Growth Reference Charts and the Nutritional Status of Indian

Preparing to load PDF file. please wait...

0 of 0
Growth Reference Charts and the Nutritional Status of Indian

Transcript Of Growth Reference Charts and the Nutritional Status of Indian

Growth Reference Charts and the Nutritional Status of Indian Children
Alessandro Tarozzi Duke University July 2008∗
Abstract We evaluate the growth performance of Indian children of age 0 to 3 using data from the 199899 National Family and Health Survey, making use of the new child growth standards developed by the World Health Organization’ Multicentre Growth Reference Study. We find that the new charts lead to an increase of 4.2 million in the estimated number of stunted children, and an increase of 2.3 million in the estimated number of wasted children. The estimated number of underweight children decreases instead by 2.1 million. We also use data on ethnic Indians living in the United Kingdom to provide evidence on the height genetic potential of Indians. We find that children of Indian ethnicity who live in the UK have anthropometric outcomes comparable to those in commonly used growth standards and that the height of ethnic South Asian in the sample is negatively related with the amount of time spent outside the United Kingdom. JEL: O1, I1 Key words: Child Nutrition, India, Growth Charts, Height, Weight.
∗I am grateful to Angus Deaton and especially to Jean Dr`eze , John Komlos (the Editor) and two anonymous referees for useful comments and suggestions and to James Nazroo for making me aware of the HSE ethnic boost dataset. I thank Macro International Inc. for granting access to the NFHS data set, and the UK Data Archive for granting access to the 2000 Health Survey for England. I am solely responsible for all errors and omissions. Address: Dept of Economics, Duke University, Social Sciences Building, PO Box 90097, Durham, NC 27708, [email protected]

1 Introduction
Child nutritional status is an essential component of a country’s overall human development. There is a growing consensus that poor nutritional status during childhood (and in utero) can have longlasting scarring consequences into adulthood, both in terms of health and mortality, and in terms of other measures of human capital such as schooling and productivity (Maluccio et al. 2005, Behrman et al. 2006, Glewwe and Miguel 2007). In the last three decades, countless studies have measured child nutritional status in developing countries using as reference growth charts introduced in 1977, estimated from a population of U.S. children by the Center for Disease Control and prevention and the National Center for Health Statistics. The use of such standards had been recommended by the World Health Organization (WHO) for international use since the late 1970s, based on studies suggesting that the growth patterns of healthy children are similar across different ethnic groups (Habicht et al. 1974, Waterlow et al. 1977, WHO Working Group 1986, Dibley et al. 1987a, Dibley et al. 1987b, Martorell and Habicht 1986, Gorstein et al. 1994). Agarwal et al. 1991 and Bhandari et al. 2002 specifically evaluate the adequacy of the U.S. charts for India, and find that the growth performance of Indian children of age five and below living in affluent families is broadly comparable to that of U.S. children, although not in all the regions analyzed.1
Estimates of the extent of child malnutrition in India invariably describe a population whose growth performance is very poor (Svedberg 2000, Borooah 2003, Pande 2003, Borooah 2005, Nandy et al. 2005, Svedberg 2006, Tarozzi and Mahajan 2007). Given that this country account for approximately one sixth of the world population, the extent of malnutrition in India is clearly important for anyone concerned with the worldwide magnitude of the problem.
In 1997, the WHO initiated the Multicentre Growth Reference Study (MGRS), with the purpose of constructing new standards for normal early childhood growth under ideal environmental conditions. The charts would be used to assess the nutritional status of children under five years of age regardless of ethnicity, socioeconomic status and feeding practice. The new charts resulting from this effort have been estimated with data from 8440 healthy breastfed infants and young children from Brazil, Ghana, India, Norway, Oman and the United States (World Health Organization
1Agarwal et al. 1991 study growth data from 1429 boys and 1206 girls aged 0-72 months from affluent sections of Bangalore, Calcutta, Delhi, Ludhiana and Varanasi. They find that child weight and height in Delhi and especially Ludhiana nearly corresponded to those in the international growth standards, while remaining at lower levels in the other three cities. Bhandari et al. 2002 use a sample of 341 children aged 12-23 months born from affluent parents in south Delhi. They find mean z-scores for weight-for-age, length-for-age and weight-for-length equal to −0.45, −0.28 and −0.32 respectively.

2006).2 New charts have also recently been introduced for the United States, to overcome some concerns with the previous references (Kuczmarski et al. 2000). Throughout the paper, we refer to the three different sets of reference charts as follows: CDC-WHO77 will indicate the US-based charts whose use was recommended until recently by the World Health Organization; WHO-2006 will denote the revised WHO charts and CDC-2000 will indicate the revised US charts.
The first purpose of this paper is to use data from the 1998-99 National Family and Health Survey, together with the revised WHO-2006 growth charts, to re-evaluate the height and weight of Indian children under 3 years of age.3 We find that, relative to estimates obtained using CDCWHO77 references, the new charts lead to an increase of 4.2 million in the estimated number of stunted children (from 42 to 48%), and an increase of 2.3 million in the estimated number of wasted children (from 15 to 18%). The largest increase occurs for the proportion of stunted boys. The estimated number of underweight children decreases instead by 2.1 million (from 44 to 41%), almost exclusively through the impact of the new charts on the extent of girl underweight.
Our second purpose is to contribute further evidence on the adequacy of international growth charts for the Indian population by using data from individuals of Indian ethnicity living in the United Kingdom. With similar purposes, other studies have used information on the nutritional status of children of different ethnicity living in rich countries. Duggan and Harbottle (1996) find anthropometric measurements comparable to those of the CDC-WHO77 references for a sample of 169 healthy children aged between 4 and 40 months of Pakistani and Bangladeshi origin living in Sheffield (UK). Yip et al. (1992) and Mei et al. (1998) study the growth patterns of children of Asian refugees in the United States, and conclude that factors such as diet and health services, rather than genetic factors, explain most of the inter-country variation in child anthropometric indices. Nguyen et al. (2004) show that growth patterns of Vietnamese infants who reside in Australia are close to the U.S. reference charts, despite their parents’ anthropometric indices being below reference. Gjerdingen et al. (1996) find, instead, that a sample of 0 to 5-year old children of Hmong ethnicity living in the United States were slightly heavier then the CDC-WHO77 reference mean for the first months after birth, and shorter thereafter. Smith et al. (2003) show that Maya-American children are taller, on average, than their counterparts in Guatemala, but are also more likely to be overweight.
Using data from the 1999-2000 Health Survey for England, we find that children of Indian
2The new charts can be found together with the corresponding documentation at childgrowth/standards/en/ap.
3The choice of the age threshold is imposed by data constraints, see Section 2.

ethnicity living in the United Kingdom have anthropometric outcomes very close to those in the WHO-2006 standards. We also find that, even after controlling for income per head, height of ethnic Indians in the sample is negatively associated with the amount of time spent outside the UK. Overall, these findings are consistent with the argument that socio-economic and epidemiological factors, and not genetic differences, are the major causes of the relatively poor growth performance of the Indian population. At the same time, these results should be interpreted with caution, because ethnic Indians who move to the UK may not be representative of the same genetic pool as that of other Indians who do not migrate.
2 Data
To evaluate the nutritional status of Indian children, we use data from the second round of the National Family and Health Survey (NFHS), completed in India between November 1998 and December 1999.4 The NFHS is one of the Demographic and Health Surveys carried out in many developing countries—adopting a largely standardized questionnaire—with the primary goal of collecting information on health, fertility and other family issues from ever married women of fertility age. The survey contains reports from approximately 90,000 women of age 15-49, and is constructed to be a representative sample for the whole of India, both rural and urban. Table 1 reports selected summary statistics. Anthropometric measurements, including height and weight, were taken for children up to three years of age. The number of women with at least one child under the age of 3 (U3) in the sample is 20,493 for rural and 7,293 for urban areas. Schooling is overall very low, especially in rural areas, where two-thirds of women are illiterate. Only about one fifth of women are either employed or self-employed. Selected indicators also indicate poor health. The hemoglobin level is below 12 grams per deciliter of blood in more than half of the sample. This threshold is sometimes used to indicate iron deficiency (Thomas et al. 2006). The body mass index (BMI) is below 18.5 for 43% of women in rural areas and 31% in urban areas.5
The second source of data is the 1999-2000 round of the Health Survey for England (HSE). This survey is carried out by the Joint Health Surveys Unit of the National Centre for Social Research and the Department of Epidemiology and Public Health at University College, London.
4The data and the documentation can be accessed at 5BMI is calculated as weight in kilograms divided by the squared of height measured in centimeters. Values of BMI below 18.5 have been often associated with higher mortality for men, while for women even higher thresholds have been often used (Waaler 1984, Payne 1992)

The 1999-2000 round was specifically formulated to also allow the analysis of health among minority ethnic groups.6 A General Population Sample was selected drawing about 6,500 addresses from the Postcode Address File in 312 postal sectors. In each sampled household, information were collected on all adults and children, unless the household included more than two children, in which case only two selected at random were surveyed. A special (Ethnic Boost) sample of about 26,500 addresses was then selected within another 340 postal sectors with the specific scope of collecting detailed health information for households belonging to the most populous six minority ethnic groups: Black Caribbean, Indian, Pakistani, Bangladeshi, Chinese and Irish. About 37,600 addresses adjacent to those sampled as described above were also covered by focused enumeration. Interviews were attempted only with respondents from the specified minority ethnic groups. Among eligible informants at an address, a maximum of four adults and three children were selected for interview, at random if necessary. For each individual included in the survey, HSE also reports weight, height, age, ethnicity, household income, and country of birth. Unfortunately, weight and height were not collected for children below two years of age, so that comparisons of child growth outcomes between NFHS and HSE data is only possible for children of age two to three.

2.1 Measures of Child Nutritional Status: Old and New Growth Charts

The three most widely used indicators of child nutritional status are weight-for-age, height-for-

age, and weight-for-height. Height-for-age is the preferred measure of long-term nutritional status.

Because weight can change in a relatively short period of time as a consequence of changes in

nutritional intake and/or health status, weight-for-age and weight-for-height are better measures

of short term nutritional status. However, the former index does not distinguish between small but

well fed children and tall but thin ones, so that weight-for-height is usually the preferred indicator

for current nutritional status. One more advantage of weight-for-height is that—unlike the other

two indicators—it does not require exact knowledge of age in months, because the reference group

is a population of children of the same gender and height, irrespective of age (Gorstein et al. 1994).

This factor can pose a significant advantage in surveys such as the NFHS, where weight and height

are measured by well trained technicians, but age recording (in months) must usually rely on the

mother’s report. In populations where illiteracy is very common, child age is often reported with

error. In Figure 1, we show the distribution of reported child age in months in the NFHS, separately

6For detailed information on the data set, see Erens et al. 2000.

The data are available at http://www.


for different levels of mother’s schooling.7 The distribution of reported child age presents a clear pattern, although it becomes somewhat less visible when mother’s education increases. Note, however, that even among illiterate mothers there is no evidence of reports clustering at “focal ages” such as 0, 6, 12, or 18 months.
Let g denote the reference group a child is being compared to, and let xig denote height or weight for child i compared to a reference g. The most common approach to evaluate growth performance transforms the anthropometric indices into a z-score, calculated as (xig − x¯g)/sg, where x¯g and sg are respectively the mean (or median) and the standard deviation of the indicator for children within the same group in the benchmark population. In the CDC-WHO77 charts, the skewness in the distribution of the anthropometric indicators in the reference population was taken into account by using different standard deviations sg, depending on whether xig remained above or below x¯g. When the corresponding nutritional indicator in the reference population is approximately normally distributed, z-scores are very easy to interpret. For instance, if a boy’s weight-for-height z-score lies below −1.645 then his weight is below that of 95% of boys in the reference population with the same height. A child is usually identified as stunted if height-for-age z-score is below −2, and as underweight or wasted if the z-score for respectively weight-for-age or weight-for-height is below the same threshold.
For several years, the CDC-WHO77 charts have represented the most widely used reference to assess child nutritional status. These standards have been introduced in 1977, after being developed in the United States by the National Center for Health Statistics (NCHS) and the Center for Disease Control and prevention (CDC). Their use has became especially common after the WHO recommended their use for the evaluation of child growth worldwide (Waterlow et al. 1977, World Health Organization 1978, Dibley et al. 1987a). The CDC-WHO77 charts have also been adopted for the calculation of the z-scores included in the NFHS data used in this paper. The use of these charts, however, has never been unanimously accepted (especially for children aged 0-24 months) neither for use within the U.S., nor as an international standard. Detractors have called attention to several limitations, perhaps more importantly the sample used for their construction. Such sample only included Caucasian infants from predominantly middle-class families (and, for younger children, coming from a single community). Other shortcomings were the long time interval between successive measurements of the children in the sample, and the fact that a large number of measured infants were bottle-fed (Kuczmarski et al. 2000).
7In this survey, child age in month is calculated as the difference between interview date and day of birth. Birth dates were reported by the mother, and the child month of birth was further probed asking “what is his/her birthday”.

The same shortcomings that led to a revision of the charts for use within the U.S. also con-

tributed to argue for a revision of the charts used for international comparisons. Of particular

concern was the fact the growth of infants fed according to practices recommended by the WHO

appeared to remain below the standards set in the CDC-WHO77 standards, especially after 4-6

months of age (World Health Organization Working Group on Infant Growth 1995). While the

charts were constructed with a sample of bottle-fed infants, a practice which appears to maximize

growth within the first year of life, exclusive breast-feeding is recommended by the WHO in the

first 4-6 months of life, with a mix of breast-feeding and supplements recommended afterwards and

up to at least two years of age. These recommendations appear to be especially important in de-

veloping countries, because several studies document a negative association between breast-feeding

and infections (Hanson et al. 1994, L´opez-Alarc´on et al. 1997, M˚arild et al. 2004 and references

therein). As a result of the combined shortcomings of the CDC-WHO77 charts, the WHO initi-

ated the Multicentre Growth Reference Study (MGRS), which between 1997 and 2003 produced a

new set of references for international comparisons that has become recently available.8 The new

charts are based on healthy children from Brazil, Ghana, India, Norway, Oman and United States,

living under conditions likely to allow the full achievement of their genetic growth potential, whose

mothers did not smoke and followed the WHO recommended feeding practices.9

Unlike the old references, the new charts have been developed using an LMS model (Cole 1988,

Cole and Green 1992), which takes explicitly into account the skewness and non-normality of the

distribution of weight and height in the reference population.10 In this approach, the z-score for a

given anthropometric measure xig is calculated using mean and standard deviation not of the same measures in the reference group, but of a Box-Cox transformation of the measures.11 In this way,

the z-score for child i compared to a reference group g is calculated as:

(xig/Mg)Lg − 1

zig =



Lg Sg

where Lg is the “power” of the Box-Cox transformation and Mg and Sg are the mean and the

standard deviation of the transformed variable in the reference population. Hence, the new charts

provide the parameters Lg, Mg and Sg necessary for the calculation of the expression in (1). Such

8For comprehensive technical descriptions of the new charts, as well as for details on the rationale of their devel-
opment, see World Health Organization 2006, which also compares the revised charts to the CDC-WHO77 references
and the new CDC-2000 charts, which have replaced the previous reference within the United States. 9It should be noted that even the new charts have not been unanimously accepted (see, for instance, Klasen and
Moradi 2000). 10The same approach is used in the revised CDC-2000 references for the U.S. (Kuczmarski et al. 2000). 11See Box and Cox 1964, or Davidson and MacKinnon 1993, Ch. 14.


parameters are gender-age specific for the construction of height-for-age and weight-for-age z-scores, while they are gender-height specific for weight-for-height.
3 Child Nutritional Status in India with New and Old Growth Charts
We now move to the evaluation of the anthropometric performance of U3s in India using the alternative reference charts described in the previous section. Figures 2 to 4 show densities of z-scores for height-for-age, weight-for-height and weight-for-age, by gender and rural or urban residence. All densities are estimated nonparametrically using standard kernel estimators, choosing the bandwidth according to the robust criterion described in Silverman (1986) for approximately normal distributions, and using a biweight kernel. In each graph, we plot two densities. The continuous lines display the densities of z-scores calculated using the new WHO-2006 charts. The dashed lines represent the estimated densities of the z-scores included in the NFHS dataset (which use the CDC-WHO77 standards). In all cases, the WHO-2006 curves appear to have more mass on very low values than the CDC-WHO77 curves. At least in part, this is due to the truncation criterion used in the NFHS, where weight-for-height z-scores are truncated below −4 or above 6, while weight-for-age and height-for-age z-scores are truncated for absolute values above 6.12 While for height-for-age and weight-for-height the CDC-WHO77 lines remain below the WHO-2006 curves over most of the range until very near the −2 threshold, the weight-for-age densities cross around −4. As a consequence, it appears that the use of the new charts increases the estimated extent of stunting and wasting among Indian U3s, while it decreases the extent of underweight.
In Table 2, we report estimates of the fraction of stunted, wasted and underweight children using the same data, by gender and sector. We also report an estimate of the total number of Indian children in each gender-sector combination.13 To enhance the comparability between estimates obtained using different references, we also report results obtained with WHO-2006 charts but using the same truncation criterion used in NFHS.
12Figures which also include densities estimated using new reference charts with the same truncation rule as NFHS are available upon request from the author.
13These totals are calculated assuming a total Indian population at the end of the 1990s of one billion, and estimating the fraction in each specific demographic group using the data in NFHS. We estimate that 73.6% of Indians lived in rural areas and that the fraction of females in the population was .486 in urban areas, and .493 in rural areas. In urban India, children below three years of age accounted for 5.9% of the male population, and 5.8% of the female population. In rural areas, the two proportions were 7.4 and 7.1 percent respectively.

In most cases, the choice of standards affects the results considerably. For instance, when height-for-age z-scores are calculated using the new reference values, the fraction of stunted boys in urban areas increases by 10 percentage points, from 34 to 44%. Clearly, the truncation criteria adopted by the NFHS for the calculation of the z-scores only explain part of the difference with respect to the figures calculated using the new WHO-2006 charts. If we use the same truncation criterion used by the NFHS together with the WHO-2006 references, we estimate an increase in stunting for urban boys by 8 percentage points, from 34 to 42%. This corresponds to an increase of 644,000 (= 8.05 million ×0.08) in the estimated number of stunted boys. The fraction of stunted girls increases by 4 percentage points, from 37 to 41%, which corresponds to 300,000 girls (7.5 million ×0.04). In rural areas, where poverty and malnutrition are more widespread than in cities, the use of the new charts increases estimated stunting from .47 to .55 among boys (that is, an increase of 27.7 × 0.08 = 2.216 million children), and from .50 to .54 among girls (that is, by about 25.9 × 0.04 = 1.036 million children). Overall, the new charts lead approximately to a 4.2 million increase in the number of stunted children in India in 1998-1999.
To a lesser extent, the new charts also increase the estimated magnitude of wasting, which for both genders and in both sectors characterizes a much smaller fraction of children than stunting. This is a common finding in developing countries, where children’s low weight is usually partly “balanced” by low height. Overall, the figures in Table 2 show that the (truncated) z-scores calculated using the WHO-2006 charts lead to an increase of 2.3 million in the number of wasted children.14
As suggested by a visual inspection of the densities in Figure 4, the effect on underweight is opposite, although the change is concentrated on girls, and the effect on severe underweight is very limited. The fraction of underweight boys decreases from 37 to 36% in urban areas, and from 48 to 47% in rural areas. However, among girls the decline is from 40 to 34% in urban India and from 52 to 47% in rural areas. Overall, these figures imply a 2.1 million decline in underweight among Indian children aged 0-35 months.15
It is also worth noting what these comparisons imply for gender differences in z-scores. The existence of preference for sons over daughters is a well-documented reality in India, particularly in the North-West. Gender inequality has been documented for different outcomes, among which are sex-selective abortion, schooling, health and health care, and child mortality (Basu 1992, Dasgupta 1993, Murthi et al. 1996 and Dr`eze and Sen 2002). Some studies have found bias against girls in
14This figure is calculated as 8.05 × 0.02 + 7.5 × 0.03 + 27.7 × 0.04 + 25.9 × 0.03. 15This figure is calculated as 8.05 × 0.01 + 7.5 × 0.06 + 27.7 × 0.01 + 25.9 × 0.05.

nutrient intakes and nutritional status (Behrman 1988a, Behrman 1988b). However, the evidence about gender bias in nutritional status remains inconclusive, as discussed in Harriss (1995). More recent evidence is provided in Tarozzi and Mahajan (2007), who use the same NFHS data used in this paper together with the previous round, completed in 1992-93. They find that in 1992-93 the distribution of z-scores for height-for-age and weight-for-age did not point to systematically worse performances among girls. However, they also find a movement towards boy advantage in rural areas in North India.
The figures in Table 2 suggest that the choice of reference leads, in some cases, to a reversal of the gender differences in nutritional status. Specifically, the new charts eliminate the boy advantage in stunting and underweight. In both rural and urban areas, the difference between boy and girl stunting changes from −3% with CDC-WHO77 charts to +1% with the new standards. The boy versus girl difference in underweight switches from −3% in urban areas to +2%. In rural areas the shift is from −4% to zero. The extent of wasting remains instead essentially unchanged. These findings point to the importance of using the same set of reference charts when making statements about differences in gender bias in anthropometric outcomes across different countries or between different periods within the same area.
Estimates of the fraction of z-scores below −2 or below −3 in India, as calculated using the CDC-WHO77 and the WHO-2006 charts, are also reported in the WHO Global Database on Child Growth and Malnutrition (World Health Organization 2008). This database is periodically updated as new data become available, and includes summary measures of child anthropometric indices for India as well as for most other countries in the world (de Onis and Bl¨ossner 2003). For some data sets prior to 2006, results obtained with both growth charts are reported. The 1998-99 Indian NFHS is one of these data sets and the tabulations report separate estimates of the extent of stunting, wasting and underweight for different gender and age groups or by sector.16 However, a direct comparison of these figures between the two references is not offered. Also, the results are not disaggregated by gender and sector, so that a direct comparison with the figures in Table 2 is not possible. Overall, these tabulations show that using the CDC-WHO77 charts 43.6% of up to 3-year old boys were stunted, 15.9% were wasted, and 45% were underweight. Among girls in the same age range, 48.5 were stunted, 15.4% were wasted, and 46.4% were underweight.17 Using instead WHO-2006 charts, 51.2% of boys are stunted, 20.5% are wasted and 45% are underweight. Among
16The tabulations and references to the data sets they refer to can be found at database/countries/ind/en/.
17See, p. 21.