Association of Poor Sanitation With Growth Measurements Among Children in India

Key Points Question Is poor sanitation associated with lower height among children older than 5 years in India? Findings This cross-sectional analysis of 134 882 children and adolescents aged 0 to 18 years in India found that poor sanitation (ie, open defecation and lack of access to boiled or filtered drinking water) was associated with lower height-for-age z scores across all ages in childhood and adolescence. Meaning Improved sanitation may be associated with greater height across all child ages.

This supplementary material has been provided by the authors to give readers additional information about their work.

eAppendix 1. Brief Description of Child Anthropometric Measurement and Coding of Exposure Variable in DLHS-4 Data Set
The data in DLHS 4 were collected from August 2012 to February 2014 and anthropometric measurements were taken for all children aged 0 to 18 years in surveyed households. In the DLHS 4 survey, child anthropometrics were measured according to the Demographic Health Survey protocols in India 1 . The Seca 874 digital scale was used to measure the weight of children and adults. The height of adults and children aged 24-59 months was measured with the Seca 213 stadiometer. The Seca 417 infantometer was used to measure the recumbent length of children under two years or less than 85 cm 2 . Fieldworkers selected for data collection of the DHLS surveys are highly qualified (minimum qualification: diploma in para-medical courses), undergo intensive pre-test training, spot checking during survey interviews and are supervised at multiple levels to maintain data fidelity 3,4 .
The DLHS questionnaire asks respondents about the type of toilet facility in their household and responses are coded as 12 mutually exclusively categories 5 . Survey responses for drinking water source at the household-level are coded into 15 categories and water treatment methods to make it safe to drink are coded into 9 types 5 .
Sample exclusions were made for children without valid age reported in months (N=77,247), outlier or missing data on child anthropometry (N=113,735), mother's height (N=39,919), father's height (N=64,555), and hemoglobin (N=31,967). While exclusions based on child age in months and child anthropometry are unavoidable, exclusions based on parental height and hemoglobin were deemed necessary to obtain unbiased associations between exposures and the outcomes. Specifically, the importance of parental height as a potential confounder superseded the loss of power due to smaller sample size 6 . Moreover, there were no systematic patterns observed in missing parental height or hemoglobin.

eAppendix 2. Description of Control Variables and Equation for Regression Analysis
In this study, control variables for regression analyses include household SES, parental (mother, father) height, education, age, child's age in months, child's hemoglobin level, and birth-year, state fixed effects. Birthyear fixed effects control for factors unique to each annual birth cohort which may vary over time. Inclusion of birth year fixed effects reduces omitted variable bias that arises from shared patterns over time in HAZ and other covariates. State fixed effects control for all unobserved time invariant state-level factors that may impact HAZ and other covariates. Examples of such factors include deeply ingrained dietary practices and differing state-level institutions, policies and programs that existed in 2013. Inclusion of birth-year and state fixed effects exploits the within birth year and within state variation in HAZ and sanitation in our sample.

Regression Equation
For child i, born in year t, to parent p, in household h, village v and state s we specified the following regression model, estimated using ordinary least squares.

eFigure. Estimated Difference in Height-for-Age z Scores From Linear
Regression-Decomposition of Socioeconomic Status for Children Aged 0 to 18 Years in India eFigure1 Legend: eFigure1 shows difference in estimated HAZ (with 95% CIs) associated with change in household-level SES from Quintile 3 (sample mean) to Quintile 5 (i.e. highest SES level). HAZ differences are estimated over narrow age intervals (≤ 3 years, >3 to ≤6 years, >6 to ≤ 9 years, >9 to ≤ 12 years, >12 to ≤ 15 years, >15 to ≤ 18 years). These age intervals were selected to approximate continuous age with equal spacing between successive age groups. HAZ estimates (by age and sex group) present the association between child height and (hypothetical) 'ideal' scenario wherein all households in the study sample moved to highest SES quintile. Standard error estimates are robust and clustered at the district-level. Models adjusted for child age, father's height, mother's height, household socio-economic status, rural residence, father's education, mother's education, mother's age, child's hemoglobin level, state fixed-effects and birth-year fixed effects. * p < 0.05, ** p < 0.01, *** p < 0.001