The objective of this study was to describe and illustrate a method for calculating fetus-specific Down syndrome risk in twins, allowing for between-fetus nuchal translucency (NT) correlation.
The objective of this study was to describe and illustrate a method for calculating fetus-specific Down syndrome risk in twins, allowing for between-fetus nuchal translucency (NT) correlation.
The between-fetus correlation coefficient of log NT, in multiples of the median, was estimated from a series of 325 unaffected twins after adjustment for sonographer bias. A bivariate log Gaussian model was used to calculate likelihood ratios for discordant and concordant Down syndrome. Applying these to the prior maternal age-specific risk yielded risks in monozygous and dizygous twins. The weighted average risk was then computed with weights relating to chorionicity, gender, assisted reproduction and ethnicity. The method was illustrated using examples.
The correlation coefficient in unaffected pregnancies was 0.45 (P < 0.0001) and estimated to be 0.12 and 0.04 in discordant and concordant twins, respectively. The examples showed very large differences in the risks obtained when the extent of correlation in NT between fetuses is taken into account and when the measurements are treated as independent.
Fetus-specific Down syndrome risks in twins should be calculated using its own NT value as well as that of the co-twin. Copyright © 2010 John Wiley & Sons, Ltd.
Maternal serum screening for Down syndrome has much lower performance characteristics in twin pregnancies compared with singletons. For example, modelling predicts that in 30-year-old women, second-trimester α-fetoprotein, unconjugated estriol (uE3), free-β human chorionic gonadotrophin (hCG) and inhibin with a 1 in 250 term cut-off risk will detect just one-quarter of affected twin pregnancies compared with two-thirds in singletons, albeit with fewer false positives (Cuckle and Benn, 2010). The reason for the poorer results is that in twins that are discordant for Down syndrome, feto-placental products from the unaffected fetus can mask the abnormal levels produced by the affected fetus.
In contrast, first-trimester ultrasound nuchal translucency (NT) screening is fetus specific and is therefore regarded as the method of choice for twins. The NT and crown–rump length (CRL) measurements of each fetus are used to calculate the chance that it is affected by Down syndrome in exactly the same way as a singleton. Even when maternal serum markers are measured concurrently, the performance is comparable with singletons. In the above example, if screening was carried out at 11 weeks with NT, free-β hCG and pregnancy-associated plasma protein-A, the modelling predicts that 72% of affected twins would be detected compared with 77% in singletons with a virtually identical false-positive rate (Cuckle and Benn, 2010).
However, the fetus-specific method of calculating Down syndrome risk in twins assumes that the NT measurements in the two fetuses are independent and this is now known to be incorrect. Wøjdemann et al. (2006) in a series of 181 unaffected twins from Denmark reported a correlation coefficient of 0.34 between the pairs of NTs, expressed in log multiples of the median (MoMs) for CRL. A similar substantial degree of correlation was found both in the 31 monochorionic (MC), with a correlation coefficient of 0.40, and in the 150 dichorionic (DC) twins, with a correlation coefficient of 0.32. The article by Wøjdemann et al. (2006) does not specify how many sonographers participated or whether there were any systematic differences between them in singleton pregnancies. Some, or potentially all, correlations could have been accounted for by differences between sonographers, with those tending to over-measure NTs doing so for both fetuses and vice versa.
In this study, we demonstrate, in a large series of twins, that there is a correlation even when NT is measured by a single sonographer. The observed correlation coefficient, allowing for sonographer differences, was then used to describe and illustrate a method for calculating the Down syndrome risk in a twin fetus using its own NT and that of the co-twin.
Between February 2003 and October 2009, a total of 325 women with unaffected twin pregnancies had NT measurements at Leeds Screening Centre, UK. The scans were carried out by four sonographers, two of whom were medically qualified; all were certified by the Fetal Medicine Foundation. NT was expressed in MoMs using a previously published formula (Logghe et al., 2003). Correlation coefficients between the log MoMs were calculated after excluding outliers exceeding the median by at least 3 standard deviations, based on the 90th and 10th centile difference divided by 2.563. A correlation coefficient allowing for sonographer differences was derived by analysis of variance.
There are four steps in the calculation: (1) prior probability of Down syndrome in fetus 1 only, fetus 2 only or both fetuses according to maternal age and zygocity; (2) likelihood ratios of Down syndrome in fetus 1 only, fetus 2 only or both, from the two NT MoMs; (3) proportion of dizygotic (DZ) and monozygotic (MZ) twins given chorionicity, gender, whether assisted reproduction or spontaneous, maternal age and ethnicity and (4) multiply the prior probabilities by likelihood ratios and take the weighted average using the zygocity distribution.
For a woman aged x, the maternal age-specific probability of Down syndrome in singletons (Px) is obtained from the birth prevalence curve derived by meta-analysis (Cuckle et al., 1987). In DZ twins, the prior probability of an individual fetus being affected is assumed to be the same as for a singleton, Px, while the probability that the other fetus is affected is assumed to be the probability of Down syndrome recurrence in singletons, Px + 0.42% at term (Cuckle and Benn, 2010). It follows that in DZ twins, the probability of both fetuses being affected (Px, DZ, both) is Px(Px + 0.42%), whereas the probability of fetus 1 being affected and fetus 2 unaffected (Px, DZ, 1only) is Px − Px, DZ, both and is the same for fetus 1 unaffected and fetus 2 affected (Px, DZ, 2only). At mid-trimester, the recurrence factor is 0.54% instead of 0.42% (Cuckle and Benn, 2010). In MZ twins, both fetuses are assumed to be affected and the probability (Px, MZ) is Px.
These are calculated for the following: fetus 1 affected and fetus 2 unaffected (LR1only); fetus 1 unaffected and fetus 2 affected (LR2only) and both fetuses affected (LRboth). The LRs are from bivariate log Gaussian distributions of the NT MoMs in fetus 1 and fetus 2 with means and standard deviations used for singleton pregnancies. The general formula is:
where s is standard deviation; r is correlation coefficient; z is (log10MoM - mean)/s; u/a is unaffected/affected; 1/2 represents fetus and e[x] represents ex.
The Down syndrome means at 11, 12 and 13 weeks of gestation were those derived by meta-analysis, 2.30, 2.10 and 1.92 MoM or 0.362, 0.322 and 0.283 log10 MoM, respectively (Cuckle and Benn, 2010). The standard deviation of log10 MoM in unaffected fetuses was 0.070, the value found at Leeds Screening Centre among unaffected singleton pregnancies, and in Down syndrome it was 0.21, derived by adding 0.04 to the unaffected variance (Spencer et al., 2003). For example, at 12 weeks of gestation, the affected pregnancy parameters when calculating LR1only are 0.322 and 0.21 for fetus 1, and 0 and 0.070 for fetus 2; the corresponding unaffected parameters are 0, 0.070, 0 and 0.070. The correlation coefficient between the log10 MoMs in unaffected pregnancies was that derived from Leeds Screening Centre data allowing for sonographer differences. In affected twins, it is assumed that the covariance is the same as for unaffected pregnancies.
The proportions of twins which are DZ and MZ are used as weights (WDZ and WMZ) in the final step of risk calculation. This is dependent on ultrasound-determined chorionicity, fetal gender differences, whether assisted reproduction or spontaneous pregnancies, maternal age and ethnicity. The proportions are summarised in Table 1.
|DC, different gender||100%||0%|
|DC, same gender|
|Spontaneous, Caucasian||0.51 × z/[0.51 × z + 0.29 × 0.4]||0.29 × 0.4/[0.51 × z + 0.29 × 0.4]|
|Spontaneous, Afro-Caribbean||1.54 × 0.51 × z/[1.54 × 0.51 × z + 0.29 × 0.4]||0.29 × 0.4/[1.54 × 0.51 × z + 0.29 × 0.4]|
Chorionicity can be determined by ultrasound examination of the fetal membranes (Stenhouse et al., 2002). The so-called λ sign, caused by invasion of the inter-twin membrane by chorionic villus, is evidence of dichorionicity and either the presence of one amniotic sac or diamniotic sacs with a ‘T’ sign is evidence of monochorionicity.
All MC twin pregnancies can be taken to be MZ: WDZ 0% and WMZ 100%. All DC pregnancies where the fetuses have different genders can be taken to be DZ: WDZ 100% and WMZ 0%.
In DC pregnancies with the same gender which are the result of IVF and embryo transfer, the proportions will depend on the number of embryos being transferred.
A study from the Society for Assisted Reproductive Technology (SART) in the United States analysed data on 39 198 pregnancies resulting from IVF in which information was available on the number of fetal hearts seen on ultrasound (Wright et al., 2004). More than 98% resulted from multiple embryo transfer (MET). There were 15 307 multiple pregnancies, 226 (1.5%) with more fetal hearts than the number of embryos transferred, classified by the authors as MZ, and the remainder with the same number or fewer fetal hearts than embryos transferred were DZ. MET was used in 185 of the presumptively MZ pregnancies and, although not stated in the report, virtually all other multiple pregnancies. In a study of zygocity and chorionicity in 300 twins, 28% (42/152) MZ twins were DC and 58% (86/148) DZ twins had the same gender (Machin et al., 1995). In the much larger East Flanders Prospective Twin Survey, the proportions were 29% (494/1720) and 49% (1487/3047), although assigning the 277 same gender twins of unknown zygocity proportionally as MZ and DZ, the latter is 51% (1615/3175) (Loos et al., 1998). Applying the East Flanders proportions to the SART results, in DC twins of the same gender following IVF with MET, the proportions of DZ and MZ can be taken to be WDZ 99.3% and WMZ 0.7%.
The SART study does not include sufficient information on zygocity following single-embryo transfer (SET). However, the expectation is that all would be MZ with WDZ 0% and WMZ 100%.
In spontaneous DC twins with the same gender, the proportions which are DZ and MZ will depend on maternal age and ethnicity. The incidence of MZ twinning in spontaneous pregnancies has been estimated to be 0.4% (Vitthala et al., 2009) and this is unrelated to maternal age (Bortolus et al., 1999) or ethnicity (National Center for Health Statistics, 1992). In contrast, DZ twinning increases with age (Meyers et al., 1997) and varies according to ethnicity. Using national data on twin live births in the United States in 1980 (Russell et al., 2003), we calculated the age-specific DZ twinning incidence by subtracting 0.4%. Standardising for the maternal age distribution in the Caucasian women, the DZ twinning incidence was 54% higher in the Afro-Americans. The DZ twining is approximately half as common in women of Oriental origin than in Caucasians (Bulmer, 1970; Elwood, 1978).
The number of maternities and births at each single year of age in England and Wales during 1960 (General Register Office, 1962), before the era of assisted reproduction, was used to calculate the incidence of twinning in Caucasians, expressed in percentage. A cubic regression of incidence on age weighted for the number of maternities yielded: 3.602 − 0.4236x + 0.01871x2 − 0.0002344x3, where x is the maternal age. The incidence of DZ with the same gender twinning is then estimated by subtracting 0.4% from the regressed incidence and multiplying by 51%, the proportion of DZ with the same gender (Loos et al., 1998), and the incidence of MZ twins that are DC is 0.4% multiplied by 29%. Then WMZ and WDZ are the relative proportions of these incidences. For example, at maternal age 20, WDZ is 60% and WMZ is 40%, whereas at age 40, WDZ is 84% and WMZ is 16%; in Afro-Caribbeans, the four proportions would be 70, 30, 89 and 11%.
The four maternal age-specific probabilities, Px, DZ, 1only, Px, DZ, 2only, Px, DZ, both and Px, MZ, are expressed as odds (i.e. in the form P: 1 − P) and multiplied by the appropriate likelihood ratios LR1only, LR2only and LRboth; the result is re-expressed in percentage and the weighted average taken using WDZ and WMZ. LRs express the relative probability of being affected compared to unaffected, for a given marker profile, hence the need to express prior probabilities as odds at that stage in the calculation.
The probability that fetus 1 is affected is then, multiplying odds and probabilities as appropriate, WDZ (Px, DZ, 1only LR1only + Px, DZ, both LRboth) + WMZPx, MZ LRboth and the probability for fetus 2 is WDZ (Px, DZ, 2only LR2only + Px, DZ, both LRboth) + WMZPx, MZ LRboth.
Figure 1 shows the individual pairs of NT MoM values for all 325 twins. The overall median was 0.99 MoM, the log10 standard deviation was 0.072 and the correlation coefficient of log MoMs after excluding outliers was 0.43 (P < 0.0001). According to operator, the correlation coefficient was 0.43 for 187 women scanned by one sonographer; 0.43, 62 by another sonographer; 0.58, 45 by a fetal medicine consultant and 0.14, 31 by another consultant. Analysis of variance yielded a correlation coefficient of 0.45 when stratified by operator and a covariance of 0.001811. There were 114 twins where the pregnancy followed IVF. The correlation coefficient was 0.52 compared with 0.38 in the remaining twins, not a statistically significant difference (P = 0.44). Ultrasound information on chorionicity was available for 299 twins; the correlation coefficient was 0.43 for the 246 DC cases compared with 0.35 for the MC, also not statistically significant (P = 0.76).
The correlation coefficients in affected twin pregnancies are calculated by dividing the covariance in the unaffected pregnancies by the appropriate standard deviations. In those discordant for Down syndrome, it is 0.12 [0.001811/(0.21 × 0.070)], and in those which are concordant, it is 0.04 (0.001811/0.212).
Table 2 shows the fetus-specific risk in three examples where the NT is raised in one fetus but not in the other, or raised to a lesser extent. For each example, for the women aged 25 years, the NT scan was taken at 12 weeks but a range of possible ultrasound and clinical findings are considered. Before taking the scan, the maternal age-specific odds of Down syndrome are 1:1357 for fetus 1 alone, 1:1357 for fetus 2 alone, 1:273 618 for both fetuses in DZ and 1:1351 in MZ twins. To illustrate the method, consider Example 3 where the LRs are 12.40, 0.1210 and 1.322 for fetus 1 alone, fetus 2 alone and both fetuses, respectively. Therefore, after the scan, the four odds become 1:109, 1:11 215, 1:206 973 and 1:1022, or 0.9091, 0.0089, 0.0005 and 0.0978%, respectively. If this twin is DC and the fetuses are of the same gender, then as the age-specific twinning incidence is 1.04%, the DZ and MZ weights are 73.88 and 26.12%. Hence, the probability of Down syndrome in fetus 1 is 73.88% of (0.9091 + 0.0005%) plus 26.12% of 0.0978%, which is 0.6976% or 1 in 140; for fetus 2, it is 73.88% of (0.0089 + 0.0005%) plus 26.12% of 0.0978%, which is 0.0325% or 1 in 3100.
|Example 1||Example 2||Example 3|
|Fetus 1||Fetus 2||Fetus 1||Fetus 2||Fetus 1||Fetus 2|
|1.7 MoM||1.5 MoM||1.7 MoM||1.0 MoM||1.5 MoM||1.0 MoM|
|DC, different gender||310||2800||6||4400||110||11 000|
|DC, same gender|
|IVF, MET||300||2000||10||2800||110||10 000|
|Independent||20||220||20||13 000||220||13 000|
The table demonstrates that there are very large differences in the risks obtained when the extent of correlation in NT between fetuses is taken into account and when the measurements are treated as independent. In some of these situations, it would have made a clinical difference and could have tipped the balance for the patient over whether to have invasive prenatal diagnosis.
In practice, when there is ultrasound evidence of an MC pregnancy, some use the average of the two NT MoM values to calculate risk. However, in the three examples, this would not have yielded a correct risk: 1.7 and 1.5 MoM, 1 in 70; 1.7 and 1.0 MoM, 1 in 1100; 1.5 and 1.0 MoM, 1 in 2800. Some take the average of the two risks but this also yields inaccurate risks, even if the geometric mean is used: 1 in 70, 1 in 510 and 1 in 1700, respectively.
We have confirmed that there is a correlation between the NT values in twin fetuses and obtained an unbiased estimate of the correlation coefficient. We have also described in detail a method to calculate Down syndrome risk in a twin fetus using its own NT and that of the co-twin. These risks can be substantially different from values obtained from the current method which incorrectly assumes that the two NTs are independent.
In addition to the series from Denmark (Wøjdemann et al., 2006) where the correlation coefficient was 0.34, and our own where allowance was made for different operators and the correlation coefficient was 0.46, there is a small series where 29 paired NT and CRL values in unaffected MC twin pregnancies are tabulated (Casasbuenas et al., 2008). Using the normal median curve for Leeds (Logghe et al., 2003) to convert results into MoMs, the correlation coefficient of log MoM was 0.18.
Second-trimester severe twin–twin transfusion syndrome is not uncommon in MC twins and a large disparity in NT thickness between the fetuses is regarded as an early marker for the syndrome (Sebire et al., 1997). This could account for the mean disparity of 0.65 mm found in 105 MC twins compared with 0.49 mm in 176 DC twins (Cheng et al., 2010). However, this does not appear to be sufficient for the correlation coefficient to be lower in MC twins because the Danish series found the reverse (Wøjdemann et al., 2006). We have therefore used a single correlation coefficient for all types of unaffected twins.
There is little information on which to assess the correlation coefficient in affected twin pregnancies, whether discordant or concordant for Down syndrome. In the largest published study, in France, the NT and CRL values of 14 affected twins, all but one discordant, were tabulated (Garchet-Beaudron et al., 2008). Using the normal median curve for Leeds to convert results into MoMs, the correlation coefficient of log MoM in the discordant pairs was 0.89 (P < 0.0001). Sixteen additional cases with NTs in twins discordant for Down syndrome are available from the literature—5 in MoMs (Spencer and Nicolaides, 2003; Goncé et al., 2005) and 11 where NT and CRL pairs are given (Pandya et al., 1995; Sperling et al., 2007; Linskens et al., 2009)—and there was one screened at Leeds Screening Centre in the same period as the current series with NTs of 1.55 and 0.90 MoM. Combining the 14 French cases with these 17 cases using the latest Fetal Medicine Foundation curve (Wright et al., 2008) to convert the NT–CRL pairs into MoMs, the correlation coefficient of log MoM was 0.29 (P = 0.13).
The French series is biased as a criterion for inclusion was that second-trimester maternal serum screening should be performed. Hence, cases with elevated NT would have tended to be excluded. Indeed, taking the largest MoM in each discordant pair to be from the affected fetus, the median value in the 15 affected fetuses was only 1.25 MoM and in the 13 unaffected fetuses was 1.03 MoM. Moreover, some of the correlations in the combined data of 31 cases would have been due to the inclusion of many sonographers, particularly as the French series was collected nationwide. In our method, we have assumed that the affected twins have the same covariance as the unaffected twins which yielded a correlation coefficient of 0.12 for discordant pairs which is consistent with the combined result taking into account these considerations.
There have been two reports in which the proportion of fetuses with NT thickness above the estimated CRL-specific 95th centile was higher for twins than in singletons. In a study of 448 twins, the proportions were 7.3 and 5.4% (Sebire et al., 1996) and in a study of 100 twins, they were 9.0 and 4.2% (Monni et al., 2000). In a third study which included 174 twins, only 5.4% of fetuses had NT above the 95th centile, although the comparable proportion in singletons was not given (Maymon et al., 2001). A high proportion of large NTs in twins would suggest that either the average value of NT was higher than in singletons or there was a wider spread of values. However, in the Leeds Screening Centre series, neither the median (0.99 MoM) nor the log10 standard deviation (0.072) is greater than for singletons. Hence, in our method, we have used the same median NT MoM and log standard deviations for affected and unaffected twin fetuses as for singletons.
There have been suggestions that the NT thickness is higher in MC twins than DC, possibly due to twin–twin transfusion (Sebire et al., 1996; Monni et al., 2000). However, others have found no significant differences according to the chorionicity in the screen-positive rate based on the NT (Maymon et al., 2001) or the median NT (Linskens et al., 2009). In a small series of 30 twins resulting from IVF or ICSI, the median NT was 10% lower than in 150 spontaneous twins, although not statistically significant (Orlandi et al., 2002). In another series, restricted to DC twins, the median NT was 1.02 in 54 resulting from assisted reproduction technologies and 1.07 in 38 spontaneous controls, again not statistically significant (Hui et al., 2006).
There are no reliable data to directly estimate the age-specific probability of having twins concordant for Down syndrome and we have used an indirect approach. In a study from the National Down Syndrome Cytogenetic Register of England and Wales for the period 1989 to 1993, among 72 affected twins registered, 9 (12.5%) were concordant for Down syndrome, all of which had the same gender (Mutton et al., 1996). Using our approach at 32.7 years, the mean registered maternal age, the proportion expected to be concordant is 13.9% and of the same gender 13.8%. In another series of 27 affected twins with median age 35.5, 7 (25.9%) were concordant (Garchet-Beaudron et al., 2008) and using our approach the expected proportion was 12.7%.
We have assumed that the prior probability of an individual twin fetus having Down syndrome is the same as for a singleton. In a French nationwide study of Down syndrome screening in twins in 1998 to 2006, 34 of 22 080 fetuses were affected, an incidence of 1.5 per 1000 compared with 1.3 per 1000 in singletons (Garchet-Beaudron et al., 2008). Taking into account the maternal age difference—1 year older on average in the twins—and concordance in DZ twins, this is consistent with our assumption. Using national registration records in England and Wales in 1979 to 1985, 42 of 74 844 twin fetuses were found to have been registered as having Down syndrome (Doyle et al., 1990). After age standardisation to a cohort of singleton births, the incidence was 0.50 per 1000 compared with 0.74 in the singletons. However, notification of congenital malformations is voluntary and there is considerable underreporting of Down syndrome. This is partly due to notification having to be made before 7 days of age, so under-ascertainment is likely to be greater for twins as a larger proportion will be stillborn. In a study of national registration records in Norway in 1967 to 1979, there were 13 affected twin fetuses out of 15 320, 0.85 per 1000 compared with 1.01 per 1000 in singletons (Windham and Bjerkedal, 1984). In Norway, registration is restricted to the time of birth allowing more chance of biased under-ascertainment. The only other large study of malformations in twins, in the United States, was based entirely on birth certificates, hence with substantial under-ascertainment of Down syndrome, and restricted to live births (Hey and Wehrung, 1970). In twins following SET, we have assumed that all are MZ. In one study of SET where zygocity is explicitly stated, the only twin is reported as MZ (Gerris et al., 1999); however, in another study, there were three twins of which one was reported as DZ and a fourth which spontaneously reduced to a singleton was also reported as DZ (Sundström and Saldeen, 2009). The most likely explanation is that these two cases were in fact DC as the authors do not discuss this unexpected finding or state how they established zygocity. If two of five were truly DZ, then applying the same logic as we used for MET, in DC twins of the same gender following IVF with SET, the weights would be taken to be WDZ 46% and WMZ 54%.
In spontaneous twins, we have used information on maternal age-specific incidence of twinning in 1960, prior to the widespread use of assisted reproduction technologies, to estimate WDZ and WMZ. Similarly, data from 1980 were used to estimate these weights in women of Afro-Caribbean origin. There is also evidence that the prevalence of twinning is higher in first-degree relatives of twins (Bortolus et al., 1999). This factor was not used in the method but it could readily be included.
Fetal gender can be determined by ultrasound at the time of the NT scan. However, if this information is not available, different WDZ and WMZ values are needed. In DC twins where IVF was carried out with MET, the values are 99.6 and 0.4%. For spontaneous pregnancies, the formulae in Table 1 are modified by deleting the 0.51 factor.
As more data become available, our method for calculating Down syndrome risk from NT in twins can be improved. For example, more directly observed values will establish more accurate estimates of the correlation coefficient in twins discordant and concordant for Down syndrome. Meanwhile, the parameters in this article are reliable enough for our method to immediately replace the current clinical practice which incorrectly assumes that the NT measurements in the two fetuses are independent. Commercial software used in Down syndrome screening will need to be modified accordingly, but until this has been done we will provide an interactive risk calculator at www.screeninfo.co.uk.
We thank the staff of Leeds Screening Centre for carrying out the ultrasound examinations of the twins and Peter Benn for his insightful comments.