Commonly used population references and standards

In most clinical and epidemiologic research, birthweight-for-gestational-age references have been used commonly. These population references were developed with very large, mostly population-based databases. They provide birthweight percentiles by each gestational week. However, evidence has shown that infants who are born preterm are more likely to be growth restricted. Thus, their birthweight does not represent all fetuses in utero at a given gestational week in preterm. The 10th percentile of the birthweight reference in preterm, for example, is substantially lower than the 10th percentile of the ultrasound-based fetal weight reference. Consequently, a population birthweight reference will significantly underdiagnose SGA infants in preterm births.

Numerous ultrasound-based fetal weight references have been published since the early 1980s. Most of them were cross-sectional references that were based on either retrospective databases or prospective data collection. In these studies, each pregnant woman contributed data from only 1 observation. The relatively large sample sizes in these studies provided relatively stable estimates.

However, the quality of the data in retrospective studies remains uncertain. Selection bias (eg, why a woman received an ultrasound examination at a gestational week when a routine ultrasound examination is not given) may have affected the representativeness of these references to an unknown degree. Prospective studies improve the data quality by confirming gestational age early, scheduling examinations systematically, and having more strict protocols with the measurements that are taken by fewer highly trained sonographers. Nonetheless, cross-sectional studies can provide a reference only for fetal size, not fetal growth velocity. Longitudinal studies with repeated measurements on the same fetus are required to study true fetal growth, which can be important for serial ultrasound measurements.

Several longitudinal ultrasound studies have been conducted in the past 25 years. Although some of them were rather small, others had reasonable sample sizes, which ranged from approximately 200 to 634 women. Most of the larger studies were performed in Europe, predominantly in white women.

The common approach to defining fetal size abnormality

Most clinicians and researchers use SGA and FGR interchangeably. However, this practice is problematic. Fetuses with a weight <10th percentile are not necessarily growth restricted (they may be constitutionally small but healthy). On the other hand, a weight of >10th percentile does not necessarily denote “normal” fetal growth. For example, the rate of fetal growth may undergo pathologic decline in late gestation. In such a case, the birthweight may still be >10th percentile, but the fetus may have experienced growth restriction and incurs an increased risk of perinatal death and morbidity.

Furthermore, it is well established that normal fetal size at birth varies significantly by race/ethnicity, sex, parity, maternal size, and other genetic and physiologic factors. Some studies even suggest that the fetal growth pattern may differ by these factors. For example, based on birthweight data, Overpeck et al showed that Mexican American fetuses seem to grow at a similar (or even faster) rate as white fetuses before term but substantially slower after term. This observation is consistent with findings in a longitudinal ultrasound study in Peruvian women. The authors speculated that this phenomenon might be due to the shorter stature of Mexican American women, which possibly constrains the uterine environment in late gestation. Therefore, ultrasound standards, which are based mostly on non-Hispanic white populations, might not be applicable to other races/ethnicities. Likewise, a given fetal size may be considered normal for a short, thin woman but may reflect FGR for a tall, large woman.

The individualized approaches to define fetal size abnormality

The key to solving these problems is to develop a method that can identify the growth potential for individual fetuses. Several approaches have been proposed over the past 20 years. Rossavik and Deter first proposed a mathematic model for fetal growth. This model assumes that all fetal biometric parameters follow a definable growth pattern throughout pregnancy. Regression analysis can be used to obtain optimal coefficient estimates for the Rossavik function (P = c[t] ^k+s[t] ), where P is the growth of the biometric parameter to be estimated, t is the time in pregnancy when the observation is made, and c, k, and s are the model coefficients. In a series of papers, the authors demonstrated that this mathematic model fit the growth of several fetal biometric parameters quite nicely. Based on this model, the authors developed an individualized growth assessment, in which an individualized fetal growth curve is created based on early ultrasound examinations. The assessment requires a minimum of 2 ultrasound examinations separated by 4–8 weeks before 26 weeks of gestation. This curve is used to predict late fetal growth in the same fetus (ie, each fetus becomes its own control). Implicitly, this model assumes that fetal growth is not affected by external factors (pathologic or environmental) before 26 weeks of gestation.

Concerns regarding this approach have been raised that fetal growth abnormality can be demonstrated as early as in the first trimester. Although measurements that are taken early in fetuses with abnormal growth may still be used to predict the weight at term, such a prediction may not recognize that the fetus has failed to reach its growth potential, because the individualized growth curve has been lowered artificially. In addition, ultrasound measurements have inherent errors, random as well as systematic. Such errors can occur in the first and/or second scans, which may affect the projection of the weight at term. When the errors occur in the opposite directions in 2 scans in early pregnancy, the deviation of term projection is amplified. So far, the literature has not provided convincing evidence that the Rossavik model is superior to computationally simpler models.

In the 1990s, Gardosi et al proposed a method that used customized birthweight norms that incorporated information about fetal growth potential. Based on the premise that birthweight varies with maternal and fetal physiologic parameters (eg, race/ethnicity, parity, sex, prepregnancy or early pregnancy body mass), they defined a new method to calculate optimal fetal weight at each gestational week that is customized by individual profile. The authors combined birthweight data at 40 weeks of gestation with EFW, based on the Hadlock EFW curve. The ultrasound EFW curves were adjusted upward or downward proportionately, according to the birthweight at 40 weeks of gestation for specific maternal and fetal profiles.

One of the important assumptions of this approach is that different fetuses follow a similar growth pattern to reach their respective birthweights at the end of the normal pregnancy. The proportionality equation ensures that differences in birthweight at term between white and Hispanic infants, for example, are formed gradually throughout pregnancy. However, findings from the study by Merialdi et al and Overpeck et al suggest that this assumption might not necessarily be true.

Earlier studies suggested that the method created by Gardosi et al appeared to have significantly improved the classification of FGR by using the risk of perinatal morbidity and death as the gold standard. Infants who were classified by the customized standard as FGR had significantly higher overall mortality and morbidity rates than did FGR infants who were classified by the birthweight standard. However, more recent analyses indicated that the large increase in perinatal mortality risk among infants who were classified as FGR based on the customized standard is due largely to the inclusion of more preterm births. Some studies have suggested that the advantages that the customized classification vs a simple ultrasound-based standard (without adjustment for maternal and fetal characteristics) are rather limited, based on risk of stillbirth and neonatal death. Further studies that use less severe outcomes are warranted to assess the clinical utility of this approach.

Over the years, other methods to individualize the growth standard have been proposed. The regression method, in particular, deserves further attention. This method uses a multivariable linear regression model to predict birthweight based on maternal and fetal characteristics (eg, body mass index, sex, gestational age). The predicted birthweight is considered to be the fetal growth potential, which is based on given maternal and fetal characteristics. One of the appealing features of this approach is that the model essentially can adjust for an unlimited number of variables. Each additional variable can improve the ability of prediction to various degrees. However, decisions about the most influential factors and how many variables should be incorporated into the model have yet to be refined. In addition, similar to the approach by Gardosi et al, this method also uses the growth pattern that was defined by Hadlock et al for all fetuses before 280 days of gestation. Whether such sophisticated models can result in clinically significant improvement in the identification of fetal growth abnormalities remains to be demonstrated.