Evaluation of Linear Growth at Higher Altitudes

Key Points Question Do children residing at higher altitudes have systematically different linear growth trajectories? Findings In this cross-sectional study including 964 299 children aged 0 to 59 months, those residing at higher altitudes were, on average, born at a shorter length and remained on a lower growth trajectory than children residing at lower altitudes. This altitude-mediated growth difference remained statistically and biologically significant after controlling for potential confounding factors or when the sample was restricted to children living in ideal home environments. Meaning The findings of this study suggest that specific attention and health care guidance are needed for the management of pregnancies and early child development in high-altitude settings; altitude-specific adjustments to 2006 World Health Organization growth standards are not recommended.

© 2020 American Medical Association. All rights reserved. 8

Appendix 1: Immediate causes of undernutrition: inadequate dietary intake and disease
Inadequate dietary intake and disease are considered as the immediate causes of child undernutrition 1,2 . As measures of inadequate dietary intake, we used exclusive breastfeeding (0-6 month old children), dietary diversity (6-23 m), and minimum meal frequency diet (6-23 m) as dependent variables. These dietary indicators were calculated following WHO 3 guidelines: see eTable 3. Questions about child's dietary diversity were introduced in DHS phase-5 (roughly mid-2000s) and therefore the sample for this indicator is restricted to surveys that were based on phase 5 or a later questionnaire. Dietary diversity is based on caregiver's responses to yes/no questions about child's consumption of different food items in the last 24 hours. These food items are then grouped in to seven food groups based on their nutritional qualities. A child who consumed from all seven food groups received a score 7, and a child that consumed only from one food group receives a score 1, and so on. These indicators were developed and validated to proxy diet quality and energy intake in children 4,5 and have been consistently associated with child growth 6,7 . To date, they are the best, mutually comparable, indicators recommended by the WHO that are available for almost all LMIC with a DHS 8 .
To measure disease risk, we used incidence of diarrhoea, fever or cough in the 2 weeks preceding the interview. DHS data collect these data for all children 0-59 m.

Indicator name Definition Exclusive breastfeeding under 6 months
Proportion of infants 0-5 months of age who are fed exclusively with breast milk.
Minimum dietary diversity Proportion of children 6-23 months of age who receive foods from 4 or more food groups.
Minimum meal frequency Proportion of breastfed and non-breastfed children 6-23 months of age who receive solid, semisolid, or soft foods (but also including milk feeds for non-breastfed children) the minimum number of times or more.

eFigure 2: The number of people (in millions) residing above 1,500 masl
The map is based on Center for International Earth Science Information Network (CIESIN, Columbia University) data on population disaggregated by altitude. Countries with large populations residing above 1,500 masl are mainly located in Asia, sub-Saharan Africa and South and Central America © 2020 American Medical Association. All rights reserved. 11 Data are odds ratios from a logistic regression. The 95% confidence intervals (CI) are reported in brackets. Statistical significance denoted at * p < 0.05, ** p <0.01, *** p < 0.001. The estimates in 'adjusted' column are based on adjusted regression that controls for child age (set of binary variables for each age-in-month) and sex, maternal age and education, household wealth and access to improved water and sanitation, binary variable capturing rural areas, and binary indicator variables capturing subnational regions (highest administrative unit in each country).

Appendix 2: Ideal Home Environments
To assess the validity of the WHO growth standards, Karra, Subramanian, Fink 10 used DHS data from 63 countries (169 surveys) to identify children living in environments in which they were more likely to achieve their full genetic growth potential. The authors used six criteria for an 'Ideal home environment': 1) Children were single births; 2) Access to safe water and sanitation; 3) Household lived in a house with finished floors, owned a television, and a car; 4) Children were born to highly educated mothers (more than 13 years of schooling); 5) Children were born in hospitals; and 6) Children had received Bacillus Calmette-Guérin, first diphtheria, pertussis, and tetanus vaccinations.  Figure 4 in the main text suggest that HAZ<0 is driven by children residing at higher altitudes.
Finally, eFigure 4 below shows that, in line with the full DHS sample (see eFigure 1 above), more than half of the children living in ideal home environments were found residing below 500 masl. In the main text, we used continuous altitude (in 1,000-masl) measure in our multivariable regression models (Table 1). This was motivated by the finding in Figure 2   Data are regression coefficient with 95% CIs in brackets. Statistical significance denoted at * p < 0.05, ** p <0.01, *** p < 0.001. The estimates in 'unadjusted' column are based on unadjusted regression. The estimates in 'adjusted' column are based on adjusted regression that controls for child age (set of binary variables for each month) and sex, maternal age and education, household wealth and access to improved water and sanitation, binary variable capturing rural areas, and subnational region (highest administrative unit in each country) fixed effects.
© 2020 American Medical Association. All rights reserved. 19  Data are regression coefficient with 95% CIs in brackets. Statistical significance denoted at * p < 0.05, ** p <0.01, *** p < 0.001. The estimates in 'unadjusted' column are based on unadjusted regression. The estimates in 'adjusted' column are based on adjusted regression that controls for child age (set of binary variables for each month) and sex, maternal age and education, household wealth and access to improved water and sanitation, binary variable capturing rural areas, and subnational region (highest administrative unit in each country) fixed effects. Additional controls include average annual rainfall (in mm) in the cluster in 1981-2010 and average annual temperature (in C) in the cluster in 1981-2010. The sample is restricted to DHS surveys for which climate data could be linked to.

eAppendix 5: Robustness to controlling for maternal stature
Low maternal height is found to be a risk factor for linear growth faltering and adverse birth outcomes 13,14 . It is possible that differences in maternal height could be a confounding factor that explains differences in child height between low and high altitude locations. However, maternal height captures both genetic and non-genetic factors 15 .
Controlling for genetic differences between low and high altitude locations is desirable but if maternal height was influenced by the environment (i.e., altitude) in which the mother herself grew up, then it is not clear whether we Mindful of this conceptual ambiguity, we did not include maternal height as a control variable in the multivariable regression model reported in the main text. To explore sensitivity in this regard, we re-estimated the top panel of Table   1 ('All children') by adding maternal height to the regression model. eTable 9 shows that estimated coefficients were similar to those reported in Table 1 of the main text. Data are regression coefficient with 95% CIs in brackets. Statistical significance denoted at * p < 0.05, ** p <0.01, *** p < 0.001. The estimates in 'unadjusted' column are based on unadjusted regression. The estimates in 'adjusted' column are based on adjusted regression that controls for child age (set of binary variables for each month) and sex, maternal age and education, household wealth and access to improved water and sanitation, binary variable capturing rural areas, and subnational region (highest administrative unit in each country) fixed effects. Additional controls include maternal height (in cm). eAppendix 6: Robustness to restricting the sample to children whose mother had lived in the same location at least since the conception Children's heights in the DHS are measured at the time of the household interview. It could be that the location in which the child lived at the time of measurement was different to the location in which the child was born. Including these 'migrant' children in the sample could cause a bias to our 'altitude deficit' estimates if the altitude of the place of birth was different to the place of the DHS cluster in which the child resided at the time of the interview.
Fortunately, some DHS surveys collect information about the households' recent migration history. More specifically, the questionnaire asks "How long have you been living continuously in (name of current city/town/village of residence)?". We have this information for 699,714 children. Using the child's current age and the response to this question, we calculated that about 73 percent of the children (for whom we have this information) were measured in the same location they were conceived in (i.e., the number of years in current location ≥ child's age in years +1 year). In eTable 10, we restricted the sample to these children whose mother had lived in the same location at least since the conception. We see that coefficients are very similar to those reported in the top panel of Table 1 ('All children'). Data are regression coefficient with 95% CIs in brackets. Statistical significance denoted at * p < 0.05, ** p <0.01, *** p < 0.001. The estimates in 'unadjusted' column are based on unadjusted regression. The estimates in 'adjusted' column are based on adjusted regression that controls for child age (set of binary variables for each month) and sex, maternal age and education, household wealth and access to improved water and sanitation, binary variable capturing rural areas, and subnational region (highest administrative unit in each country) fixed effects. The sample is restricted to children who have lived in the same location at least since the conception. Data are regression coefficient with 95% CIs in brackets. Statistical significance denoted at * p < 0.05, ** p <0.01, *** p < 0.001. The estimates in 'unadjusted' column are based on unadjusted regression. The estimates in 'adjusted' column are based on adjusted regression that controls for child age (set of binary variables for each month) and sex, maternal age and education, household wealth and access to improved water and sanitation, binary variable capturing rural areas, and subnational region (highest administrative unit in each country) fixed effects. The sample is restricted to countries that have clusters above 1,500 masl.  Note: Data are regression coefficient with 95% CIs in brackets. Statistical significance denoted at * p < 0.05, ** p <0.01, *** p < 0.001. The estimates in 'unadjusted' column are based on unadjusted regression. The estimates in 'adjusted' column are based on adjusted regression that controls for child age (set of binary variables for each month) and sex, maternal age and education, household wealth and access to improved water and sanitation, binary variable capturing rural areas, and subnational region (highest administrative unit in each country) fixed effects.