Sonographic examination does not predict twin growth discordance accurately

Sonographic Examination Does Not Predict Twin Growth Discordance Accurately JOHN W. CARAVELLO, MD, SUNEET P. CHAUHAN, MD, JOHN C. MORRISON, EVERETT F. MAGAiVN, MD, JAMES N. MARTIN, Jr, MD, AND LAWRENCE D. DEVOE, MD

Objective: To assess the accuracy of estimating birth weight among twins with discordancy (intra-pair difference in actual birth weight of more than 25%) and to determine the relative accuracy of an intra-pair difference in abdominal circumference (A AC) of 20 mm or more or in estimated fetal weight (A EFW) of 25% or more for the identification of discordant growth in twins. Methods: Over a 6-year period, we identified all nonanomalous twin pairs with gestational ages greater than 23 weeks and sonographic examinations within 3 weeks of birth. Ultrasonographic biometry of both twins included AC, head circumference, and femur length; these indices were used to estimate fetal weight by Hadlock’s formula. Likelihood ratios, receiver-operating characteristic curves, and prediction limits were applied to assess the accuracy of the two diagnostic methods to predict an abnormal outcome. Results: A total of 242 twin pairs were studied. The mean gestational age among the 21 twins with abnormal growth (30.6 f 4.6 weeks) was significantly less than among the 221 twins with normal growth (33.2 f 4.0 weeks) (P < .005). The biometric measurements of fetal parts, sonographic estimate of fetal weight, and actual birth weight for both fetuses were significantly less for discordant twin pairs (P c .05). The accuracy of predicting birth weight, as determined by mean error and percentage of the estimate within 10% of the actual weight, was similar between the groups. Receiver-operating characteristic curves showed that both diagnostic tests yielded areas under the two curves not significantly different from the area under the nondiagnostic line (P > .05). Most important, prediction limit calculations indicated that a 90% certainty that the actual birth weight discordance was at least 25% was achievable only if A AC was 172 mm or greater or A EFW was 112% or more. Conclusion: The most popular current methods (difference in AC or EFW) for predicting discordant growth in twin From the Departmenf $ Obstetrics and Gynecology, Medical College $Georgia, Augusta, Georgia; and the Lhiuersity $Mississippi, Jackson, Mississippi. This study was supported in part by the Vicksburg Hospital Medical Foundation.

VOL.

89,NO,4,APRIL

1997

MD,

gestations have limited accuracy when held to a standard for discordance that requires a birth weight difference of at least 25%. (Obstet Gynecol1997;89:529-33. 0 1997 by The American College of Obstetricians and Gynecologists.)

Potentially serious complications are more common in twin gestations with discordant fetal growth, defined as an intra-pair birth weight difference that exceeds an established threshold of 25% or more.rV3Although twin pair birth weight differences exceeding 25% are infrequent,4 they are linked with a high risk for perinatal death and developmental handicaps. Erkkola et al’ reported that discordant twin birth weight exceeding 25% was associatedwith a sixfold increase in the risk of death of one fetus and an overall 2.5-fold increase in perinatal mortality. Babson et al2noted that at a median age of 8 years, the smaller twin with discordant growth had persistent and significant differences in height, head circumference, scores on intelligence tests, and language comprehension and expression. Similarly, Ylitalo et al3 reported that at a mean age of 9.4 years, the larger twin of each pair performed significantly better in fine motor performance, balance, and coordination and in visual-motor perception. Based on earlier reports (Table l), it has been concluded that comparing the difference in abdominal circumference (AC) or in estimated fetal weight (EFW) (expressed as a percentage of the larger fetus) can lead to accurate identification of twin fetal growth discordance.5-10A critical analysis of these studies indicates that they were hampered by small sample size and suboptimal statistical approaches to address the validity of this conclusion. We performed this study to assessthe relative accuracy of estimating birth weight in discordant and nondiscordant twins and to determine the relative accuracy

0029-7844/97/$17.00 PII 5.0029.7844(97)00010-O

529

Table

1. Published Accuracies Growth Among Twin

of Difference Gestations

Discordance (A BW)

Authors

in Abdominal

Incidence

Blickstein et al’ Divan et al”

15%> 15%

24/54 (44%) 18/58 (31%)

Chauhan et al’ Hill et al’

157” 20%

20/78 (26%) 18/49 (36.7%)

Storlaazi

20%

lo/43

et al’

Chamberlain

et al”’

25%

(23%)

B/85 (9.570)

Circumference

Diagnostic A h h h A A 6 h A h

test

AC 218 mm EFW ~15% AC ~20 mm EFW ~15% EFW ~20% AC ~20 mm EFW ~20% AC 220 mm EFW ~25% AC 220 mm

or in Estimated Sensitivity (74 67 47 53 50 93 83 80 80 37 43

Fetal

Weight

Specificity (%) 78 81 77 85 86 90 93 85 98 91

to Identify

Discordant

I-TV (%I

NPV (70)

69 56 53 53 72 83 80 62 75 53

76 74 77 83 97 90 93 93 93 77

A BW = difference in the actual birth weight, expressed as a percentage predictive value; A AC = absolute difference in abdominal circumferences expressed as a percentage of the larger fetus.

of the iarger twin; PPV = positive predictive value; NPV = negative of two fetuses; a EFW = difference in the estimated fetal weight,

of intra-pair differences in AC and EFW (A EFW) to identify twins with actual birth weight discordancy of 25% or more.

abnormal condition (true discordance) and parturients considered to have an abnormal condition but a normal outcome (false discordance). Receiver-operating characteristic curves were generated for an AC difference (O-40 mm, at 2-mm intervals) or an EFW difference (O-40%, at 2% intervals) to identify accurately those twins with 25% or more discordant growth. These curves were developed by plotting the sensitivity (true-positive rate) against 1 specificity (false-positive rate) for an intra-pair difference in AC or discordant growth (based on EFW) to predict the adverse outcomes. The area under the curve (w) and the standard error were estimated by the point-to-point trapezoidal method of integration. The critical ratio z test for a paired statistical design was used to determine statistical significance of the curve.i’ Prediction limits were calculated by regression analysis of predicted discordance (based on intra-pair AC difference or EFW) versus actual discordance and by calculating 50th and 90th percentile ranges for the discordance at birth.i3

Materials and Methods We identified all live-born twin pairs born at a tertiary center during a 6-year period who met the inclusion criteria of gestational age greater than 23 weeks, no anomalies, and sonographic examination within 3 weeks of delivery. The following data were obtained for each pregnancy: maternal demographics, sonographic measurements of AC and femur length, presentation of the two fetuses at delivery, and actual birth weights. The EFW was derived from measurements of AC and femur length. i’ The sonographic examinations were performed by obstetric residents or sonographic technologists using the Ultramark IV-A System (Advanced Technologies Laboratory, Bothell, WA). The accuracy of predicting birth weight was assessed by mean error (actual estimated birth weight), mean standardized error (absolute error [gl/actual birth weight [kg] * loo), and percentage of the estimate within 10% of the actual birth weight. Statistical tests included Student t test, 2 analysis, or Fisher exact test, where appropriate. Because mean absolute error is not normally distributed, analysis of variance followed by Dunnett’s test was used. P < .05 was considered significant. The relative risk (RR) and 95% confidence interval (CI) were determined for contingency data. Sensitivity, specificity, and positive and negative predictive values were calculated for AC difference of 20 cm or more or EFW difference of 25% or more to identify twins with discordance of at least 25%. Likelihood ratios were calculated by determining the proportions of patients with an adverse outcome and an

530

Caravello

et al

Twin

Grozuth

Discordance

Results Among the 242 parturients with twin gestation, the mean (2 standard deviation) maternal age was 24.3 ? 5.4 years, and gravidity and parity were 2.7 -t 1.7 and 1.3 + 1.4, respectively. The mean gestational age at sonographic examination was 32.1 -+ 4.1 weeks and at delivery was 32.8 ? 4.1 weeks. The mean birth weights of twins A and B were 1928 i 650 and 1917 2 682 g, respectively. The incidence of twins with at least 25% discordant growth was 8.6% (21 of 242). Fetal presentations were similar among the discordant and nondiscordant twins: vertex-vertex, nine of 21 versus 63 of 221 (P = .21; RR 1.77; 95% CI 0.78, 4.01); vertex-nonvertex, six of 21 versus 86 of 221 (P = .48; RR 0.65; 95% CI 0.26,

Obstetrics

6 Gynecology

Table

2. Biometric Measurements (mm) and Relative Accuracy of Estimated Fetal Weight Among Discordant and Nondiscordant Twins*

Measurement GA at delivery (wk) Twin A BPD HC AC FL SEFW (g) Actual BW (g) Error (g) St Ab Er (g/kg) 510% Twin B BPD HC AC FL SEFW (g) Actual BW (g) Error fg) St Ab Er (g/kg) klO%

Discordant twins (n = 21)

Nondiscordant twins (n = 221)

30.6 2 4.6

33.2 i 4.0

93%, respectively.

P

.005

74.7 266.1 252.5 56.7 1551 1566 -14.5 150.4 11/21

t 10.9 -c 38.6 i 40.9 i 9.3 2 635 2 728 k 364.7 k 142.5 (52%)

80.5 285.6 271.9 60.7 1912 1963 -50.6 131.3 107/221

i2 2 k f + + 2

9.6 35.5 41.0 9.2 675 633 351 121.9 (48%)

,009 .02 .03 .04 .02 ,007 .65 .93 .82

73.1 262.3 237.5 54.5 1367 1367 9.9 164.1 lo/21

+ 11.4 + 39.5 k 44.2 t 9.7 L 651 + 616 2 321.1 2 155.3 (48%)

79.5 285.5 271.9 60.9 1937 1971 -33.6 121.3 119/221

+ 2 t 2 f z z ?

9.5 34.3 41.0 8.6 708 675 372.4 111.9 (54%)

,004 .03 ,002 ,001 .0005 .OOOl .60 .35 .65

GA = gestational age; BPD = biparietal diameter; HC = head circumference; AC = abdominal circumference; FL = femur length; SEFW = sonographic estimate of birth weight, based on formula proposed by Hadlock et al ‘I. , BW = birth weight; error = actual BW SEFW; St Ab Er = absolute value of error/BW (kg) . 100; t-10% = number of estimate within 10% of the actual birth weight. *Discordant twins: actual birth weight difference 25% or more; nondiscordant twins: actual birth weight difference less than 25%.

1.62); and nonvertex-others, five of 21 versus 72 of 221 (P = .06; RR 0.40; 95% CI 0.15, 1.07). Table 2 shows that the mean gestational age at delivery was significantly different between the groups. Regardlessof whether discordance was identified antenatally, biometry of fetal parts, the sonographic estimate of birth weight, and actual birth weights were significantly different. Based on the sonographic estimate of birth weight, the mean discordance among 21 twins with abnormal growth was 20.4 2 18.5%, versus 11.1 ? 9.3% for the 221 twins with normal growth (P < .OOOl). At birth, the actual discordance for the two groups was 34.3 ? 9.5% and 8.9 + 6.1%, respectively (P < .OOOl). For both fetuses, the accuracy of the sonographic estimate of birth weight (as determined by mean error, mean standardized absolute error, and percentage of the estimate within 10% of the actual birth weight) was similar among the discordant and nondiscordant twins (Table 2). The sensitivity, specificity, and positive and negative predictive values for intra-pair AC difference of 20 mm or more to identify twins with an actual birth weight difference of 25% or greater were 43%, 68%, ll%, and

VOL.

89,

NO.

4, APRIL

1997

Similar

values

for EFW difference

of

25% or more to predict the abnormal growth were 33%, 94%, 33%, and 94%. The likelihood ratios for these two diagnostic tests to predict an adverse outcome were 1.35 and 5.26, respectively. Figure 1 depicts the receiver-operating characteristic curves for a difference in intra-pair AC (O-40 mm at 2-mm intervals) and for discordance based on sonographic EFW (O-40% at 2% intervals) to identify twins with actual discordance of 25% or more. In both curves, the false-positive rate exceeds the true-positive rate, precluding a cutoff point. The areas under the curves are significantly less than the area under the nondiagnostic line (P > .05). The linear relation between A AC and discordance is described by the following equation: Actual discordance = 7.8 + (0.20) A AC. The correlation coefficient of this equation was 0.11, and the relation was not significant (P > .05). Similarly, for calculated discordance (A EFW) to predict actual discordance, the relation is described by: Actual discordance = 7.3 + (0.32) A EFW. The correlation coefficient of this equation was 0.13, with P > .05 for the relation. Tables 3 and 4 list the ranges (50th and 90th percentiles) of actual discordance for a given A AC or A EFW, respectively.

Discussion This is the largest study that has assessedthe accuracy of sonographic examination to diagnose discordant growth among twin pairs. We found that, although the gestational age and the biometric measurementsamong the discordant twin pairs differed significantly from those of nondiscordant twin pairs, the degree of accuracy of EFW was similar for the two fetuses regardless of the growth pattern. Admittedly, the sample size was not sufficient to determine whether the accuracy of the sonographic estimate of birth weight was similar among discordant and nondiscordant twins. If 50% of the estimates of birth weight are within 10% of the actual weight, l4 then 408 sets of twins would be required in each group to detect a minimum difference of 10%. Considering that the incidence of discordance of 25% or more is 9.5%,l” 4295 sets of twins would be required to provide 408 discordant twins for analysis. Such a large sample size is a formidable undertaking. However, the receiver-operating characteristic curve, which obviates the need for sample size calculations, indicated that neither the difference in AC nor that in EFW appeared to predict accurately an actual birth weight discordance of 25% or more, the arbitrary standard adopted for this study as well as for others. The most important findings of this study are derived from the prediction limitation calculations, which indi-

Caravello

et al

Twin

Growth

Discordance

531

Figure acteristic

1. Receiver-operating curves for abdominal

charcir-

cumference (A AC) difference (O-40 mm, at Z-mm intervals) (open squares) among twins and estimated

a e 0.6

fetal weight 40%, at 2% to

(A EFW) intervals)

identify

twins

difference (O(closed circles) with

25%

or

greater discordant growth. The area (i standard error) under the curves (0.41 5 0.09 for A AC and 0.37 + 0.2

0.08 for A EFW) is not significantly different from the area under nondiagnostic line (P > .05).

the

0.0 0.4

0.6

False positive rate

cated that for any diagnostic threshold, the range of actual birth weight discordance is too wide to be clinically useful. For example, if the difference in AC is 20 mm, then 50% of the time the actual birth weight discordance among the newborns could range from 6% to 18% (Table 3). If a 50% accuracy in diagnosing a 25% birth weight discordance is desired, then the A AC would have to be 116 mm or greater (Table 3) and the A EFW 76% or more (Table 4); these conditions are rarely found. Although it could be argued that not all studies were performed or repeated by the same individual, thus making intra- and inter-observer variation a possible Prediction Limit Calculations for Absolute Difference in Abdominal Circumference to Predict Actual Discordance

Table

3.

A AC

(mm)

50%

90%

2

2-14

8 14

3-16 4-17

-7-23 -6-24

20 26

6-18 7-19

32 38

8-20 9-22

44 50 56

lo-23 12-24 13-25

62 68

14-27 15-28

74 80

16-29 18-30

86

19-32

AC

532

= abdominal

Caravello

A AC

50%

90%

92 98

20-33 21-34

-4-26 -3-27

104 110

22-35 24-37

11-42 12-44 13-45

-2-28 -1-29

116 122 128

25-38 26-39 27-41

134 140

28-42 29-43

146 152

30-44 32-46

158 164

33-47 34-48

170 176

35-50 36-51

o-31 2-32 3-33 4-34 5-36 6-37 7-38 S-40 10-41

(mm)

circumference.

et al

Twin

Growth

Discordance

14-46 15-48 16-49 17-50 18-52 19-53 20 -54 21-56 22-57 23-59 25-60 26-62

confounding issue, previous work from the same institution indicated that the accuracies of predicting birth weight and identifying discordant growth were similar regardless of whether the sonographic examination was done by obstetric house staff or by registered medical sonographers.7Z’4 Moreover, the only study” that defined discordance as a difference in birth weight of 25% or more had a similar sensitivity (37-43%) in detecting the abnormal condition as found in the current report (33-43%). These observations suggest that the findings of the present study could be generalized to other settings.

Table

4.

A EFW

(%)

2.0

Prediction Limit Calculations for Difference in Estimated Fetal Weight* to Predict Actual Discordance 50%

90%

A EFW

(%)

2-14 3-15

-7-23 -6-24

62.0 64.0

4-17 6-18

-4-25 -3-27

68.0 72.0

7-19

-2-28

22.0 26.0

8-20 9-22

30.0 34.0

11-23 12-24

-1-29 l-31 2-32

76.0 80.0

38.0 42.0

13-26 14-27

46.0 50.0 54.0

16-28 17-29 18-31

58.0

19-32

6.0 10.0 14.0 18.0

3-33 4 -34 5-36 7-37 8-38 9-40 10-41

EFW = estimated fetal weight. * Expressed as a percentage of the

84.0 88.0 92.0 96.0 100.0 104.0 108.0 112.0 116.0

larger

50%

90%

21-33

11-43

22-35 23-36

13-44 14-45

24-37 25-39

15-47 16-48

27-40 28-41

17-50 18-51

29-43 30-44

19-52 20-54

31-45 33-47

22-55 23-57

34-48 35-49 36-51

24-58 25- 60 26-61

37-52

27-62

twin.

Obstetrics

6 Gynecology

In conclusion, the current methods for antenatal prediction of discordant twin growth, ie, intra-pair differences in AC or EFW, appear to be too inaccurate as relevant indicators for clinical intervention. Therefore, surrogate markers for discordant growth should be sought, such as umbilical artery Doppler velocimetry, which could improve the detection of this high-risk condition.

R$erences 1. Erkkola R, Ala-Mel10 S, Piiroinen 0, Abad L. Growth discordancy in twin pregnancies: A risk factor not detected by measurements of biparietal diameter. Obstet Gynecol 1985;66:203-6. 2. Babson SG, Kanagas J, Young N, Bramhall JL. Growth and development of twins of dissimilar size at birth. Pediatrics 1964; 33:32733. 3. Ylitalo V, Kero I?, Erkkola R. Neurological outcome of twins dissimilar in size at birth. Early Hum Dev 1988;17:245-55. 4. O’Brien WF, Knuppel RA, Scerbo JC, Rattan I’. Birth weight in twins: An analysis of discordancy and growth retardation. Obstet Gynecol 1986;67:483-6. 5. Blicksteln I, Friedman A, Caspi B, Lancet M. Ultrasonic prediction of growth discordancy by intertwin difference in abdominal circumference. Int J Gynaecol Obstet 1989;29:121-4. 6. Divon MY, Girz BA, Sklar A, Guidetti DA, Langer 0. Discordant twins-a prospective study of the diagnostic value of real-time ultrasonography combined with umbilical artery velocimetry. Am J Obstet Gynecol 1989;161:757-60. 7. Chauhan SP, Washburne JF, Martin JN Jr, Roberts WE, Roach H, Morrison JC. Intrapartum assessment by house staff of birth weight among twins. Obstet Gynecol 1993;82:523-6. 8. Hill LM, Guzick D, Chenevey I’, Boyles D, Nedzesky I’. The sonographic assessment of twin growth discordancy. Obstet Gynecol 1994;84:501-4.

VOL.

89,

NO.

4, APRIL

1997

9. Storlazzi E, Vintzileos AM, Campbell WA, Nochimson DJ, Weinbaum PJ. Ultrasonic diagnosis of discordant fetal growth in twin gestation. Obstet Gynecol 1987;69:363-7. 10. Chamberlain I’, Murphy M, Comerford FR. How accurate is antenatal sonographic identification of discordant birthweight in twins? Eur J Obstet Gynecol Reprod Biol 1991;40:91-6. 11. Hadlock FP, Harrist RB, Carpenter RJ, Deter RL, Park SK. Sonographic estimation of fetal weight. Radiology 1984;150:535-40. 12. Beck JR, Shultz EK. The use of relative operating characteristic (ROC) curves in lest performance evaluation. Arch Path01 Lab Med 1986;110:13-20. 13. Pollack RN, Pollack-Hauer G, Divon MY. Macrosomia in postdates pregnancies: The accuracy of ultrasonographic screening. Am J Obstet Gynecol 1992;167:7-11. 14. Chauhan SF’, Cowan BD, Brost BD, Washburne JF, Magann EF, Morrison JC. Estimating birth weight in twins: Comparison of eight sonographic models. J Reprod Med 1996;41:403-8.

Address reprint requests to: Suneef P. Chauhan, MD Medical College of Georgia 1120 25th Street Room No. BAA 735 Augusta, GA 30922-3350

Received August 27, 1996. Received in revised form November Accepted December 5, 1996.

13, 1996

Copyright 0 1997 by The American College of Obstetricians Gynecologists. Published by Elsevier Science Inc.

Caravello

et al

Twin

Growth

Discordance

and

533