First‐trimester screening for pre‐eclampsia and small for gestational age: A comparison of the gaussian and Fetal Medicine Foundation algorithms

Abstract Objective Pre‐eclampsia (PE) and small for gestational age (SGA) can be predicted from the first trimester. The most widely used algorithm worldwide is the Fetal Medicine Foundation (FMF) algorithm. The recently described Gaussian algorithm has reported excellent results although it is unlikely to be externally validated. Therefore, as an alternative approach, we compared the predictive accuracy for PE and SGA of the Gaussian and FMF algorithms. Methods Secondary analysis of a prospective cohort study was conducted at Vall d'Hebron University Hospital (Barcelona) with 2641 singleton pregnancies. The areas under the curve for the predictive performance for early‐onset and preterm PE and early‐onset and preterm SGA were calculated with the Gaussian and FMF algorithms and subsequently compared. Results The FMF and Gaussian algorithms showed a similar predictive performance for most outcomes and marker combinations. Nevertheless, significant differences for early‐onset PE prediction favored the Gaussian algorithm in the following combinations: mean arterial blood pressure (MAP) with pregnancy‐associated plasma protein A, MAP with placental growth factor, and MAP alone. Conclusions The first‐trimester Gaussian and FMF algorithms have similar performances for PE and SGA prediction when applied with all markers within a routine care setting in a Spanish population, adding evidence to the external validity of the FMF algorithm.

plasma protein A [PAPP-A]) can predict PE and SGA. [1][2][3][4] The Fetal Medicine Foundation (FMF) and Gaussian algorithms can identify 80%-90% of pregnant women who will develop PE with delivery <32/<34 weeks of gestation 1,5 and 60%-70% of women who will develop PE with delivery <37 weeks, 1,6 at a 10% false-positive rate (FPR). These algorithms can also predict 50%-60% of SGA with delivery <32 weeks and 30%-40% of SGA with delivery <37 weeks. 2,4 Both algorithms use a similar methodology to assess the risk for PE: they combine the a priori risk (based on maternal characteristics and obstetric and medical history) with the results of various biochemical and biophysical markers, to estimate the individual a posteriori risk for PE, which is used to classify a pregnant person as at high or low risk for PE. In both algorithms, risk for PE can be obtained based on maternal factors alone and in combination with any of the biochemical and/or biophysical markers.
Despite the FMF algorithm being the most used and validated worldwide, the Gaussian algorithm has some features that confer advantages in the clinical setting, which is why it has been used for routine first-trimester PE screening in most maternities in Spain since 2018. First, blood samples for measurements of biochemical markers (PAPP-A and PlGF) can be drawn between 8 ± 0 weeks and 13 ± 6 weeks, as with routine aneuploidy screening, while in the FMF algorithm biomarkers should be assessed only between 11 ± 0 and 13 ± 6 weeks. 6 Second, UtAPI assessment can be done both transabdominally and transvaginally, rendering the algorithm more versatile to different clinical settings, as the UtAPI for the FMF algorithm can be assessed only transabdominally. Third, likelihood ratios for the a priori risk calculation were not derived from the study population in which the algorithm was investigated but from a larger metaanalysis that included >25 million pregnancies. 7 This may render the Gaussian algorithm less overfitted to a given population and, therefore, more adaptable for populations with different characteristics.
The FMF algorithm has been developed and prospectively validated in large populations, showing comparable predictive performances to the original study. [8][9][10][11][12] By contrast, the Gaussian algorithm has been investigated only in a single cohort of participants. In the past few years, routine PE screening has been implemented in most hospitals, leaving virtually no women at high risk for PE without aspirin treatment to prospectively assess the external validity of the Gaussian algorithm. Therefore, an indirect approach to test the performance of the Gaussian algorithm is to compare it with the most externally validated combined screening tool for PE worldwide: the FMF algorithm.
The aim of this study was to compare the predictive accuracy for PE and SGA of the Gaussian and FMF algorithms. Obstetric history variables included parity (nulliparous/multiparous); gestational age at birth (weeks) in the last pregnancy; interval between the last delivery and the beginning of the current one (years); and personal or family history of PE (yes/no). Biochemical markers, including serum PAPP-A and PlGF, were measured at the firsttrimester routine blood test for aneuploidy screening (from 8 ± 0 to 13 ± 6 weeks) by the fully automated Elecsys assays for PAPP-A and PlGF on an immunoassay platform (cobas e analyzers, Roche Diagnostics). Biophysical markers, including MAP and UtAPI, were assessed at the first-trimester scan. Blood pressure was measured automatically using a calibrated device according to a standard procedure: single measurement in one arm (right or left) while women were seated and after a 5-min rest. MAP was calculated as: diastolic blood pressure + (systolic-diastolic blood pressure)/3. UtAPI was measured following the recommendations of the FMF. 14 All examiners were certified by the FMF for PE risk assessment and Doppler ultrasound assessment.

| MATERIAL S AND ME THODS
SGA newborns were defined as having a birthweight below the 10th centile according to customized local charts. 15     Abbreviations: CI, confidence interval; DR, detection rate; FMF, Fetal Medicine Foundation; FPR, false-positive rate; MAP, mean arterial pressure; PAPP-A, pregnancy-associated plasma protein A; PE, pre-eclampsia; PlGF, placental growth factor; UtAPI, mean uterine artery pulsatility index.

| Statistical Analysis
The statistical software RStudio    Abbreviations: CI, confidence interval; DR, detection rate; FMF, Fetal Medicine Foundation; FPR, false-positive rate; MAP, mean arterial pressure; PAPP-A, pregnancy-associated plasma protein A; PlGF, placental growth factor; SGA, small for gestational age; UtAPI, mean uterine artery pulsatility index. Characteristics of the study population are summarized in Table 1 and Table 2.    The results of this study are relevant since the Gaussian algorithm is already being implemented in other countries aside from Spain.

| RE SULTS
Additionally, as seen in previous studies, 23 we confirm that PAPP-A does not increase the predictive accuracy of any of the algorithms when PlGF was being used; however, when PlGF is not available, PAPP-A could increase DR by 5% with some marker combinations.
Finally, we observed that a single measurement of MAP could decrease the predictive accuracy of the FMF algorithm; therefore, the appropriate methodology (the average of two measurements in both arms simultaneously) should be performed when using this algorithm.
One of the main strengths of the study includes the prospective enrollment of patients. Furthermore, this study was performed within the context of routine clinical practice and patients were seen by their usual physicians, making the results more reliable and applicable in routine care settings. Moreover, this is the first study assessing the performance of the FMF algorithm exclusively in a Spanish cohort and in a clinical setting where MAP was measured once and only in one arm, showing similar results to those reported in the original study, for most combinations of markers.
Despite a previous study showing that prediction of PE is similar when biomarkers are measured before or after 11 weeks, 6 the FMF algorithm was designed with biomarkers assessed between 11 ± 0 and 13 ± 6 weeks. In this study, biomarkers were measured before 11 ± 0 weeks in 1675 (63.4%) women. Therefore, another remarkable strength of our work is that it provides evidence of the applicability of the FMF and Gaussian algorithms before and after 11 weeks for predicting PE and SGA.
The main limitation of our study is the low number of cases with early-onset SGA and early-onset PE and the relatively low number of cases with preterm SGA and preterm PE. Additionally, indication for elective delivery of SGA fetuses based on Doppler and cardiotocogram findings may be different when using other fetal growth restriction protocols. However, Doppler and cardiotocogram classification is uniform in Spain, where the Gaussian algorithm is widely used. Another limitation to be noted is that the technique for MAP measurements may potentially reduce the FMF algorithm's performance and could explain its lower AUC versus the Gaussian algorithm for some marker combinations.

| CON CLUS IONS
This study shows that the first-trimester Gaussian and FMF algorithms have similar predictive performances for PE and SGA in a Spanish population within a routine care setting. The accuracy of the FMF algorithm in our study was similar to that reported in previous studies, adding evidence to its external validity.