Skip to main content

The association of early-life exposure to ambient PM2.5 and later-childhood height-for-age in India: an observational study

Abstract

Background

Children in India are exposed to high levels of ambient fine particulate matter (PM2.5). However, population-level evidence of associations with adverse health outcomes from within the country is limited. The aim of our study is to estimate the association of early-life exposure to ambient PM2.5 with child health outcomes (height-for-age) in India.

Methods

We linked nationally-representative anthropometric data from India’s 2015–2016 Demographic and Health Survey (n = 218,152 children under five across 640 districts of India) with satellite-based PM2.5 exposure (concentration) data. We then applied fixed effects regression to assess the association between early-life ambient PM2.5 and subsequent height-for-age, analyzing whether deviations in air pollution from the seasonal average for a particular place are associated with deviations in child height from the average for that season in that place, controlling for trends over time, temperature, and birth, mother, and household characteristics. We also explored the timing of exposure and potential non-linearities in the concentration-response relationship.

Results

Children in the sample were exposed to an average of 55 μ g/m3 of PM2.5 in their birth month. After controlling for potential confounders, a 100 μg/m3 increase in PM2.5 in the month of birth was associated with a 0.05 [0.01–0.09] standard deviation reduction in child height. For an average 5 year old girl, this represents a height deficit of 0.24 [0.05–0.43] cm. We also found that exposure to PM2.5 in the last trimester in utero and in the first few months of life are significantly (p < 0.05) associated with child height deficits. We did not observe a decreasing marginal risk at high levels of exposure.

Conclusions

India experiences some of the worst air pollution in the world. To our knowledge, this is the first study to estimate the association of early-life exposure to ambient PM2.5 on child height-for-age at the range of ambient pollution exposures observed in India. Because average exposure to ambient PM2.5 is high in India, where child height-for-age is a critical challenge in human development, our results highlight ambient air pollution as a public health policy priority.

Peer Review reports

Background

India experiences some of the worst particulate air pollution in the world, with mean PM2.5 concentrations consistenly above World Health Organization guidelines [1,2,3]. Due to these high exposures, the Disease Burden of India study recently estimated that 12.9 (11.4–14.4) million disability adjusted life years (DALY) and 149.8 (132.3–167.6) thousand deaths annually were attributable to PM2.5 in children under 5 years of age [4]. The child mortality burden due to household PM2.5 exposure in India is also large, estimated at 50 (30–60) thousand deaths per year. These estimates, however, rely on concentration-response functions that were developed from epidemiological studies carried out primarily in high-income countries. Moreover, the empidemiological studies used for constructing these estimates do not consider impacts on a number of child health outcomes that have been linked to PM2.5 exposure, including sudden infant death syndrome [5], low birth weight [6, 7], intrauterine growth retardation [8] and reduced size [9].

Unlike many developed countries, India does not have a vital registration system, making it difficult to study mortality, a commonly used outcome variable in the air pollution literature. Another widely-studied marker of early-life health insults is the average height of children. Children in India are unusually short compared to international standards [10]. Many causes of this child height deficit have been proposed in the demographic, epidemiological, and econometric literatures, including poor sanitation and maternal nutrition [11, 12]. Exposure to PM2.5 from household solid fuel use for cooking and heating has also been associated with child growth in India [13, 14], but to our knowledge, no study from India has explored the link between ambient air pollution and child growth. One study from Bangladesh [8] observed that the risk of child stunting and wasting was positively associated with higher levels of in utero exposure.

In this study, we examine the association of in utero and early-age ambient PM2.5 exposure on child height-for-age in India. We do so in a representative sample of Indian children – a population exposed to a large range of ambient PM2.5 – using data from the 2015–2016 Demographic and Health Survey (DHS), matched to air pollution data, as measured by satellite remote sensing. The association between child height-for-age and early-life exposure to air pollution is estimated using an approach that accounts for fixed differences across villages, secular trends over time, and district-specific seasonal patterns. We also investigate the shape of the concentration-response function.

Data and methods

India’s 2015–2016 demographic and health survey

Data on child height and potential confounders are taken from India’s most recent DHS survey (note that in India the DHS is also known as the National Family Health Survey). These data were collected from a nationally-representative sample of women of reproductive age. The survey visited all 640 Indian districts that existed at the time of the 2011 Census, and was designed to be representative at the district level. These data were collected between January 2015 and November 2016.

In our analysis, the outcome (dependent) variable is a child’s height-for-age z-score, scaled according to the World Health Organization 2006 reference population mean and standard deviation by sex and age-in-months [15]. In the DHS, height is measured for children less than 5 years old at the time of the survey. The sex and month of birth (e.g. August 2011) is also recorded for each child with measured height.

Air pollution data by district-month

Each child was assigned the average ambient PM2.5 exposure in his or her district of residence during the month in which he or she was born. This matching assumes that the district where children live at the time of the survey is the same as the district where children lived when they were born.

Because India lacks ground-based PM2.5 measurements at a spatial resolution sufficient for our study design, we used satellite-derived PM2.5. Specifically, we use the Multiangle Imaging SpecroRadiometer (MISR) retrieved daily aerosol optical depth (AOD) V22 product at 17.6 km × 17.6 km spatial resolution to estimate PM2.5 with the help of a spatially and temporally varying conversion factor (ƞ). ƞ is derived from GEOS-Chem chemical transport model simulations and depends on aerosol vertical distribution, emissions, and meteorological factors like temperature, relative humidity, and precipitation. Details about the conversion factor ƞ are discussed elsewhere [2, 16, 17]. The MISR AOD product was previously and extensively evaluated for the Indian subcontinent [18]. The satellite-retrieved PM2.5 was bias-corrected using coincident ground-based quality controlled measurements following our earlier study and has ~ 10% uncertainty [2, 19]. The district-level statistics are extracted using the shape files of the district boundaries in ArcGIS. We generated a monthly PM2.5 exposure database for 15 years (2001–2015), although because height is only measured in the DHS for children under five, no child in our sample was born before 2010.

Temperature data by district-month

Considering the large spatio-temporal heterogeneity in temperature across India [20], we control for temperature in the month and district of birth. Monthly temperature data at the 0.125° × 0.125° (approximately, a 12 km × 12 km grid) resolution was obtained from the European Centre for Medium Range Weather Forecast (ECMWF) ERA-INTERIM dataset. Mean district-level temperature was estimated using a spatially weighted average of the 0.125° × 0.125° grid cells in the district.

Main statistical approach

The central empirical strategy of this paper is fixed effects regression, with child height-for-age as the outcome (dependent) variable, and early-life district-month exposure to ambient PM2.5 as the independent variable (exposure) of interest. Fixed effects regression has been identified as a useful tool in epidemiological analyses to control for unobserved characteristics that are common across observations within groups, time periods, or individuals [21], and has been successfully applied in prior epidemiological studies of air pollution-health relationships [8]. In our analysis, we include fixed effects for birth place, seasonal patterns in the district, and year. We therefore study whether deviations in seasonal average PM2.5 in a particular place (village or urban block) are associated with deviations in child height from the average for that same season in that same place. As a result, our study asks the question: is exposure to PM2.5 in the month of birth that is higher than the seasonal average associated with heights that are shorter than average for that place and season of birth?

In India, and other countries where environmental risks are widespread, the average height-for-age z-score declines in the first 2 years of life, reflecting the accumulating impact of early-life health insults on a child’s growth [22]. Because age is predictably correlated with height-for-age, each regression also controls for 119 age-in-months-by-sex indicators, one for each age in months from zero to 59, for girls and boys separately, and excluding one to avoid perfect multicollinearity. This type of adjustment is standard in the literature on child height [10, 11].

Our main models take the following form:

$$ {h}_{ipdmy}=\beta {x}_{dm y}+{\mu}_1{t}_{dm y}+{\mu}_2{t}_{dm y}^2+\rho\ {momh}_{ipdmy}+{\boldsymbol{\alpha}}_{pd}+{\boldsymbol{\gamma}}_{dm}+{\boldsymbol{\delta}}_y+{\boldsymbol{X}}_{ipdmy}\boldsymbol{\theta} +{\varepsilon}_{ipdmy} $$
(1)

where i indexes individual children, p places (survey primary sampling units – PSUs – such as urban blocks or rural villages), d districts, m calendar month of birth (such as February), and y calendar year of birth (such as 2012). The dependent variable, h, is child i’s height-for-age z-score. The independent variable of interest, xdmy, is PM2.5 in district d in month m of year y, corresponding to child i’s birth month. Similarly, tdmy is temperature in that same district-month. We include temperature as a quadratic in order to allow for nonlinearities in its association with child height. momhipdmy is the height of the mother’s child, in centimeters, a proxy for the health and socioeconomic status of the mother. Fixed effects are αpd, 27,266 local places (PSUs); γdm, 7679 categories of district-month (such as for Februarys in Sitapur district, or Aprils in Kanpur district); and δy, 6 calendar years, to capture any secular time trend. Child-level covariates Xipdmy include age-by-sex fixed effects and other covariates that have been associated with child height. These include birth characteristics (mother’s age at birth [23], birth order [24], whether the delivery occurred in a hospital or health facility [25], and whether it was a multiple birth [23]), mother characteristics (whether she smokes [23], the total number of children born to her by the time of the survey [24], and her relationship to the household head [26]), and household-level covariates (caste [12], religion [27], solid fuel use [13], open defecation [11], and drinking water source [8]).

This statistical strategy was designed to address several potential sources of confounding. The strategy allows us to add fixed effects and covariates in stages to verify that the main effect estimate, \( \hat{\beta} \), is robust to respecification. In particular, we first estimate the model without PSU fixed effects. PSU fixed effects would account for any fixed geographic differences in factors known to affect child height, such as the presence and quality of markets [28], local open defecation [29], or the religious composition of the neighborhood [30]. We then add birth, mother, and household characteristics, as described above, which are intended to control for other known determinants of child height. While birth-level covariates reflect characteristics at the time of birth, mother and household level characteristics are observed at the time of the survey. Many of these characteristics, such as mother’s height, caste, and religion, are not likely to have changed over time. However, other characteristics, such as mother smoking, mother’s relationship to the household head, and water source may have changed. The use of solid fuels for cooking and open defecation has been changing relatively slowly in India over time [12, 31], and so these variables are likely to be highly correlated with household behaviors at the time of the child’s birth, even though they are only observed at the time of the survey. Our a priori preferred specification, however, includes birth characteristics only because these variables reflect the environment at the time of birth, and not at the time of the survey. All subsequent analyses build off of this preferred specification.

All of our main model specifications control for district-month fixed effects, a tool which has been used in the literature to control for seasonal trends [32]. This strategy allows each district to have any distinct seasonal pattern, and identifies effects off of deviations from each district’s seasonal patterns. Controlling for seasonal trends is important because pollution in India is highly seasonal [33], and later life outcomes such as educational attainment are also known to be predicted by seasonal patterns [34]. Since child height is also correlated with these outcomes [35], season is likely to be a confounder. A month fixed effect by itself would control for seasonality that is common across all of India. However, since pollution patterns are highly localized, we include district-month fixed effects, which allow seasonal patterns to be different in each district, and therefore we control for local seasonality. We also include a sensitivity check with finer (PSU-month) controls for seasonality.

Finally, we conduct a falsification test, in which we control for ambient PM2.5 in the same district-month, but 2 years before the month of birth; if our identification strategy is credible, this control should not predict height nor change our estimate. Standard errors are clustered by 640 districts, to permit arbitrary correlation of error terms over space and time within districts [36].

DHS data include sampling weights, to be used to generate estimates that are representative of the population of Indian children under five. Although we use weights for our summary statistics in Table 1, sampling weights are not appropriate for estimating associations [37], so we do not use them in any of our statistical models. In Additional file 1: Table S1, we also calculate summary statistics without sampling weights, which are very similar to the statistics presented in Table 1.

Table 1 Summary statistics describing sample of children with measured height from India’s 2015–16 DHS, reported for the full sample and by PM2.5 quintiles in the month and district of birth

Age of exposure

Our primary goal in this study is to investigate the effect on child height of exposure to ambient PM2.5 in the month of birth. The most vulnerable period of exposure is an active area of research [38, 39], but we chose exposure in the month of birth a priori because it represents an important period for early-life development [40]. Nevertheless, in additional analyses we also consider exposure at other ages by averaging over three-month periods, from − 8 to − 6 months before birth (approximately the first trimester of pregnancy) to 10 to 12 months after birth (approximately the last quarter of the first year of life). Average PM2.5 in each age of exposure is used as the independent variable in a separate regression:

$$ {h}_{ipdmy}=\beta \left(\frac{x_{dm y}^{+0}+{x}_{dm y}^{+1}+{x}_{dm y}^{+2}}{3}\right)+{\mu}_1{t}_{dm y}+{\mu}_2{t}_{dm y}^2+\rho\ {momh}_{ipdmy}+{\boldsymbol{\alpha}}_{pd}+{\boldsymbol{\gamma}}_{dm}+{\boldsymbol{\delta}}_y+{\boldsymbol{X}}_{ipdmy}\boldsymbol{\theta} +{\varepsilon}_{ipdmy} $$
(2)

where indices and fixed effects are as in regression eq. (1), but the covariates X include only the age-in-months-by-sex indicators and birth characteristics, as these variables reflect attributes at the time of birth, and not at the time of the survey. Therefore, this model builds off of the preferred specification discussed in the previous section.

Shape of the concentration-response function

The shape of the concentration-response function has been a focus of the air pollution literature, in light of its importance for policy responses [41]. Although the prior literature has suggested the possibility of diminishing marginal risks at higher levels of exposure, there is little well-identified evidence on exposure to PM2.5 at levels as high as in India during the period studied, or for child height as the outcome [42, 43]. Therefore, we perform three analyses in which we allow the shape of the concentration-response function to differ from the linear form in eq. (1):

$$ {h}_{ipdmy}=\beta f\left({x}_{dm y}\right)+{\mu}_1{t}_{dm y}+{\mu}_2{t}_{dm y}^2+\rho\ {momh}_{ipdmy}+{\boldsymbol{\alpha}}_{pd}+{\boldsymbol{\gamma}}_{dm}+{\boldsymbol{\delta}}_y+{\boldsymbol{X}}_{ipdmy}\boldsymbol{\theta} +{\varepsilon}_{ipdmy} $$
(3)

First, we substitute in the natural log of PM2.5 in one specification, and a linear spline at the median of PM2.5 in another. Then we allow polynomial shapes of the concentration-response curve, of degree 1 through 5. Finally, we implement a Box-Cox power transformation of the form f(x) = xλ, for coefficients λ in steps of 0.1 from 0.1 to 2.0. We implement each power transformation in a separate model, and plot the resulting log-likelihoods. If likelihood is maximized near λ = 1, then this procedure would suggest that a linear concentration-response function best fits the data. As in the age of exposure analysis, the covariates X include only the age-in-months-by-sex indicators and birth characteristics, as these variables represent attributes at the time of birth, and not at the time of the survey.

Informed consent

Because we study publicly-available, anonymized data on child height, our study is classified as “not human subjects research” and informed consent is not required.

All analyses in the paper were computed with Stata 12.1.

Results

Summary and descriptive statistics

Height was measured for 225,002 children under five in the DHS. We were able to match air pollution data to children born from February 2010 to December 2015, who are 97% of those with measured height, resulting in a final sample of 218,152 children (Fig. 1). Summary statistics as sample means for these children are presented in Table 1, reported for the full sample, as well as by quintiles of ambient PM2.5 exposure. Across the whole sample, children were exposed to an average of 55 μ g/m3 in their month of birth, although with substantial variation. 92% of children were born while the mother was living in her current residence. The results also highlight that children who are exposed to higher ambient PM2.5 tend to be disadvantaged in other ways: they come from larger families, have shorter mothers, live in households that are more likely to defecate in the open, and use solid fuels for cooking.

Fig. 1
figure 1

Study sample with excluded or missing observations. Note: In Table 2, some samples are smaller than 193,040 because the regression models ignore categories within which there is no variation in the independent variable

Location [44, 45], time of year [33], and mother’s height, as seen in Table 1, are correlated with air pollution concentrations. Therefore, Fig. 2 plots crude associations between pollution and height, stratified by rural/urban, season, and mother’s height. Each panel in Fig. 2 presents locally-weighted kernel regressions of the relationship between ambient PM2.5 in the district-month of birth and height-for-age z-score residuals (after controlling only for age-by-sex, see methods for more discussion). We do not control for any other covariates in this figure. Panel A reveals a negative (downward) gradient – which is approximately linear - between ambient PM2.5 exposure and child height for both rural and urban children. Although the range of PM2.5 exposure is similar in both rural and urban areas, the former are shorter, on average, because they are more exposed to other factors associated with growth faltering [12, 29]. Panel B demonstrates that ambient PM2.5 reaches the highest levels in winter (November through January), and that a similar downward gradient is present in all seasons. Panel C suggests that the association is not driven by mother’s height, since the downward gradient exists for each quintile of mother’s height.

Fig. 2
figure 2

Crude associations between child height and exposure to PM2.5 in the month of birth. Note: Panel a stratifies by urban/rural, Panel b by season of birth, and Panel c by mother’s height. Curves are kernel-weighted local regressions. The vertical axis in all panels is the residual of child height-for-age, after controlling for age-in-months by sex indicators. In Panel B, observations of PM2.5 that are greater than the 95th percentile for each season are left out because a sufficient number of observations are not available to construct means for pollution levels that are very high for the season

Exposure in the month-of-birth

Table 2 presents our main results: fixed effects regression results following Eq. 1. For ease of interpretation, results are presented for a 100 μ g/m3 increase in PM2.5. Column 1 shows coefficients from a regression that includes age-by-sex fixed effects, district-month fixed effects, and year of birth fixed effects. Column 2 adds PSU fixed effects, temperature, and mother’s height. Columns 3, 4, and 5 progressively add birth, mother, and household characteristics, respectively. Across the alternative specifications in columns 1 through 5, a 100 μ g/m3 increase in ambient PM2.5 exposure is associated with an approximately 0.05 standard deviation decrease in child height-for-age z-score. Because columns 4 and 5 include covariates that are measured at the time of the survey, and do not necessarily reflect the environment at the time of birth, Column 3 represents our a priori preferred specification. Column 6, reports the falsification test: as expected, ambient PM2.5 exposure 2 years before the child is born does not predict height and does not change the coefficient of interest. Columns 7 and 8 are tests of the linearity assumption and are discussed further below.

Table 2 Association of district-level PM2.5 (per 100 μg/m3) in month of birth with child height-for-age z-score

Age of exposure

Figure 3 presents the association of PM2.5 and child growth given different time periods of exposure (see Eq. 2 above for modeling details). Of the seven time periods explored, two show significant (p < 0.05) adverse effects on child growth – the last trimester in utero and the period just after birth. No other periods of exposure are significantly associated with child height. These results are consistent with evidence in the literature that shocks in utero and early-life are critical for child development outcomes [40, 46].

Fig. 3
figure 3

Effects of PM2.5 exposure at various ages. Note: Dots denote point estimates and lines denote 95% confidence intervals. Each result shown is from a separate fixed effects regression of child height-for-age on the average exposure to PM2.5 in the months, relative to birth, specified along the horizontal axis

Shape of concentration-response function

The three tests for non-linear concentration-response functions each failed to reject that a linear shape best fits the data. Moreover, each approach suggests that, if anything, effects may be steeper at higher concentration levels. Specifically, column 7 of Table 2 shows that a natural log functional form – consistent with a concentration-response function exhibiting diminishing marginal costs – fits the data less well than the linear form. Column 8 includes a linear spline that allows a different slope above the median level of ambient PM2.5; although the two PM2.5 terms are jointly statistically significant at the 10% level (F = 2.72; p = 0.067), neither is individually significantly different from zero. Although this model does not fit the data better than a simple linear form, the negative sign on the coefficient suggests the possibility of a steeper concentration-response function at higher levels of exposure. Additional file 1: Figure S1 demonstrates that none of the polynomial forms we tested (quadratic through quantic) improve on a linear functional form, while Additional file 1: Figure S2 – the Box-Cox transformation – indicates that a model with slightly increasing marginal effects may best fit the data.

In Additional file 1: Table S2, we present results from statistical analyses similar to columns 6 through 8 of the the main Table 2, the difference is that models presented in the supplementary table include all coverates, including birth, mother, and household characteristics, rather than birth characteristics only. The inclusion of these additional control variables does not change the interpretation of these analyses. We also show that the model is robust to replacing district-month fixed effects with PSU-month fixed effects, a finer measure of seasonality.

Discussion

We report the first evidence of an association between ambient PM2.5 exposure and child height in India by using the country’s most recent DHS, which measures children under 5 years old in a nationally representative sample of reproductive age women. We find that an increase in PM2.5 of 100 μ g/m3 in the month of birth is associated with a decrease of 0.05 height-for-age standard deviations; for an average 5 year old girl, this would equate to a height deficit of 0.24 cm. Consistent with evidence in the literature that shocks in utero and early-life are critical for child development outcomes [40, 46], we find evidence that exposure to PM2.5 during the last few months in utero and the first few months of life are associated with height deficits.

The average child in our data is exposed to a PM2.5 concentration of 55 μ g/m3 in her month of birth. Using the estimates from our analysis, this means that the average child is about 0.027 height-for-age standard deviations shorter than she would be if exposed to very low levels of air pollution at birth. For an average 5 year old girl, this represents a height deficit of 0.13 cm. Although this effect is small relative to other environmental factors affecting child health, such as open defecation [11], it influences all of the almost 30 million births per year that occur in India.

Moreover, the difference between the children in our sample most exposed to PM2.5 (at the 95th percentile) and the children least exposed (at the 5th percentile) is 116 μ g/m3. Therefore, based on our findings, the most exposed children in India are about 0.06 height-for-age standard deviations shorter than they counterfactually would be if they were exposed only at the lowest levels in our sample. This projected difference — 0.06 height-for-age standard deviations — is of the same order of magnitude as other height differences that have received sustained attention in the literature on the demography of child height: it is about half as large as the well-studied India-Africa height gap [11], and is about one-tenth of the height gap between children of literate versus illiterate mothers. Since child growth is highly correlated with early-life mortality [47], the associations we observed in this study are suggestive of an association between PM2.5 exposure and early-life survival. In the data we use for this analysis, a district where children are 0.06 height-for-age standard deviations shorter would be expected, on average, to have an infant mortality that is larger by 5 infant deaths per 1000 live births: a large difference that is approximately equal to Canada’s overall infant mortality rate.

Although child height has traditionally been interpreted as a measure of “malnutrition,” it is increasingly recognized to reflect the totality of early-life health insults, including both net available nutrition and losses due to diseases. Our study does not allow us to observe disease directly; however, mechanisms in the literature are consistent with the association that we document. For example, exposure to particulate matter is associated with lower birth weight [6, 7], which is in turn linked to stature in childhood [48]. Similarly, exposure to ambient air pollution is associated with the incidence of pneumonia [49, 50]. Respiratory infections, like pneumonia, sometimes occur with fevers which can suppress the appetite, and reduce nutrient intake [51]. Moreover, infection and inflammation are metabolically demanding and may reallocate resources at the expense of growth [52].

This study has several limitations. One is the possibility of residual confounding. For example, we were not able to control for potential co-pollutants such as ozone or NO2, for which data is not available. In addition, some variables included in our models were measured at the time of the survey rather than at the time of the child’s birth, such as open defecation and household solid fuel use. However, we have no reason to believe that these practices would have changed for a large proportion of households. Similarly, we assumed that surveyed mothers delivered their children in the same district in which they were surveyed. This assumption seems sound considering that 92% of children were recorded as being born while the mother was living in her current residence, and because migration across districts is relatively rare. Finally, we rely on district-level measures of exposure derived from satellite data, thus raising the possibility of measurement error. However, assuming this error is random, the consequence would be attenuation towards the null, meaning that the true size of the effect of PM2.5 on child height may be larger than we observe here. In light of these limitations, we encourage additional research on this topic. If possible, this would include other study designs (cohort studies, natural experiments etc.) and, when available, finer-resolution estimates of exposure.

Conclusions

To our knowledge, this is the first study to directly estimate the impact of early-life exposure to ambient PM2.5 on child height-for-age at the range of exposures found in India. Because average exposure to ambient PM2.5 is high in India, where child height-for-age is a critical challenge in human development, our results highlight ambient air pollution as public health policy priority. Ambient PM2.5 exposure is likely to increase in India in the near future [1, 53]. Therefore, the health burden that we quantify here could potentially increase unless appropriate policy action is taken to reduce air pollution throughout India. In particular, although policy conversations often focus on Delhi (and, to a lesser extent, other big cities), we find effects throughout India, and on both rural and urban children, suggesting that the policy challenges are broader than is commonly understood. Because child height has lasting consequences for human capital [10, 40], this is a problem with potential ramifications throughout the Indian society and economy.

Availability of data and materials

The DHS 2015–2016 survey data for India are publicly available free of charge, and archived at https://www.dhsprogram.com/data/available-datasets.cfm. MISR AOD data used to generate the PM2.5 exposure are archived at https://misr.jpl.nasa.gov/getData/accessData/. The Stata do files used to arrive at the results depicted in this study are located at https://riceinstitute.org/wp-content/uploads/2019/07/hfa_PM25_public.txt.

References

  1. GBD MAPS Working Group. Burden of disease attributable to major air pollution sources in India, special report 21. Boston: Health Effects Institute; 2018.

    Google Scholar 

  2. Dey S, Di Girolamo L, van Donkelaar A, Tripathi SN, Gupta T, Mohan M. Variability of outdoor fine particulate (PM2.5) concentration in the Indian subcontinent: a remote sensing approach. Remote Sens Environ Elsevier Inc. 2012;127:153–61. https://doi.org/10.1016/j.rse.2012.08.021.

    Article  Google Scholar 

  3. Apte JS, Marshall JD, Cohen AJ, Brauer M. Addressing global mortality from ambient PM2.5. Environ Sci Technol. 2015;49(13):8057–66.

    Article  CAS  Google Scholar 

  4. Dandona L, Dandona R, Kumar GA, Shukla DK, Paul VK, Balakrishnan K, et al. Nations within a nation: variations in epidemiological transition across the states of India, 1990–2016 in the global burden of disease study. Lancet. 2017;6736:1–24.

    Google Scholar 

  5. Litchfield IJ, Ayres JG, Jaakkola JJK, Mohammed NI. Is ambient air pollution associated with onset of sudden infant death syndrome: a case-crossover study in the UK. BMJ Open. 2018;8 Available from: http://bmjopen.bmj.com/content/8/4/e018341.abstract.

  6. Balakrishnan K, Ghosh S, Thangavel G, Sambandam S, Mukhopadhyay K, Puttaswamy N, et al. Exposures to fine particulate matter (PM2.5) and birthweight in a rural-urban, mother-child cohort in Tamil Nadu. India Environ Res. 2018;161:524–31 Available from: http://www.sciencedirect.com/science/article/pii/S0013935117312276.

    Article  CAS  Google Scholar 

  7. Payam D, Jennifer P, BM L, Matteo B, Michael B, DL A, et al. Maternal exposure to particulate air pollution and term birth weight: a multi-country evaluation of effect and heterogeneity. Environ Health Perspect. 2013;121:267–373. https://doi.org/10.1289/ehp.1205575.

    Article  CAS  Google Scholar 

  8. Goyal N, Canning D. Exposure to ambient fine particulate air pollution in utero as a risk factor for child stunting in Bangladesh. Int J Environ Res Public Heal. 2018;15(1):22.

    Article  Google Scholar 

  9. Anna S, Kees de H, Marie P, Payam D, David M, Gerard H, et al. Ambient air pollution and newborn size and adiposity at birth: differences by maternal ethnicity (the born in Bradford study cohort). Environ Health Perspect Environmental Health Perspectives. 2015;123:1208–15. https://doi.org/10.1289/ehp.1408675.

    Article  Google Scholar 

  10. Spears D. Height and cognitive achievement among Indian children. Econ Hum Biol. 2012;10:210–9 Available from: http://www.sciencedirect.com/science/article/pii/S1570677X11000888.

    Article  Google Scholar 

  11. Spears D. Exposure to open defecation can account for the Indian enigma of child height. J Dev Econ. 2018; Available from: http://www.sciencedirect.com/science/article/pii/S0304387818300439.

  12. Coffey D, Spears D. Child height in India. Econ Polit Wkly. 2018;53(31):87–94 Available from: https://www.epw.in/journal/2018/31/special-articles/child-height-india.html.

  13. Mishra V, Retherford RD. Does biofuel smoke contribute to anaemia and stunting in early childhood? Int J Epidemiol. 2007;36:117–29. https://doi.org/10.1093/ije/dyl234.

    Article  Google Scholar 

  14. Balietti A, Datta S. The impact of indoor solid fuel use on the stunting of Indian children. 2017. Working paper.

    Google Scholar 

  15. World Health Organization, Department of Nutrition for Health and Development. WHO Child Growth Standards. 2006. Available from: https://www.who.int/childgrowth/standards/Technical_report.pdf?ua=1.

  16. van Donkelaar A, Martin RV, Brauer M, Boys BL. Use of satellite observations for long-term exposure assessment of global concentrations of fine particulate matter. Environ Health Perspect. 2014;110:135–43 Available from: http://ehp.niehs.nih.gov/1408646.

    Google Scholar 

  17. van Donkelaar A, Martin RV, Brauer M, Kahn R, Levy R, Verduzco C, et al. Global estimates of ambient fine particulate matter concentrations from satellite-based aerosol optical depth: development and application. Environ Health Perspect. 2010;118:847–55.

    Article  Google Scholar 

  18. Dey S, Di Girolamo L. A climatology of aerosol optical and microphysical properties over the Indian subcontinent from 9 years (2000-2008) of multiangle imaging Spectroradiometer (MISR) data. J Geophys Res Atmos. 2010;115:1–22.

    Google Scholar 

  19. Chowdhury S, Dey S. Cause-specific premature death from ambient PM2.5 exposure in India: estimate adjusted for baseline mortality. Environ Int Elsevier Ltd. 2016;91:283–90 Available from: http://linkinghub.elsevier.com/retrieve/pii/S0160412016300848.

    CAS  Google Scholar 

  20. Mondal A, Khare D, Kundu S. Spatial and temporal analysis of rainfall and temperature trend of India. Theor Appl Climatol. 2015;122:143–58.

    Article  Google Scholar 

  21. Gunasekara FI, Richardson K, Carter K, Blakely T. Fixed effects analysis of repeated measures data. Int J Epidemiol. 2013;43(1):264–9.

    Article  Google Scholar 

  22. Victora CG, de Onis M, Hallal PC, Blössner M, Shrimpton R. Worldwide timing of growth faltering: revisiting implications for interventions. Pediatrics. 2010;125:e473 LP–e480 Available from: http://pediatrics.aappublications.org/content/125/3/e473.abstract.

    Article  Google Scholar 

  23. Pietiläinen KH, Kaprio J, Räsänen M, Winter T, Rissanen A, Rose RJ. Tracking of body size from birth to late adolescence: contributions of birth length, birth weight, duration of gestation, parents' body size, and twinship. Am J Epidemiol. 2001;154(1):21–9.

    Article  Google Scholar 

  24. Coffey, D, and Spears D. Neonatal death in India: birth order in a context of maternal undernutrition. IZA discussion paper series. 2019. IZA DP No. 12288.

  25. Coffey D. The association between neonatal death and facility birth in regions of India. Demogr Res. 2019;40:417–30.

    Article  Google Scholar 

  26. Coffey D, Khera R, and Spears D. Women’s status and children’s height in India: Evidence from joint rural households. Working paper. r.i.c.e. 2016. Available from: https://riceinstitute.org/research/womens-status-and-childrens-height-in-india-evidence-from-joint-rural-households/.

  27. Subramanian SV, Ackerson LK, Smith GD, John NA. Association of maternal height with child mortality, anthropometric failure, and anemia in India. Jama. 2009;301(16):1691–701.

    Article  CAS  Google Scholar 

  28. Darrouzet-Nardi AF, Masters WA. Urbanization, market development and malnutrition in farm households: evidence from the demographic and health surveys, 1986–2011. Food Secur. 2015;7:521–33. https://doi.org/10.1007/s12571-015-0470-9.

    Article  Google Scholar 

  29. Coffey D, Spears DE. Where India goes : abandoned toilets, stunted development and the costs of caste. India: Harper Collins; 2017.

  30. Geruso M, Spears D. Neighborhood sanitation and infant mortality. Am Econ J Appl Econ. 2018;10:125–62.

    Article  Google Scholar 

  31. Gupta A, Vyas S, Hathi P, Khalid N, Srivastav N, Spears D, Coffey D. Persistence of solid fuel use despite increases in LPG ownership: New survey evidence from rural north India. Working paper. r.i.c.e. 2019. Available from: https://riceinstitute.org/research/persistence-of-solid-fuel-use-despite-increases-in-lpg-ownership-new-survey-evidence-from-rural-north-india/.

  32. Barreca A, Deschenes O, Guldi M. Maybe next month? Temperature shocks and dynamic adjustments in birth rates. Demography. 2018;55:1269–93. https://doi.org/10.1007/s13524-018-0690-7.

    Article  Google Scholar 

  33. Vyas S, Srivastav N, Spears D. An experiment with air purifiers in Delhi during Winter 2015-2016. PLoS One. 2016;11(12):e0167999.

    Article  Google Scholar 

  34. Buckles KS, Hungerman DM. Season of birth and later outcomes: old questions, new answers. Rev Econ Stat. 2013;95:711–24 Available from: http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3777829/.

    Article  Google Scholar 

  35. Case A, Paxson C. Stature and status: height, ability, and labor market outcomes. J Polit Econ. 2008;116(3):499–532.

    Article  Google Scholar 

  36. Colin Cameron A, Miller DL. A practitioner’s guide to cluster-robust inference. J Hum Resour. 2015;50:317–72 Available from: http://jhr.uwpress.org/content/50/2/317.abstract.

    Article  Google Scholar 

  37. Solon G, Haider SJ, Wooldridge J. What are we weighting for? J Hum Resour. 2015;50:301–16 Available from: http://www.nber.org/papers/w18859.pdf.

  38. Šrám RJ, Binková B, Dejmek J, Bobak M. Ambient air pollution and pregnancy outcomes: a review of the literature. Environ Health Perspect. 2005;113:375–82.

    Article  Google Scholar 

  39. Herr CE, Dostal M, Ghosh R, Ashwood P, Lipsett M, Pinkerton KE, Sram R, Hertz-Picciotto I. Air pollution exposure during critical time periods in gestation and alterations in cord blood lymphocyte distribution: a cohort of livebirths. Environ Health. 2010 Dec;9(1):46.

    Article  Google Scholar 

  40. Currie J, Vogl T. Early-life health and adult circumstance in developing countries. Annu Rev Econ. 2013;5(1):1–36.

    Article  Google Scholar 

  41. Nasari MM, Szyszkowicz M, Chen H, Crouse D, Turner MC, Jerrett M, et al. A class of non-linear exposure-response models suitable for health impact assessment applicable to large cohort studies of ambient air pollution. Air Qual Atmos Heal. 2016;9:961–72.

    Article  CAS  Google Scholar 

  42. Burnett R, Chen H, Szyszkowicz M, Fann N, Hubbell B, Pope CA, Apte JS, Brauer M, Cohen A, Weichenthal S, Coggins J. Global estimates of mortality associated with long-term exposure to outdoor fine particulate matter. Proc Natl Acad Sci. 2018;115(38):9592–7.

    Article  CAS  Google Scholar 

  43. Burnett RT, Pope CA III, Ezzati M, Olives C, Lim SS, Mehta S, Shin HH, Singh G, Hubbell B, Brauer M, Anderson HR. An integrated risk function for estimating the global burden of disease attributable to ambient fine particulate matter exposure. Environ Health Perspect. 2014;122(4):397–403.

    Article  Google Scholar 

  44. Crouse DL, Peters PA, van Donkelaar A, Goldberg MS, Villeneuve PJ, Brion O, Khan S, Atari DO, Jerrett M, Pope CA III, Brauer M. Risk of nonaccidental and cardiovascular mortality in relation to long-term exposure to low concentrations of fine particulate matter: a Canadian national-level cohort study. Environ Health Perspect. 2012;120(5):708–14.

    Article  Google Scholar 

  45. Cesaroni G, Badaloni C, Gariazzo C, Stafoggia M, Sozzi R, Davoli M, Forastiere F. Long-term exposure to urban air pollution and mortality in a cohort of more than a million adults in Rome. Environ Health Perspect. 2013 Jan 8;121(3):324–31.

    Article  Google Scholar 

  46. Rangel M, Vogl T. Agriculture, fire, and infant health. Natl Bur Econ Res. 2016. NBER Working Paper No. 22955. https://www.nber.org/papers/w22955.pdf

  47. Bozzoli C, Deaton A, Quintana-Domeque C. Adult height and childhood disease. Demography. 2009;46:647–69 Available from: http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2809930/.

    Article  Google Scholar 

  48. Binkin NJ, Yip R, Fleshood L, Trowbridge FL. Birth weight and childhood growth. Pediatrics. 1988;82(6):828–34.

    CAS  Google Scholar 

  49. Ram PK, Dutt D, Silk BJ, Doshi S, Rudra CB, Abedin J, et al. Household air quality risk factors associated with childhood pneumonia in urban Dhaka, Bangladesh. Am J Trop Med Hyg. 2014;90:968–75.

    Article  Google Scholar 

  50. Dherani M, Pope D, Mascarenhas M, Smith KR, Weber M, Bruce N. Indoor air pollution from unprocessed solid fuel use and pneumonia risk in children aged under five years: a systematic review and meta-analysis. Bull World Health Organ. 2008;86:390–4.

    Article  Google Scholar 

  51. Dewey KG, Mayers DR. Early child growth: how do nutrition and infection interact? Matern Child Nutr. 2011;7:129–42.

    Article  Google Scholar 

  52. Crimmins EM, Finch CE. Infection, inflammation, height, and longevity. Proc Natl Acad Sci. 2006;103(2):498–503.

    Article  CAS  Google Scholar 

  53. Chowdhury S, Dey S, Smith KR. Ambient PM2.5 exposure and expected premature mortality to 2100 in India under climate change scenarios. Nat Commun. 2018;9:318.

Download references

Acknowledgments

MISR aerosol data are archived in NASA Langley Research Atmospheric Science Data Center. All the authors acknowledge Dr. Aaron van Donkelaar, Dalhousie University for sharing the conversion factors from GEOS-Chem CTM which are used to estimate PM2.5. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Funding

This study received no specific funding; however, some of the authors are supported by research center funding. SD acknowledges DST-FIST grant (SR/FST/ESII-016/2014) for upgradation of computing facility at IIT Delhi. DS and SV acknowledge support for research time and computing resources from Bill and Melinda Gates Foundation grant OPP1125318. Research by DS and SV was supported by grant, P2CHD042849, Population Research Center, awarded to the Population Research Center at The University of Texas at Austin by the Eunice Kennedy Shriver National Institute of Child Health and Human Development. SV acknowledges that this material is based upon work supported by the National Science Foundation Graduate Research Fellowship Program under Grant No. DGE-1610403.

Author information

Authors and Affiliations

Authors

Contributions

DS and SD developed the idea, DS carried out the main analysis. SC carried out the satellite data analysis to generate the exposure data. DS and SD wrote the paper with inputs from all the authors. NS and SV revised the article, with inputs from all authors. All authors have read and approved the final version.

Corresponding author

Correspondence to Dean Spears.

Ethics declarations

Ethics approval and consent to participate

Because we study publicly-available, anonymized data on child height, our study was classified as “not human subjects research” by the RICE Institute, Inc. IRB (registered with the NIH as IRB00010425) and informed consent is not required.

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.

Additional information

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Additional file

Additional file 1:

Table S1. Summary statistics describing sample of children with measured height from India’s 2015–16 DHS, computed without sampling weights. Table S2. Association of district-level PM2.5 (per 100 μg/m3) in month of birth with child height-for-age z-score with all covariates. Figure S1. Projected effects of PM2.5 on child height-for-age z-score, at increasing non-linearity. Each curve is the projected effect from a separate fixed effects regression where PM2.5 in the month of birth is specified as a polynomial of degree 1 through 5. p-values report joint F tests that all PM2.5 terms are zero. Figure S2. Box-Cox transformation of PM2.5 in month of birth: Each point plots the log likelihood of a separate fixed effects regression of PM2.5 transformed according to the coefficient on the horizontal axis. (DOCX 83 kb)

Rights and permissions

Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Spears, D., Dey, S., Chowdhury, S. et al. The association of early-life exposure to ambient PM2.5 and later-childhood height-for-age in India: an observational study. Environ Health 18, 62 (2019). https://doi.org/10.1186/s12940-019-0501-7

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1186/s12940-019-0501-7

Keywords