Fractional fetal thigh volume in the prediction of normal and abnormal fetal growth during the third trimester of pregnancy

BACKGROUND: Currently, 2‐dimensional ultrasound estimation of fetal size rather than fetal growth is used to define fetal growth restriction, but single estimates in late pregnancy lack sensitivity and may identify small for gestational age rather than growth restriction. Single or longitudinal measures of 3‐dimensional fractional thigh volume may address this problem. OBJECTIVE: We sought to derive normal values for 3‐dimensional fractional thigh volume in the third trimester, determine if fractional thigh volume is superior to 2‐dimensional ultrasound biometry alone for detecting fetal growth restriction, and determine whether individualized growth assessment parameters have the potential to identify fetal growth restriction remote from term delivery. STUDY DESIGN: This was a longitudinal prospective cohort study of 115 unselected pregnancies in a tertiary referral unit (St Mary's Hospital, Manchester, United Kingdom). Standard 2‐dimensional ultrasound biometry measurements were obtained, along with fractional thigh volume measurements (based on 50% of the femoral diaphysis length). Measurements were used to calculate estimated fetal weight (Hadlock). Individualized growth assessment parameters and percentage deviations in longitudinally measured biometrics were determined using a Web‐based system (iGAP; http://iGAP.research.bcm.edu). Small for gestational age was defined <10th and fetal growth restriction <3rd customized birthweight centile. Logistic regression was used to compare estimated fetal weight (Hadlock), estimated fetal weight (biparietal diameter–abdominal circumference–fractional thigh volume), fractional thigh volume, and abdominal circumference for the prediction of small for gestational age or fetal growth restriction at birth. Screening performance was assessed using area under the receiver operating characteristic curve. RESULTS: There was a better correlation between fractional thigh volume and estimated fetal weight ((biparietal diameter–abdominal circumference–fractional thigh volume) obtained at 34‐36 weeks with birthweight than between 2‐dimensional biometry measures such as abdominal circumference and estimated fetal weight (Hadlock). There was also a modest improvement in the detection of both small for gestational age and fetal growth restriction using fractional thigh volume–derived measures compared to standard 2‐dimensional measurements (area under receiver operating characteristic curve, 0.86; 95% confidence interval, 0.79–0.94, and area under receiver operating characteristic curve, 0.92; 95% confidence interval, 0.85–0.99, respectively). CONCLUSION: Fractional thigh volume measurements offer some improvement over 2‐dimensional biometry for the detection of late‐onset fetal growth restriction at 34‐36 weeks.


Introduction
3][4][5][6] This problem is particularly important in late-onset FGR, which is usually defined by ultrasound estimation of fetal size; however, single estimates of fetal size have a low sensitivity for the detection of FGR. 7 It is also difficult to distinguish between a fetus that is constitutionally small for gestational age (SGA) and one that has pathological FGR.This is important in clinical practice as it is recognized that growthrestricted fetuses are most at risk for adverse perinatal outcomes such as admission to neonatal intensive care unit, low 5-minute Apgar scores, neurological injury, and even stillbirth or early neonatal death. 8,9Many different methods have been described to differentiate between healthy and pathologically small fetuses including the use of customized growth charts, 10 analysis of placental and fetal Doppler blood flow, 11 analysis of fetal growth velocity, 12 and the use of placentally derived biomarkers. 13However, these methods have been evaluated mainly in the context of early-onset FGR <34 weeks. 14Doppler ultrasound of umbilical artery flow, which is the mainstay of FGR diagnosis and management in early-onset FGR, is of limited use in identifying term FGR, as this will often be normal. 14][17][18][19] Despite the limitations of fetal sizee based assessment, investigators have continued to examine the use of estimated fetal weight (EFW) calculated using 2-dimensional (2D) ultrasound measurements such as that described by Hadlock et al. 20 This screening is based on estimations of fetal size at a single point rather than growth velocity, so that fetuses >10th centile, but not achieving their growth potential, will not be identified as at risk of FGR.Unsurprisingly, detection of SGA is better the nearer the ultrasound scan is performed to delivery (70-75% at 35-37 weeks and 50-60% at 30-34 weeks for the same 10% falsepositive rate). 21These single estimates of fetal size are also hampered by the lack of adjustment for differences in individual growth potential and therefore a method designed to adjust for individual growth was developed: individualized growth assessment (IGA). 12In IGA, second-trimester growth velocity data are used to determine Rossavik growth models that predict individual thirdtrimester growth trajectories.Threedimensional (3D) sonography may allow more accurate assessment of fetal weight and improved differentiation between normal and pathological growth because it includes soft-tissue volume. 224][25][26] Second-trimester TVol measurements can also be used to generate Rossavik models for predicting TVol and EFW growth trajectories during the third trimester. 27,28he aim of this study was to create reference centiles for TVol measurements and to examine whether 3D measurements of fetal soft tissues can detect pathological deviations in fetal growth more accurately than conventional 2D measurements.

Study design
This was a longitudinal prospective study of fetal biometry using 2D and 3D ultrasound in 1 tertiary referral unit (St Mary's Hospital, Manchester, United Kingdom).From November 2013 through July 2015, women with healthy uncomplicated singleton pregnancies were invited to participate.Exclusion criteria were: fetuses subsequently shown to have a major congenital abnormality, multiple pregnancies, and maternal medical conditions known to affect fetal growth such as maternal diabetes, renal disease, and chronic hypertension.Ultrasound scans were performed for research purposes and were not part of routine antenatal care, and did not include clinically indicated examinations.Scan results were available to view in the clinical records.

Participant recruitment
The study was approved by the Greater Manchester East National Research Ethics Committee in 2010 (Ref 10/ H1013/9) and all participants were enrolled under signed informed consent.Original Research OBSTETRICS

ajog.org
Women meeting study criteria were approached, and given verbal and written information when they attended for their booking 12-week scan appointments.Pregnancies were dated using fetal crown-rump length measurements <14 weeks' gestation. 29EFW centile was calculated using Hadlock et al. 20,31 Participants were not excluded from the study if growth abnormalities were detected at the later scans, but underwent closer monitoring of fetal growth.Clinicians were not blinded to clinical information/women's history at the time of examination, and both participants and clinicians were not blinded to the results of the 2D measurements as these appeared on the screen and were reported in the woman's notes at the time of the scan.Automated volume measurements were obtained using the Voluson 4D view tool (GE Healthcare).TVol was calculated as described by Lee et al. 32 The 3D images were archived for later analysis in random order with clinicians blinded to patient identity.Each research scan took 20 minutes with postimage analysis for TVol measurements taking 5 minutes for each volume.Reproducibility of TVol measurements has previously been examined in 25 women in the second trimester and 33 women in the third trimester, and across both trimesters in 40 women in separate studies. 30,32The intraclass correlations were 0.992 for intraexaminer variability and 0.943 for interexaminer variability. 30

Generation of TVol centiles
Reference centiles for TVol from 14-37 weeks' gestation were produced using the LMS method based on 357 scans from fetuses with normal birthweight outcomes (normal growth potential and birthweight >10th centile). 33This method summarizes the changing distribution of a measurement with age by 3 curves representing the median (M), coefficient of variation (S), and skewness (L), the latter expressed as a Box-Cox power.Using penalized likelihood, the 3 curves can be fitted as cubic splines by nonlinear regression, and the extent of smoothing required can be expressed in terms of smoothing parameters or equivalent degrees of freedom. 33A key assumption of the method is that skewed data can be rendered normal with suitable power transformations.The analysis was performed using the computer software LMSchartmaker. 34Fitted L, M, and S curves were obtained by setting the equivalent degrees of freedom to 1, 8, and 4, respectively, and measurements were converted to centiles using LMSgrowth. 35

Biometry analysis
All data were assessed for normality of distribution by visual inspection of histograms and the Shapiro-Wilk test.
Positively skewed data were transformed to base 10 logarithms.Linear regression analysis was used to assess the relationship with birthweight of 2D and 3D ultrasound measures at 26-28 and 34-36 weeks' gestation, adjusting for gestational age at the time of the scan and at delivery.Hindmarsh et al 36 reported that 2D biometry at 30 weeks' gestation explained 40% of the variance in birthweight.To assess the relationship with birthweight using 2D and 3D biometry obtained at 34-36 weeks, a sample size of 112 patients was calculated as sufficient to identify a 20% increase in the variance explained (from 40% for 2D to 60% for 3D), assuming a type 1 error of 5% and power of 80% using a Fisher 1 correlation z test.SGA was defined as final customized birthweight (individual birthweight ratio [IBR]) <10th centile and FGR final customized birthweight (IBR) <3rd centile using Gestation-Related Optimal Weight (GROW) software (Version 6.7.6.1_15[UK]). 37,38Logistic regression was used to compare the utility of 2D and 3D ultrasound measures (EFW [Hadlock], EFW [BPDeACeTVol], AC, TVol) to predict SGA and FGR.The EFW prediction model that was used for BPD, AC, and TVol was based on a previous study by Lee et al 26,39 and a subsequent validation study.Screening performance was compared using the area under the receiver operating characteristic curve (AUC) (method of DeLong et al 40 ).This analysis was performed on 115 women with complete data.Using data from Roma et al, 41 we calculated that 84 participants would be needed to detect a change in the detection of FGR from 60% for 2D to 72.5% for 3D (power 80%, type I error 5%). 42Statistical analysis was performed using SPSS, Version 22 (IBM Corp, Armonk, NY); Stata 13.1 (StataCorp, College Station, TX); and GraphPad Prism 6 for Windows (GraphPad Software, San Diego, CA).

Individualized growth assessment
IGA was assessed using the publicly available iGAP software package (http:// iGAP.research.bcm.edu), which allows the prediction of third-trimester growth trajectories and birthweight based on a Rossavik model of fetal growth.Growth models for HC, AC, femur diaphysis length, thigh circumference, and EFW (BPDeACeTVol) are based on linear trends of the corresponding secondtrimester growth curves.Each individual growth curve was based on a minimum of 2 second-trimester scans <28 weeks separated by 4-8 weeks.These velocity values were then compared to reference ranges previously calculated in a cohort of fetuses with normal growth outcomes. 28Participants (n ¼ 0) delivering before their 34-to 36week scan were excluded.Expected and observed measurements at 34-36 weeks were compared as follows: percent deviation ¼ 100 Â (observed-predicted)/ predicted.
The mean of this percent deviation was taken as a measure of the systematic prediction error and its SD as a measure of the random prediction error 43 among normally grown infants in this cohort.The modified Prenatal Growth Assessment Score (mPGAS), a composite growth index that combines the percent deviations of HC, AC, femur diaphysis length, thigh circumference, and EFW (BPDeACe TVol), was calculated for each fetal growth curve.The negative version of mPGAS was used to classify fetuses with failing growth potential.Fetuses outside the previously reported 95% reference range of 0% to e0.17% 44 were retrospectively identified as having abnormal growth.The growth potential realization index for birthweight (GPRIWT) is the ratio of the actual birthweight to the predicted birthweight multiplied by 100 and was calculated using measured birthweight

Scatterplots of fractional thigh volumes
Scatterplots of fractional thigh volumes (TVol) (n ¼ 357) during second and third trimesters of pregnancy with fitted percentiles (5th, 10th, 25th, 50th, 75th, 90th, and 95th).Original Research OBSTETRICS ajog.orgwithin 24 hours of delivery and the predicted birthweight from the thirdtrimester growth curve (Supplementary Table 2).The predicted gestation at delivery is at the actual gestation if delivery is 38 weeks or at 38 weeks for deliveries occurring >38 weeks. 44The 95% reference range for this measurement is 84-118%; infants outside of this range were also defined as having abnormal growth following birth. 45he GROW customized birthweight centile and IGA were both included to enable an assessment of the agreement between the 2 methods.The GROW method uses final birthweight only and thus defines FGR by this parameter alone (adjusted for population effects of maternal characteristics and fetal sex), 46 whereas IGA takes into account the individual growth potential of the fetus.

Results
Clinical characteristics of the 115 participants are shown in Table 1.In the normal birthweight group with an IBR >10th centile (n ¼ 91), there was an unexplained stillbirth and a preterm delivery at 36 weeks.Three babies were classified as large for gestational age with an IBR >90th centile.A total of 24 babies (21%) were SGA and 10 babies (9%) were FGR.The study was not powered to assess perinatal outcomes; however, there were no cesarean deliveries for nonreassuring fetal status in the normal birthweight group and 2 each in both the SGA and FGR groups.There were 3 admissions to neonatal intensive care unit in the normal birthweight group and 1 each in the FGR and SGA groups.There were no cases of neonatal metabolic acidosis.
TVol measures were taken in all participants and were normally distributed.Figure 2 shows a scatterplot of TVol during the second and third trimesters with fitted percentiles (5th, 10th, 25th, 50th, 75th, 90th, and 95th).The values of the TVol centiles are given in Table 2 for reference.

Prediction of birthweight using 2D and 3D ultrasound parameters
In isolation, 2D and 3D ultrasound measures at 26-28 weeks' gestation were Table 4 demonstrates the screening performance of ultrasound at 34-36 weeks using the 2 EFW formulae, TVol, and AC for the prediction of SGA and FGR at birth.For the prediction of SGA, the AUC values for EFW (Hadlock), EFW (BPDeACeTVol), TVol, TVol centile, and AC were similar at 0.87, 0.91, 0.86, 0.86, and 0.85, respectively.The only significant difference between prediction models was between TVol and EFW (BPDeACeTVol) (P ¼ .04),but EFW (BPDeACeTVol) was not significantly better than EFW (Hadlock).
For the prediction of FGR, the AUC values were also similar apart from the AUC for EFW (BPDeACeTVol), which was significantly greater than the AUC for AC alone (0.939 vs 0.847; P ¼ .03)(Figure 3).For individual measurements, the specificity of TVol and AC were similar (98% vs 100%), but the sensitivity was higher for TVol (40% vs 20%).Comparison of TVol centile to the absolute TVol measurement did not improve the prediction.

Comment Principal findings of the study
This study is the first to define reference TVol centiles across the third trimester of pregnancy using the LMS method.The fitted centiles from 18-37 weeks are similar to a previous study on TVol centiles using a different method. 32In our study at 20 weeks, measurements at the 5th, 50th and 95th centiles were 3.0 cm 3 , 4.0 cm 3 , and 5.8 cm 3 and 45.6 cm 3 , 57.5 cm 3 and 74.3 cm 3 at 36 weeks, respectively.This compares with those previously reported (20 weeks: 3.7, 5.3, and 7.4 cm 3 ; 36 weeks: 43.6, 62.6, and 89.9 cm 3 ). 32We have also demonstrated that TVol is more closely correlated to birthweight than standard 2D measurements and comparable to 2D EFW and AC measurements for the detection of SGA and FGR at birth.In addition, this is also the first report of a clinical evaluation of the use of IGA in the prediction of SGA and FGR.TVol was better than AC for the prediction of FGR although the Original Research OBSTETRICS ajog.org the prediction of FGR.

Results in context
With regard to SGA and FGR prediction, few studies have examined the correlation between ultrasound and birthweight at 34-36 weeks with more focus on ultrasound EFW estimations within 7 days of delivery.Linear regression analysis of 2D and 3D ultrasound measures at 34-36 weeks demonstrated a good correlation with birthweight, especially for EFW (BPDeACeTVol) (R 2 ¼ 0.76).This improves on the R 2 values of 0.40 reported by Hindmarsh et al 36 obtained with 2D biometry at 30 weeks.A previous study examined EFW (BPDeACe TVol) at 34-36 weeks' gestation in 125 gestational diabetic pregnancies and found it to be more accurate than EFW (Hadlock). 47Other studies predicting birthweight using 2D ultrasound EFW and fetal thigh volume reported similar R 2 values for EFW (Hadlock) (R 2 0.42) at 37 weeks, but no significant association of TVol to birthweight. 48,49However, the latter study was small (n ¼ 42) and observer repeatability was not tested. 49ur results demonstrated a similar performance of TVol and 2D measurements at 34-36 weeks to that of Roma et al 41 and Sovio et al. 50Roma et al 41 obtained AUC at 36 weeks of 0.82 and 0.86 for the detection of FGR (birthweight <10th centile) and severe FGR (birthweight <3rd centile) using 2D EFW, with detection rates of 39% and 61%, respectively.Sovio et al 50 reported AUC at 36 weeks of 0.87 and 0.91 for SGA <10th centile and severe SGA <3rd centile using 2D EFW, with sensitivities of 57% and 77%, respectively.
The use of iGAP software demonstrated low systematic and random prediction errors for EFW (BPDeACe TVol) (e1.0%AE 6.1%), although the systematic error is slightly higher than previously published values using a similar method (0.12% vs 1.0%). 26wever, in the study by Lee et al multiple regression analysis was required to determine model coefficients, whereas in the iGAP model coefficients are known but Rossavik models are needed to provide appropriate parameter values.In this study, we also used the individually derived parameters (mPGAS and/or GPRIWT outside 95% reference range for birthweight) to distinguish FGR from SGA, which has the potential advantage of including fetuses failing to grow appropriately born >10th centile.In this study, there was good agreement between the different classifications of fetal growth using IGA and GROW customized centiles; 9/10 infants born <3rd centile had abnormal scores, 11/14 SGA infants had abnormal scores, and 88/91 appropriate-for-gestational-age infants had normal scores.There were also 3 cases of macrosomia detected by IGA in agreement with macrosomia as defined by conventional customized birthweight centiles.

Clinical implications
We previously demonstrated that TVol is highly reproducible in clinical practice 30 and this holds true for measurements obtained late in the third trimester. 51ew studies evaluated the interobserver reliability of ultrasound-derived EFW 52 but this can vary by as much as 15-20% from actual birthweight. 53Interobserver variability for HC and AC measurements also increases with gestation due to fetal position, reduced liquor, or fetal breathing movements that make obtaining correct ultrasound measurement planes difficult. 54With the advent of semiautomated measurement acquisition and good reproducibility, TVol may therefore be clinically useful in the detection of FGR in late pregnancy. 32lthough small numbers limited our study, we have demonstrated that TVol alone is comparable to 2D biometry measures (EFW [Hadlock] and AC), for the detection of FGR fetuses at 34-36 weeks and demonstrated that sensitivity could perhaps be improved by the inclusion of this measurement.This may be related to exponential increase in softtissue mass >28-29 weeks in normally grown, but not FGR fetuses, which is

Research implications
Several studies demonstrated the potential value of IGA for the detection of pathological growth.A study by Deter et al, 55 which examined the detection of FGR fetuses using a combination of birthweight, placental assessment, and IGA, found that 14/16 fetuses (87.5%) retrospectively classified as FGR had abnormal GPRIWT values for birthweight.More recently, another study by the same group concluded that on a retrospective analysis of 184 SGA fetuses the majority (67%) with both abnormal fetal pathology growth scores (similar to mPGAS) and negative GPRIWT values were <5th centile for birthweight. 56hile the iGAP online software package is straightforward to use and could be introduced into clinical practice, further validation of the IGA method for the identification of late pregnancy FGR is required.Ideally this would include placental markers indicative of poor placental function FGR (eg, placental growth factor 57,58 ), histological examination of the placenta, and postnatal markers of FGR such as the presence or absence of neonatal catch-up growth. 59,60urther work is also required to understand the importance of optimal fetal growth in the context of ongoing childhood development; this research can only progress with more accurate tools for the detection of abnormal fetal growth. 61,62

Strengths and limitations
The strength of our study method is in the standardized image acquisition on a low-risk obstetric population by 2 investigators.This necessarily limited the size of the cohort studied such that differences in perinatal outcome could not be assessed, although previous studies have proven enhanced detection of FGR/ SGA to be a desirable outcome. 3,50In addition, although TVol measurements were concealed and analyzed without knowledge of the pregnancy outcome, the 2D biometry measurements results were made available to the sonographers

FIGURE 1
FIGURE 1 Study design and participant recruitment and follow-up flow diagram Simcox et al.Fractional thigh volume use in third-trimester fetal assessment.Am J Obstet Gynecol 2017.

3
FIGURE 3Area under ROC for ultrasound examination

TABLE 1
Baseline demographics of study participants(n [ 115)

TABLE 3
Correlation of different ultrasound parameters at 26e28 and 34e36 weeks to birthweightAll linear regression models adjusted for gestational age at time of scan and birth.AIC used to compare models confirming better model fit with EFW (BPDeACeTVol).

TABLE 4
Screening performance for small for gestational age and fetal growth restriction based on fetal biometry and fractional thigh volume at 34e36 weeks TVol, fractional thigh volume.Simcox et al.Fractional thigh volume use in third-trimester fetal assessment.Am J Obstet Gynecol 2017.ConclusionsIn conclusion, our study suggests that 3D TVol on its own is equivalent or better to 2D biometry for detecting SGA and FGR at 34-36 weeks' gestation.Moreover, this study demonstrated the potential clinical utility of IGA for the identification of abnormal fetal growth.There are potential advantages in measurement acquisition, positive predictive value, and identification of at-risk fetuses that warrant larger scale study.n

TABLE 5
Comparison of birthweight and ultrasound metrics for those infants defined as fetal growth restriction based on individual birthweight ratio <3rd centile, modified Prenatal Growth Assessment Score <e0.17%, and/or growth potential realization index for birthweight <84% Measurements below threshold demonstrating difference between different criteria.34e36 week TVol threshold of <10th centile or EFW <10th centile (Gestation-Related Optimal Weight [GROW]) calculated using TVol would have identified 7/10 and 8/10 babies <3rd customized birthweight centile, respectively.In contrast to <10th centile (GROW) threshold using EFW (Hadlock) <10 th , which would have identified 2/10.Simcox et al.Fractional thigh volume use in third-trimester fetal assessment.Am J Obstet Gynecol 2017.