Abstract
Objective
To determine whether the improved precision of nuchal translucency (NT) measurement used in antenatal screening for Down’s syndrome observed over time as evidenced by a decrease in the multiple of the median (MoM) standard deviation requires a modification to the NT MoM truncation limits to maintain accurate risk estimation.
Methods
Probability plots were derived from the measurements of NT MoM values used in a 2018 audit of 22,362 unaffected pregnancies. The plots were used to determine whether the NT MoM upper truncation limit should be lowered. Validation plots were used to assess the screening accuracy of Down’s syndrome risk estimates calculated from observed NT MoM values in the 22,362 unaffected pregnancies and 69 Down’s syndrome pregnancies for original and revised NT MoM truncation limits.
Results
Probability plots indicated that with improved precision of NT measurements, there was deviation from a Gaussian distribution at less high MoM values than with less precise measurements. Validation plots showed that using the current NT MoM upper truncation limit of 2.5 MoM with improved precision NT measurements overestimates the Down’s syndrome risk (median risk in highest risk category expressed as an odds was 53.3:1 and observed prevalence was 1:1.1). The large discrepancy was corrected by changing the NT upper truncation limit to 2.0 MoM (median risk in highest risk category expressed as an odds was 1:1.78 and observed prevalence 1:2.7).
Conclusion
The NT MoM upper truncation limit should be reduced from 2.5 to 2.0 MoM.
Keywords: Down’s syndrome, antenatal screening, nuchal translucency, truncation limit, prenatal screening, validation plot
Introduction
Measurement of nuchal translucency (NT) is an important marker in antenatal screening for Down’s syndrome, trisomy 18 and trisomy 13. Over time, the precision of measuring NT has improved leading to a reduction in the standard deviation. For example, at 12 completed weeks’ gestation, the log10 NT MoM (multiple of the median) standard deviation was 0.1329, 0.1105 and 0.0855 in 2003,1 20102 and 2018,3,4 respectively, as reported from the Wolfson Institute Screening Service. The distribution of NT is positively skewed; after log transformation, the distribution is reasonably Gaussian, but there is still a degree of positive skew.5 To deal with screening markers deviating from Gaussian distributions in the tails of the distributions, it is standard practice to specify truncation limits in which values above or below these limits are assigned values at the truncation limits.6,7
As NT measurement has become more precise, there is a case for lowering the upper truncation limit from the 2.5 MoM value2 set in 2010 to take account of the positively skewed distribution of log-transformed NT MoM values. We here examined data from the Wolfson Institute of Preventive Medicine Antenatal Screening Programme to determine whether there was a need to revise the NT upper truncation limit.
Methods
Data from a 2018 audit of antenatal screening were used based on measurements from 22,362 unaffected singleton pregnancies and 69 Down’s syndrome pregnancies.8 A probability plot was generated for the NT MoM values and inspected to determine if the upper truncation limits should be lowered from the currently specified value of 2.50 MoM.2 A term Down’s syndrome risk estimate was calculated from the NT measurements for each of the 22,362 unaffected singleton pregnancies and 69 affected singleton pregnancies. The risk estimates were calculated by previously described standard methods9 from the overlapping log10 NT MoM Gaussian distributions of affected and unaffected pregnancies, using published sets of NT distribution parameters2,4,10 (see Appendix 1), the observed overall Down’s syndrome prevalence in the sample population and the currently specified NT upper truncation limits of 2.50 MoM and 2.0 MoM.
To assess the accuracy of risk estimation using the two NT upper truncation limits, validation plots11,12 were created in which the risk estimates were ranked and put into seven categories with approximately equal numbers of affected pregnancies in each category. Within each category, the median risk and the observed prevalence of Down’s syndrome were calculated. The accuracy of risk estimation is displayed by plotting the median risk against the observed prevalence in each category. Points lying close to the diagonal line of identity (where predicted risk equals observed prevalence) indicate the accuracy of risk estimation.
Statistical analyses were performed using RStudio (RStudio Team (2019). RStudio: Integrated Development for R. RStudio, Inc., Boston, MA URL http://www.rstudio.com/). NT MoM values calculated at the time of screening with the Alpha software (Logical Medical Systems, London, UK) were used.
Results
Figure 1 shows the NT MoM probability plot. Deviation from a log-Gaussian distribution is apparent at an NT measurement of about 2.0 MoM rather than 2.5 MoM.
Figure 1.
Probability plot of NT measurements in unaffected pregnancies from 2018 study.8 The two horizontal lines show the positions of the NT upper truncation limits at 2.0 and 2.5 MoM.
MoM: multiple of the median.
Figure 2(a) and (b) shows validation plots for Down’s syndrome risk estimates calculated using the current NT upper truncation limits of 2.5 MoM and 2.0 MoM, respectively. Table 1 gives the numerical results shown in Figure 2(a) and (b). When the NT upper truncation limit is 2.5 MoM, the risk estimates are inaccurate at high risks (greater than about 1 in 10). In the highest risk category, over 50 times more affected pregnancies are predicted than observed: the predicted risk estimate and the observed prevalence, expressed as an odds (affected:unaffected) are 53.1:1 and 1:1.1, respectively.
Figure 2.
Validation plots for Down’s syndrome risk estimates expressed as probabilities (1 in x) calculated using NT measurements from the 2018 audit,8 with the current reduced unaffected NT standard deviation.4 (a) Validation plot with NT upper truncation limit of 2.5 MoM. (b) Validation plot with NT upper truncation limit of 2.0 MoM. (The points plotted in the two highest risk categories are nearly superimposed.).
Table 1.
Numerical results for the validation plot in Figure 2.
|
Predicted risk |
Observed |
|||||
|---|---|---|---|---|---|---|
|
Median |
Prevalence |
|||||
| Category | Probability | Odds | Down’s Syndrome | Unaffected | Proportion | Odds |
| (a) NT upper truncation limit = 2.5 MoM | ||||||
| >1 in1.0195 | 1 in 1.0188 | 53.3:1 | 10 | 11 | 1 in 2.1 | 1:1.1 |
| 1 in 1.0195– | 1 in 1.166 | 6.0:1 | 10 | 62 | 1 in 7.2 | 1:6.2 |
| 1 in 1.26– | 1 in 12.0 | 1:11.0 | 10 | 180 | 1 in 19.0 | 1:18.0 |
| 1 in 35.4– | 1 in 141.2 | 1:140.2 | 10 | 630 | 1 in 64.0 | 1:63.0 |
| 1 in 245– | 1 in 472 | 1:471 | 10 | 1571 | 1 in 158 | 1:157 |
| 1 in 658– | 1 in 1077 | 1:1076 | 10 | 4439 | 1 in 445 | 1:444 |
| <1 in1352 | 1 in 1816 | 1:1815 | 9 | 15469 | 1 in 1720 | 1:1719 |
| All | 1 in 1661 | 1:1660 | 69 | 22362 | 1 in 325 | 1:324 |
| (b) NT upper truncation limit = 2.0 MoM | ||||||
| >1 in 2.7864 | 1 in 2.7831 | 1:1.7831 | 10 | 27 | 1 in 3.7 | 1:2.7 |
| 1 in 2.7864– | 1 in 2.796 | 1:1.796 | 10 | 30 | 1 in 4.0 | 1:3.0 |
| 1 in 2.91– | 1 in 10.7 | 1:9.7 | 10 | 196 | 1 in 20.6 | 1:19.6 |
| 1 in 35.4– | 1 in 141.2 | 1:140.2 | 10 | 630 | 1 in 64.0 | 1:63.0 |
| 1 in 245– | 1 in 472 | 1:471 | 10 | 1571 | 1 in 158 | 1:157 |
| 1 in 658– | 1 in 1077 | 1:1076 | 10 | 4439 | 1 in 445 | 1:444 |
| <1 in 1352 | 1 in 1816 | 1:1815 | 9 | 15,469 | 1 in 1720 | 1:1719 |
| All | 1 in 1661 | 1:1660 | 69 | 22,362 | 1 in 325 | 1:324 |
NT: nuchal translucency; MoM: multiple of the median.
Reducing the NT upper truncation limit to 2.0 MoM brings the plotted points closer to the diagonal line of identity and thereby improves the accuracy of the risk estimates. With an upper truncation limit of 2.0 MoM in the highest risk category, the predicted risk estimate and the observed prevalence (both expressed as an odds) are 1:1.78 and 1:2.7, respectively.
Discussion
This study shows that the NT MoM upper truncation limit should be revised from 2.5 to 2.0 MoM as a result of the improved prevision of NT measurements over time. The less high upper truncation limit improves the accuracy of risk estimation in antenatal screening for Down’s syndrome. Retaining the 2.5 MoM upper truncation limit overestimates risk estimation for pregnancies with high NT MoM values. Although such pregnancies would probably be screen-positive, a truncation limit of 2.0 MoM avoids an excess of very high inaccurate risk estimates among women with screen-positive results.
The distribution of NT log10 MoMs is positively skewed, so that the distribution of NT measurements at higher MoM values (those above about 2.0 MoM) is not well estimated by a Gaussian distribution (applicable to about 0.4% of unaffected pregnancies). With a smaller standard deviation, the expected number of pregnancies with high NT MoMs is substantially less than that observed. Consequently, the likelihood ratio calculated from the overlapping Gaussian distributions of affected and unaffected pregnancies at high NT MoMs will be overestimated, and hence the risk of having an affected pregnancy will be overestimated. This is illustrated in the validation plot of Figure 2(a) where the NT MoM upper truncation limit is set at 2.5 MoM, which shows how the error is limited to women with the highest risk estimates.
Figure 3 shows a partial histogram of NT MoM values (between 1.5 and 3.0 MoM) in unaffected pregnancies. The figure also shows the proportions using a Gaussian distribution with and without truncation at 2.0 MoM. With an NT upper truncation limit of 2.0 MoM, the truncated NT distribution fits the observed data reasonably well, illustrating how it leads to improved accuracy of risk estimation.
Figure 3.
Partial histogram of NT MoM values (between 1.5 and 3.0 MoM) in unaffected pregnancies from the 2018 audit.8 The solid dotted line shows the Gaussian distribution of NT measurements. To show a single line, a weighted average for gestational age weeks 10 to 13 was used to calculate the NT MoM standard deviation. The horizontal line is specified by the truncation limit (2.0 MoM).
MoM: multiple of the median.
Revising the NT upper truncation limit from 2.5 MoM to 2.0 MoM will not reduce the detection rate because the effect is limited to pregnancies with very high risk estimates (greater than about 1 in 10) that would anyway be classified as screen positive. It, however, avoids the overestimation of risk among some pregnancies with the highest risk estimates and being limited in this way is robust to centre-to-centre variation in NT precision.
The improvement in the accuracy of risk estimation using a less high NT upper truncation limit of 2.0 MoM will apply to screening for trisomy 18 and trisomy 13 as well as for Down’s syndrome because the over estimation of risk for high NT MoM values arises from the persistence of right skewness in the distribution of log10 NT MoM values in unaffected pregnancies.
In conclusion, the NT MoM upper truncation limit should be revised from 2.5 to 2.0 MoM when used with the smaller NT standard deviation associated with the improved precision of current NT measurements.
Acknowledgements
We thank Jon Bestwick for his helpful comments.
Appendix 1. Summary of published NT distribution parameters.2,4,10
| Down’s syndrome pregnancies | Unaffected | |
|---|---|---|
| Median (log10 MoM) | 0.8554679–0.0064686 × gestational age in days | 0.0000 |
| Standard deviation (log10 MoM) | ||
| 10 completed weeks | 0.2313 | 0.1193a |
| 11 completed weeks | 0.0976 | |
| 12 completed weeks | 0.0855 | |
| 13 completed weeks | 0.0855 | |
| Lower truncation limits (MoM) | ||
| 10 completed weeks | 0.5 | |
| 11 completed weeks | 0.7 | |
| 12 completed weeks | 0.8 | |
| 13 completed weeks | 0.85 |
Footnotes
Declaration of conflicting interests: The author(s) declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: SV and NW are employees of Logical Medical Systems which produces the Alpha software for the interpretation of antenatal screening tests.
Funding: The author(s) received no financial support for the research, authorship, and/or publication of this article.
ORCID iD: Stephen H Vale https://orcid.org/0000-0001-6000-3746
References
- 1.Wald NJ, Rodeck C, Hackshaw AK, et al. First and second trimester antenatal screening for Down’s syndrome: the results of the Serum, Urine and Ultrasound Screening Study (SURUSS). J Med Screen 2003; 10: 56–104. [DOI] [PubMed] [Google Scholar]
- 2.Bestwick JP, Huttly WJ, Wald NJ. Distribution of nuchal translucency in antenatal screening for Down’s syndrome. J Med Screen 2010; 17: 8–12. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Wald NJ, Bestwick JP, Huttly WJ, et al. Prenatal screening for Down syndrome in twin pregnancies: estimates of screening performance based on 61 affected and 7302 unaffected twin pregnancies. Prenat Diagn 2018; 38: 1079–1085. [DOI] [PubMed] [Google Scholar]
- 4.Wald NJ, Bestwick JP, Huttly WJ, et al. Corrigendum for Wald NJ, Bestwick JP, Huttly WJ, Aldis J, Borrell A, Goodburn S and Mills I (2018) 10.1002/pd.5381 Prenat Diagn 2019;39: 489–491. [DOI] [PubMed]
- 5.Schuchter K, Wald NJ, Hackshaw AK, et al. The distribution of nuchal translucency at 10-13 weeks of pregnancy. Prenat Diagn 1998; 18: 281–286. [PubMed] [Google Scholar]
- 6.Wald NJ, Cuckle HS, Densem JW, et al. Maternal serum screening for Down’s syndrome in early pregnancy. Br Med J 1988; 297: 883–887. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Barnes IM, Bestwick JP, Larsen SO. Setting truncation limits for marker values in medical screening. J Med Screen 2007; 14: 103–106. (erratum in J Med Screen 2008; 15: 159) [DOI] [PubMed] [Google Scholar]
- 8.Wald NJ, Huttly WJ, Bestwick JP, et al. Prenatal reflex DNA screening for trisomies 21, 18 and 13. Genet Med 2018; 20: 825–830. [DOI] [PubMed] [Google Scholar]
- 9.Wald N, Hackshaw A. Tests using multiple markers. In: Wald NJ, Leek (eds) Antenatal and neonatal screening. 2nd ed. Oxford: Oxford University Press, 2000, pp.23–57. [Google Scholar]
- 10.Wald NJ, Bestwick JP. and Huttly WJ. Improvements in antenatal screening for Down’s syndrome. J Med Screen 2013; 20: 7–14. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Wald NJ, Hackshaw AK, Huttly WJ, et al. Empirical validation of risk screening for Down’s syndrome. J Med Screen 1996; 3: 185–187. [DOI] [PubMed] [Google Scholar]
- 12.Wald NJ, Bestwick JP, Huttly WJ, et al. Validation plots in antenatal screening for Down’s syndrome. J Med Screen 2006; 13: 166–171. [DOI] [PubMed] [Google Scholar]