Ultrasound diagnosis of fetal macrosomia: a comparison of weight prediction models using computer...

4

Click here to load reader

description

Original research by Dr Max Mongelli and Prof Ron Benzie on the use of ultrasound for detecting macrosomia

Transcript of Ultrasound diagnosis of fetal macrosomia: a comparison of weight prediction models using computer...

Page 1: Ultrasound diagnosis of fetal macrosomia: a comparison of weight prediction models using computer simulation

Ultrasound Obstet Gynecol 2005; 26: 500–503Published online in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/uog.1989

Ultrasound diagnosis of fetal macrosomia: a comparison ofweight prediction models using computer simulation

M. MONGELLI* and R. BENZIE†*Division of Women and Children’s Health, Western Clinical School, University of Sydney, Nepean Hospital, and †Department of PerinatalUltrasound, Nepean Hospital, Penrith NSW, Australia

KEYWORDS: birth-weight prediction; computer modeling; gestational age; macrosomia

ABSTRACT

Objective To assess the frequency of the diagnosis ofmacrosomia in relation to differing weight estimationformulae in unselected pregnancies.

Methods Computer modeling techniques were employed.Computer modeling software generated correlated fetalbiometry measurements according to published Britishstandards, from 37 to 41 weeks’ gestation. For each set ofmeasurements, estimated fetal weights were obtained bya panel of 18 ultrasound weight formulae. The diagnosisof macrosomia was made if the fetal weight estimate wasgreater than 4500 g. Cohorts of 5000 pregnancies foreach week of gestation were studied.

Results The frequency of diagnosis of macrosomiaincreased progressively with advancing gestational age,with large increases between 40 and 41 weeks. The typeof weight estimation formula had a profound influenceon the frequency of diagnosis of macrosomia. Five of theformulae tested almost never returned a weight estimategreater than 4500 g. Three formulae yielded false positiverates in excess of 15%. The Hadlock group of formulaeyielded frequencies of 0.3% to 14.6%.

Conclusions Most formulae tend to over-diagnosemacrosomia at term. Intervention rates for suspectedfetal macrosomia may be influenced by gestational ageat the time of scan and the type of fetal weight estimationformula in use. Copyright 2005 ISUOG. Published byJohn Wiley & Sons, Ltd.

INTRODUCTION

The temporal trend towards higher birth weights isleading to increased rates of fetal macrosomia, a

condition associated with increased neonatal and mater-nal morbidity1,2. This excess morbidity is significantlyhigher in pregnancies complicated by gestational diabetesand maternal obesity. In most units the antenatal diagno-sis of fetal macrosomia is based on ultrasound-estimatedfetal weight, and some clinicians may regard this an indi-cation for intervention, such as induction of labor orelective Cesarean section3.

Ultrasound biometry used to detect macrosomia ischaracterized by low sensitivity, low positive predictivevalue and high negative predictive value4. A particularconcern is the prevention of shoulder dystocia, but thecorrelation with birth weight is rather poor5. Hence it islikely that most elective deliveries for this indication areunnecessary. It is important therefore to determine theeffect of differing fetal weight estimation formulae on thefrequency of diagnosis of macrosomia.

The aim of this study was to estimate the frequencyof the ultrasound diagnosis of macrosomia in relation toultrasound weight formulae and gestational age at the timeof ultrasound scan using computer modeling techniques.

METHODS

The computer model simulates ultrasound screening ofunselected populations of singleton pregnancies between37 and 41 weeks’ gestation. Macrosomia was definedas an estimated fetal weight greater than 4500 g6. Thesimulation was based on a modification of a previouslypublished model of fetal growth screening7, which hasbeen independently validated by field studies8. Thesoftware generated random numbers, which were thentransformed to normally-distributed z-scores. Since the z-scores for head measurements, abdominal circumference(AC) and femur length (FL) are correlated, the Choleskymatrix decomposition method was used to generate inter-correlated data sets9. This is a computerized mathematical

Correspondence to: Prof. M. Mongelli, Division of Women and Children’s Health, Western Clinical School, University of Sydney, NepeanHospital, Penrith NSW, Australia (e-mail: max [email protected])

Accepted: 27 June 2005

Copyright 2005 ISUOG. Published by John Wiley & Sons, Ltd. ORIGINAL PAPER

Page 2: Ultrasound diagnosis of fetal macrosomia: a comparison of weight prediction models using computer simulation

Ultrasound diagnosis of fetal macrosomia 501

algorithm that uses matrix algebra to generate datasets with desired correlation coefficients. The correlationcoefficients for biometric variables were derived froma UK database of fetal ultrasound measurements in alow-risk population10. For each week of gestational agethe z-scores were converted to sets of measurementsfor head circumference (HC), biparietal diameter (BPD),AC and FL, according to the medians and standarddeviations of British ultrasound reference standards11–13.These variables were then converted to a fetal weightestimate (EFW) using a total of 18 weight estimationformulae14–28 retrieved from a literature search. Thepopulation characteristics of the original formulae andtheir ultrasound variables are listed in Table 1.

Simulations were run on cohorts of 5000 pregnanciesfor each week of gestation, using a Pentium III PC andTurbo Pascal programming software (Borland SoftwareCorporation, Scotts Valley, CA, USA). The statisticaldistributions of the output of the computer modelwere tested to check for conformity with the publishedinput data. The frequency of weight estimates greaterthan 4500 g was converted to a percentage using theweekly cohort as denominator. The output of Campbell’sand Wilkin’s formula as a function of the abdominalcircumference is shown in Figure 1.

Differences in proportions were tested using the PearsonChi-square test, at the 0.05 significance level. Statisticalanalyses were performed with SPSS for Windows (SystatSoftware Inc., Richmond, CA, USA)29.

RESULTS

The results of the model validation exercise are shown inTable 2; we used medians in order to compare the outputwith the original published studies. In most instances thecomputer-generated fetal biometry medians were within10% of the reference standard, with very similar standarddeviations.

EFW

(g)

0

500

1000

1500

2000

2500

3000

3500

4000

4500

20 25 30 35 40 45 50

AC (cm)

Figure 1 Estimated fetal weight (EFW) from Campbell’s andWilkin’s formula (on y-axis) as a function of the fetal abdominalcircumference (AC) (x-axis).

The percentage of cases diagnosed with macrosomiaaccording to the week of gestation and the fetalweight estimation formula is shown in Table 3. Formost formulae the frequency of ultrasound diagnosesof macrosomia increased progressively with advancinggestational age. It was generally very low at 37–38 weeks,with large increases occurring between 40 and 41 weeks.It ranged from nearly zero in the case of the formulae ofCampbell and Wilkin14, Scott et al.22, Thurnau et al.24,Weinberger et al.27 and Woo and Wan (b)28, to amaximum of 55% for the formula of Vintzileos et al25.Nine of the formulae yielded diagnoses of macrosomia formore than 10% of the population at 41 weeks’ gestation.This was even higher for the formulae of Jordaan19

and Higginbottom et al.18 – as well as that of Vintzileoset al.25 – with rates in excess of 15% at 41 weeks.

DISCUSSION

The advantages of computer modeling in this area includethe avoidance of biases due to differences between centersand operators, selection bias, avoidance of errors related

Table 1 Characteristics of fetal populations used to derive original weight estimation formulae

Authors ReferenceSample

sizeBirth weight;

range (g)Weight

> 4000 g (%)Gestational age;range (weeks)

Campbell & Wilkin 14 140 790–5460 7 N/ACombs et al. 15 380 500–4600* N/A N/ADudley et al. 16 779 1300–5200 10 N/AHadlock et al. 17 109 < 1500 to > 4000 17 N/AHigginbottom et al. 18 50 1200–4300* N/A N/AJordaan 19 98 1000 to > 3500 N/A 26–41Ott et al. 20 464 < 1500 to > 3999 N/A N/APersson & Weldner 21 89 330–5070 N/A N/AScott et al. 22 142 400–1000 N/A N/AShepard et al. 23 73 710–4125 N/A N/AThurnau et al. 24 62 500–2500 N/A 25.5–35.5Vintzileos et al. 25 89 570–4678 N/A 24–42Warsof et al. 26 85 400–4800* 6 17–41Weinberger et al. 27 41 510–1999 N/A N/AWoo & Wan 28 125 850–4350 N/A 25–42

*, data estimated from published graphs. N/A, not available.

Copyright 2005 ISUOG. Published by John Wiley & Sons, Ltd. Ultrasound Obstet Gynecol 2005; 26: 500–503.

Page 3: Ultrasound diagnosis of fetal macrosomia: a comparison of weight prediction models using computer simulation

502 Mongelli and Benzie

Table 2 Model validation: comparison of output with reference standard at different gestational ages; median (SD) in cm

Gestational age (weeks)

37 38 39 40 41

Generated BPD 9.26 (0.37) 9.43 (0.37) 9.59 (0.37) 9.74 (0.37) 9.87 (0.36)BPD reference standard 9.29 (0.42) 9.47 (0.42) 9.48 (0.38) 9.78 (0.45) 9.86 (0.33)Generated HC 33.04 (1.36) 33.48 (1.44) 33.85 (1.52) 34.17 (1.61) 34.42 (1.71)HC reference standard 32.93 (1.15) 33.31 (1.81) 33.51 (1.18) 34.19 (1.86) 34.98 (1.58)Generated AC 32.5 (1.81) 33.6 (1.86) 34.6 (1.91) 35.65 (1.98) 36.67 (2.01)AC reference standard 32.64 (1.74) 33.5 (1.79) 34.4 (1.83) 35.25 (1.88) 36.08 (1.93)Generated FL 6.94 (0.32) 7.09 (0.32) 7.23 (0.33) 7.37 (0.33) 7.48 (0.33)FL reference standard 6.97 (0.32) 7.11 (0.33) 7.24 (0.33) 7.36 (0.34) 7.46 (0.35)

AC, abdominal circumference; BPD, biparietal diameter; FL, femur length; HC, head circumference.

Table 3 Diagnosis of macrosomia (% of cases) in relation to gestational age and fetal weight formula

Gestational age (weeks)

Authors ReferenceUltrasoundvariables 37 38 39 40 41 P

Campbell & Wilkin 14 AC only 0 0 0 0 0 —Combs et al. 15 HC, AC, FL 0 0 0 0.26 2.1 < 0.001Dudley et al. 16 HC, BPD, AC, FL 0 0 0.06 1.56 7.3 < 0.001Hadlock et al. (a) 17 AC, FL 0 0 0.3 3.2 13.8 < 0.001Hadlock et al. (b) 17 BPD, AC, FL 0 0 0.5 3.4 14.6 < 0.001Hadlock et al. (c) 17 HC, BPD, AC, FL 0 0 0.40 2.72 11.4 < 0.001Higginbottom et al. 18 AC only 0.1 0.8 3.3 11 24.9 < 0.001Jordaan 19 HC, AC 0.1 0.6 2.6 7.2 15.4 < 0.001Ott et al. 20 HC, AC, FL 0 0 0.1 1.5 8.1 < 0.001Persson & Weldner 21 BPD, AC, FL 0 0 0.02 0.96 5.0 < 0.001Scott et al. 22 HC, AC, FL 0 0 0 0 0 —Shepard et al. 23 BPD, AC 0 0.02 0.7 3.5 13.8 < 0.001Thurnau et al. 24 BPD, AC 0 0 0 0 0 —Vintzileos et al. 25 BPD, AC 0.4 2.6 12.1 31.3 55.2 < 0.001Warsof et al. 26 BPD, AC 0 0 0.64 3.2 13.4 < 0.001Weinberger et al. 27 BPD, AC 0 0 0 0 0 —Woo & Wan (a) 28 BPD, AC, FL 0 0 0.5 3.1 13.3 < 0.001Woo & Wan (b) 28 BPD, AC, FL 0 0 0 0 0.2 < 0.001

AC, abdominal circumference; BPD, biparietal diameter; FL, femur length; HC, head circumference.

to time intervals from scanning to delivery, and largesample sizes. However a model requires validation inorder for its results to be acceptable. The output of ourmodel in terms of the biometric variables is close to theinput reference standard, and well within range of similarbiometric standards for Western populations. The resultsof a very similar model were validated by independentfield studies8. Hence it is likely that it would approximateresults from a prospective clinical study of similar design.

We noted increasing rates in the diagnosis of macro-somia with advancing gestational age. The large increasefound between 40 and 41 weeks for most formulae couldbe an artifact arising from the intrinsic properties of theformulae, or possibly related to an overestimate of thefetal growth kinetics in the post-term period in the ref-erence standards of Chitty et al.11–13. It is very unlikelyto reflect a true biological phenomenon, since in the EastMidlands Obstetric Database the frequency of macro-somia was approximately 1.3% at 40 weeks and 2.9%at 41 weeks30, whereas many of the formulae we testedgave frequencies higher than 3% at 40 weeks and 10%

at 41 weeks. At the other extreme in performance, withzero rates of macrosomia, diagnosis is due either to thecharacteristics of the original populations from which theformulae were derived, or to the intrinsic properties ofthe formulae. The equations by Scott et al.22, Thurnauet al.24 and Weinberger et al.27 were targeted at small-for-dates fetuses, and thus the zero rates of diagnosis ofmacrosomia are expected.

The mathematical properties of the formula can have amajor influence on the accuracy of the weight estimates.The example of Campbell’s and Wilkin’s formula is shownin Figure 1, which shows the weight estimate of this for-mula as a function of the fetal abdominal circumference.It is grossly non-linear in that the estimate is never higherthan about 4100 g irrespective of the input fetal AC.

In general, these findings are consistent with previouswork, suggesting that ultrasound weight estimation is lessaccurate at term than at 34–37 weeks’ gestation31, andthat it becomes less accurate for larger fetuses4. They alsoshow that formulae perform best in populations similarto those from which they were originally derived. The

Copyright 2005 ISUOG. Published by John Wiley & Sons, Ltd. Ultrasound Obstet Gynecol 2005; 26: 500–503.

Page 4: Ultrasound diagnosis of fetal macrosomia: a comparison of weight prediction models using computer simulation

Ultrasound diagnosis of fetal macrosomia 503

great variation in performance between different formu-lae suggests that choice of fetal weight estimation formulamay significantly affect intervention rates for macroso-mia. The formulae of Campbell and Wilkin14, Thurnauet al.24, Scott et al.22, Weinberger et al.27 and Woo andWan (b)28 will almost never yield a diagnosis of macro-somia, and hence should not be used for this purpose.On the other hand, the formulae of Jordaan19, Vintzileoset al.25 and Higginbottom et al.18 will yield high falsepositive rates and are thus likely to be of limited clinicalutility. The equations of Ott et al.20, Dudley et al.16 andPersson and Weldner21 yield macrosomia frequencies thatare somewhat closer to the UK birth-weight distributionand with lower false positive rates.

Our computer model cannot estimate the sensitivitiesof differing ultrasound weight formulae; field data wouldbe required to address this issue. Previous work andour studies suggest that the diagnosis of macrosomia ismost accurate if the ultrasound examination is performedbefore 40 weeks’ gestation. This estimate can then be pro-jected forwards to more advanced gestational ages usingthe gestation-adjusted (GAP) method32. This is a simplemathematical technique that allows extrapolation of afetal weight estimate to any gestational age by using fetalgrowth formulae.

CONCLUSIONS

The detection of a large fetus depends on ultrasound fetalweight estimation, and the way this technique is appliedwill have a significant impact on clinical intervention rates.Hence sonographers should be aware of the ultrasoundformula that has been programmed in their machineswhen assessing cases of suspected macrosomia.

REFERENCES

1. Orskou J, Kesmodel U, Henriksen TB, Secher NJ. An increasingproportion of infants weigh more than 4000 grams at birth. ActaObstet Gynecol Scand 2001; 80: 931–934.

2. Brieger GM, Rogers MS, Rushton AW, Mongelli M. Are HongKong babies getting bigger? Int J Gynecol Obstet 1997; 57:267–271.

3. Haram K, Pirhonen J, Bergsjo P. Suspected big baby: a difficultclinical problem in obstetrics. Acta Obstet Gynecol Scand 2002;81: 185–188.

4. Ben-Haroush A, Yogev Y, Hod M. Fetal weight estimation indiabetic pregnancies and suspected fetal macrosomia. J PerinatMed 2004; 32: 113–121.

5. Sandmire HF, Woolley RJ. Macrosomia: can we prevent bigproblems with big babies? Birth 1998; 25: 263–267.

6. Ferber A. Maternal complications of fetal macrosomia. ClinObstet Gynecol 2000; 43: 283–297.

7. Mongelli M, Ek S, Tambyrajia R. Screening for fetal growthrestriction: A mathematical model of the effect of time intervaland ultrasound error. Obstet Gynecol 1998; 92: 908–912.

8. Owen P, Maharaj S, Khan KS, Howie PW. Interval betweenfetal measurements in predicting growth restriction. ObstetGynecol 2001; 97: 499–504.

9. Press WH, Teukolsky SA, Vetterling WT, Flannery BP. Numer-ical Recipes in C: the Art of Scientific Computing (2nd edn).Cambridge University Press: New York, NY, 1992; 97.

10. Mongelli M, Gardosi J. Longitudinal study of fetal growth insubgroups of a low risk population. Ultrasound Obstet Gynecol1995; 6: 340–344.

11. Chitty LS, Altman DG, Henderson A, Campbell S. Charts offetal size: 4. Femur length. Br J Obstet Gynaecol 1994; 101:132–135.

12. Chitty LS, Altman DG, Henderson A, Campbell S. Charts offetal size: 3. Abdominal measurements. Br J Obstet Gynaecol1994; 101: 125–131.

13. Chitty LS, Altman DG, Henderson A, Campbell S. Charts offetal size: 2. Head measurements. Br J Obstet Gynaecol 1994;101: 35–43.

14. Campbell S, Wilkin D. Ultrasonic measurement of fetalabdomen circumference in the estimation of fetal weight. BrJ Obstet Gynaecol 1975; 82: 689–697.

15. Combs CA, Jaekle RK, Rosenn B, Pope M, Miodovnik M,Siddiqi T. Sonographic estimation of fetal weight based ona model of fetal volume. Obstet Gynecol 1993; 82: 365–370.

16. Dudley JN, Lamb MP, Hatfield JA, Copping C, Sidebottom K.Estimated fetal weight in the detection of the small-for-menstrual-age fetus. J Clin Ultrasound 1990; 18: 387–393.

17. Hadlock FP, Harrist RB, Sharman RS, Deter RL, Park SK.Estimation of fetal weight with the use of head, body and femurmeasurements – a prospective study. Am J Obstet Gynecol1985; 151: 333–337.

18. Higginbottom J, Slater J, Porter G, Whitefield CR. Estimationof fetal weight from ultrasonic measurement of trunkcircumference. Br J Obstet Gynaecol 1975; 82: 698–701.

19. Jordaan HV. Estimation of fetal weight by ultrasound. J ClinUltrasound 1983; 11: 59–66.

20. Ott WJ, Doyle S, Flamm S, Wittman J. Accurate ultrasonicestimation of fetal weight. Prospective analysis of new ultrasonicformulas. Am J Perinatol 1986; 3: 307–310.

21. Persson PH, Weldner BM. Intrauterine weight curves obtainedby ultrasound. Acta Obstet Gynecol Scand 1986; 65: 169–173.

22. Scott F, Beeby P, Abbott J, Edelman D, Boogery A.New formula for estimating fetal weight below 1000 g:comparison with existing formulas. J Ultrasound Med 1996;15: 669–672.

23. Shepard MJ, Richards VA, Berkowitz RL, Warsof SL, Hob-bins JC. An evaluation of two equations for predicting fetalweight by ultrasound. Am J Obstet Gynecol 1982; 142: 47–54.

24. Thurnau GR, Tamura RK, Sabbagha R, Depp OR 3rd, Dyer A,Larkin R, Lee T, Laughlin C. A simple estimated fetal weightequation based on real-time ultrasound measurements of fetusesless than thirty-four weeks’ gestation. Am J Obstet Gynecol1983; 145: 557–561.

25. Vintzileos AM, Campbell WA, Rodis JF, Bors-Koefoed R,Nochimson DJ. Fetal weight estimation formulas with head,abdominal, femur, and thigh circumference measurements. AmJ Obstet Gynecol 1987; 157: 410–414.

26. Warsof SL, Gohari P, Berkowitz RL, Hobbins JC. The estima-tion of fetal weight by computer-assisted analysis. Am J ObstetGynecol 1977; 128: 881–892.

27. Weinberger E, Cyr DR, Hirsch JH, Richardson T, Hanson JA,Mack LA. Estimating fetal weights less than 2000 g: an accurateand simple method. AJR Am J Roentgenol 1984; 142: 973–977.

28. Woo JS, Wan MC. An evaluation of fetal weight predictionusing a simple equation containing the fetal femur length. JUltrasound Med 1986; 5: 453–457.

29. Norusis MJ. SPSS 10.0 Guide to data analysis. PearsonEducation: Upper Saddle River, NJ, 2000.

30. Wilcox M, Gardosi J, Mongelli M, Ray C, Johnson I. Birthweight from pregnancies dated by ultrasonography in amulticultural British population. BMJ 1993; 307: 588–591.

31. Best G, Pressman EK. Ultrasonographic prediction of birthweight in diabetic pregnancies. Obstet Gynecol 2002; 99:740–743.

32. Mongelli M, Gardosi J. Gestation-adjusted projection of esti-mated fetal weight. Acta Obstet Gynecol Scand 1996; 75:28–31.

Copyright 2005 ISUOG. Published by John Wiley & Sons, Ltd. Ultrasound Obstet Gynecol 2005; 26: 500–503.