Evaluation of a Proposed Mixture Model to Specify the Distributions of Nuchal Translucency Measurements in Antenatal Screening for Down's Syndrome

J P Bestwick,W J Huttly,N J Wald

doi:10.1258/jms.2010.009108

Abstract

ObjectivesA mixture model of crown–rump length (CRL)-dependent and CRL-independent nuchal translucency (NT) measurements has been proposed for antenatal screening for Down's syndrome. We here compare the efficacy of the mixture model method with the standard method, which uses NT multiple of the median (MoM) values in a single distribution.SettingsA routine antenatal screening programme for Down's syndrome comprising 104 affected and 22,284 unaffected pregnancies.MethodsThe ability of NT to distinguish between affected and unaffected pregnancies was compared using the mixture model method and the standard MoM method by using published distribution parameters for the mixture model of NT and parameters derived from these for the standard MoM method. The accuracy of the two methods was compared for NT and maternal age by comparing the median estimated risk with the prevalence of Down's syndrome in different categories of estimated risk.ResultsUsing NT alone observed estimates of discrimination using the two methods are similar; at a 70% detection rate the false-positive rates were 12% using the mixture model method and 10% using the MoM method. Risk estimation was marginally (but not statistically significantly) more accurate using the standard MoM method.ConclusionsThe mixture model method offers no advantage over the standard MoM method in antenatal screening for Down's syndrome, is more complicated and less generalizable to other data-sets. The standard MoM method remains the method of choice.

Highlights

I t has been proposed that nuchal translucency (NT) measurements in antenatal screening for Down’s syndrome be fitted to two distributions in affected pregnancies and two distributions in unaffected pregnancies at each crown –rump length (CRL) measurement – a mixture model.[1]
It was suggested that this model better describes NT measurements than the standard method of using a single distribution of NT for affected pregnancies and a single distribution for unaffected pregnancies at each day of gestation but this has not been shown
We compared the proposed mixture model method with the standard multiple of the median (MoM) method using data on the 104 Down’s syndrome and the 22,284 unaffected pregnancies screened at the Wolfson Institute of Preventive Medicine with CRL measurements between 45 and 84 mm

Summary

Introduction

I t has been proposed that nuchal translucency (NT) measurements (in mm) in antenatal screening for Down’s syndrome be fitted to two distributions in affected pregnancies and two distributions in unaffected pregnancies at each crown –rump length (CRL) measurement – a mixture model.[1]. A concern with the use of a mixture model to describe the distribution of NT measurements is that the model may be too tailored to the data-set from which it was derived and may not be generalizable to other data-sets This concern and the lack of comparison with the standard MoM method in the report proposing its use[1] prompted us to perform a quantitative comparison of the two methods using an independent data-set to determine whether the mixture model offers an improvement in antenatal screening for Down’s syndrome

Objectives

Methods

Results

Conclusion