Screening for trisomies in dichorionic twins by measurement of fetal nuchal translucency thickness according to the mixture model.

Abstract

To examine the distribution of fetal nuchal translucency (NT) thickness in dichorionic twins and investigate the effect of the correlation between NT measurements in each twin pair on the performance of screening for trisomies.
The distribution of fetal NT for crown-rump length (CRL) was examined in 5646 dichorionic twin pregnancies, including 103 with fetal trisomies 21, 18 or 13. The correlation in fetal NT in each euploid twin pregnancy was estimated.
The distribution of NT in both euploid and trisomic fetuses was consistent with the mixture model in singleton pregnancies. In the euploid pregnancies, there was a correlation in log NT measurements in each twin pair (r = 0.42, 95% CI: 0.39-0.45) and, after removal of the effect of the operator, this correlation was reduced to 0.34. Allowing for this correlation in risk assessment for trisomies had a major impact on the estimated patient-specific risk but had little effect on the overall performance of screening.
In dichorionic twin pregnancies, the mixture model of distributions of NT can be applied as in singletons. In screening for trisomies, the correlation in NT measurements between the fetuses should be taken into account in the estimation of patient-specific risks.

Full text

Page 1

PRENATAL DIAGNOSIS
Prenat Diagn 2011; 31: 16–21.
Published online in Wiley Online Library
(wileyonlinelibrary.com) DOI: 10.1002/pd.2643
Screening for trisomies in dichorionic twins by measurement
of fetal nuchal translucency thickness according to the
mixture model
David Wright1, Argyro Syngelaki2,3, Ismini Staboulidou2, Jader de Jesus Cruz2,3
and Kypros H. Nicolaides2,3*
1Department of Mathematics and Statistics, University of Plymouth, Plymouth, UK
2Harris Birthright Research Centre of Fetal Medicine, King’s College Hospital, London, UK
3Department of Fetal Medicine, University College Hospital, London, UK
Objective
investigate the effect of the correlation between NT measurements in each twin pair on the performance of
screening for trisomies.
To examine the distribution of fetal nuchal translucency (NT) thickness in dichorionic twins and
Methods
pregnancies, including 103 with fetal trisomies 21, 18 or 13. The correlation in fetal NT in each euploid twin
pregnancy was estimated.
The distribution of fetal NT for crown–rump length (CRL) was examined in 5646 dichorionic twin
Results
in singleton pregnancies. In the euploid pregnancies, there was a correlation in log NT measurements in each
twin pair (r = 0.42, 95% CI: 0.39–0.45) and, after removal of the effect of the operator, this correlation was
reduced to 0.34. Allowing for this correlation in risk assessment for trisomies had a major impact on the
estimated patient-specific risk but had little effect on the overall performance of screening.
The distribution of NT in both euploid and trisomic fetuses was consistent with the mixture model
Conclusions
in singletons. In screening for trisomies, the correlation in NT measurements between the fetuses should be
taken into account in the estimation of patient-specific risks. Copyright  2011 John Wiley & Sons, Ltd.
In dichorionic twin pregnancies, the mixture model of distributions of NT can be applied as
KEY WORDS: first-trimester screening; trisomy 21; trisomy 18; trisomy 13; dichorionic twins; nuchal translucency
INTRODUCTION
In twin pregnancies, effective screening for trisomies
is provided by a combination of maternal age and
fetal nuchal translucency (NT) thickness (Pandya et al.,
1995; Sebire et al., 1996a,b; Maymon et al., 2001). In
dichorionic twins, patient-specific risks for trisomy 21
are calculated for each fetus on the basis of maternal age,
and the fetal NT and detection rate (75–80%) and false
positive rate (5% per fetus or 10% per pregnancy) are
similar to those in singleton pregnancies (Sebire et al.,
1996a). Therefore, effective screening and diagnosis of
major aneuploidies can be achieved in the first trimester,
allowing the possibility of earlier and therefore safer
selective fetocide in cases where one fetus is euploid
and the other is abnormal (Sebire et al., 1996b).
In the assessment of patient-specific risks for aneu-
ploidies, the a priori maternal-age-related risk is mul-
tiplied by the likelihood ratio determined from the
deviation of the measured NT from the expected median
for crown–rump length (CRL). There are essentially two
approaches for such quantification of the deviation in
*Correspondence to: Kypros H. Nicolaides, Harris Birthright
Research Centre for Fetal Medicine, King’s College Hospital,
Denmark Hill, London SE5 9RS, UK.
E-mail: kypros@fetalmedicine.com
NT. One approach is to subtract the normal median from
the NT measurement and to produce a deviation in mil-
limetres referred to as delta NT (Pandya et al., 1995;
Spencer et al., 2003). The other approach is to divide
NT by the normal median to produce a multiple of the
median (MoM) value (Nicolaides et al., 1998). Recently,
a new approach has been proposed, which is based on
the observation that in both aneuploid and euploid preg-
nancies fetal NT follows two distributions, one that is
CRL-dependent and another that is CRL-independent
(Wright et al., 2008). In this mixture model, the distri-
bution in which NT increases with CRL is observed in
approximately 95% of euploid fetuses, 5% with trisomy
21, 30% with trisomy 18 and 15% with trisomy 13. The
median CRL-independent NT is 2.0 mm for the euploid
group and 3.4, 5.5 and 4.0 mm for trisomies 21, 18 and
13 respectively.
In the calculation of risk for trisomies in dichorionic
twins, it was assumed that in each pregnancy the mea-
surements of NT for CRL between the two fetuses were
independent of each other. However, Wøjdemann et al.
(2006) examined 150 dichorionic twins with euploid
fetuses and reported that there was an association in the
measurements of NT in each twin pair with a correla-
tion coefficient of 0.32. Similarly, Cuckle and Maymon
(2010), reported a correlation coefficient of 0.43 in 246
dichorionic twins with euploid fetuses.
Copyright  2011 John Wiley & Sons, Ltd.
Received: 18 August 2010
Revised: 6 September 2010
Accepted: 6 September 2010

Page 2

NUCHAL TRANSLUCENCY IN TWINS
17
The aims of this study are to examine the distribution
of NT measurements in dichorionic twin pregnancies to
investigate the correlation between NT measurements in
twins, allowing for the effect of CRL and sonographer,
and to apply the mixture model to risk assessment
independently to each fetus, also allowing for the
correlation in the CRL-dependent process.
METHODS
At our centre, early screening for trisomies in twins
is based on maternal age and the measurement of
fetal NT at 11 to 13+6weeks’ gestation (Snijders
et al., 1998). Essentially, the maternal- and gestational-
age-related risk is multiplied by a likelihood ratio
corresponding to the deviation in fetal NT thickness
from the appropriate normal median for CRL. Fetal NT
and CRL are measured in each fetus and, in addition,
chorionicity is determined by examining the inter-twin
membrane at its junction with the placenta (Sepulveda
et al., 1996).
Demographic characteristics and ultrasound findings
were recorded in a fetal database at the time of the
examination. In the cases undergoing chorionic villous
sampling or amniocentesis, both fetuses were sampled
and the results of fetal karyotype were also entered
in the database when they were made available. Data
on pregnancy outcome were obtained from the patients
themselves, their general practitioners or the maternity
units in which they delivered.
A computer search was made to identify all dichori-
onic twin pregnancies that had NT screening between
February 1993 and February 2010. The selection cri-
teria were that, firstly, in each pregnancy both fetuses
were alive at the 11 to 13+6weeks’ scan and, sec-
ondly, the fetal karyotype was determined by prena-
tal invasive testing or the pregnancy outcome was
known.
Statistical analysis
In the twin pregnancies with two euploid fetuses, the
distribution of fetal NT for CRL was compared with
the percentiles from the mixture model (Wright et al.,
2008). Two models were developed for the calculation
of risk for trisomy 21. In model A, it was assumed
that in each pregnancy the measurements of NT for
CRL between the two fetuses were independent of each
other and the risks were computed according to the
mixture model with no modification for twins. The
parameters used to determine the risk were the same
as previously published (Wright et al., 2008). In model
B, the correlation of the CRL-dependent NT between
the two fetuses in each twin pair was estimated using the
method of moments. This correlation was used to modify
the mixture model for calculation of risks. Risks were
computed under the two models and used to produce
estimates of false positive rates and detection rates. For
model A, the calculations were identical to those for
singletons. For model B, risks were assigned to the pair
of twins using Bayes theorem incorporating a bivariate
Gaussian distribution for the CRL-dependent process.
A technical description of the model is given in the
Appendix.
RESULTS
The computer search identified 5646 dichorionic twin
pregnancies. The median maternal age was 34.7 (range
16–54) years, the median CRL of the bigger twin was
64.2 (range 45–84) mm and that of the smaller twin was
61.1 (range 45–84) mm. The median NT thickness was
1.7 (range 0.5–12.9) mm. There were 5530 pregnancies
with normal fetal karyotype or birth of phenotypically
normal babies, 103 with trisomies in one or both fetuses
(trisomy 21 in one fetus, n = 65; trisomy 18 in one
fetus, n = 22; trisomy 13 in one fetus, n = 7; trisomy
21 in both fetuses, n = 5; trisomy 13 in both fetuses,
n = 2; trisomy 21 in one fetus and trisomy 18 in the
other, n = 1; trisomy 21 in one fetus and trisomy 13
in the other, n = 1) and 13 with other aneuploidies
in one fetus that were excluded from further analysis
(Turner syndrome, n = 7; 47, XYY, n = 1; addition or
mosaicism, n = 5).
The relation between fetal NT and CRL in the euploid
and trisomic fetuses is shown in Figures 1 and 2. The
empirical estimates of the median and 95th and 99th
percentiles for the euploid fetuses, obtained from non-
parametric quantile regression, and the distribution of
values in the trisomic fetuses are consistent with those
from the mixture model in singleton pregnancies (Wright
et al., 2008).
After removing the effect of CRL, the estimated
correlation between log NT measurements in euploid
twins was 0.42 (95% CI: 0.39–0.45). After removal of
the effect of operator, this correlation was reduced to
0.34.
The relation of log MoM values for NT relative to the
CRL-dependent process between the two fetuses in each
twin pair is shown in Figures 3 and 4. The curves show
the likehood ratio contours for a trisomy 21 (in twin 1
or twin 2) relative to unaffected pregnancies under the
mixture model according to model A, which assumes
that the measurements in the two euploid fetuses are
independent of each other (Figure 3), and model B,
which assumes a correlation in the measurements of the
fetuses (Figure 4). In model B, the likelihood ratio for
trisomy 21 is relatively low compared to that in model
A when the NT is large in both fetuses, and it is high
when the NT in one fetus is large and that in the co-twin
is normal or small.
The performance of screening with a risk cut-off of
1 in 100 at the time of screening applied to each twin
under the two versions of the mixture model was similar
(Table 1). The estimated risk for trisomy 21 at 12 weeks’
gestation in twin 1 in a dichorionic twin pregnancy
where the maternal age is 35 years and the CRL of both
fetuses is 60 mm as shown in Table 2. The risk increases
with the NT thickness of twin 1 but this is modified by
the NT measurement of twin 2.
Copyright  2011 John Wiley & Sons, Ltd.
Prenat Diagn 2011; 31: 16–21.
DOI: 10.1002/pd

Page 3

18
D. WRIGHT et al.
Figure 1—Relation of fetal nuchal translucency thickness with crown–rump length in euploid fetuses. Percentiles obtained from applying
non-parametric quantile regression (full lines) and from the mixture model for singleton pregnancies (broken lines)
Figure 2—Nuchal translucency thickness in fetuses with trisomy 21 (left), trisomy 18 (middle) and trisomy 13 (right) plotted on the percentiles
(5th, 50th, 95th and 99th) for euploid twin fetuses
DISCUSSION
The findings of this study in more than 5500 dichorionic
twin pregnancies confirm the results of two previous
smaller studies (Wøjdemann et al., 2006; Cuckle and
Maymon, 2010) that NT measurements in euploid twins
are correlated and this correlation is not a simple
reflection of the common effect of sonographers.
High NT is associated with aneuploidies, cardiac
defects and a wide range of other fetal malformations
and genetic syndromes (Souka et al., 2005). However,
in most of the cases the fetuses are normal and it is likely
that the individual measurements would be affected by
parentally derived genetic factors as well as pregnancy-
specific environmental factors. Recent evidence, for
example, indicates that there is a significant association
between fetal NT and birth weight, with a small NT
increasing the risk of delivering small babies and a
large NT increasing the risk of delivering macrosomic
neonates (Poon et al., 2010, 2011). In this respect, the
finding that in normal dichorionic twins there is an
association in the NT of the two fetuses is not surprising.
In euploid dichorionic twins, the distribution of NTs
was consistent with the mixture model in singleton
pregnancies (Wright et al., 2008). In screening for
trisomies in twins, the standardised false positive rate for
each fetus is very similar to that in singleton pregnancies
but the overall false postive rate for twin pregnacies is
approximately double that of singleton pregnancies.
Copyright  2011 John Wiley & Sons, Ltd.
Prenat Diagn 2011; 31: 16–21.
DOI: 10.1002/pd

Page 4

NUCHAL TRANSLUCENCY IN TWINS
19
Figure 3—Scatter diagram of fetal nuchal translucency (NT) mea-
surements in twins expressed as multiples of the median (MoM) of
the crown–rump-length-dependent component of the mixture model
of distributions of NT in euploid fetuses. Sample of 1000 euploid
pregnancies (hollow circles) and the full sample of twin pregnancies
with a trisomy 21 fetus (full diamonds). The contours are for likeli-
hood ratios (one or more trisomy 21 twins/both euploid twins) of 1,
5 and 10 on the assusmption that the NT measurements in the two
fetuses are independent of each other
In our dataset, the detection rate of trisomies was
very high irrespective of whether the model used for
screening took into account the correlation in NT
measurements between the fetuses or not. Similarly,
inclusion of the correlation between NT measurements
had a very small impact on the overall false positive
rates. However, the correlation in NT measurements in
twins had a substantial impact in the estimated patient-
specific risk for trisomies. In the example given in
Table 2, the estimated risk of trisomy 21 for twin 1
with NT of 2.5 mm was 1 in 116 according to the
independent model, whereas with the dependent model
the risk increased to 1 in 48 if the NT in twin 2 was
1.0 mm and this decreased to 1 in 255 if the NT of twin
2 was 3.0 mm.
The difference between our model for estimation of
fetus-specific risk of aneuploidy, compared to the one
proposed by Cuckle and Maymon (2010), relates to the
Figure 4—Scatter diagram of fetal nuchal translucency (NT) mea-
surements in twins expressed as multiples of the median (MoM) of
the crown–rump-length-dependent component of the mixture model
of distributions of NT in euploid fetuses. Sample of 1000 euploid
pregnancies (hollow circles) and the full sample of twin pregnancies
with a trisomy 21 fetus (full diamonds). The contours are for likeli-
hood ratios (one or more trisomy 21 twins/both euploid twins) of 1,
5 and 10 on the assusmption that the NT measurements in the two
fetuses are correlated
method of estimating both the prior risk and likelihoods.
In the estimation of the prior risk, we took the pragmatic
approach of assuming that all dichorionic twins are dizy-
gotic and the risk for aneuploidies in each fetus is the
same as the maternal- and gestational-age-related risk
in singleton pregnancies. Cuckle and Maymon (2010),
attempted to adjust the risk according to the estimated
proportions of dizygotic and monozygotic twins depend-
ing on maternal age, racial origin, method of conception
and fetal gender. In the calculation of likelihood ratios,
we assumed that as in singleton pregnancies the distri-
bution of NT arises from a mixture of a CRL-dependent
and a CRL-independent process, because this provides a
better fit to the empirical data on both affected and unaf-
fected pregnancies. Cuckle and Maymon (2010) derived
likelihoods using the MoM approach assuming that log
MoM values follow Gaussian distributions for affected
and unaffected fetuses.
Table 1—Empirical screening performance in dichorionic twins with the mixture model of distributions in nuchal translucency
(NT) thickness and a risk cut-off of 1 in 100 at the time of screening applied to each twin
Screen
positive
Euploid/euploid,
n = 5530
Trisomy
21/euploid,
n = 65
Trisomy
18/euploid, n = 22
Trisomy
13/euploid,
n = 7
Trisomy/trisomya,
n = 9
Model A
Twin 1
Twin 2
Either twin
Model B
Twin 1
Twin 2
Either twin
362 (6.5)a
435 (7.9)b
618 (11.2%)c
62 (95.4%)
9 (13.8%)
62 (95.4%)
14 (63.6%)
4 (18.2%)
16 (72.7%)
6 (85.7%)
0
6 (85.7%)
9 (100%)
7 (77.8%)
9 (100%)
340 (6.1%)d
432 (7.8%)e
624 (11.3%)f
62 (95.4%)
8 (12.3%)
62 (95.4%)
14 (63.6%)
4 (18.2%)
16 (72.7%)
6 (85.7%)
0
6 (85.7%)
8 (88.9%)
7 (77.8%)
8 (88.9%)
In model A, it is assumed that the measurements of NT in the two euploid fetuses are independent of each other and, in model B, the correlation
in NT between the fetuses is taken into account.
Standardised rate: a = 3.3%, b = 4.7%, c = 6.6%, d = 2.9%, e = 4.5%, f = 6.5%.
aTrisomy 21/trisomy 21 (n = 5), trisomy 21/trisomy 18 (n = 1), trisomy 21/trisomy 13 (n = 1), both trisomy 13 (n = 2).
Copyright  2011 John Wiley & Sons, Ltd.
Prenat Diagn 2011; 31: 16–21.
DOI: 10.1002/pd

Page 5

20
D. WRIGHT et al.
Table 2—Estimated risk for trisomy 21 at 87 days of gestation
in the presenting fetus in a dichorionic twin pregnancy where
the maternal age was 35 years and the crown–rump length of
both fetuses was 60 mm
Estimated risk (1 in number below) for twin 1
Model B with different measurements
of NT in twin 2
NT twin 1
(mm)
Model
A
NT
1.0 mm
(0.61)
NT
2.0 mm
(1.21)
NT 3.0 mm
(1.82)
1.6 (0.97)
1.7 (1.03)
1.8 (1.09)
1.9 (1.15)
2.0 (1.21)
2.1 (1.27)
2.2 (1.33)
2.3 (1.39)
2.4 (1.46)
2.5 (1.52)
2.6 (1.58)
2.7 (1.64)
2.8 (1.70)
2.9 (1.76)
3.0 (1.82)
3.1 (1.88)
3.2 (1.94)
3.3 (2.00)
3.4 (2.06)
3.5 (2.12)
3.6 (2.18)
3.7 (2.24)
3.8 (2.30)
3.9 (2.37)
4.0 (2.43)
1685
1488
1246
991
750
543
380
259
174
116
79
55
40
30
24
20
17
15
14
13
12
11
10
9
9
1388
996
682
454
299
198
133
91
65
48
37
30
25
22
19
17
16
14
13
12
11
11
10
9
9
1598
1552
1418
1224
999
773
568
400
273
182
120
80
54
38
29
22
19
16
14
13
12
11
10
9
9
1476
1353
1199
1031
864
707
566
443
339
255
189
139
101
74
54
40
31
24
19
16
14
12
11
10
9
The risk increases with the nuchal translucency (NT) thickness of
twin 1. In model A, it is assumed that there is independence in
the measurements of NT in the two fetuses. In model B, there is
a correlation between the measurements of NT in the two fetuses.
The prior risk for a singe fetus with maternal age of 35 years is 1
in 305. MoM values relative to the crown–rump-length-dependent
process are shown in brackets.
The implications of the results of this study in
dichorionic twins are that, firstly, the mixture model
of distributions of NT can be applied as in singleton
pregnancies and, secondly, in screening for trisomies
the correlation in NT measurements between the fetuses
should be taken into account in the estimation of patient-
specific risks.
ACKNOWLEDGEMENT
This study was supported by a grant from the Fetal
Medicine Foundation (Charity No: 1037116).
APPENDIX: RISK CALCULATIONS FOR
DICHORIONIC TWIN PREGNANCIES
If we assume that each fetus either has trisomy 21 or
is euploid, the possible outcomes for twin 1 and twin
2 can be represented by the four cells in a 2 × 2 table.
The final or posterior risk of the four possible outcomes
are obtained from applying Bayes theorem to combine
prior probabilities with the likelihoods for the four table
entries. If another condition, such as trisomy 18, is to be
included, then a 3 × 3 table is necessary.
Prior probabilities
In the calculation of the prior probabilities, it is assumed
that trisomy 21 arises independently and the probability
of each twin being affected is the same as that of a sin-
gleton pregnancy and depends on the maternal age and
gestation. This leads to a prior probability distribution
over the possible outcomes given in Table A1, where π
denotes the maternal- and gestational-age-specific risk
of trisomy 21.
Likelihoods
The likelihoods for the four cell entries are obtained
from the mixture model. Each cell is a mixture of
Gaussian probability densities according to whether or
not the NT measurement for the fetus arises from
the CRL-dependent or CRL-independent component as
described in Wright et al. (2008). The mixing proba-
bilities, which we denote by p1and p2for twin 1 and
twin 2 respectively, that the measurement arises from
the CRL-independent component give rise to the entries
in Table A2.
The likelihood for the four possible outcomes is all of
the form
(1 − p1)(1 − p2)ϕD1D2+ p1(1 − p2)ϕI1ϕD2
+ (1 − p1)ϕ2ϕD1ϕI2+ p1p2ϕI1ϕI2
In the above expression, ϕD1D2is a bivariate Gaussian
probability density for the situation where both NT mea-
surements arise from the CRL-dependent process. The
mean and standard deviations describing this distribution
are the same as those of Wright et al. (2008). In model
B, it is assumed that the bivariate Gaussian distribution
has a non-zero correlation coefficient.
(1)
Table A1—Prior probabilities under the independence model
Twin 2
Twin 1
Euploid Trisomy 21
Euploid
Trisomy 21
(1 − π)(1 − π)
(1 − π)π
π(1 − π)
π2
Table A2—Mixture probabilities
Twin 2
Twin 1
CRL-dependent CRL-independent
CRL-dependent
CRL-independent
(1 − p1)(1 − p2)
(1 − p1)p2
p1(1 − p2)
p1p2
Copyright  2011 John Wiley & Sons, Ltd.
Prenat Diagn 2011; 31: 16–21.
DOI: 10.1002/pd