OBJECTIVE: The objective of this study was to create and validate new intrauterine weight, length, and head circumference growth curves using a contemporary, large, racially diverse US sample and compare with the Lubchenco curves.
METHODS: Data on 391 681 infants (Pediatrix Medical Group) aged 22 to 42 weeks at birth from 248 hospitals within 33 US states (1998–2006) for birth weight, length, head circumference, estimated gestational age, gender, and race were used. Separate subsamples were used to create and validate curves. Smoothed percentile curves (3rd to 97th) were created by the Lambda Mu Sigma (LMS) method. The validation sample was used to confirm representativeness of the curves. The new curves were compared with the Lubchenco curves.
RESULTS: Final sample included 257 855 singleton infants (57.2% male) who survived to discharge. Gender-specific weight-, length-, and head circumference-for-age curves were created (n = 130 111) and successfully validated (n = 127 744). Small-for-gestational age and large-for-gestational age classifications using the Lubchenco curves differed significantly from the new curves for each gestational age (all P < .0001). The Lubchenco curves underestimated the percentage of infants who were small-for-gestational-age except for younger girls (≤36 weeks), for whom it was more likely to be overestimated; underestimated percentage of infants (≤36 weeks) who were large-for-gestational-age; and overestimated percentage of infants (>36 weeks) who were large-for-gestational-age.
CONCLUSIONS: The Lubchenco curves may not represent the current US population. The new intrauterine growth curves created and validated in this study, based on a contemporary, large, racially diverse US sample, provide clinicians with an updated tool for growth assessment in US NICUs. Research into the ability of the new definitions of small-for-gestational-age and large-for-gestational-age to identify high-risk infants in terms of short-term and long-term health outcomes is needed.
- growth curves
- intrauterine growth curves
- head circumference-for-age
- small for gestational age
- large for gestational age
WHAT'S KNOWN ON THIS SUBJECT:
Intrauterine growth curves for preterm infants are limited by small, older, homogeneous samples; non-US data; combined genders; a lack of length and head circumference curves and/or are based on different samples for weight, length, and head circumference curves.
WHAT THIS STUDY ADDS:
The new intrauterine growth curves created and validated in this study using a contemporary, large, racially diverse US sample provide clinicians and researchers with an updated tool for growth assessment in US NICUs.
Intrauterine growth curves are the standard for assessing the growth of preterm infants1 and are widely used in the NICU setting. Intrauterine curves, which are based on cross-sectional birth data, differ from longitudinal postnatal curves2,–,6 in that these illustrate “ideal” or fetal growth versus actual growth of preterm infants over time, respectively. Newer intrauterine growth curves7,–,14 have been published to improve on earlier curves,3,15,–,18 by using more current, larger, and more diverse samples of infants; however, these more recent curves still have limitations, such as the lack of length- and head circumference-for-age curves to accompany the weight-for-age curves,7,8,10,12,14 different samples for growth measurements (eg, different sample of infants for weight-for-age than length-for-age),9 and/or the curves are based on samples of infants from outside the United States.9,11 The goals for this study were to develop a new set of growth curves for the assessment of weight, length, and head circumference of preterm infants in the NICU setting using a contemporary, large, racially diverse US sample; validate these curves; and compare them with 1 of the older but still commonly used growth curves in NICUs, the Lubchenco curves.15,16
This study used a deidentified, cross-sectional sample of birth data from Pediatrix Medical Group, Inc (data collection details previously summarized19). The sample included 391 681 insured and uninsured infants aged 22 to 42 weeks at birth born in 1 of 248 hospitals within 33 US states (1998–2006). Available data included birth weight (measured on electronic scale to nearest gram), length and head circumference (using measuring tape to nearest millimeter), estimated gestational age (GA; by neonatologist best estimate using obstetric history, obstetric examinations, prenatal ultrasound, and postnatal physical examinations; in completed weeks) gender, and race (using maternal race).
Infants were excluded for missing weight, length, or head circumference (HC); unknown gender; and factors with a known or suspected negative impact on intrauterine growth (eg, multiple births, congenital anomalies, mortality before discharge). These data were divided into male and female because of differences in birth size found in the current sample20 as well as in a previous study of Pediatrix data19 and other data sets.11,12,16 Infants with physiologically improbable growth measurements (“extreme outliers”) were excluded from the gender-specific samples. As in previous growth studies,21,22 extreme outliers were defined as values >2 times the interquartile range (25th to 75th percentiles) below the first quartile and above the third quartile for each GA.23 For each gender, these data were randomly divided into 2 subsamples: the “curve samples” were used to create the curves, and the “validation samples” were used to validate the resultant curves. SAS SURVEYSELECT procedure was used to create stratified random samples; gender-specific stratification was by GA, race, and state in which the birth hospital was located.
Gender-specific weight-, length-, and HC-for-age intrauterine curves were created in this study. LmsChartMaker Pro 2.324 was used to create smoothed percentile curves for the 3rd, 10th, 25th, 50th, 75th, 90th, and 97th percentiles from these raw data. Cole's Lambda Mu Sigma (LMS) method estimates 3 age-specific parameters: a Box-Cox power transformation of skewness (L), median (M), and coefficient of variation (S) that correspond to the relationships in the following formulas: z = [(X/M)L − 1]/LS, where X is the measured value of weight (in kg), length, or HC; and Centile = M(1 + LSZ)1/L, where Z is the z score that corresponds to a given percentile. The curves were developed so that the resulting z scores follow a normal distribution and then were smoothed and converted to percentiles for clinical use.
Several techniques were used to assess goodness of fit for each curve. Worm plots25 were used for visual inspection of the fit of the smoothed curves, and the “best” version was confirmed by visual inspection of the empirical percentiles (on the basis of these raw data) superimposed on the smoothed curves. Z scores were calculated for each infant in the curve samples by using the aforementioned calculations and grouped by GA. Because standard percentile curves such as ours are based on a normal distribution of z scores, the overall mean and SD for the calculated z scores for the sample should be 0.0 ± 1.0. Inspection of the calculated z score distributions by GA was used to determine whether the curves fit these data well at all GAs.
The new growth curves were validated by using gender-specific validation samples. For each growth measure, z scores, SDs, and confidence intervals (CIs) for all validation infants were calculated using the LMS parameters from the new growth curves for each GA. Means and CIs were compared with 0, and SDs were compared with 1 (using an adjusted α of .003 for 19 comparisons within each gender).
The new curves were further evaluated by examining the percentage of infants whose growth measurements fell within the size-for-age categories routinely used in NICUs. By definition, ∼10% of a population should be <10th percentile (small-for-GA [SGA]), ∼80% between the 10th and 90th percentiles (appropriate-for-GA [AGA]), and ∼10% >90th percentile (large-for-GA [LGA]).
Comparison of Lubchenco Curves With New Curves and Validation Data Set
The new curves were also compared with the Lubchenco curves15,16 because these are commonly used in NICUs and met the following a priori criteria: included curves for weight-, length-, and HC-for-age, which were created from the same sample; and the curves were based on US data. The validation infants (as a whole and stratified by GA and GA groups 23–26 weeks, 27–31 weeks, 32–36 weeks, and 37–41 weeks) were classified as SGA or LGA by using the Lubchenco curves. The amount of misclassification was summarized, and a likelihood ratio χ2 was calculated to assess statistical significance of differences with a Bonferroni adjusted α.
The worm plots were generated by using S-PLUS 8.0, and SAS 9.1 was used for all other statistics.
From the 391 681 infants, exclusions were made for missing growth data (n = 31 319; 8% of total sample), unknown gender (n = 215; 0.05%), factors that negatively affect growth (n = 95 962; 24.5%), and physiologically improbable growth measurements (1.6% of total sample: n = 3578 or 0.9% boys; n = 2752 or 0.7% girls). The final sample included 257 855 infants (57.2% male) with a race distribution of 50.6% white, 15.7% black, 24.4% Hispanic, and 9.3% other, similar to US birth data available from National Vital Statistics System 2005 (55.1% white, 14.1% black, 23.8% Hispanic, and 7.0% other). The “curve samples” included 74 390 boys and 55 721 girls, and the “validation samples” included 73 175 boys and 54 569 girls.20 The curve and validation samples had similar racial, GA, and state distributions.
Curve Creation and Presentation
Final weight-, length-, and HC-for-age curves for girls and boys are presented in Fig 1. 20 These curves met the selection criteria described already, and the resulting z score distributions for each GA were 0.0 ± 1.0 (mean ± SD).
Data from all GAs were used in the development of the curves; however, there were a small number of infants in the 22-week group (n < 20), and the 42-week group did not seem to represent healthy 42-week infants given the drop in their average size compared with those at 41 weeks. Although these data from the 22- and 42-week infants provided information about the shape of the overall curve, inspection of the curves compared with the empirical distributions did not show a good fit; therefore, the final curves were truncated to describe infants 23 to 41 weeks' GA.
Descriptive statistics (mean, SD, and 3rd to 97th percentiles) for the new curves are presented in Tables 1 and 2. New definitions for SGA (<10th percentile for age) and LGA (>90th percentile for age) are included in these tables and illustrated by the curves. The LMS parameters for the curves are presented in Table 3.
The newly created growth curves were validated on separate samples. Gender-specific group (all ages combined) z score means and CIs were calculated for all 3 measures (weight, length, and HC) and at each GA. As expected for a normal curve, the combined group means and medians were essentially 0 (≤0.01), all SDs were within 0.01 of 1.00, skewness was essentially 0, and the z scores exhibited a Gaussian distribution.
Similar results were observed when these data were stratified by GA for which all but 1 of the means were <0.1 away from 0. Figure 2 shows an example of this, illustrating that all CIs include 0. In addition, the age-specific z scores had means and medians within ≤0.02 of each other, and each had a skewness of ≤0.009. Weight z scores were always normally distributed, and length and HC z scores were approximately normally distributed. The only exception was that HC z scores for infants <26 weeks approached a normal distribution except for 24-week boys, which showed a slight skew with a mean of 0.02, a median of 0.22, and skewness of 0.24.
Figure 3 shows the gender-specific percentile distributions in the validation samples by using the curve LMS values, illustrating a normal distribution for each measure within each gender, and percentiles within expected ranges. These indicate that the curves are a good fit to the validation samples and would be expected to show a similar good fit to the population from which the samples were drawn.
The percentage of the validation sample classified as SGA or LGA for the group overall (∼10% SGA and ∼10% LGA), for individual GAs (∼10% for each), and for GA groups was determined. For GA groups, boys and girls were combined when calculating CIs for the percentage of SGA/LGA as a result of small sample size at the younger GAs. In all cases, the CIs and percentages were consistent with ∼10%, as would be expected if the curves were correctly classifying infants. Figure 4 shows an example of 1 of these analyses for GA groups.
Comparison of Lubchenco Curves With New Curves and Validation Data Set
The new gender-specific curves were plotted with the unisex Lubchenco curves (gender-specific US curves not available) for comparison. Figure 5 presents the weight-for-age curves as an example of these comparisons. Generally, the new curves had lower average weights, lengths, and HC at younger GAs than the Lubchenco curves until 30 to 36 weeks; the new curves had higher average growth measurements at older GAs. When the validation samples were categorized as SGA/AGA/LGA by using the Lubchenco curves, we found a statistically significant difference for each GA, GA group, and the group as a whole (all P < .0001). Table 4 presents the overall comparisons (boys and girls combined). Overall, the percentage of SGA infants was underestimated by the Lubchenco curves except for younger girls (approximately ≤36 weeks), for whom it was more likely to be overestimated. The percentage of LGA infants was consistently underestimated at younger ages (≤36 weeks) and overestimated at older ages (>36 weeks). In fact, for the younger GAs (24–26 weeks), none of the infants in the validation or curve samples (n = 7399) were categorized as LGA for weight, indicating substantial lack of fit with the Lubchenco curves.
We developed a new set of weight-, length-, and HC-for-age intrauterine growth curves for the assessment of preterm infants in NICUs. The advantages of these curves are the inclusion of the 3 routine NICU growth measurements (weight, length, and HC) on the same infants at birth; a contemporary, large, racially diverse US sample that is racially similar to national birth statistics; and validation on separate gender-specific subsamples to confirm that these new curves are accurate. The misclassification of SGA and LGA infants in the validation samples by using the Lubchenco curve definitions supports the need for updated and gender-specific growth curves. These new intrauterine growth curves offer an updated set of curves that should be generalizable to infants in US NICUs.
Of the available intrauterine growth curves that include weight-, length-, and HC for age 9,15,–,17,26, the Lubchenco curves offer advantages of a reasonable sample size (>5000 infants), the same sample of infants for all 3 curves; small grid increments (ie, weekly for GA; every 200 g of weight; every 1 cm of length/HC); percentiles (versus SDs from mean) for ease of interpretation; and a measure of body proportionality.27 These advantages help to explain why the Lubchenco curves are still used in NICUs today and why we chose them to compare with our new curves. These curves are used at birth for the assessment of intrauterine growth to identify infants who are large or small for their age and allow for the comparison of postnatal growth with that of the gold standard, fetal growth.1
Our exclusions resulted in a sample of healthy, singleton infants to create standard-type growth curves that represent an estimate of optimal growth.28 Concerns have been published29,–,32 regarding intrauterine-type curves that are based on cross-sectional birth data, in particular of preterm infants. Cross-sectional birth data do not represent growth over time; however, these do represent intrauterine growth of fetuses up to the time of birth, which is the American Academy of Pediatrics's recommended goal for preterm infants.1 Preterm infants at birth are smaller in size than fetuses of the same GA,16,33,34 and error in GA dating is a widely known concern because of the questions that it brings to the accuracy of the size/age relationships represented in growth curves31,35,36; however, the use of ultrasound data for estimating size and GA dates has been associated with error,9,37,38 it is not a direct measurement of fetal size,9,32 and it is not feasible in a large study of infants. Despite the limitations, cross-sectional birth data from infants of varying ages remain the generally accepted best sources for creating growth curves for the assessment of infant size at birth and postnatally.30,35,39
Similar to previous studies, we found that birth size differs by gender. Lubchenco et al16 noted small but significant differences in weight between boys and girls 38 to 41 weeks. Thomas et al19 found that girls on average were 95 g lighter, were 0.6 cm shorter, and had 0.6 cm smaller HC. The average differences in our study were comparable to those of the study by Thomas et al19 (96 g, 0.7 cm, and 0.6 cm, respectively); all were statistically different by age group, and most were considered clinically different enough (≥48 g weight, ≥0.5 cm length/HC) to justify separate curves.
SGA and LGA are commonly used to define high-risk groups of infants who receive extra attention in the NICU. SGA infants are at risk for adverse outcomes such as growth40,41 and neurodevelopmental delays,42 and LGA infants are at risk for early hypoglycemia43 and later metabolic syndrome.44 The misclassification that we found by using the old Lubchenco unisex curves may lead to many SGA infants not receiving the care that they need because they are misclassified as AGA, younger LGA infants may be overlooked because they are misclassified as AGA, and older infants may be mistargeted for extra NICU attention because they are misclassified as LGA. This could lead to misdirection of NICU resources and suboptimal outcomes in NICU infants. Research is needed to validate our new definitions of SGA and LGA with short-term and long-term health outcomes.
The differences found between our new curves and the Lubchenco curves may be explained in part by differences in the infants who were from 1998 to 2006 versus from 1948 to 1963, respectively. Given advancements in prenatal care that extend high-risk pregnancies, some of the infants who were born at very preterm ages in the new curves may be more likely to have experienced extended periods of growth restriction than the young infants in the older curves; this could help to explain in part why the new curves are shifted down at this end of the age spectrum. Higher maternal body weight and pregnancy weight gain in recent years45,46 may help to explain why the new curves are shifted up in the older GAs.
Our data also illustrate that the Lubchenco weight curve may not accurately reflect the current US population at GAs <27 weeks. More than 16 000 infants who were <27 weeks were in the Pediatrix sample (before our exclusions), and only 87 of these were classified as LGA by Lubchenco (all of whom were removed as extreme outliers, consistent with previous curves21,22). It was extremely unlikely to have so few infants considered LGA in our sample if the Lubchenco curve were correct at these ages. The Lubchenco weight curve was based on an extremely small number of 24- to 26-week infants (n = 24, 27, and 68, respectively). Lubchenco's 90th percentile was defined by only two 24-week infants, three 25-week infants, and seven 26-week infants. Removing any 1 of these infants would have changed the curves dramatically. Thus, we propose that our curves may provide a more accurate representation of infant size than the Lubchenco curves.
One limitation of these curves was that the sample included NICU admissions only. As a result, our sample may not represent all infants, especially those at older GAs; however, our sample was made up of all NICU admissions to a large number of NICUs from around the United States and was racially similar to recent US birth data, and the size of our term infants were similar to those in the 2000 Centers for Disease Control and Prevention's growth charts.47
The new gender-specific intrauterine growth curves created and validated in this study using a contemporary, large, racially diverse US sample provide clinicians with an updated tool for the assessment of growth status of preterm infants in US NICUs. Research into the ability of our new definitions of SGA and LGA to identify high-risk infants in terms of short-term and long-term health outcomes is needed.
This project was funded, in part, under a grant from the Pennsylvania Department of Health. Dr Olsen is a member of the Biobehavioral Research Center in the School of Nursing at the University of Pennsylvania.
Part of the results presented were included in a master's thesis (New Preterm Infant Growth Curves Influence of Gender and Race on Birth Size) by Ms Groveman in partial fulfillment of the Masters of Science in Human Nutrition degree requirements at Drexel University (July 2008).
We thank Amy Sapsford, RD, and Deirdre Ellard, MS, RD, for sharing clinical expertise on the curves' design and Lauren Ackerman and Paul Hamby for statistical assistance.
- Accepted August 3, 2009.
- Address correspondence to Irene E. Olsen, PhD, RD, LDN, c/o Louise Lawson, PhD, Kennesaw State, Department of Math and Stats, Box 1204, building 12, 1000 Chastain Rd, Kennesaw, GA 30144-5591. E-mail:
FINANCIAL DISCLOSURE: The authors have indicated they have no financial relationships relevant to this article to disclose.
- GA =
- gestational age •
- HC =
- head circumference •
- SGA =
- small-for-gestational-age •
- AGA =
- appropriate-for-gestational-age •
- LGA =
- large-for-gestational-age •
- CI =
- confidence interval
- 1.↵American Academy of Pediatrics, Committee on Nutrition. Nutritional needs of low-birth-weight infants. Pediatrics. 1977;60(4):519–530
- Ehrenkranz RA,
- Younes N,
- Lemons JA,
- et al
- Wright K,
- Dawson JP,
- Fallis D,
- Vogt E,
- Lorch V
- Kramer MS,
- Platt RW,
- Wen SW,
- et al
- Lubchenco LO,
- Hansman C,
- Boyd E
- Lubchenco LO,
- Hansman C,
- Dressler M,
- Boyd E
- Thomas P,
- Peabody J,
- Turnier V,
- Clark RH
- Groveman SA
- Tukey JW
- Johnston FE,
- Zemel B,
- Eveleth PB
- Cameron N
- Cooke RW
- Sauer PJ
- Doubilet PM,
- Benson CB,
- Wilkins-Haug L,
- Ringer S
- Ehrenkranz RA
- Hack M,
- Schluchter M,
- Cartar L,
- Rahman M,
- Cuttler L,
- Borawski E
- Hediger ML,
- Overpeck MD,
- Maurer KR,
- Kuczmarski RJ,
- McGlynn A,
- Davis WW
- Monset-Couchard M,
- de Bethmann O,
- Relier JP
- Groh-Wargo S,
- Thompson M,
- Cox JH,
- Hartline JV
- Hooy SK,
- Amorde-Spalding K
- Boney CM,
- Verma A,
- Tucker R,
- Vohr BR
- Kuczmarski RJ,
- Ogden CL,
- Grummer-Strawn LM,
- et al
- Copyright © 2010 by the American Academy of Pediatrics