Development of a Research Child Growth Reference and Its Comparison With the Current International Growth Reference FREE

GROWTH REFERENCES are used widely in the developed and developing worlds, both as a clinical tool to monitor growthin individual children and as a public health indicator to compare nutritional status over time or among different populations.^1- 2 It is widely accepted that an international reference is useful, since the growth in height and weight of well-fed, healthy children younger than 5 years from different ethnic backgrounds and different continents is reasonably similar.^3- 5 In 1978, the World Health Organization adopted the growth reference developed by the National Center for Health Statistics based on the US children as an international growth reference both for monitoring individual child growth and for assessing the nutritional status of populations.² This international growth reference has served a valuable purpose by providing a common basis for the analysis of growth data.

Although the current National Center for Health Statistics and World Health Organization international growth reference continues to serve a unique and valuable role, it has some important technical limitations. The current international reference was developed from 2 data sources. For children younger than 36 months, the length-for-age, weight-for-age, and weight-for-length references were based on data from a longitudinal growth study from 1929 to 1975 of the Fels Research Institute.^6- 8 For children aged 2 to 18 years, the height-for-age, weight-for-age, and weight-for-height references were based on 3 nationally representative surveys: National Health Examination Survey cycle II for ages 6 to 11 years (1963-1965), National Health Examination Survey cycle III for ages 12 to 17 years (1966-1970), and the first National Health and Nutrition Examination Survey (NHANES I) for ages 2 to 17 years (1971-1974).^6- 8

In recent years, several studies have shown that the current international reference based on the Fels sample (<2 years of age) and the part based on the national sample (≥2 years of age) are not comparable. When the 2 curves are compared, there is a clear disjunction of the height for age and weight for height at 24 months of age.^8- 11 Also, the weight-for-age curve has a disjunction at 24 months for upper centiles. These disjunctions make it difficult to compare the growth status of children younger than 2 years with that of those older than 2 years. In addition, several studies have found that the infancy portion of the current international reference, which is based on the Fels sample, does not fit with the growth pattern of infants regardless of whether they were breast-fed or formula-fed.^12- 15

Because of the limitations of the current international growth reference, a new growth reference was developed for research purposes by using a subset of growth data from the Centers for Disease Control and Prevention (CDC) Pediatric Nutrition Surveillance System (PedNSS). The main advantages of the new research reference are that it is based on a large sample (162163 observations) and that all the data are drawn from the same data source. It thus overcomes many of the irregularities seen in the current international growth reference, especially the discontinuity or disjunction in the growth curves at 24 months of age.^8- 11

In this article, we describe the characteristics of the research growth reference and the methods used to construct it. We then compare its characteristics and performance with those of the current international growth reference. The research reference provides a relatively stable and consistent basis for evaluating and comparing growth characteristics of children in research studies. It therefore should help investigators better understand the biologic characteristics of child growth. The research reference may also be useful in evaluating further the limitations of the current international growth reference as well as evaluating future modifications of growth references.

The NHANES I and NHANES II were conducted by the US National Center for Health Statistics to provide representative data from samples of the civilian, noninstitutionalized US population.^16- 17 The NHANES I examined persons aged 1 to 74 years during 1971 through 1974¹⁶; NHANES II examined persons aged 6 months to 74 years during 1976 through 1980.¹⁷ The problem with using NHANES I and II to create a national reference is that no children younger than 6 months were included in the data sets. Furthermore, the sample size for children aged 6 to 11 months is rather small (only 356 infants).

The CDC PedNSS monitors the general health and nutritional characteristics of low-income US children who participate in publicly funded health and nutrition programs. It provides a framework for tabulating and interpreting state-specific information on the nutritional characteristics of low-income children.^18- 20 Data for the majority of the infants and children monitored by the PedNSS come from clinic service records of the Special Supplemental Nutrition Program for the Women, Infants, and Children Program. This program was initiated in 1972 and is administered by the Food and Nutrition Service of the US Department of Agriculture.^21- 22 Since its inception, the number of states participating in the PedNSS has increased from 5 in 1973 to 38 states, plus the District of Columbia, Puerto Rico, and seven Indian reservations, in 1995.

According to the protocols used by the Women, Infants, and Children Program, height or length is measured to the nearest 1/8 in or 0.1 cm. A measuring board is used to measure the child's recumbent length if the child is younger than 24 months; a standing height is measured for children older than 24 months. Weight is measured to the nearest 1/4 lb or 0.1 kg by means of a pediatric scale or other beam balance scale. All the height and weight measurements require trained public health nurses, nutritionists, or dietitians. Also, the protocols require that 2 measurements of length or height of a child agree within 1/4 in, and 2 respective readings in weight agree within 1/4 lb.^20,23 All records are entered onto a standardized paper form or onto an automated computer system in the clinics. Once the records are computerized at the state level, they are transferred to the Centers for Disease Control and Prevention for inclusion in the PedNSS database.

A drawback of using the PedNSS database for creating a reference is that the data are not nationally representative. Also, since the data were collected in public health clinics, there is no guarantee that the protocols for measurement techniques were always followed, and thus the accuracy of the measurements may vary. To counteract these drawbacks, we based the new reference on a subset of the PedNSS data from 1975 to 1995 for which we chose only clinics in which the growth of children matched closely with those of national samples for ages 12 to 59 months. Clinics submitting data to the CDC PedNSS were selected only if they met 3 criteria: First, the mean height and weight of the children in the clinic's population were within ±0.5 cm or kg of the mean from the combined NHANES I and II for every 6-month interval from 12 to 59 months of age. Second, the selected clinics' population had to have an SD within ±0.2 cm or kg of the SD of the combined NHANES I and II. Third, the skewness in weight distribution of the selected clinics' population had to be within ±0.3 of the skewness in weight of the combined NHANES I and II. We assumed that the growth of infants aged 0 to 11 months in these clinics was reflective of national patterns.

The final data set for this analysis included 162163 observations coming from 52 clinics in 18 states (Alabama, Connecticut, Florida, Georgia, Illinois, Indiana, Iowa, Kansas, Kentucky, Massachusetts, Minnesota, New Hampshire, North Carolina, North Dakota, New Jersey, New York, Pennsylvania, and Washington). The sample size for each single month of age in infancy exceeds 2000. Both the height and the weight curves from the PedNSS closely match the NHANES curves for children 12 to 59 months of age (Figure 1). Even at 6 to 11 months of age, the curves match well also, indicating that the assumption of similar growth in infancy is reasonable. Also, as Table 1 shows, the 2 data sets were fairly similar, although the PedNSS data set had a higher proportion of younger children.

Of the many statistical-smoothing techniques available for reference construction,^24- 44 we chose the LMS method for this research reference construction. Cole^35- 36 developed the LMS method in 1988 to generate smoothed age- and/or height-specific percentile curves. It assumes that data can often be normalized by means of a power transformation, to stretch 1 tail of the distribution and shrink the other, thereby removing any skewness.^35- 36 The optimal power of a Box-Cox transformation,⁴⁵ which obtains approximate normality, is calculated for each of a series of age groups and the trend is summarized by a smooth (L) curve. Trends in the mean (M) and coefficient of variation (S) are similarly smoothed. The resulting L, M, and S curves contain the information to draw any centile curve and to convert measurements into exact standardized scores or z scores.^35- 36

The z score system expresses anthropometric values as a number of SDs or z scores below or above the reference mean or median value. Because the z score scale is linear, summary statistics such as means, SDs, and SEs can be computed from z score values. Also, the z score summary statistics are helpful for grouping growth data by age and sex. The summary statistics can be compared with the reference, which has an expected mean z score of 0 and an SD of 1.0 for all the normalized growth indices.^9,14

In computing the height-for-age and weight-for-age references, we grouped the reference data by sex into half-month intervals after birth to 59.9 months of age. For example, we grouped the ages at 0 to 0.4 months, at 0.5 to 0.9 months, etc. Similarly, for the weight-for-height reference, we grouped the data by centimeter of height. Within each age (or height) group, we computed the skewness (L), median (M), and coefficient of variation (S), by means of formulas supplied by Cole.³⁶ We then used regression models to smooth the L, M, and S values. For each height-for-age, weight-for-age, and weight-for-height, we used linear regression to smooth the L curves according to Cole's suggestion,³⁶ fifth-degree polynomial regression to smooth the M curves, and cubic polynomials to smooth the S curves. The 3 quantities provide any required centiles (C) by using the following equations³⁶: C=M×(1+L×S×Z_α)^1/L, where Z_α is the normal equivalent deviate corresponding to the centile. The z score (Z) of an individual measurement can be computed as Z=[(measurement/M)^L−1]/(L×S).

To eliminate any discrepancy caused by length vs stature measurements for each reference across age, we adjusted all the recumbent lengths to standing stature by subtracting 0.4 cm. This 0.4-cm adjustment was based on the difference between the length and stature measurements in the NHANES II data for children aged 24 to 35 months for whom both measurements were taken. After the smoothing procedure for the final reference data, we converted the stature to length by adding 0.4 cm. The measurement protocols used to create the reference are therefore equivalent to standard procedures developed to measure children in clinic and field settings.¹⁴

We used 3 data sets to test the research reference and to compare it with the current international reference: combined NHANES I and II data, PedNSS 1995 data, and the data set from the Davis Area Research on Lactation, Infant Nutrition, and Growth (DARLING) study. The PedNSS 1995 data were from 38 states and the District of Columbia, Puerto Rico, and 7 Indian reservations; there were 7017585 records for children younger than 5 years. The DARLING study, conducted in California from 1986 to 1991, included a cohort of 73 infants with a total 665 measurements who had been breast-fed for at least 12 months as well as 46 formula-fed infants with 499 measurements.⁴⁶

Figure 2 compares our research growth reference for boys' height- (length-)for-age from birth to 59 months with the current international reference. For length-for-age (0-23 months), the median of our reference is consistently lower than that of the current international reference, but the SD is larger than the current international one, especially in infancy. For 24 to 33 months, the height-for-age median is higher than the current international reference, and for 34 to 59 months it is lower. The new curve has a small disjunction at 24 months to account explicitly for the difference in measured recumbent length and standing stature. However, it is less marked than the disjunction in the current international reference. Girls' length-for-age and height-for-age z score curves (not shown) follow the same pattern as those of boys, but the difference between the 2 references is relatively smaller.

The difference between the medians for the 2 weight-for-age references is relatively small, especially for children younger than 2 years (Figure 3). The 2 references are also similar below the median. Above the median, however, our reference accounts for a greater degree of skewness in the weight distributions. The upper z score curves (except +1 SD curve) of the new reference are slightly higher than those of the current international reference. Also, the current international reference shows an obvious discrepancy at 24 months at upper z score curves, which is consequent to the combination of 2 data sources. Our new curve has no disjunction.

The difference between the 2 references in weight-for-height (length) for boys is shown in Figure 4. The new reference starts at 45 cm rather than 49 cm because the average length of younger infants in developing countries is sometimes less than 49 cm. For weight-for-length (45-84 cm), the new reference curves are consistently higher at the upper z score curves than those of the current international reference for both sexes. For weight-for-height (85-115 cm), the new reference curves are higher at the lower z scores and slightly lower at the upper z scores compared with those of the current international reference for both sexes. The new curve has a small disjunction at 85 cm to account explicitly for the difference in measured recumbent length and standing stature. However, it is less marked than the disjunction in the current international reference.

We transferred the smoothed L, M, and S curve parameter estimates into statistical analysis system (SAS) codes for height- (length-)for-age, weight-for-age, and weight-for-height (length) for children from birth to 59 months (Table 2). Using these SAS codes, researchers can easily assign z scores and/or percentiles to each child in their studies. In addition, the equations given in the "Subjects and Methods" section of this article can be used to generate percentile or z score growth charts.

To compare the research growth reference and the current international reference, we used each reference to assign z scores to the heights and weights of children in 3 data sets: the combined NHANES I and II data, the PedNSS 1995 data, and the DARLING study data.

Although the combined NHANES I and II data were used in selecting the PedNSS clinics for this reference, they can also be used to validate the research reference. Figure 5 summarizes the mean height-for-age, weight-for-age, and weight-for-height for the combined NHANES I and II data. Mean z scores based on the research reference show a pattern that more consistently adheres to the expected 0 z score line. The fact that the combined NHANES I and II data do not adhere to the current international reference is surprising, given that it was created on the basis of NHANES I data and there are no trends in height or weight between NHANES I and NHANES II.

The second comparison of the current international reference and the research reference was based on the entire year of data of the CDC PedNSS collected in 1995 (Figure 6). The large sample size in PedNSS 1995 data allows us to show mean z score by month of age and thus better demonstrates the marked disjunctions of height-for-age and weight-for-height at 24 months of age for the international reference. Growth status based on the current international reference gives a lower length-for-age z score before 24 months of age but shifts dramatically upward at 24 months. This same shift is also evident for weight-for-height; mean z scores based on the current international reference are higher than our new reference means before 24 months of age. All 3 mean z score curves based on the new reference are more stable across age than are the curves based on the current international reference.

The infants of the DARLING study were taller and heavier at birth compared with either reference (Figure 7). In the comparison of length-for-age, the research reference consistently resulted in a higher mean z score than the current international reference. Interestingly, there was an important difference in how each reference characterized the mean weight-for-age z score of the 2 infant feeding groups. The current international reference gave a higher z score at early months but became lower by 8 months. This crossover behavior occurred for both breast-fed and formula-fed infants, giving the impression of growth faltering, especially among the breast-fed infants. The absolute decline for breast-fed infants in mean weight-for-age z score from birth to 12 months of age was about 1.2 z scores based on the current international reference, and about 0.7 z scores assigned from the research reference. This finding indicates that neither reference adequately reflects the weight pattern of breast-fed infants, although the research reference is better. In the case of formula-fed infants, the weight-for-age z score from the current international reference implied growth faltering with a net reduction of 0.6 z scores, but the research reference shows no growth faltering. This finding indicates that the current international reference generates an impression of growth faltering among formula-fed infants, whereas the research reference does not.

We also observed a growth faltering in weight-for-length among breast-fed infants, although it was more pronounced for the current international reference (Figure 7). For formula-fed infants, the z scores assigned by both references were relatively stable, even though the research reference gave a slightly increasing trend.

A reference is defined as a tool for grouping and analyzing data and provides a common basis for comparing populations. In practice, however, researchers use a reference as a standard and, in most cases, for judging the nutritional status of individuals and populations. Therefore, it would be desirable to use reference curves that properly describe the normal pattern of growth.^14,47

There is currently no single source of data, other than PedNSS, that describes the growth of the US children from birth to 5 years of age. The PedNSS data have 2 main drawbacks: (1) they represent only low-income children and (2) measurement techniques and equipment calibration are not as tightly controlled as in a survey or special study, and thus the accuracy of measurements may vary. However, by choosing only clinics whose distributions matched the NHANES I and II, we approximated nationally representative data (Figure 1). This similarity supports the validity of the data.

It has been suggested that the current international reference is inadequate for capturing the growth of breast-fed infants.^12- 15 In fact, we have shown that it is also inadequate for capturing the growth of formula-fed infants. The formula-fed infants in the DARLING study showed a faltering in the weight-for-age z score after 6 months when interpreted with the current international reference. However, when the research reference was applied, the mean weight-for-age z score exhibited a stable pattern without the faltering seen during infancy. The faltering with the current international reference is probably related to problems with the Fels data. The Fels data do not accurately describe the characteristics of infant growth, perhaps because there were too few data observations during infancy (the children were measured at birth and at 1, 3, 6, 9, and 12 months) to fully capture the rapid growth pattern.

Previous studies show that breast-fed children have a rapid decline in the mean weight-for-age compared with formula-fed children during infancy.^13- 15 Our analysis from the DARLING study observed the same growth pattern regardless of which reference was used. However, the current international reference shows a more dramatic decline in the mean weight-for-age z score, giving the impression that breast-fed infants exhibit greater "growth faltering" than formula-fed infants (Figure 7).

The LMS method has been used in Europe to construct several anthropometric references.^48- 51 It has several advantages: the skewness can be incorporated while retaining many of the advantages of a normal distribution; it can generate any required centiles, not just the conventional set of 7 (fifth, 10th, 25th, median, 75th, 90th, and 95th); and individual measurements of height and weight can be accurately and directly converted to z scores or centiles.^35- 36

The research childhood growth reference was based on a single data source with a large sample. The data were chosen to match the nationally representative US data sets. It eliminates the marked disjunction of height-for-age and weight-for-height z score at the 24-month age point of the current international reference, and it more accurately characterizes growth of formula-fed infants and better captures growth of breast-fed infants.

The principal purpose of making this research reference available is to provide researchers who wish to study changes of growth in infants and young children across age a more accurate tool to characterize such changes. This research reference is not intended to replace the current international reference for routine clinical-based growth monitoring or international comparison. An international working group under the auspices of the World Health Organization is working toward a new international reference to substitute for the current international reference adopted from the United States. This research reference will be useful for the evaluation of the new international reference. The methods and the data source for the research reference development can potentially be useful for the future development of a new US national reference. The recently completed NHANES III survey will be able to provide a more expanded national sample for older infants and children.⁵² However, there will still be a gap from birth to 3 months when no national representative data are available, and the methods developed in this article may be useful to fill this gap.

Accepted for publication November 12, 1997.

We extend thanks to all state nutrition programs participating in the CDC PedNSS, especially those that provided the data for this study. We also thank Ellen Borland and Jimmy Simmons for providing data management support and Kelley Scanlon, PhD, and Bettylou Sherry, PhD, for reviewing the document. We are grateful to Kathryn Dewey, PhD, for providing the data from the DARLING study and for her helpful comments. We are also indebted to Tim Cole, PhD, for his methodological advice.

Editor's Note: Periodically, evaluation tools also need to be evaluated and updated, if indicated. After all, the only sure things in life are death, a polite IRS, and change.—Catherine D. DeAngelis, MD

Reprints: Zuguo Mei, MD, MPH, Centers for Disease Control and Prevention, Mailstop K-25, 4770 Buford Hwy, Atlanta, GA 30341-3724 (e-mail: zam0@.cdc.gov).