Click on the links below to access all the ArticlePlus for this article.
Please note that ArticlePlus files may launch a viewer application outside of your web browser.
Most perinatal monitoring systems identify newborns at high risk based solely on low birth weight (usually a birth weight <2500 g). This approach has led to the disadvantage of capturing a mixture of preterm and growth-restricted babies.1 The criterion of “small-for-gestation” (SGA) identifies babies in the bottom 5% or 10% of the distribution of birth weight at any given gestational age. This dichotomy will identify most babies with severe impairment of fetal growth but also some babies who may be small despite having achieved their growth potential. Conversely, many babies who have failed to achieve their growth potential will not be identified by the SGA criterion. To better identify babies at risk, we need accurate and individualized measures as close to real fetal growth restriction as possible. Studies from Norway2–5 show that birth weights correlate in families and that a small second baby whose older sibling was large has a higher mortality than a small baby whose older sibling was small. These findings suggest that small size per se is an inadequate marker of risk.
At present, we can only indirectly estimate deviation from the supposed growth potential, since most of the genetic determinants of fetal growth are unknown. Skjærven et al4 developed equations to estimate the expected mean birth weight of a second baby based on the birth weight of the older sibling. The size of a baby relative to the size of its older sibling predicted mortality better than SGA. Because disruption in fetal growth may be related to diseases later in life,6,7 it is of interest to explore whether this approach could be useful for fetal outcomes in addition to perinatal mortality.
In this report, we apply the coefficients estimated by Skjærven et al4 to a segment of second-born babies from the Danish National Birth Cohort, with a modification aimed at taking into account the sex of the older sibling. We identify the lowest 10% of babies falling short of their predicted birth weight based on older sibling and compare this group with that identified by the common definition of SGA. We examine the predictive value of these categories in relation to motor development at 6 months. Finally, we discuss some of the advantages and limitations of this approach in etiologic research.
The Danish National Birth Cohort enrolled women nationwide early in pregnancy.8 The present analysis study makes use of data from the first and third of 4 interviews among women giving birth between 1998 and 2001. Depending on the participation of general practitioners, approximately 60% of eligible women were invited to participate and, among these, approximately 60% enrolled.
To be eligible for this analysis, women in the cohort had to be pregnant with their second baby at the time of the first interview. Because we had access only to a segment of the Danish Medical Birth Registry, covering the period 1997 to 2001, the first children in this study were born between 1997 and 2000 and the second children between 1998 and 2001. We required both infants to be live births, as birth weight and gestational age are less prone to errors in this case. Furthermore, we included only babies born at 28 weeks or later because Skjærven’s method of predicting birth weight did not apply to earlier gestational ages. We extended this requirement to first babies as well. We excluded any intervening stillbirth (defined as the birth of a dead fetus from the 28th completed gestational week), but we had no information on whether there had been an intervening spontaneous abortion. There were 10,577 second-born babies fulfilling the aforementioned criteria and not missing key variables, as illustrated in Figure 1.
Assessment of Gestational Age
For second babies, we had 3 estimates of gestational age, based on (1) the date of last menstrual period (LMP) reported by the women when entering the study, (2) the estimated date of delivery (EDD) reported at the first interview, and (3) the gestational age recorded in the Medical Birth Registry (MBR). We do not know whether ultrasound fetal measurements had been taken before the interview. However, for the 10,461 women in which both the LMP and the EDD were available, the values differed by at least 1 day in 65% of the cases, which suggests that, by the time of interview, most women had received an estimate of gestational age based upon an ultrasound assessment. To avoid potential influence of fetal growth on the assessment of gestational age,9 we preferentially used the LMP gestational age as long as it did not differ by more than 13 days from the EDD or MBR (96% of all cases). For the remaining 399 cases we used EDD, which did not differ from MBR by more than 13 days (n = 379), and MBR if EDD was not recorded (n = 16). If all disagreed by more than 2 weeks (n = 4), we used EDD. According to this assessment procedure, all interviews were administered between the 8th and 37th completed week of gestation; 50% of the women in this study were interviewed by the 16th week, and 95% by the 25th.
Calculation of Predicted Birth Weight
The expected birth weight of the second baby was calculated through a slightly modified version of the algorithm suggested by Skjærven et al.4 In their algorithm the predictions differed for the various combinations of sex and gestational age of the second child, according to the following rules: (a) If the second baby was born before the 32nd week of gestation, then the prediction was made independently of the birth weight of the first baby. (b) If the first baby had a birth weight <2500 g, then the prediction was made independently of how small the first baby actually was. (c) For all other babies, the predicted birth weight of the second baby was a linear function of the birth weight of the first baby.
This algorithm does not take sex of the first baby into account and, because girls have smaller birth weight than boys, the predicted birth weight is affected by the sex of the first child. In our data the difference between the actual birth weight and the predicted birth weight was +10 g if the first child was a boy and +77 if the first child was a girl. We then modified the algorithm as follows. We estimated the average difference in birth weight between boys and girls using 27,637 first live-born babies in the cohort at different gestational ages. If the first baby was a girl, before applying (c) we added to her birth weight half the difference between the birth weight of firstborn boys and that of girls. If the first baby was a boy, we subtracted the same quantity before applying (c) (see appendix, available with the electronic version of this article). After this correction, the difference between the actual and predicted birth weight became virtually independent of the sex of the first baby (40 g if the first child was a boy and 46 g if the first child was a girl).
We calculated the degree to which the babies had achieved their predicted birth weight by dividing the birth weight of each baby by the predicted birth weight and expressing it as a percentage (birth–weight ratio). This measure provided an estimate of how close a baby was to its expected birth weight, regardless of the absolute birth weight. We then identified the sex-specific lowest decile of this distribution, and defined these babies as those who had failed to achieve their target birth weight. By using the sex- and gestational age-specific z-scores of birth weight calculated from Norwegian data,10 we identified the lowest decile of this distribution and divided the babies into 4 groups: normal by both classifications, small according solely to the birth-weight ratio, small according solely to the z-score, and small by both classifications.
We describe the characteristics of the 4 groups and examine how they predicted motor development at 6 months. We also report results in relation to infant mortality, although the data were too sparse to provide meaningful results. The results for motor development were available for the babies whose mother responded to a telephone interview about 6 months after delivery (as part of the Danish National Birth Cohort follow-up). Twenty-five babies died before the age of 6 months. Approximately 80% of the remaining mothers responded to the interview (n = 8524). The interview was scheduled to occur when the baby was 6 months of age, but the actual age range was between 152 and 426 days (mean ± SD = 188 days ± 18.3). We restricted this analysis to babies aged between 152 and 210 days (n = 7874) at the time the mother was interviewed. We further excluded 12 babies with missing values for any of the questions of interest, thus leaving 7862 for analysis.
We used 5 questions to assess motor development: (1) “Can the baby hold up his/her head when you pick him/her up?” (2) “Does the baby sit up straight when he/she sits in your lap?” (3) “Can he/she roll from back to stomach?” (4) “Can the baby sit on the floor without rolling over?” and (5) “Can the baby creep (ie, crawl on his/her stomach)?” All questions could be answered by “yes,” “no,” “don’t know,” or “do not wish to answer.” For question (2), there was also the possibility of answering “yes, with a little help.” We classified any answer different from “yes” as a “no” for the purpose of this analysis. A few mothers replied “do not wish to answer” (n = 4) and “don’t know” (n = 24). All but 1 of the babies could hold their head up, but we used this question anyway, as we had made the decision a priori.
The questions on child development were devised by a child neuropsychologist based on known scales, milestones, and field experience in evaluating babies of this age. We defined babies who could perform only 1 (2%) or 2 (17%) of these actions as having a delayed motor development because this reflected approximately the lowest quintile of the distribution. This definition was purely arbitrary because our intent was to compare the predictive value of the 2 criteria of “smallness” rather than to identify clinically pathologic babies. By means of logistic regression adjusted for age of the baby and of the mother, we estimated how the 4 categories of birth weight predicted motor development. We compared the −2log-likelihood of the models after adding the birth-weight ratio to SGA, and vice versa, to explore how the 2 measures affected model fit.
We further describe how the classification was associated with mother’s height and smoking during pregnancy. We chose these 2 factors to exemplify a situation where a predictor remains stable between the 2 pregnancies and a situation where the predictor may change. (We also examined body mass index (in kg/m2), although we do not present these results in the tables.) We chose these 2 factors because they are among the best-known predictors of birth weight, beyond gestational age, sex of the baby, and size of a sibling, which are taken into account by the prediction. Information on smoking, height, and body mass index was available only for second babies.
The second-born babies in this analysis exceeded, on average, their predicted birth weight by approximately 43 g (±395). The mean ratio between the actual birth weight and the predicted birth weight was 101% (±11%). The mean z-score, based on gestational age- and sex- adjusted standards from Norway,10 was 0.21 (±1.00).
The cut-off point for the lowest 10% of the distribution of the birth-weight ratio was 87.7% for boys and 87.6% for girls. For the SGA z-score, the analogous values were −1.00 for boys and −1.02 for girls. A total of 1058 babies were classified as small by the birth-weight ratio and 1059 by the z-score, respectively, but only 738 (70%) by both simultaneously.
Table 1 shows the birth weight characteristics of the babies according to the 4 groups defined by the 2 classifications. Babies with a normal z-score but a low birth-weight ratio were, on average, 679 g lighter than their older siblings; their older siblings were, however, 427 g heavier than the average. However, babies classified as small solely by the SGA z-score were 242 g heavier than their older siblings, although they weighted substantially less than both their predicted birth weight and the average. Babies identified as being small by both criteria had the lowest mean birth weight and the largest difference with their expected birth weight.
A total of 28 babies died within the first year of life. In the category of normal weight by both criteria, 13 of 9198 babies died; in each of the discordant groups (consisting of 320 and 321 babies, respectively) 1 baby died; and in the category where both criteria defined the babies as small 13 of 738 died. The babies who died in the latter category had a mean z-score of −2.35 and an average birth-weight ratio of 0.69.
Table 2 shows the association between the 4 categories and motor skills at around 6 months of age for the 7862 babies available for this analysis. The overall odds ratio associated with being classified as small by the SGA z-score was 1.39 (95% confidence interval [CI] = 1.16–1.66), after adjustment for baby’s and mother’s age. Adding the dichotomous birth-weight ratio to this model resulted in a change in −2log-likelihood of 5.8 (∼χ12, P = 0.016). In the analogous model including the dichotomous birth-weight ratio, the estimated OR was 1.47 (1.23–1.75). Adding SGA to this model resulted in a smaller improvement in model fit (change in −2log-likelihood: 1.11∼χ12, P = 0.292).
Table 3 shows how the babies’ actual and predicted birth weights differed between categories of mother’s height and smoking. Despite the fact that children of tall women had a higher predicted birth weight, tall mothers had even heavier babies than expected. Women smoking during pregnancy had given birth to smaller first babies than nonsmoking mothers (3350 vs. 3525 g) and, consequently, they had a lower predicted birth weight (3554 vs. 3648 g) for their second birth. As with height, babies born of mothers who smoked during pregnancy were smaller than predicted.
Table 3 also illustrates how an excess of babies born to smokers or to short mothers were classified as small solely by the SGA z-score, while such pattern was substantially less evident in the category defined as small solely by the birth-weight ratio. The results for body mass index were similar to those for height (data not shown).
Our analysis shows that there are substantial differences in the newborns we classify as “too small,” depending on whether they are small compared with the population average (SGA z-score) or according to family history (birth-weight ratio). This may have consequences for prediction. In our example, the criterion of SGA and the one based on failure to achieve the “target” birth weight had similar predictive values as risk factors for delayed motor development at 6 months. The 2 classifications produced groups that overlapped only partially, suggesting that a combination of the 2 approaches might be a more sensitive tool in risk assessment and prevention. Programming theories hypothesize that fetal growth impairment may affect organ development6,7 and actual birth weight will often be a poor correlate of growth restriction.
Predicting Birth Weight: Strengths
Given the high correlation between the birth weight of babies born to the same mother, using a previous child as comparison for evaluating the birth weight of the next is clearly an important tool in studying the variability of birth weight and in clarifying the complex relation between birth weight and mortality.2–5 Furthermore, information on a previous baby would highlight important discrepancies that might go unnoticed if a baby is within the norm of its weight for gestational age. Deviation from an individualized expected birth weight is likely as important (or perhaps more important) an indicator of fetal growth restriction as the deviation from a population average.
A number of maternal factors are taken into account when using family history to predict birth weight. Ideally, we would want this predicted birth weight to represent the genetic potential for fetal growth, but it reflects more aspects than that. Several factors, including variation in uterine environment and gene expression, interfere with this paradigm.
The babies identified as too small solely based on their older siblings’ birth weight were not characterized by having short or thin mothers, who tend to have smaller babies.11–14 The group identified in this manner would, therefore, be an interesting target for research on outcomes occurring later in these babies’ life, or for specific monitoring and intervention programs. Also, fewer short or thin women were classified as having small babies with the birth-weight ratio than with the SGA criterion. Although our data on infant mortality were sparse, the babies at highest risk of mortality were the ones defined as small by both criteria simultaneously.
Because the approach based on family data provides a more individualized measure of growth restriction, a measure of deviation from the predicted birth weight may be used as an outcome when studying the impact on fetal growth of time-specific exposures (such as specific diseases in pregnancy). Because many factors influencing birth weight are incorporated in the prediction, this can achieve a better control of known and unknown confounders.
Predicting Birth Weight: Limitations
There are several limitations in using the deviation from the birth weight predicted by family history. It is a surrogate measure for what we really would like to know, and it requires an older sibling. In the absence of a sibling, other individualized birth-weight ratios could be used, such as that proposed by Skjærven et al,4 based on the mother’s birth weight, or the one proposed by Wilcox et al,14 which has been used to identify “growth-retarded babies.”15
The babies in our study were, on average, slightly heavier than predicted, a discrepancy likely attributable to the fact that we applied coefficients estimated from a different source of data. Regardless of the differences between the babies in our study and those used to estimate the coefficients, their application assumes the same association between the birth weights of the 2 babies, which might not be a valid assumption in all circumstances.
Taking family history into consideration when estimating fetal growth restriction has merits, but the deviation from the predicted birth weight should be used as an outcome in etiologic research with caution, since it “overadjusts” for the exposures under study that were also present in the previous pregnancy.
Measurement errors are probably more likely to occur among highly discordant siblings. Furthermore, calculations are based upon the assumption that the first child (if heavier than 2500 g) had reached its growth potential but, if this was not the case, the second baby may reach the predicted birth weight and still be growth restricted.
Strengths and Limitations of This Study
This analysis was based on a relatively large dataset with longitudinal data on a number of determinants of birth weight and it included information on the baby’s development collected about 6 months after birth. The information in the birth records is recorded routinely and thus not prone to differential errors.
Because of the restrictions on date of birth, the babies in this analysis had to be born relatively close to each other. The median interval between birth of the 1st child and conception of the next was about 20 months. This restriction could have had an impact on both the birth weight and on how well the prediction worked. We modified the applied method to incorporate the sex of the first baby. Although it is possible that the coefficients would have been different, had sex of the first baby been taken into consideration by the original model, we do not think that it introduced any bias. Furthermore, this measure performed better than SGA for assessment of motor development.
The babies in this study were slightly heavier than predicted by the standards for Norway,10 as indicated by the mean z-score of 0.21. This may be because the standards were transferred to a different population, or because they were not parity-specific. The differences, however, were small and (since we identified the lowest 10% of the distribution) the criterion is internally robust.
Skjærven et al4 did not take change-of-partner into consideration, nor did we. In our data, 97% of the babies had the same father recorded for both births, and the correlation between the 2 birth weights was higher for these babies than for those estimated when the father was different (1.5%) or missing (1.5%) in one or both deliveries (r = 0.47 vs. 0.39 and 0.36, respectively). Having children with a different man will affect not only the genetic contribution to birth weight but also other factors, such as interpregnancy interval, social class, smoking, and nutrition, all of which may influence birth weight. While the prediction will be less accurate in these cases, we found a very high degree of correlation between the 2 births (even though we did not study this aspect in depth, having too few changes in paternity for a meaningful analysis). We estimated the odds ratios of delayed motor development in babies whose father was the same as that of their older sibling (n = 7652); the OR associated with the birth-weight ratio was 1.52 (95% CI = 1.27–1.81) and that associated with SGA was 1.41 (1.17–1.69).
We had too few deaths for informative analyses on infant mortality, and the measure of motor development we used was of a screening type and arbitrarily defined. However, our main aim was to compare 2 definitions of “smallness” in relation to this outcome, and we doubt that either selections in our sample or the accuracy of the outcome measure would affect the findings of this internal comparison.
Using family data to assess fetal growth of a baby to predict subsequent risk could identify a thus far neglected group of at-risk babies. Combining this measure with SGA may provide further insights in the complex relations between fetal growth and newborns’ health, as well as providing a more sensitive tool for identifying babies at increased risk.