Researching new methods of screening for adverse pregnancy outcome: lessons from pre-eclampsia

Abstract

Gordon Smith argues for more and better research in screening for pregnancy outcomes, using the example of previous trials in pre-eclampsia.

Figure 1. The effect of study design on sample size calculations and conclusions that can be drawn from screening studies.

(A) Women are randomised to having or not having the screening test performed. (B) Women have the screening test performed and those who screen as high risk are randomised to having an intervention or having the result concealed. The number of women required (indicated in the top of each figure) is from a sample size calculation for 90% power to detect an effect at p<0.05 (two sided). In panel (A), it is the number of women who need to be consented to be randomised to being screened or not screened and the calculation is based on the rate of the primary outcome comparing screened versus not-screened. In panel (B), it is the number of women who need to be screened and it is based on the rate of the primary outcome comparing intervention versus no intervention among women who screened positive. Both calculations assume an outcome with a background incidence of 1%, a screen positive rate of 5%, and a positive predictive value of 10% (hence a positive likelihood ratio of 11). Based on these parameters, the 95% of the population who screened negative would have a 0.53% incidence, giving a negative predictive value of 99.5% and a negative likelihood ratio of 0.52. The intervention is assumed to reduce the risk of the outcome by 60%. See Table S1 for power calculations for both designs using other values of screening performance or intervention efficacy.

Figure 2. Timing of screening tests and the subsequent risk of complications, in relation to gestational age at delivery.

(A) Area under the receiver operating characteristic curve (AUROCC) for pre-eclampsia at different gestational ages in relation to uterine artery Doppler flow velocimetry performed at around 23 weeks gestational age. The origin of the y-axis is set to 0.5 as this is the value where the test is non-informative (data from [3]). (B) Adjusted odds ratio for stillbirth at different gestational ages among women with maternal serum alpha fetoprotein levels in top 1% at 15–20 weeks. (C) As (B) except top 1% of maternal serum levels of human chorionic gonadotrophin (data from [13]).