Changes in Child Undernutrition and Overweight in India from 2006 to 2019: An Analysis of 22 States

Objectives: India has historically displayed high levels of child stunting and low levels of child overweight. Using newly released data, we evaluated changes in priority indicators of child growth from 2006 to 2019 and examined the role of human development measures in these changes. Methods: We estimated cumulative and annualized changes in state- and district-level child growth indicators using three rounds of National Family Health Surveys (2005-06, 2015-16, 2019-20) in 22 states. Outcomes included stunting, underweight, wasting, and overweight. Human development was measured using a principal components analysis of nine survey-based items. We contrasted expected versus observed changes in district-level growth indicators between 2015 and 2019 based on changes in development measures using two-way Blinder Oaxaca decomposition. Results: From 2006 to 2019, the prevalence of stunting and underweight decreased by 10.9 percentage points (pp) and 7.1 pp, respectively, while the prevalence of wasting and overweight increased by 2.8 pp and 2.2 pp, respectively. Annualized rates of change for stunting, wasting, and underweight were lower from 2015 to 2020 compared with the 2006 to 2015 period, while rates of change in overweight were higher. Simultaneously, all nine human development indicators improved between 2006 and 2020. A unit increase between 2015 and 2020 in the human development score predicted a -4.7 pp (95% CI: - 5.7, -3.6) change in stunting, yet stunting declined by just -0.3 pp. Conclusions: Population-level reductions in child undernutrition have stalled and the rise in child overweight has accelerated between 2015 and 2020 relative to the 10 years preceding this period.


Introduction
Since the rst cross-national studies of child growth, India has consistently ranked among the countries with highest levels of growth faltering (1)(2)(3)(4). India's pernicious child undernutrition problem has been attributed to a complex interaction among structural factors such as inadequate food systems, poor water, sanitation, and hygiene infrastructure (WASH), and household poverty, which ultimately determine individual dietary intake and morbidities that impact child nutrition (5,6). The period from 2005 onwards witnessed the launch of several ambitious national level development programs targeting many of these structural factors through income, education, sanitation, and direct nutrition campaigns. Whether and how this dynamic development policy landscape has impacted trends in child undernutrition-and child overweight-over the past 15 years is not clear.

Methods
We analysed aggregate data from three cross-sectional surveys conducted over 15 years in 22 states and 341 districts in India. We tracked changes in child undernutrition, overweight and household development indicators. Some of the indicators listed are used in monitoring India's progress towards achieving the Sustainable Development Goals (9). The study was determined to be not human subjects research by the Emory University Institutional Review Board.

Data Sources
Nutrition and programming indicators are from state-and district-level data released by the National

Child Growth Outcomes
State-and district-level prevalence of child growth indicators were treated as the outcomes. All outcomes are expressed as the percentage of children under the age of 5 years exhibiting the indicator. Stunting was de ned as height-for-age z-score [HAZ] < -2 standard deviations [SD] from the median based on World Health Organization (WHO) growth standards (13). Underweight was de ned as weight-for-age [WAZ] < -2SD (13). Wasting was de ned as weight-for-height [WHZ] < -2 SD (13). Overweight or obesity in children was de ned as weight-for-height > 2 SD (13). Additional information on measurement protocol and quality control is provided in Supplementary Note 2.

Human Development Indicators
We focus on two domains of human development: human capital and standard of living. Human capital was measured by four indicators (% women who are literate; % of women with 10 or more years of schooling, sex ratio at birth, and women aged 20-24 years who were unmarried at age 18), and standard of living was measured by ve indicators (households with improved sanitation, improved drinking water, electricity, clean cooking fuel, or covered by a health insurance scheme).

Statistical Analysis
We conducted an ecological analysis of child malnutrition and development indicators for 22 states and 297 districts (2015 boundaries) using published estimates at the state and district levels, respectively. The mean prevalence and standard deviation of child malnutrition was weighted by under-ve population, and the mean prevalence and standard deviation of development indicators was weighted by the total population. Both sets of weights were normalized to the number of units (i.e., the sum of state weights was 22 and the sum of district weights was 297). We estimated annualized state-level change between surveys as average absolute change in percentage points per year (pp/y). We tested for difference in state-level annualized change between 2006 to 2015 and 2015 to 2019 using a paired Wilcoxon signed-rank test. We also estimated district-level changes in growth indicators between 2015 and 2019.
In addition, we examined expected changes in child growth indicators from 2015 to 2019 based on changes in the set of human development indicators by estimating a two-way Blinder-Oaxaca decomposition (unweighted, bivariate). In the bivariate decomposition, the expected change re ects the change in the child growth indicator that is predicted ("explained") by the human development indicator. The difference between the expected and observed changes re ect the change in the nutrition outcome that was not predicted ("unexplained") by the human development indicator. Our primary measure of human development was a weighted composite score of nine indicators of district-level human development that was derived using principal component analysis of these indicators. In addition, we report the decomposition ndings for each of the human development indicators individually.
We conducted a sensitivity analysis to examine the impact of sampling uncertainty on annualized changes of growth and development indicators across time. To do so, we computed the standard errors of state-level estimates of nutrition and development indicators from an individual-level analysis of NFHS-3 and NFHS-4; individual-level data were not available for NFHS-5 at the time of analysis. Because the sampling frame, survey design, and sample size were similar in NFHS-4 and NFHS-5, we applied the NFHS-4 standard errors to NFHS-5 point prevalence to generate estimates of con dence intervals for the annualized change from 2015 to 2019 (Supplementary Tables S1 and S2) .
All analysis was conducted using R 3.6.1 and Stata 13.0 statistical software.

Child growth outcomes
From 2006 to 2020, the population-weighted prevalence of stunting declined from 46.9-36.0% (10.9 pp reduction) and that of underweight declined from 41.4-34.3% (7.1 pp reduction) ( Table 1). In the same period, the prevalence of wasting rose from 18.9-21.7% (2.8 pp increase) and that of overweight rose from 1.6-3.8% (2.2 pp increase). The annualized rates of change in stunting (-0.3 pp/y versus − 1.1 pp/y; p < 0.01) and underweight (-0.2 pp/y versus − 0.7 pp/y; p < 0.01) were lower in 2015 to 2019 compared to those from 2006 to 2015, respectively ( Supplementary Fig. 2). The annualized rate of change in overweight was higher from 2015 to 2019 compared with 2006 to 2015 (0.03 pp/y versus + 0.5 pp/y, respectively). There was no statistically signi cant difference in the annual rate of change in wasting across the two reference periods. Consistent with the national pattern, we observed slowing reductions in stunting and underweight and faster increases in overweight in rural settings. In urban settings, we observed positive rates of change for stunting, underweight and overweight between 2015 and 2019. Relative changes are presented in Supplementary Table S3.  154, 126, and 137 districts experienced some reductions in stunting, wasting, and underweight, respectively. Far fewer -49, 32, and 28 -experienced reductions of 7 pp or more (the rate needed needed to achieve targets) in stunting, wasting, and underweight. Nearly all districts (n = 223, 75%) experienced an increase in overweight. We provide the uncertainty estimates for state and district prevalence of child malnutrition and development indicators in Supplementary Files 2 and 3. Analyses accounting for sampling uncertainty of estimates of child malnutrition for each survey round into the estimates of annualized change (Supplementary Table S1) were consistent with state-level paired differences in annualized change reported in Table 1 Table S2). However, paired tests of annualized changes at state-level were different only among 3 out of these 5 indicators (Table 2), and for 5 out of 9 indicators overall suggesting that there was heterogeneity across state-level annualized changes between the two periods. The largest absolute annualized changes from 2015-19 were observed in the percentage of households with improved sanitation facilities (4.3 pp/y) and households using clean fuel for cooking (4.8 pp/y).    Child stunting was de ned as height-for-age < -2SD, child wasting was de ned as weight-for-height < -2SD, child underweight was de ned as weight-for-age < -2SD, and child overweight was de ned as weight-for-height > 2SD. All values are coe cient ± bootstrapped standard error. Δ obs − exp : Difference between observed and expected change between 2015 (NFHS-4) and 2020 (NFHS-5). Contributions are modelled through two-way Blinder-Oaxaca decomposition with each indicator entered separately. Equal weights (w = 0.5) were applied for both survey rounds. The decomposition model gives each of the 297 districts analysed equal weight, and district grand mean will not sum to the state populationweighted mean.  Child stunting was de ned as height-for-age < -2SD, child wasting was de ned as weight-for-height < -2SD, child underweight was de ned as weight-for-age < -2SD, and child overweight was de ned as weight-for-height > 2SD. All values are coe cient ± bootstrapped standard error. Δ obs − exp : Difference between observed and expected change between 2015 (NFHS-4) and 2020 (NFHS-5). Contributions are modelled through two-way Blinder-Oaxaca decomposition with each indicator entered separately. Equal weights (w = 0.5) were applied for both survey rounds. The decomposition model gives each of the 297 districts analysed equal weight, and district grand mean will not sum to the state populationweighted mean.
With respect to individual development indicators, 7 of 9 predicted statistically signi cant reductions in one or more child undernutrition indicators and 6 of 9 predicted signi cant increases in overweight. In most cases, reductions in undernutrition were smaller than expected while the rise in overweight was larger than expected.

Discussion
As compared to the period between 2006 to 2015, the pace of reductions in child stunting and underweight across Indian states have slowed whereas the rise of child overweight has accelerated between 2015 and 2020. These changes disrupt favourable trends that included large reductions in undernutrition and stability of overweight observed in the preceding time period between 2006 to 2015. Our ndings also indicate that India is likely to fail in achieving the national targets set by the National Nutrition Mission (POSHAN Abhiyaan) for stunting, wasting, and underweight of ~ 23%, ~ 7%, and ~ 21%, respectively, by year 2022 (15,18). While there is no national target for the prevalence of child overweight in 2022, and the projected prevalence in 2022 (5.2%) is estimated to be close to the LMIC average in 2017 (6.0%), the rise in the proportion of overweight children is expected to offset any reductions in undernutrition (19)(20)(21). The total prevalence of unhealthy weight among children in India will likely remain a concern (22, 23). We also found that both undernutrition and overweight increased in urban areas. If current trends continue, the gap between child stunting in urban and rural areas will reduce by nearly 4 pp by 2022.
The stalled progress in undernutrition occurred against the backdrop of improvements in 7 of 9 development indicators from 2015 to 2020. Our ndings highlight a potentially worrisome disconnect between the apparent expansion of human development and improvements in child nutrition outcomes (24). Speci cally, the bivariate decomposition analysis predicted moderate to large declines in population-level undernutrition based on the rise in favourable development indicators such as women's literacy and households with electricity. Our analyses showed that these predicted reductions in undernutrition did not occur. Moreover, overweight is rising at a rate that is greater than what would be expected by improvements in human development observed in the past 5 years. The higher than expected prevalence in all four measures of unhealthy weight in children may imply that overall improvements in standard of living and human capital did not materialize in gains for child nutrition.
The observed stagnation in reductions in child undernutrition are consistent with data suggesting a reversal in the decline in annual infant mortality rates (25). Our results are also consistent with studies conducted in other contexts that have shown limited reductions in the prevalence of stunting despite implementation of successful nutrition-sensitive and nutrition-speci c interventions (26).
Child wasting, measured as weight-for-height, re ects current de cits in tissue and fat mass. Wasting has also demonstrated a small but statistically signi cant lagged association with linear growth status, measured as length-for-age especially for children aged 1 year or less, which is a marker of long-term undernutrition (27,28). Therefore, the apparent rise in wasting may be a predictor of future stunting at the population level. Prior literature suggests that the prevalence of stunting in India is not susceptible to seasonal variability while the prevalence of wasting tends to be 5-10% higher between June to December compared to January to May, due to seasonal variation in food availability and infectious disease (29). Given that phase 1 of NFHS-5 (which is all that is currently available) was conducted in most states from June to December 2019, it is possible that the observed prevalence of wasting is an overestimate.
Nevertheless, the unchanged national prevalence of stunting between rounds with one-fourth of districts experiencing a rise in prevalence of more than 5 percentage points is concerning.
Our analysis has several strengths. We used the latest available data to evaluate and benchmark ongoing efforts to reduce child malnutrition in India, and priority setting for global reductions in child undernutrition. The application of decomposition analysis permitted us to explore the hypothetical reductions that should have been achieved based on improvements in human development indicators in the population. However, our analyses must be interpreted in light of data limitations. In order to produce timely results that can inform relevant decisions including equitable allocation of funds to programs and states, we analysed aggregated data released at the state and district levels with varying precision of estimates as seen for NFHS-3 and NFHS-4 (Supplementary File 2, Supplementary File 3). This precluded more sophisticated hypothesis testing regarding the secular trends in child nutrition indicators. Such testing may be done once individual-level data become available in the future. Data on indicators from states with a high burden of undernutrition, such as Uttar Pradesh, Madhya Pradesh, Rajasthan, Chhattisgarh, Jharkhand and Uttarakhand, were not collected in Phase 1 of NFHS-5 and were excluded from the current analysis (30). This may skew predicted national level prevalence for 2022 as these high-burden states would be expected to demonstrate the most gains from a comprehensive intervention package as proposed by POSHAN Abhiyaan. We could not explore the association of changes in micronutrient supplementation and of provision of supplementary nutrition with changes in malnutrition outcomes due to data unavailability (31). Finally, our decomposition analysis described associations between human development indicators and child nutrition measured over the same time period. In doing so, we were not able to evaluate a lag between human development and improvements in child undernutrition.
A recent analysis of ve LMICs that achieved substantive reductions in child stunting suggest a combination of direct (nutrition-speci c) and indirect (nutrition-sensitive) health and nutrition interventions (32,33). Our ndings underscore the importance of targeted monitoring of nutrition outcomes, and support of nutrition sensitive and speci c interventions, to address the high and persistent burden of child undernutrition (34,35). Attention to child nutrition will be even more important in the coming years in light of the unprecedented economic and psychosocial strain of the ongoing COVID-19 pandemic (36). Projections show that in the absence of additional social safety net programs to tackle mounting food insecurity and nancial distress incurred through theCOVID-19 related economic shut down, there may be a doubling of the number of wasted and stunted children in South Asia alone by 2022 (37). Given that parts of India which experience the highest burden of child undernutrition also tend to be the most vulnerable to loss of life from COVID-19, it is important to focus on inter-ministerial efforts to curb the pandemic while at the same time avoiding reversal of gains made in alleviation of undernutrition (38).
In summary, the observed slowing progress in measures of child undernutrition warrants further exploration when considered against apparent indicators of progress. Further investigation is prudent to determine whether inequitable human development across segments of the population, or broader social and economic shocks (such as demonetization) are driving the stalling rates of reductions in undernutrition. Furthermore, investigation of outcomes beyond anthropometric growth, such as cognition, would provide a more comprehensive picture of the impact of changes in socioeconomic conditions on child social and developmental outcomes. WAZ -Weight-for-age z-score; WHZ -Weight-for-Height z-score Declarations Ethics approval and consent to participate: Wasting; C: Underweight; D: Overweight Note: The designations employed and the presentation of the material on this map do not imply the expression of any opinion whatsoever on the part of Research Square concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. This map has been provided by the authors.