Correspondence to: Dr W. Lee, Texas Children's Pavilion for Women, Department of Obstetrics and Gynecology, Baylor College of Medicine, 6651 Main Street, Suite 1020, Houston, TX 77030, USA (e-mail: firstname.lastname@example.org)
To prospectively validate the use of fractional limb volume measurements for estimated fetal weight (EFW) during the second and third trimesters of pregnancy and to summarize the medical literature regarding application of fractional limb volume for fetal weight estimation.
One hundred and sixty-four women prospectively underwent three-dimensional ultrasonography within 4 days of delivery. Birth weights (BWs) ranged from 390 to 5426 g. Fetal measurements were extracted using volume datasets for biparietal diameter, abdominal circumference, femur diaphysis length, fractional arm volume and fractional thigh volume. Fractional limb volumes were manually traced from a central portion of the humerus or femur diaphysis. Mean percentage differences and SDs of the percentage differences were calculated for EFW. The proportion of newborns with EFW within 5 or 10% of BW were compared with an estimate obtained using a Hadlock formula that was modified using model coefficients from the same local population sample.
Ultrasound scans were performed between 21.7 and 42 weeks' menstrual age. Optimal model performance (1.9 ± 6.6%) resulted from using a combination of biparietal diameter, abdominal circumference and fractional thigh volume. The precision of this model was superior to results obtained using a modified Hadlock model (1.1 ± 8.4%), although accuracy of these predictions was slightly decreased for female infants. For all fetuses, the prediction model that incorporated fractional thigh volume correctly classified a greater proportion of EFW within 5% (55.1 vs 43.7%; P = 0.03) or 10% (86.5 vs 75.9%; P < 0.05) of BW when compared with the modified Hadlock model.
Fractional thigh volume can be added to two-dimensional sonographic measurements of the head and trunk to improve the precision of fetal weight estimation. This approach permits the inclusion of soft tissue development as part of a weight estimation procedure for the assessment of generalized fetal nutritional status.
Neonatal nutritional status is routinely evaluated by comparing birth weight (BW) with age-related reference ranges for a given population. Since it is not feasible to directly weigh fetuses, obstetricians have used sonographic measurements of the head, trunk and limbs to estimate fetal weight. Unfortunately, estimated fetal weight (EFW) is not as precise as actual BW and is typically associated with random errors ranging from 8.1 to 11.8%. We have previously reported a weak correlation between EFW and neonatal adiposity among late third-trimester infants; only 28–30% of the variation in neonatal percentage body fat was explained by EFW. Furthermore, conventional two-dimensional (2D) sonographic parameters such as biparietal diameter (BPD), abdominal circumference (AC) and femur diaphysis length (FDL) only accounted for 4.0, 24.8 and 14.2% of the variation in neonatal body fat percentages, respectively. Moyer-Mileur et al. also reported that newborn adiposity is not reliably predicted by 2D measurements of the fetus. These observations collectively support a need for refining fetal weight estimation models that also incorporate a soft tissue parameter. Prenatal characterization of soft tissue may more precisely separate small or large, but otherwise normal, fetuses from those that are malnourished.
Fractional limb volume is a fetal soft tissue parameter that includes fractional arm volume (AVol) or fractional thigh volume (TVol), and is based on 50% of the long bone diaphysis length. Such measurements are reproducible among blinded examiners and can be manually calculated from three-dimensional (3D) volume datasets within approximately 2 min. Normal reference ranges for AVol and TVol have been established and fractional limb volume has been used with conventional fetal biometry to improve the precision of EFW[5, 6]. Soft tissue parameters can also be used to specify second-trimester Rossavik models for accurately predicting expected AVol or TVol fetal growth trajectories during the third trimester of pregnancy[7, 8].
The main objective of this investigation was to perform prospective validation regarding accuracy and precision of fetal weight estimation using fractional limb volume over a wide range of BWs. A review of other studies that used fractional limb volume to estimate fetal weight was also carried out.
This was a prospective, cross-sectional study of pregnant women with singleton fetuses in the second and third trimesters of pregnancy. All patients had been enrolled in research protocols approved by the Human Investigation Committees at Beaumont Hospitals, Wayne State University, and the Institutional Review Board of the National Institute of Child Health and Human Development. The inclusion criterion consisted of newborn infants that were delivered during the second and third trimesters of pregnancy; exclusion criteria were pregnancies with poor menstrual dating data, multiple gestations and fetuses with congenital anomalies. Gestational age was based on the first day of the last normal menstrual period or menstrual age confirmed by a first- or early second-trimester dating scan. Maternal age, gravidity, menstrual age at time of scan, fetal gender and ethnicity were also documented. Fetal presentation was not systematically documented for this investigation.
Women were prospectively scanned by 2D and 3D ultrasonography (GE Voluson Expert, GE Healthcare Ultrasound, Milwaukee, WI, USA) within 4 days of delivery. The study population primarily consisted of uncomplicated pregnancies but also included women with gestational diabetes (n = 8), hypertension (n = 8), tobacco exposure (n = 3) and Type I diabetes (n = 2). All fetal measurements were obtained from 3D volume datasets for the following parameters: BPD, AC, FDL, AVol and TVol. Fractional limb volume measurements were manually traced around a central portion of the humerus or femur diaphysis (4D View 9.0, GE Healthcare Ultrasound)[4, 6].
Mean percentage differences (systematic weight estimation error) and SD of the percentage differences (random weight estimation error) were used to compare the accuracy and precision of fetal weight estimation based on our local population sample. The proportion of newborns with estimated BWs within ± 5% or ± 10% of actual BW were compared using McNemar's test for paired observations. Results were compared with those derived using a modified Hadlock formula (using BPD, AC, FDL) that was customized using previously published model coefficients for a Michigan cohort.
The systematic error of each model was examined using a one-sample sign or Student's t-test to determine if the mean percentage difference of each model from actual BW was significantly different from zero. Random errors of various models were compared using the Pitman test for correlated variances. Statistical analysis was performed using the SAS system for Windows (version 9.2, SAS Institute, Cary, NC, USA), and P < 0.05 was considered to be statistically significant.
The study population comprised 164 women who were prospectively scanned within 4 days of delivery between June 2005 and December 2009. Sonographic examinations were performed between 21.7 and 42.0 weeks' menstrual age. Most fetuses were scanned after 35 weeks' gestation (20–24 weeks, n = 6; 25–29 weeks, n = 7; 30–34 weeks, n = 18; 35–39 weeks, n = 107; 40–42 weeks, n = 26). The mean maternal age was 28.5 ± 6.4 years, with an average gravidity of 2.4 ± 1.5 pregnancies. Ethnicities included the following: 53.6% White, 36.0% Black, 6.7% Asian and 2.4% Hispanic. Newborn infants (54.9% female, 45.1% male) were delivered at a mean ± SD gestational age of 37.1 ± 4.1 weeks. BWs were normally distributed with a mean ± SD of 3057 ± 1102 g (range, 390–5426 g).
Table 1 summarizes the accuracy and precision of BW predictions for the original Hadlock models (OH1 and OH2) from a Houston population and sample-specific modified Hadlock models (MH1 and MH2) that were previously developed in Michigan. Predicted BWs were slightly overestimated for all groups (range, 4.4–8.6%) when OH1 and OH2 were used for Michigan research subjects. Sample-specific versions of modified Hadlock models (MH1 and MH2) were associated with improved systematic errors that were not significantly different from zero in all groups (range, 0.7–2.4%). However, random errors of these predictions for the original and modified Hadlock models were similar for all subjects (range, 7.0–11.9%).
Table 1. Summary of systematic (signed mean percentage difference) and random (SD of percentage differences) errors, with respect to birth weight, for different Hadlock and volume-based fetal weight estimation models in our study group with ultrasound within 4 days of delivery
< 2000 g
> 4000 g
Data given as systematic error ± random error. Volume-based models from Lee et al.. Signed mean percentage difference = ((predicted birth weight – actual birth weight)/birth weight) × 100. *Systematic error value significantly different from zero based on a one-sample t-test, P < 0.05. AC, abdominal circumference; AVol, fractional arm volume; BPD, biparietal diameter; FDL, femur diaphysis length; TVol, fractional thigh volume.
Table 1 also summarizes the accuracy and precision of BW prediction models that included fractional limb volume in their weight estimation procedure. For all fetuses, optimal performance was associated with Model 6 (BPD, AC, TVol; 1.9 ± 6.6%), although the systematic error was slightly greater than zero (P < 0.001) (Figure 1).
Weight prediction models that incorporated soft tissue parameters uniformly improved the precision of these estimates, although occasionally at the expense of slightly greater systematic error. For infants with a BW of < 2000 g, Model 6 appeared to provide the lowest random error (0.4 ± 7.8%) as opposed to the corresponding modified Hadlock MH2 model (BPD, AC, FDL; 1.0 ± 10.0%) (P = 0.05). For infants with a BW of 2000–4000 g, random errors were also significantly reduced from the 8.0% range (MH2) when compared with corresponding three-parameter prediction models that included either AVol (Model 3, 6.2%; P < 0.05) or TVol (Model 6, 6.4%; P < 0.05). For infants having BW > 4000 g, modified two- (AC, FDL) and three-parameter (BPD, AC, FDL) Hadlock models provided the following systematic and random errors: MH1 (2.4 ± 7.0%) and MH2 (0.5 ± 8.3%). Corresponding Models 3 (BPD, AC, AVol) and 6 (BPD, AC, TVol) had slightly larger systematic errors (−3.8 and + 4.3, respectively), but smaller random errors ranging from 5.8 to 6.6%. The use of Model 6 for larger infants (> 4000 g) also demonstrated random errors that were significantly lower than those of MH2 (BPD, AC, FDL) (5.8 vs 8.3%, P < 0.05) (Table 1).
Both Models 3 (BPD, AC, AVol) and 6 (BPD, AC, TVol) identified a greater proportion of infants to within 5 or 10% of their actual weight than did MH2 (BPD, AC, FDL). Specifically, proportions correctly identified to within 5% by Model 3 and MH2 were 51.6 vs 43.7% (P = 0.079, suggestive of a trend). The corresponding proportions were 55.1 vs 43.7% (P = 0.027) for Model 6 and MH2. Similarly, a greater proportion of infants were correctly classified to within 10% of actual weight using either Model 6 (86.5%, P < 0.05) or Model 3 (84.7%, P < 0.05) as compared with MH2 (75.9%).
A sub-analysis of the results of Model 6 by gender indicates mildly increased systematic error in weight estimation for females (3.11%) when compared with males (0.46%) (P = 0.01, t-test). However, no gender differences were noted for random errors between 90 females (6.3%) and 74 males (6.7%) (P = 0.58, F-test).
Fractional limb volume can now be added to 2D sonographic measurements of the head and trunk to improve the precision of fetal weight estimation. The performance of a fetal weight estimation procedure requires a careful examination of systematic (accuracy) and random (precision) errors. Systematic error is a measurement component that when replicated remains constant or varies in a predictable manner. The reasons for this type of error can be either known or unknown, although a correction factor may be required to compensate for systematic estimation biases. Random error is a component that varies in an unpredictable manner when the measurement is replicated. Unlike systematic errors, random errors cannot be corrected for because they are inherent in the technique being used to acquire this information. In our study, systematic error was defined as the signed mean percentage difference between predicted and actual BW while random error was calculated from the SDs of all mean percentage differences for a given model.
An ideal fetal weight estimation prediction model should provide results with minimal systematic error and low random error. Our prior retrospective study suggested that the precision of fetal weight estimation is improved by adding fractional limb volume to 2D sonographic measurements of the head and trunk. The current prospective validation study was based on a wide range of BWs and the results were compared with those of sample-specific Hadlock prediction models. Model 6 provided the most precise weight estimates with the lowest random errors for all fetuses (6.6%) as well as for infants with BW < 2000 g (7.8%), BW 2000–4000 g (6.4%) and BW > 4000 g (5.8%). This weight prediction model correctly classified a greater proportion of newborns with predicted BWs within 5 or 10% of actual BW when compared with MH2.
A review of the medical literature indicates that fractional limb volume has been used for EFW in four countries, although when making these comparisons it is important to recognize that different prediction models were variously applied (Table 2)[4, 6, 11-14]. Most studies primarily examined late third-trimester fetuses and many report acceptable reproducibility of these manually traced TVol measurements (Table 3). Only one study evaluated AVol measurements, rather than TVol, for fetal weight estimation. In the current study, Model 2 and Model 3 yielded clinically acceptable accuracy with random errors in the 7.2–7.7% range for all fetuses as opposed to results from the modified Hadlock models (MH1 = 8.7%, MH2 = 8.4%; Table 1). Systematic and random errors for all six investigations are summarized in Table 4. With the exception of that of Lindell and Marsal, most prospective studies have reported improved precision with their limb volume based models. This Swedish study of prolonged pregnancies compared much earlier versions of the fractional thigh volume prediction model by Lee et al.[4, 15] (2001, 2006) with their local reference formula. They also introduced a new sample-specific weight prediction model that included both fractional thigh volume and a 3D volume measurement of the fetal abdomen.
Table 2. Summary of models used in previous studies evaluating fetal weight prediction including volume-based fetal measurements
Optimal weight estimation for macrosomic fetuses is particularly important because of increased risk of birth injury and operative delivery[16-19]. Unfortunately, weight prediction for such fetuses is typically associated with the greatest random errors, as demonstrated by the following studies. First, Melamed et al. retrospectively evaluated the performance of 26 weight prediction models using 3705 weight estimations performed within 3 days of delivery in a population of Israeli women. Mean systematic and random errors were documented for the following weight classes: 4000–4499 g (−1.9 ± 7.0%, n = 360) and ≥ 4500 g (−6.2 ± 8.1%, n = 41). Second, Hart et al. introduced a novel prediction model that added maternal weight at clinic enrollment to 2D sonographic measurements of head circumference, AC and femur length for 424 macrosomic fetuses. A measurement cut-off (AC = 35.1 cm) was used to decide whether to apply the model for this weight estimation procedure that yielded a mean error of −0.03 ± 4.6%. In our validation study, infants with BW > 4000 g had random errors that ranged from 6.6 (Model 3; BPD, AC, AVol) to 5.8% (Model 6; BPD, AC, TVol) despite small decreases in accuracy. A similar inclusion of maternal weight at enrollment for fractional limb volume-based prediction models may improve the precision of fetal weight estimation for macrosomic infants as well.
Some of our results are similar to those of a recent retrospective study by Melamed et al., who reported greater systematic weight prediction errors in female fetuses using published models that did not adjust for gender. Gender-specific prediction models improved the accuracy of fetal weight estimation in a manner that was independent of the adjustment of model coefficients to a local population sample, as we have previously described. This interesting observation may have resulted from using a different set of model coefficients for each gender or because the process of combining biometric parameters for the optimal fit for BW may be different between males and females. The potential benefit of gender-specific prediction models that also incorporate fetal soft tissue assessment warrants further investigation.
The combination of fractional limb volume and 2D fetal measurements provides a soft tissue component to a weight estimation procedure for a more robust assessment of fetal nutritional status. A major step towards translation of these results into clinical practice will depend on the development of automated fractional limb volume measurements that are easily calculated using commercially available computer software.
The authors wish to acknowledge the technical assistance of Melissa Powell, RDMS and Beverley McNie, BS, CCRP. This research was supported (in part) by the Perinatology Research Branch, Division of Intramural Research, Eunice Kennedy Shriver National Institute of Child Health and Human Development, NIH, DHHS. Dr Romero contributed to this work as part of his official duties as an employee of the United States Federal Government.