A mixture model of nuchal translucency thickness in screening for chromosomal defects: validation of a single operator dataset

Abstract

Objective: (1) To validate the mixture model in a single operator dataset and (2) to compare the detection rates for fetal chromosomal defects obtained from the mixture model with those obtained from either the delta nuchal translucency (NT) or log multiple of the median (MoM) approach.

Methods: Database query, viable singletons [crown-rump length (CRL) 45-84 mm corresponding to 11-13(+6) weeks], December 1997 to November 2006, examined by Adam Gasiorek-Wiens, the statistical mixture model was applied.

Results: Seventy-four of 4171 were lost to follow-up (1.8%), 4097 singleton pregnancies included trisomy 21 (n = 34, 0.8%), trisomy 18 (n = 20, 0.5%), trisomy 13 (n = 8, 0.2%), Turner syndrome (n = 9, 0.2%) and other chromosomal abnormalities (n = 14, 0.3%). The main findings are that (1) the log-transformed NT measurements follow a mixture of two Gaussian distributions and (2) the criteria to apply either the delta-NT or log MoM models are not met. In the normal group, the majority of NT measurements were dependent on the CRL, a small group showed a median independent of the CRL. In the abnormal group it was the opposite. For a 5% false-positive rate (FPR), the trisomy 21 detection rate was 83%.

Conclusions: The use of the mixture model in a single operator dataset produces results compatible with the original study. The mixture model has thus been validated.