Effect of screening algorithm, parameter values and median smoothing on patient-specific risk estimates for Down's syndrome screening

Abstract

The efficiency of a screening programme for Down's syndrome is usually expressed in terms of the detection rate for a given false positive rate. Two programmes with approximately equal detection rates and false positive rates would then be regarded as being broadly equivalent. While this might be the case at a population level, we show in this paper that it may be far from true at the individual patient level. Different algorithms, or even different implementations of the same algorithm can lead to calculation of very different individual risks and can result in different decisions for individual patients. Even though two programmes might lead to the same number of referrals, they could be referring quite different women. We consider the effect of using different parameter values within an algorithm, different algorithms and alternative ways of estimating the gestational age-dependent medians of the marker values. We show that the combined effects of these factors could explain much of the range of risks reported in the UK National External Quality Assessment Scheme for Down's syndrome screening.