First‐trimester screening for pre‐eclampsia: time for reflection

Over the last years, many papers have been published on the development and validation of first-trimester screening algorithms for pre-eclampsia (PE)1234. They essentially offer a multivariable panel of predictors and some equations able to estimate the a-posteriori risk for different phenotypes of PE, including early and late PE. The predictors include a set of ‘maternal factors’ and a step-by-step addition of uterine artery pulsatility index, maternal mean arterial pressure and biochemical markers. Since the inclusion of each marker in the prediction model requires additional economic resources, a key question that should be asked is the real improvement that such multimarker panels offer in the prediction of PE. Two points that should be more clearly defined on this issue are the detection rate (DR) of each additional marker alone and not only in combination with maternal factors, and, more importantly, the improvement of the discriminant power achieved by the subsequent addition of each marker. The most recent papers seem to indicate that the addition of a single marker does not improve significantly predictive performance or DR, as shown by overlapping 95% CIs (when reported); therefore, no real benefit seems to be achieved by a screening program involving just one extra variable in addition to maternal factors. This is exemplified in Table 1, in which data from two models for prediction of PE at < 32 and < 37 weeks' gestation (total cohort, n = 35 948; PE < 32 weeks, n = 66; PE < 37 weeks, n = 292)5 and from another model assessing PE at < 34 weeks (total cohort, n = 9462; PE < 34 weeks, n = 57)6 are presented. This emphasizes the clinical importance of estimating the improvement in screening performance after addition of successive biomarkers over and above the detection provided by the a-priori odds. Another issue that should be addressed is the choice of the cut-off to define a screen-positive woman. Although most studies on first-trimester prediction of PE report DRs at a false-positive rate (FPR) of 5% or 10%, very few evaluate the screening performance of the models for specific cut-off values of PE risk (data from the most relevant ones are summarized in Table 2)5 789. At a FPR of 10%, the risk cut-offs of these models seem to be quite close to, and sometimes the same as, the incidence of PE, raising doubts about their real clinical value when applied to prospective populations. In fact, in this way, the cut-off value to define a screen-positive woman would be the same as the risk of the study population used to develop the algorithm. In the model of Akolekar et al.7, it seems odd that, to predict PE < 34 weeks at a FPR of 10%, the risk cut-off used was almost the same as the incidence of PE < 34 weeks in the general population (1:269 vs 1:275), while the cut-off used to predict PE < 37 weeks was slightly lower than the estimated incidence in the general population (1:67 vs 1:104). In the same study, 70% of the PE cases delivered before 37 weeks lie in a risk category of > 1 in 50 and 87% have a risk of > 1 in 100, therefore, 13% of such cases are in the low-risk group of < 1 in 100. In a more recent paper from the same group, a single risk cut-off of 1:70 was used to predict PE < 37 weeks without further subclassification into PE < 32 or < 34 weeks5. Moreover, a randomized controlled trial investigating the effect of aspirin in reducing the incidence of PE < 37 weeks identified high-risk women based on a risk cut-off of 1:10010, which is again different from the previously proposed cut-offs. In this trial population, the 1:100 risk cut-off performed similarly to the 1:70 cut-off used in the model-building study5, 9. Conversely, another randomized controlled trial of aspirin use after first-trimester screening adopted the same prediction algorithm but with a risk cut-off of 1:811. This is analogous to the evolution of changes in risk cut-offs for screening for trisomy 21, which have been modified over the years in different countries depending on improvements in screening performance and what is considered to be an acceptable screen-positive rate. 5 93.4 1 in 128 1 in 275 10 96.3 1 in 269 5 61.1 1 in 36 10 76.6 1 in 67 5 69.2 1 in 73 10 80.8 1 in 178 15 96.2 1 in 278 These apparent incongruences make clinical counseling of patients difficult. Instead of using a risk cut-off, it might be easier to introduce some form of adimensional score which would allow prediction models to focus on the FPR irrespective of the estimated a-posteriori risk, even if only to make the test easier for physicians to understand than for patients. A careful health economic analysis should support any changes in prevention strategies.