table,
electric fan, television, sewing machine, internet, computer, air condi-
tioner, and washing machine (IIPS, 2018). The distribution of house-
hold wealth scores was divided into quintiles. The correlation between
our wealth quintile and the NFHS-4 wealth quintile (which had in-
corporated information on source of drinking water and toilet facility)
was very high (r = 0.94). The source of drinking water was considered
safe if ‘piped into dwelling or yard/plot’, ‘public tap/standpipe’, ‘tube
well or borehole’, ‘protected well or spring’, or ‘rain water’. The sani-
tation facility was defined as improved if households had ‘flush to piped
sewer system’, ‘flush to septic tank or pit latrine’, ‘ventilated improved
pit latrine’, ‘pit latrine with slab’, or ‘composting toilet’. Household air
quality was characterized as use of non-solid fuels (i.e., best quality),
solid fuels in separate kitchen, and solid fuels in non-separate kitchen
(i.e., worst quality) (Bassani et al., 2010). Binary variables were used
for disposal of child stools (safe/unsafe) and use of iodized salt in the
household (yes/no).
At maternal level, education, height, body mass index (BMI), and
age at marriage were considered. Maternal education was categorized
in five levels: no schooling, primary, secondary, higher secondary, and
college education and above. Women's weight and height were mea-
sured by field interview teams using digital Secascales and adjustable
Shorr measuring boards (IIPS, 2018). Maternal height was categorized
as < 145, 145–149.9, 150–154.9, 155–159.9, and 160+ cm. Maternal
BMI was calculated by taking weight in kg and dividing by height in
meters squared, and was categorized as < 18.5, 18.5–24.9, and
25+ kg/m2. Mother's age at marriage was defined using a dichotomous
variable of < 18 years or ≥18 years.
Five variables concerning nutrition and health at child level were
included. Based on a 24-h recall of food intake in the NFHS ques-
tionnaire, a score for child's dietary diversity was developed by as-
signing 1 point for consumption of milk, meat, legumes and nuts,
starchy staples (grains, roots and tubers), vitamin A rich fruits and
vegetables, other fruits and vegetables, dairy, oils/fats/butter (Ruel and
Menon, 2002). A higher value on this score indicated more diverse
dietary intake, and the score was grouped into quintiles. While the use
of 24-h recall for dietary diversity makes a critical assumption that
recent intake is a good proxy for longer-term dietary pattern, it is a
useful indicator in the context where detailed dietary assessment is not
R. Kim, et al.
Fig. 1. Flow diagram showing exclusions and final sample sizes of the study population, Indian National Family Health Survey 2016.
feasible (Arimond and Ruel, 2004). Binary variables were created for
early breastfeeding initiation (≥1 h of birth/ < 1 h of birth), having
experienced infectious disease (e.g., diarrhea, cough/fever) two weeks
prior to the survey (yes/no), and whether the child was given vitamin A
supplementation (yes/no).
Other MCHN coverage indicators were also considered as correlates
of child anthropometric failures. Unmet need for family planning was
coded as 1 if woman reported unmet need for spacing or limiting, and 0
otherwise. The number of ANC visits was categorized as < 4 or ≥4
based on the WHO recommendation at the time of the survey (WHO,
3
Social Science & Medicine 238 (2019) 112374
2016). While the newer guideline recommends a minimum of 8 ANC
contacts during pregnancy to improve pregnancy experience (WHO,
2016), we utilize the older standard to be consistent with the policy
context at which the survey was collected. An indicator variable was
created for births attended by skilled health personnel (doctor, nurse, or
midwife). We also considered whether ORT was given to a child with
diarrhea (yes/no) and whether care was sought for a child with cough
as a proxy measure for care seeking for pneumonia (yes/no). Based on
vaccination data collected from the child's health card and direct re-
porting from the mother, a binary indicator was constructed to assess
R. Kim, et al. Social Science & Medicine 238 (2019) 112374

whether the child was fully vaccinated with measles, BCG, DPT 3, and Table 1
Polio 3 (yes/no). For more than half of our sample (55%), the health Distribution of anthropometric failures by selected correlates among children
card was available to be seen at the time of the interview. Another 25% aged 6–59 months, India NFHS 2016
of our sample had the health card, but was not seen at the time of the N Stunting (%) Underweight (%) Wasting (%)
interview, and the remaining 20% had either lost it or never had one.
Finally, paternal characteristics were collected for a sub-sample of Total sample 140,444 38.62 34.19 19.56
Child's age
the NFHS-4 population. Paternal education (no schooling, primary,
6–11 months 17,573 23.08 27.04 25.51
secondary, higher secondary and above), height (< 155, 155–159.9, 12–23 months 31,192 40.90 32.60 20.84
160–164.9, 165–169.0, and 170+ cm), and BMI (< 18.5, 18.5–24.9, 24–35 months 28,137 40.48 35.25 19.37
and 25+ kg/m2) were additionally assessed in our secondary analysis. 36–47 months 32,291 43.17 36.58 17.45
48–59 months 31,251 38.72 36.36 17.28
Child's sex
2.4. Statistical analysis Male 73,939 39.31 34.85 20.62
Female 66,505 37.85 33.45 18.37
We ran a series of logistic regression models to evaluate the in- Birth order
dependent and joint associations between the identified correlates and First 50,723 33.33 29.75 18.74
Second, Third 67,550 38.95 34.74 19.81
each anthropometric failure. The association with each of the correlates
Fourth, Fifth 16,629 48.26 41.85 20.70
was first assessed in separate models where only child's age and sex Sixth or more 5542 54.15 45.07 20.44
were adjusted for. Then, all 20 MCHN variables were simultaneously Type of residence
included in a mutually adjusted model where age and sex of the child, Urban 37,957 31.46 27.80 17.80
Rural 102,487 41.27 36.55 20.21
birth order, and place of residency (urban/rural) were further con-
Household wealth quintile
trolled for. The statistical significance and strength of magnitude (as Quintile 1 31,462 52.00 48.31 24.48
measured by odds ratio[OR]) associated with each correlate from the (poorest)
mutually adjusted model were compared. As a secondary analysis, we Quintile 2 27,736 45.48 39.54 20.33
also evaluated the importance of paternal characteristics for a sub- Quintile 3 26,819 38.41 33.08 18.73
Quintile 4 27,727 30.85 27.46 17.91
sample of children. Finally, whether the association between correlates
Quintile 5 26,700 24.02 20.07 15.50
and anthropometric failures differ by age (6–23 vs 24–59 months) and (richest)
sex (girls vs boys) were tested in stratified analyses. As a sensitivity Maternal education
analysis, we also re-ran the mutually adjusted model including low No schooling 40,833 50.78 45.80 22.47
birth weight (< 2500 g). Low birth weight was not included in the main Primary 19,506 44.55 39.10 20.12
Secondary 50,209 34.95 30.78 18.82
analysis due to a large number of missing cases (31,866 children or 23% Higher Secondary 14,773 27.36 23.58 17.21
of the primary analytic sample) and almost half of the birth weight data College 15,123 21.37 18.16 15.68
(48%) being maternal reports, which tend to cause misclassification in Maternal height
low-income and middle-income countries (Shenkin et al., 2017). We < 145 cm 15,495 59.61 51.97 21.97
145–149.9 cm 37,058 46.68 41.15 20.89
used Stata (version 13.1) for all analyses procedures and accounted for
150–154.9 cm 47,940 36.29 32.18 19.46
the cluster survey design. All statistical tests were two-sided and 155–159.9 cm 28,990 27.95 24.95 17.98
p < 0.05 was considered to determine statistical significance. 160 + cm 10,961 20.15 18.74 16.26
Maternal BMI
2.5. Ethics statement < 18.5 kg/m2 34,718 47.20 48.11 25.94
18.5–24.9 kg/m2 85,039 38.08 32.15 18.91
2
25.0 + kg/m 20,687 26.45 19.21 11.50
The study was reviewed by Harvard T.H. Chan School of Public Maternal marriage
Health Institutional Review Board and was considered exempt from full Age at marriage 50,427 44.72 40.09 20.65
review because the study was based on an anonymous public use data < 18
Age at marriage 90,017 35.21 30.88 18.95
set with no identifiable information on the study participants.
≥18
Dietary diversity score
3. Results Quintile 1 (lowest) 48,828 38.69 35.19 20.31
Quintile 2 53,222 40.37 35.74 19.29
Of the total 140,444 children in the final analytic sample, 38.62% Quintile 3 12,078 33.65 29.67 22.40
Quintile 4 6369 30.19 25.37 20.32
were stunted and 34.19% were underweight. Around one fifth of the Quintile 5 19,947 39.50 33.12 16.45
children (19.56%) suffered from wasting (Table 1). A clear patterning (highest)
in the prevalence of anthropometric failures was shown by household Breastfeeding initiation
wealth quintile and environmental factors (eg., stool disposal, ≥1 h of birth 77,946 39.70 35.34 19.63
< 1 h of birth 62,498 37.27 32.75 19.46
sanitary facility, and air quality) as well as maternal and paternal
Infectious disease in past 2 weeks
education level, height and BMI. For instance, more than half of the Yes 31,505 37.57 33.91 19.73
children in the poorest household wealth quintile were stunted (52%) No 108,939 38.93 34.27 19.51
whereas less than a quarter in the wealthiest quintile were stunted Drinking water source
(24.02%). Not improved 22,800 37.50 33.60 20.32
Improved 117,644 38.84 34.30 19.41
Stool disposal
3.1. Stunting Unsafe 87,990 43.43 39.30 21.40
Safe 52,454 30.56 25.61 16.46
In age- and sex-adjusted logistic models for stunting, all correlates, Sanitary facility
Not improved 68,832 46.27 42.49 22.46
except for infectious disease and unsafe drinking water source, were
Improved 71,612 31.27 26.21 16.77
significantly associated with higher odds of stunting (Fig. 2; Exact es- Household air quality
timates in Supplementary Table 1). In these separate models, the Solid fuels in non- 47,311 48.79 44.06 22.06
strongest associations were observed with short maternal stature (OR: separate kitchen
5.96; 95% confidence interval[CI]: 5.42, 6.54) followed by no maternal (continued on next page)
education (OR: 4.16; 95% CI: 3.88, 4.47) and poor household wealth

4
R. Kim, et al. Social Science & Medicine 238 (2019) 112374

Table 1 (continued) 4.28, 5.15) conditional on age and sex of the child. Socioeconomic
factors such as poor household wealth, lack of maternal education, and
N Stunting (%) Underweight (%) Wasting (%)
low maternal BMI were also strongly associated with underweight
Solid fuels in 40,916 39.63 35.30 20.55 (ORs > 3.50) in these models. When simultaneously adjusted for all
separate kitchen correlates, short maternal stature and low maternal BMI were each
Non-Solid fuels 52,217 28.62 24.37 16.51 associated with 3.6-fold and 2.6-fold increased odds of underweight,
Iodized salt
followed by lack of maternal education (OR: 1.68, 95% CI: 1.55, 1.83),
Not used 7768 44.16 40.44 21.78
Used 132,676 38.30 33.82 19.43 poor household wealth (OR: 1.40, 95% CI: 1.29, 1.53), and poor air
Full vaccination quality (OR: 1.15, 95% CI: 1.08, 1.21) (Fig. 3A). Unimproved sanita-
No 61,969 39.30 35.20 20.47 tion, unsafe stool disposal, poor dietary diversity, < 4 ANC visits, and
Yes 78,475 38.09 33.39 18.84 absence of SBA were also significant correlates of underweight with
Vitamin A supplementation
ORs ranging from 1.07 to 1.11 in mutually adjusted model.
No 61,874 39.71 34.97 19.32
Yes 78,570 37.77 33.57 19.75
Family planning needs met 3.3. Wasting
No 40,348 39.42 34.07 19.53
Yes 100,096 38.30 34.23 19.57
When independently evaluated in separate age- and sex-adjusted
Skilled birth attendance at the delivery
No 29,519 47.84 41.91 20.63 logistic models, low maternal BMI, poor household wealth, no maternal
Yes 110,925 36.17 32.13 19.27 education, short maternal stature, unimproved sanitation facility, poor
Antenatal care air quality, unsafe stool disposal, poor dietary diversity, < 4 ANC visits,
visits no iodized salt, no care seeking for cough, and absence of SBA at de-
Less than four 52,387 41.92 38.37 22.38
livery were significantly associated with higher odds of wasting with
At least four 49,198 30.45 26.99 18.81
Oral rehydration therapy for children diarrhea ORs ranging from 1.07 to 2.33 (Fig. 2). In the mutually adjusted model,
No 4903 41.16 38.47 21.35 only five correlates were significantly associated with higher odds of
Yes 7868 37.99 34.95 21.86 wasting: low maternal BMI (OR: 2.22; 95% CI: 2.05, 2.41), short ma-
Care seeking for pneumonia
ternal stature (OR: 1.25; 95% CI: 1.13, 1.37), poor household wealth
No 128,856 38.90 34.39 19.63
Yes 11,588 35.55 31.89 18.73
(OR: 1.24; 95% CI: 1.12, 1.37), lack of maternal education (OR: 1.16;
Paternal education 95% CI: 1.06, 1.28), and poor dietary diversity (OR: 1.10; 95% CI: 1.00,
No schooling 3767 49.99 45.05 22.80 1.20) (Fig. 3A).
Primary 3658 44.89 41.50 22.12
Secondary 14,349 35.05 30.93 18.38
3.4. Secondary analysis with paternal characteristics
Higher Secondary 3829 25.75 22.04 15.88
and above
Paternal height In a subset of 25,603 children with information on fathers' char-
< 155 cm 2648 49.24 46.19 26.02 acteristics, we found that paternal education and anthropometry were
155–159.9 cm 4876 45.18 40.32 21.12
consistently associated with all anthropometric failures. For instance,
160–164.9 cm 7038 39.91 34.37 18.91
165–169.9 cm 6276 32.62 28.81 17.75
short paternal stature was associated with significantly higher odds of
170 + cm 4765 24.70 22.69 15.74 stunting (OR: 3.07; 95% CI: 2.57, 3.65), underweight (OR: 2.87; 95%
Paternal BMI CI: 2.42, 3.41) and wasting (OR: 1.85, 95% CI: 1.53, 2.25) in age- and
2
< 18.5 kg/m 4461 47.01 45.51 22.86 sex-adjusted models (Fig. 2). Similarly, lack of paternal education was
18.5–24.9 kg/m2 16,561 37.56 33.23 19.49
associated with almost three-fold increase in odds of stunting and un-
25.0 + kg/m2 4581 26.68 21.04 14.56
derweight and almost 50% increase in odds of wasting conditional age
Unweighted N(%). and sex of the child. When jointly assessed with all other MCHN factors
Paternal characteristics (paternal education, height, BMI) were available for a in mutually adjusted models, short maternal stature continued to be the
subset of 25,603 children. strongest predictor of stunting along with short paternal stature (OR:
1.83; 95% CI: 1.53, 2.18) and low paternal BMI (OR: 1.37; 95% CI:
(OR: 3.70; 95% CI: 3.49, 3.92). The magnitude of associations with 1.18, 1.59) (Fig. 3B; Exact estimates in Supplementary Table 3). In
stunting substantially attenuated, and many were no longer statistically terms of child underweight, maternal factors (height, BMI, education)
significant, in mutually adjusted model (Fig. 3A; Exact estimates in were the top correlates followed by short paternal stature (OR: 1.74;
Supplementary Table 2). Conditional on all other correlates, short 95% CI: 1.46, 2.09) and low paternal BMI (OR: 1.64; 95% CI: 1.40,
maternal stature was still the strongest predictor of stunting (OR: 4.39; 1.91). In the mutually adjusted model for wasting, only low maternal
95% CI: 4.00, 4.81) followed by lack of maternal education (OR: 1.74; BMI (OR: 1.82; 95% CI: 1.44, 2.29), short paternal stature (OR: 1.63;
95% CI: 1.60, 1.89), low maternal BMI (OR: 1.64; 95% CI: 1.54, 1.75), 95% CI: 1.34, 1.99), and low paternal BMI (OR: 1.33; 95% CI: 1.10,
poor household wealth (OR: 1.25; 95% CI: 1.15, 1.35) and poor 1.60) were statistically significant. Poor household wealth was no
household air quality (OR: 1.22; 95% CI: 1.16, 1.29). Unimproved sa- longer significantly associated with all anthropometric failures after
nitation facility (OR: 1.12; 95% CI: 1.07, 1.18), failure to meet the WHO adjusting for paternal characteristics.
recommended four ANC visits (OR: 1.12, 95% CI: 1.07, 1.17), absence
of SBA at delivery (OR: 1.08, 95% CI: 1.04, 1.13), unsafe stool disposal 3.5. Additional analysis: stratification by age and sex
(OR: 1.08, 95% CI: 1.04, 1.13), and delayed breastfeeding initiation
(OR: 1.05, 95% CI: 1.01, 1.09) were also significantly associated with Some differences were found in the ranking and magnitude of cor-
higher odds of stunting. relates for younger (6–23 months) and older (24 + months) children
(Fig. 4). For instance, short maternal stature was more strongly asso-
3.2. Underweight ciated with stunting among older children (OR: 4.87; 95% CI: 4.37,
5.42) compared to younger children (OR: 3.58; 95% CI: 3.07, 4.18) in
All correlates, except for unsafe drinking water source, were sig- the mutually adjusted models. Similarly, lack of maternal education
nificantly associated with higher risk of underweight in separate age- (OR: 1.79 vs OR: 1.66), low maternal BMI (OR: 1.71 vs OR: 1.48) and
and sex-adjusted models (Fig. 2). Short maternal stature was associated poor household wealth (OR: 1.28 vs OR: 1.19) were more strongly as-
with almost 5-fold increased odds of underweight (OR: 4.70; 95% CI: sociated with stunting among older children than younger children,

5
R. Kim, et al. Social Science & Medicine 238 (2019) 112374

Fig. 2. Age- and sex-adjusted odds ratio (OR) and 95% confidence interval (CI) for each maternal and child health and nutrition (MCHN) and paternal factors, Indian
National Family Health Survey 2016. ORs estimated from separate age- and sex-adjusted logistic models accounting for the cluster survey design. Full estimates are
reported in Supplementary Table 1. BMI: body mass index; ANC: antenatal care visits; SBA: skilled birth attendant; ORT: oral rehydration therapy. Paternal char-
acteristics (paternal education, height, BMI) were available for a subset of 25,603 children.

6
R. Kim, et al. Social Science & Medicine 238 (2019) 112374

Fig. 3. Mutually adjusted odds ratio (OR) and 95% confidence interval (CI) for (A) 20 maternal and child health and nutrition (MCHN) factors among 140,444
children and (B) 23 MCHN and paternal factors among a subset of 25,603 children, Indian National Family Health Survey 2016. ORs estimated from mutually
adjusted models accounting for age- and sex-of the child, birth order, place of residence (urban/rural) and the cluster survey design. Full estimates are reported in
Supplementary Tables 2 and 3. BMI: body mass index; ANC: antenatal care visits; SBA: skilled birth attendant; ORT: oral rehydration therapy.

while the contrary was observed for poor sanitation (OR: 1.14 vs OR:
1.10). But the overall finding on the importance of maternal factors and
household socioeconomic and environmental factors remained the
same for both age groups. We found highly consistent results from sex-
stratified models (Supplementary Fig. 1).
3.6. Sensitivity analysis: low birth weight
In our primary analytic sample, 12% had birth weight < 2,500 g,
65% had birth weight ≥2,500 g, and the remaining 23% had no data
(either not weighed or responded ‘don't know’). Adjusting for age and
sex of the child, low birth weight was significantly associated with
higher odds of stunting (OR: 1.50; 95% CI: 1.43, 1.58), underweight
(OR: 1.88; 95% CI: 1.78, 1.98), and wasting (OR: 1.55; 95% CI: 1.46,
1.64). When its relative importance was assessed in light of all the other
MCHN factors, low birth weight was found to be the fourth most im-
portant correlate of stunting (OR: 1.37; 95% CI: 1.29, 1.44), third for
underweight (OR: 1.73; 95% CI: 1.63, 1.83), and second for wasting
(OR: 1.49; 95% CI: 1.40, 1.58) (Supplementary Fig. 2). Inclusion of low
birth weight had no substantial effect on the ranking and magnitude of
other correlates.
4. Discussion
This study provides three salient findings to inform the NNS-related
policy discussion in India. First, when a comprehensive set of MCHN
correlates were simultaneously evaluated in mutually adjusted models,
maternal factors (height, BMI, education) and household socio-
economic and environmental factors (wealth, air quality, sanitation)
were found to have the most consistent and strongest associations with
increased risk of anthropometric failures. Additional consideration of
paternal characteristics showed that father's height, BMI and education
were also important predictors of anthropometric failures, but to a
lesser degree than maternal factors. Second, properly accounting for the

7
R. Kim, et al. Social Science & Medicine 238 (2019) 112374

Fig. 4. Mutually adjusted odds ratio (OR) and 95% confidence interval (CI) for maternal and child health and nutrition (MCHN) factors among (A) 6–23 months old
children and (B) 24–59 months old children, Indian National Family Health Survey 2016. ORs estimated from mutually adjusted models accounting for age- and sex-
of the child, birth order, place of residence (urban/rural) and the cluster survey design. BMI: body mass index; ANC: antenatal care visits; SBA: skilled birth attendant;
ORT: oral rehydration therapy.

role of confounding by socioeconomic status and maternal health in-
dicators demonstrated the small effect of remaining factors that were
individually associated with anthropometric failures in prior literature.
We found weaker and less consistent results for correlates such as
dietary diversity, vitamin A supplementation and breastfeeding initia-
tion, indicating a rather limited role of nutrition-specific interventions
that are not accompanied by nutrition-sensitive programmes. Third,
while prior literature emphasized the age dynamics in the different
manifestations of growth failure (Victora et al., 2010), we found no
evidence of systematic differences in the relative importance of corre-
lates of anthropometric failures by age of children. Taken together, our
findings suggest that breaking the vicious cycle of multi-generational
poverty and improving environmental factors are promising invest-
ments to improve child development.
There are several data-related limitations to our study. Given the
cross-sectional nature of our analysis, we do not make any causal in-
ference. Moreover, we are unable to identify children that may have
experienced a decrease in growth velocity but recovered by the time of
the survey nor children that suffered an insult at the time of the survey
but have gained weight/height later on. Another concern relates to the
fact that most of the correlates considered in our analysis were self-
reported by mothers, and therefore are prone to potential measurement
error. However, anthropometric measures for children and parents
were objectively taken by field investigators, and NFHS is generally
known for high quality data with standardized and representative
sampling of participants and high response rates (∼98%) (Corsi et al.,
2012). Lastly, while our complete case analysis excluded children with
missing data on the primary outcomes and correlates, we found no
evidence for systematic differences in respect to age, place of residence,
household wealth, and maternal education when comparing the dis-
tribution among total eligible children versus our final analytic sample
for the primary analysis (Supplementary Table 4).
Our findings were highly consistent with a prior study that used the
NFHS-3 (2005–2006) to estimate relative importance and population

8
R. Kim, et al.
attributable risk and fractions of 15 relevant risk factors of stunting
(Corsi et al., 2016b). Compared to their mutually adjusted estimates for
the five most important predictors (short maternal stature, mother
having no education, households in lowest wealth quintile, poor dietary
diversity, and maternal underweight), in the latest NFHS-4 data we
found increase in the relative importance of maternal height and BMI;
decrease in the magnitude of associations for maternal education and
household wealth; and a statistical significance of household air quality
instead of dietary diversity. We additionally found a weaker role of
household wealth, maternal education and dietary diversity when fa-
thers' characteristics were further adjusted for. By assessing a more
comprehensive set of correlates, we also identified the importance of
ANC visits and SBA for stunting and underweight even after controlling
for all other MCHN factors. Another notable contribution of our work
was in confirming the consistency in the significance of maternal fac-
tors and household socioeconomic and environmental factors to reduce
child anthropometric failures for both younger (6–23 months) and older
(24–59 months) children as well as for girls and boys.
When interpreted in light of existing evidence (Bhutta et al., 2008;
R. Black et al., 2013; R. E. Black et al., 2008; Corsi et al., 2016a; Kim
et al., 2017; Ruel et al., 2013), our findings suggest that past and pre-
sent socioeconomic conditions are the most important determinants of
child undernutrition. Household wealth and maternal/paternal stature
and body composition ranked the highest in relative ordering for all
anthropometric failures. Short stature of the mother and father may be
interpreted as a measure of economic welfare that is sensitive to the
consumption of basic necessities or the quality of life during childhood
(Perkins et al., 2016; Steckel, 1983). Maternal height had a relatively
stronger influence than paternal height, suggesting potential benefits of
investing in earlier nutrition in girls to ameliorate faltered growth and
development in the next generation (Subramanian et al., 2009). Par-
ental BMI may be a marker of current standard of living (Subramanian
et al., 2010). Prior studies have found similar associations between
paternal and maternal BMI and childhood undernutrition (Green et al.,
2018; Hasan et al., 2016; Khan and Raza, 2016), suggesting that in-
tergenerational associations in nutritional status are probably not en-
tirely driven by intrauterine biological processes, but rather by con-
tinuity of adversity (Subramanian et al., 2010).
In essence, our findings re-emphasize the need to take a broader
systemic perspective and incorporate multidimensional explanations
for child undernutrition. While the UNICEF conceptual framework of
‘context, causes and consequences’ of childhood stunting offers such
perspective (Stewart et al., 2013), current epidemiological studies and
interventions tend to target specific causes of this framework, such as
maternal nutrition and infant and young child feeding practices, one at
a time without fully incorporating the community and societal factors.
At the same time, this does not necessarily mean that increases in
macroeconomic growth will automatically translate to reductions in
child stunting, underweight, and wasting in India (Subramanian et al.,
2016). Without balanced growth across different sectors and proper
infrastructure in place, economic development and improvement in
individuals' incomes by itself will not change the individual behavioral
or ‘proximal’ risk factors (Subramanian et al., 2016). The relative im-
portance of household air quality and sanitation facility, as well as in-
dicators such as ANC visits and SBA at delivery, implies that structural
changes to improve environment and health service provision will
likely make meaningful contributions to reduce acute and chronic child
undernutrition in India.
To accelerate India's commitment to reduce child undernutrition,
the NNS recognizes the need for convergence of multiple sectors, in-
stitutions and local bodies to support preventive actions that straddle
nutrition-specific interventions and address underlying determinants in
ways that are nutrition-sensitive (NITIAayog, 2017). In our effort to
identify the main correlates of child undernutrition using the latest
nationally representative sample of children in India, we found that
improving socioeconomic circumstances will likely be most effective in
9
Social Science & Medicine 238 (2019) 112374
preventing and reducing all types of anthropometric failure in early
infancy and childhood. Future research and policy discussions should
explicitly link efforts to improve living conditions with nutrition and
health outcomes to ensure children's attainment of optimal growth
potential.
Declaration of interests
The authors have declared that no competing interests exist.
Role of funding sources
The authors received no specific funding for this work.
Acknowledgements
Author contribution: Dr Kim conceptualized and designed the study,
carried out the initial analyses, drafted the initial manuscript, and re-
viewed and revised the manuscript. Mr Rajpal and Drs Joe and Corsi
contributed to data analyses and interpretation of data, and reviewed
and revised the manuscript. Drs Sankar and Mr Kumar contributed to
interpretation of data from a policy perspective and critically reviewed
the manuscript for important intellectual content. Prof Subramanian
conceptualized and designed the study, critically reviewed the manu-
script for important intellectual content, and provided overall super-
vision. All authors approved the final manuscript as submitted and
agree to be accountable for all aspects of the work.
Appendix A. Supplementary data
Supplementary data to this article can be found online at https://
doi.org/10.1016/j.socscimed.2019.112374.
References
Arimond, M., Ruel, M.T., 2004. Dietary diversity is associated with child nutritional
status: evidence from 11 demographic and health surveys. J. Nutr. 134, 2579–2585.
Barros, A.J., Victora, C.G., 2013. Measuring coverage in MNCH: determining and inter-
preting inequalities in coverage of maternal, newborn, and child health interventions.
PLoS Med. 10, e1001390.
Bassani, D.G., Jha, P., Dhingra, N., Kumar, R., 2010. Child mortality from solid-fuel use in
India: a nationally-representative case-control study. BMC Public Health 10, 491.
Baye, K., 2017. The Sustainable Development Goals cannot be achieved without im-
proving maternal and child nutrition. J. Public Health Policy 38, 137–145.
Bhattacharyya, A.K., 2006. Composite index of anthropometric failure (CIAF) classifica-
tion: is it more useful? Bull. World Health Organ. 84 335-335.
Bhutta, Z.A., Ahmed, T., Black, R.E., Cousens, S., Dewey, K., Giugliani, E., et al., 2008.
What works? Interventions for maternal and child undernutrition and survival.
Lancet 371, 417–440.
Black, R., Alderman, H., Bhutta, Z., Gillespie, S., Haddad, L., Horton, S., 2013. In:
Executive Summary of the Lancet Maternal and Child Nutrition Series. Maternal and
Child Nutrition Study Group. Maternal and Child Nutrition, pp. 1–12.
Black, R.E., Allen, L.H., Bhutta, Z.A., Caulfield, L.E., De Onis, M., Ezzati, M., et al., 2008.
Maternal and child undernutrition: global and regional exposures and health con-
sequences. Lancet 371, 243–260.
Corsi, D.J., Mejía-Guevara, I., Subramanian, S., 2016a. Improving household-level nu-
trition-specific and nutrition-sensitive conditions key to reducing child under-
nutrition in India. Soc. Sci. Med. 1982, 157–189.
Corsi, D.J., Mejía-Guevara, I., Subramanian, S., 2016b. Risk factors for chronic under-
nutrition among children in India: estimating relative importance, population attri-
butable risk and fractions. Soc. Sci. Med. 157, 165–185.
Corsi, D.J., Neuman, M., Finlay, J.E., Subramanian, S., 2012. Demographic and health
surveys: a profile. Int. J. Epidemiol. 41, 1602–1613.
De Onis, M., Blössner, M., Borghi, E., 2012. Prevalence and trends of stunting among pre-
school children, 1990–2020. Publ. Health Nutr. 15, 142–148.
De Onis, M., Dewey, K.G., Borghi, E., Onyango, A.W., Blössner, M., Daelmans, B., et al.,
2013. The World Health Organization's global target for reducing childhood stunting
by 2025: rationale and proposed actions. Matern. Child Nutr. 9, 6–26.
Green, M.A., Corsi, D.J., Mejía-Guevara, I., Subramanian, S., 2018. Distinct clusters of
stunted children in India: an observational study. Matern. Child Nutr. 14, e12592.
Haddad, L.J., Hawkes, C., Achadi, E., Ahuja, A., Ag Bendech, M., Bhatia, K., et al., 2015.
Global Nutrition Report 2015: Actions and Accountability to Advance Nutrition and
Sustainable Development. Intl Food Policy Res Inst.
Hasan, M.T., Soares Magalhães, R.J., Williams, G.M., Mamun, A.A., 2016. Long-term
changes in childhood malnutrition are associated with long-term changes in maternal
R. Kim, et al.
BMI: evidence from Bangladesh, 1996–2011. Am. J. Clin. Nutr. 104, 1121–1127.
Horton, S., Steckel, R., 2011. Copenhagen consensus on human challenges. Malnutrition-
Global Economic Losses Attributable to Malnutrition 1900-2000 and Projections to
2050. (Assessment paper).
IIPS, 2017. NFHS-4 (National Family Health Survey-4) Fact Sheet. International Institute
for Population Studies.
IIPS, 2018. National Family Health Survey, India. International Institute for Population
Sciences, Mumbai.
Khan, R.E.A., Raza, M.A., 2016. Determinants of malnutrition in Indian children: new
evidence from IDHS through CIAF. Qual. Quantity 50, 299–316.
Kim, R., Mejía-Guevara, I., Corsi, D.J., Aguayo, V.M., Subramanian, S., 2017. Relative
importance of 13 correlates of child stunting in South Asia: insights from nationally
representative data from Afghanistan, Bangladesh, India, Nepal, and Pakistan. Soc.
Sci. Med. 187, 144–154.
Kruk, M.E., Freedman, L.P., 2008. Assessing health system performance in developing
countries: a review of the literature. Health Policy 85, 263–276.
Martorell, R., Young, M.F., 2012. Patterns of stunting and wasting: potential explanatory
factors–. Adv. Nutr. 3, 227–233.
Menon, P., Bamezai, A., Subandoro, A., Ayoya, M.A., Aguayo, V., 2015. Age-appropriate
infant and young child feeding practices are associated with child nutrition in India:
insights from nationally representative data. Matern. Child Nutr. 11, 73–87.
NITIAayog, 2017. Nourishing India - national nutrition strategy. In: India, G.o. (Ed.),
India.
Perkins, J.M., Kim, R., Krishna, A., McGovern, M., Aguayo, V.M., Subramanian, S., 2017.
Understanding the association between stunting and child development in low-and
middle-income countries: next steps for research and intervention. Soc. Sci. Med. 193,
101–109.
Perkins, J.M., Subramanian, S., Davey Smith, G., Özaltin, E., 2016. Adult height, nutri-
tion, and population health. Nutr. Rev. 74, 149–165.
Ruel, M.T., Alderman, H., Maternal, & Group, C.N.S., 2013. Nutrition-sensitive inter-
ventions and programmes: how can they help to accelerate progress in improving
maternal and child nutrition? Lancet 382, 536–551.
Ruel, M.T., Menon, P., 2002. Creating a Child Feeding Index Using the Demographic and
10
Social Science & Medicine 238 (2019) 112374
Health Surveys: an Example from Latin America. International Food Policy Research
Institute, Washington, DC.
Shenkin, S., Zhang, M., Der, G., Mathur, S., Mina, T., Reynolds, R., 2017. Validity of
recalled v. recorded birth weight: a systematic review and meta-analysis. J. Dev.
Origins Health Dis. 8, 137–148.
Steckel, R.H., 1983. Height and per capita income. Hist. Methods J. Quant. Interdiscip.
Hist. 16, 1–7.
Stewart, C.P., Iannotti, L., Dewey, K.G., Michaelsen, K.F., Onyango, A.W., 2013.
Contextualising complementary feeding in a broader framework for stunting pre-
vention. Matern. Child Nutr. 9, 27–45.
Subramanian, S., Ackerson, L.K., Smith, G.D., 2010. Parental BMI and childhood under-
nutrition in India: an assessment of intrauterine influence. Pediatrics, peds 2010-
0222.
Subramanian, S., Ackerson, L.K., Smith, G.D., John, N.A., 2009. Association of maternal
height with child mortality, anthropometric failure, and anemia in India. Jama 301,
1691–1701.
Subramanian, S., Mejía-Guevara, I., Krishna, A., 2016. Rethinking policy perspectives on
childhood stunting: time to formulate a structural and multifactorial strategy.
Matern. Child Nutr. 12, 219–236.
Sudfeld, C.R., McCoy, D.C., Danaei, G., Fink, G., Ezzati, M., Andrews, K.G., et al., 2015.
Linear growth and child development in low-and middle-income countries: a meta-
analysis. Pediatrics 135, e1266–e1275.
UNICEF, 2009. Tracking Progress on Child and Maternal Nutrition: a Survival and
Development Priority.
Victora, C.G., de Onis, M., Hallal, P.C., Blössner, M., Shrimpton, R., 2010. Worldwide
timing of growth faltering: revisiting implications for interventions. Pediatrics, peds
2009-1519.
WHO, 2006. World Health Organization (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. World Health Organization.
WHO, 2016. WHO Recommendations on Antenatal Care for a Positive Pregnancy
Experience. World Health Organization.