These regression equations, derived from a population based sample of more than 62,000 singleton Caucasian pregnancies without the major risk factors for intrauterine growth anomaly, demonstrate the method used at our Institute to assess appropriateness of fetal growth. The method is applicable to all populations with suitable data available. Our results may be directly applicable to other populations besides Western Australian Caucasian and Aboriginal singletons, and we consider it likely to be applicable to all Caucasian populations, but applicability should be verified as suggested below.
Advantages of the ratio method
The use of ratios, such as POBW, in the measurement of intrauterine growth is not a novel idea, [
6,
20] but has not been universally adopted despite the many advantages of ratios over the more commonly used percentile positions. The following discussion applies to all ratios of optimal dimensions, but, for simplicity POBW will be used as the example throughout. These advantages are:
a) Ratios, such as POBW, represent continuous interval measures.
b) Estimations of POBW require only a single standard value, the predicted optimal birth weight, rather than values at several points on the birth weight distribution. The precision of estimating a percentile position varies inversely with observation density and, since the majority of distributions have fewer observations at the extremes, extreme observations will be the least precise, whatever the size of the sample generating the distribution. Extreme observations are also most subject to error. When verification of individual observations is not possible (as with de-identified data), it is a common practice to exclude extreme values on the assumption that they
are in error, significantly altering the estimated value of extreme percentile positions. The positions of percentile extremes are therefore both imprecise and sensitive to actual and perceived data quality. The most precise percentile estimates are those at the highest observation densities, which, since many distributions are akin to Gaussian (particularly those of birth dimensions), is often the 50
th percentile or median[
5].
c) Births affected by growth disturbing factors are over-represented in the extremes of the growth distribution. Hence the positions of extreme percentiles are sensitive to the incidence of growth affecting pathologies in the reference population and vary with the health of the reference population. For example, a newborn with a POBW of 85% might be at the 20
th percentile position of the birth weight distribution for a population with a high burden of growth restricting pathologies, but the 8
th percentile of a population with optimal fetal growth. For an extreme percentile position to be meaningful therefore, the health status of the population from which it is derived needs to be defined, whereas the predicted birth weight is less sensitive to disease burden. Though less sensitive, the proportion of the reference population with growth disturbing factors will also affect the predicted birth weight, except in the unlikely situation where pathological restriction is balanced by pathological acceleration. For this reason we sought to identify a population without growth disturbing factors. The ratio of observed birth weight to predicted birth weight is more generalisable than extreme percentile positions, and the ratio of observed to predicted
optimal birth weight is even more generalisable[
5].
d) POBW is a continuous scale that correlates with weight deficit, whereas percentile position is an ordinal scale that does not. For example, Table
8 considers a population sample with a normal (Gaussian) birth weight distribution, mean birth weight of 3,400 g and standard deviation of 345 g. Being Gaussian, the predicted weight equals the mean (and 50
th percentile position) or 3,400 g. In Table
8, changes in percentile position of 4 or 5 percentile points are shown to represent changes in weight of between 43 g and 151 g depending on the particular percentile positions, whereas, within a population, there is a linear correlation between differences in weight and change in POBW. Equivalent changes in percentile position do not represent equivalent changes in weight. Furthermore, in a total population the presence of growth restricting factors creates a negatively skewed (non-Gaussian) birth weight distribution, so the observed range of birth weights covered by extreme percentiles is broader than indicated in Table
8 and is unpredictable.
Table 8
Comparison of changes in percentile position and in POBW for selected changes in birth weight, for a neonate with an estimated optimal birth weight of 3,400 g.
50th-45th
| 5 | 3400-3357 | 43 | 100-98.74 | 1.3 |
10th-5th
| 5 | 2958-2832 | 126 | 87.0-83.3 | 3.7 |
5th-1st
| 4 | 2832-2681 | 151 | 83.3-78.9 | 4.4 |
Failure to utilise the advantages of a ratio may in part be due to clinical unfamiliarity. In contrast to percentile position, there is little literature describing the clinical associations of appropriateness of growth expressed as proportions of a desirable birth dimension [
21]. For this reason we have included Table
7 which gives the estimated mean, over gestational weeks 38 – 41 inclusive and each gender and parity group, proportion of optimal ratio values of the 3
rd, 10
th and 90
th percentile positions, of each distribution of weight, length and head circumference at birth. This table of equivalences enables an approximate
translation of the literature using percentile positions to percentages of optimal dimensions. The populations from which percentiles are derived will seldom be confined to those without factors affecting growth pathologically. The proportion of optimal equivalences in Table
7 will over-estimate the appropriateness of growth of percentile defined groups to an extent depending on the burden of growth restricting pathologies in their reference sample. When POBW becomes familiar, its numerical values will convey more precise and generalisable clinical meaning than the traditional percentile positions.
Sample selection
The sample was limited to singleton births to Caucasian mothers. It is not useful to classify all determinants of intrauterine growth according to whether or not they have a pathological effect on growth. For example, twin pregnancy slows fetal growth particularly in the third trimester; and gestation-specific perinatal outcomes for multiple births delivered at term are not as good as those for singletons [
22‐
24]. However, it is seldom desirable to reduce twin pregnancies to singleton pregnancies and reasonable to ask whether a twin fetus is growing appropriately, given that it is a twin. Multiplicity-specific fetal growth standards would be required to answer this question.
Maternal race is also a problematic factor. The observed variation in intra-uterine growth rates between ethnic groups [
7,
6] may reflect genetically determined differences in optimal rates and/or systematic differences in incidence of growth restricting pathologies and/or environmental exposures. That is, the association between growth and maternal race may arise as a result of either or both non-pathological and pathological determinants of fetal growth. If racial variation in intrauterine growth arises purely as a result of pathological determinants, maternal race is merely associated with growth rate, rather than being a determinant, and should not be controlled. The balance between non-pathological and pathological influences is likely to vary between ethnic groups and between locations. For example, in Western Australian (WA) Indigenous communities the tendency to slower fetal growth relative to Caucasians is believed to be primarily a result of a higher incidence of growth restricting pathologies and environmental exposures [
11]. In south east Asian communities living in WA women also tend to have small babies but their perinatal outcomes are similar to those of WA Caucasians. It is reasonable to differentiate birth weight distributions by race only if race is itself a (non-pathological) determinant of fetal growth. Whether this is the case may be determined by comparing the estimations for optimal fetal growth, adjusted for non-pathological determinants, between populations of different races, after excluding pregnancies with evidence of exposure to pathological growth determining factors. If the estimations are significantly and systematically different, race specific standards are required. If they are not, the same standards for optimal fetal growth may be used even if the observed distributions in birth weight differ.
The aim of many of our exclusion criteria was to select a sample of births that had not been exposed to factors that have a pathological effect on intrauterine growth. Although the selection criteria only consider causes of growth anomaly we anticipated that selected births would be more likely to be born at term, be larger and born to taller, older and less disadvantaged mothers for several reasons. For example, pathologically affected growth is more often restricted than accelerated and is also associated with preterm birth and maternal smoking, the most prevalent pathological determinant of intrauterine growth, is associated with maternal age and socio-economic circumstances. Table
2 shows these anticipations to be realised. Selected births were also somewhat more likely to be female, supporting the general observation of the female advantage during gestation.
We believe that the curves shown in Figure
2 demonstrate that the exclusion of pathologically affected intrauterine growth, and of erroneous gestational estimates, has been reasonably successful. Biological growth (of which fetal growth is an example) typically proceeds to produce a Gaussian distribution at any point in time, with the standard deviation being proportional to the mean. Charts of unselected birth weights against gestational duration usually demonstrate increasing dispersion with decreasing gestation of delivery for deliveries before 40 weeks. There are two reasons for this: (i) the proportion of erroneously reported gestational age values increases with decreasing gestation, simply because the number of births actually delivered at any gestation week decreases the further it is from the modal value. Typically, erroneous preterm gestational reporting underestimates true gestational duration, hence the birth weight associated with an erroneously reported preterm gestational age is typically higher than those of births actually delivered at the reported gestation.
(ii) Birth much before term often has a pathological cause that also influences fetal growth. Thus the distributions of birth weights delivered at preterm gestations are no longer Gaussian, but typically, because pathological restriction occurs much more frequently than acceleration, are negatively skewed. Thus weights of neonates born at preterm gestations tend to be lower than fetuses of the same gestational age who go on to deliver at term.
In this study we address (i) by using the best estimate of gestational age that can be derived from all available data[
16] and to exclude the '4-week errors' arising from gestational break-through bleeding. If we have succeeded in excluding pregnancies exposed to factors known to pathologically affect fetal growth we have addressed (ii), though the cost is the exclusion of a disproportionate number of infants born preterm, particularly very preterm, as can be seen in Table
2. The observation that the POBWs of the percentile positions are independent of gestational age, Figure
2, indicates that the dispersion is proportional to the mean across gestational age, and is compatible with both (i) and (ii) having been successfully addressed.
Selection of independent variables
Some may consider our selection of predictor variables incomplete as it does not include measures of paternal size, maternal weight or maternal weight gain.
Paternal size
While paternal size is known to influence fetal growth it was not included because the biological father cannot routinely be identified and therefore measures of paternal size are not available on our database. The proportion of variability accounted for by the regression equations would, no doubt, be increased by the inclusion of paternal height as an independent variable.
Maternal weight
It has been suggested that maternal size affects fetal growth because it correlates with the area of uterine endometrium available for placentation. Since this area is not directly measurable, it is logical to seek and adjust for the maternal dimension(s) that correlates most closely with it. Maternal height measures skeletal size in the vertical dimension only, while maternal weight is associated with skeletal size and soft tissue mass, including adipose tissue. Skeletal height tends to correlate with skeletal size, but the proportion of weight consisting of soft tissue, particularly adipose tissue, is very variable, weakening the correlation between maternal weight and skeletal size. Data from the 5 month Dutch famine suggest that maternal pre-natal weight for height, a measure of soft tissue mass, is not a strong determinant of birth weight [
25]. We therefore suggest that maternal height is likely to correlate better with the uterine area available for placentation than is prenatal maternal weight or weight for height.
Maternal weight gain
Maternal weight gain is occasionally considered to be a determinant of birth weight [
26]. However fetal weight can be expected to correlate with maternal weight gain, because fetal weight, and its correlate placental weight, are significant components of maternal weight gain. Thus rather than being a non-pathological
determinant of fetal weight, maternal weight gain partially
measures fetal growth, whether or not it is optimal and should therefore not be adjusted for when estimating appropriateness of growth.
The non-pathological determinants of growth used in these models accounted for 40.5%, 32.2% and 25.2% of the variance in birth weight, length and head circumference respectively. The variation between these proportions may result from variation in the accuracy with which each birth dimension can be measured. Birth weight is routinely measured to within 5 g, representing about 0.15% of a median weight baby. Compared with birth weight birth, length is more difficult to measure reliably due to the tendency of the neonate to flex and the facility with which it may be stretched. Measured head circumference at delivery may be influenced by moulding of the head during passage through the birth canal. The effect of moulding on head circumference may be largely avoided by waiting until 2 days after birth before measurement. However with early discharge policies, such a wait risks failing to obtain any measurement of head circumference and in WA head circumference is routinely measured in the delivery room.
The highly significant, though small, dependence of head circumference on maternal age, despite adjustment for parity, was unexpected and requires confirmation in independent samples
Inclusion of maternal height and age as linear variables
All three dependent variables were found to have a linear dependence on maternal height in the range 147–183 cm. Outside this range there was a tendency for regression to the mean value of the dimension. This may occur because we could not include a term for paternal height and there will be a tendency for women at the extremes of height to have partners with heights that are less extreme. However since only 269 (~0.4%) of our selected sample had a height outside this range, maternal height was included as a linear variable.
Of the three dependent variables only head circumference had an association with maternal age, which was found to be linear up to age 45 years. Only 16 (0.03%) of our selected sample were older than 45 at the time of delivery, therefore maternal age was also included as a linear variable.
Comparisons with previous methods of assessing intrauterine growth
In 1963 Lubchenco and colleagues [
27] presented the first percentile charts of gender- and gestation-specific birth weights for an unselected population of live births and thereby initiated the modern study of intrauterine growth. The next major innovation in the methods of assessing intrauterine growth was the development of customised computer generated charts for individual neonates [
4] by adding to gender and gestational duration the following predictor variables: maternal height, weight, ethnic group, parity and the birth weight, gestational duration and gender of any previous siblings, with the option of further adding measures of growth taken during the index pregnancy. The charts were again presented as percentiles. These charts were designed to predict birth weight rather than assess appropriateness of growth, as not all the independent variables (eg. sibling growth and ethnic group) are necessarily non-pathological determinants of intrauterine growth.
In 1993, Wilcox and colleagues [
6] introduced the concept of the birth weight ratio, the ratio of the observed birth weight to the birth weight predicted given gestational duration, fetal gender, maternal height, weight, parity and ethnic group. Their study sample excluded multiple births, stillbirths and congenitally abnormal babies, and limited their analysis to term births, but did not attempt to exclude pregnancies affected by other pathological determinants of growth. Poor growth was defined on the basis of a percentile position of the birth weight ratio, thereby retaining the problems inherent in the use of percentile positions as standards.