Background
Preeclampsia (PE) affects 2–3% of all pregnancies and is a major cause of maternal and perinatal morbidity and mortality. The traditional approach to screening for PE is to use a risk-scoring system based on maternal demographic characteristics and medical history (maternal factors), but the performance of such an approach is very poor.
Objective
To develop a model for PE based on a combination of maternal factors with second-trimester biomarkers.
Study Design
The data for this study were derived from prospective screening for adverse obstetric outcomes in women attending their routine hospital visit at 19–24 weeks’ gestation in 3 maternity hospitals in England between January 2006 and July 2014. We had data from maternal factors, uterine artery pulsatility index (UTPI), mean arterial pressure (MAP), serum placental growth factor (PLGF), and serum soluble fms-like tyrosine kinase-1 (SFLT) from 123,406, 67,605, 31,120, 10,828, and 8079 pregnancies, respectively. Bayes’ theorem was used to combine the a priori risk from maternal factors with various combinations of biomarker multiple of the median (MoM) values. The modeled performance of screening for PE requiring delivery at <32, <37, and ≥37 weeks’ gestation was estimated. The modeled performance was compared to the empirical one, which was derived from 5-fold cross validation. We also examined the performance of screening based on risk factors from the medical history, as recommended by the American Congress of Obstetricians and Gynecologists (ACOG).
Results
In pregnancies that developed PE, the values of MAP, UTPI, and SFLT were increased and PLGF was decreased. For all biomarkers the deviation from normal was greater for early than for late PE, and therefore the performance of screening was inversely related to the gestational age at which delivery became necessary for maternal and/or fetal indications. Screening by maternal factors predicted 52%, 47%, and 37% of PE at <32, <37, and ≥37 weeks’ gestation, respectively, at a false-positive rate of 10%. The respective values for combined screening with maternal factors and MAP, UTPI, and PLGF were 99%, 85%, and 46%; the performance was not improved by the addition of SFLT. In our population of 123,406 pregnancies, the DR of PE at <32, <37, and ≥37 weeks with the ACOG recommendations was 91%, 90%, and 91%, respectively, but at a screen positive rate of 67%.
Conclusion
The performance of screening for PE by maternal factors and biomarkers in the middle trimester is superior to taking a medical history.
Preeclampsia (PE) affects 2–3% of all pregnancies and is a major cause of maternal and perinatal morbidity and mortality. The traditional approach to screening for PE is to identify risk factors from maternal demographic characteristics and medical history (maternal factors). According to the American Congress of Obstetricians and Gynecologists (ACOG), taking a medical history to evaluate for risk factors is currently the best and only recommended screening approach for PE. In the UK, the National Institute for Health and Clinical Excellence (NICE) has issued guidelines recommending that women should be considered to be at high risk of developing PE if they have any 1 high-risk factor or any 2 moderate-risk factors. However, the performance of such an approach, which essentially treats each risk factor as a separate screening test with additive detection rate (DR) and screen positive rate, is poor, with DR of only 35% of all PE and 40% of preterm PE requiring delivery at <37 weeks’ gestation, at a false-positive rate (FPR) of about 10%.
An alternative approach to screening, which allows estimation of individual patient-specific risks of PE requiring delivery before a specified gestation, is to use Bayes’ theorem to combine the a priori risk from maternal factors, derived by a multivariable logistic model, with the results of various combinations of biophysical and biochemical measurements made at different times during pregnancy. We have previously reported that first-trimester screening by a combination of maternal factors with mean arterial pressure (MAP), uterine artery pulsatility index (UTPI), and serum placental growth factor (PLGF) can predict 75% of preterm PE and 47% of term PE, at 10% FPR.
The objective of this study of singleton pregnancies with data on MAP, UTPI, PLGF, and serum soluble fms-like tyrosine kinase-1 (SFLT) at 19–24 weeks’ gestation is to examine the potential improvement in performance of screening by maternal factors alone with the addition of each biomarker and combinations of biomarkers. We also examined the performance of screening based on risk factors from the medical history, as recommended by ACOG.
Methods
Study design and participants
The data for this study were derived from prospective screening for adverse obstetric outcomes in women attending for routine pregnancy care at 11 +0 to 13 +6 and 19 +0 to 24 +6 weeks’ gestation in 3 maternity hospitals in the UK (King’s College Hospital between January 2006 and July 2014, Medway Maritime Hospital between February 2007 and July 2014, and University College London Hospital between April 2009 and September 2013). Maternal characteristics and medical history were recorded at the visit at 11 +0 to 13 +6 weeks (n = 123,406) and measurements of UTPI, MAP, PLGF, and SFLT at 19 +0 to 24 +6 weeks. Screening evolved over time in 2 respects. Firstly, there was a change in participating hospitals; although all 3 hospitals were providing routine screening of their local populations, there were differences in the distribution of the racial origin of the study populations, which would affect the prior risk for PE. Secondly, there was a change in the content of the clinics; in the first phase of the study, only UTPI was measured (n = 67,605), then measurement of MAP was added (n = 31,120); and in the final phase serum concentration of PLGF was measured (n = 10,828) and then SFLT was added (n = 8,079). Measurements of all 4 biomarkers were obtained from 7748 pregnancies.
The left and right UTPI were measured by transvaginal color Doppler ultrasound and the mean pulsatility index was calculated. Measurements of MAP were obtained by validated automated devices and a standardized protocol. Measurement of serum concentration of PLGF and SFLT were by an automated biochemical analyzer within 10 minutes of blood sampling (Cobas e411 system; Roche Diagnostics, Penzberg, Germany). The inter-assay coefficients of variation for low and high concentrations were 5.4% and 3.0% for PLGF, and 3.0% and 3.2% for SFLT-1, respectively. Gestational age was determined from measurement of fetal crown-rump length (CRL) at 11–13 weeks or the fetal head circumference at 19–24 weeks. The women gave written informed consent to participate in the study, which was approved by the NHS Research Ethics Committee.
The inclusion criteria for this study were singleton pregnancy delivering a nonmalformed live birth or stillbirth at ≥24 weeks’ gestation. We excluded pregnancies with aneuploidies and major fetal abnormalities and those ending in termination, miscarriage, or fetal death at <24 weeks.
Outcome measures
Data on pregnancy outcome were collected from the hospital maternity records or the general medical practitioners of the women. The obstetric records of all women with pre-existing or pregnancy-associated hypertension were examined to determine if the condition was PE or pregnancy-induced hypertension (PIH), as defined by the International Society for the Study of Hypertension in Pregnancy. Outcome measures were PE delivering at <37 weeks’ gestation (preterm PE), PE delivering at ≥37 weeks (term PE), and subgroups of PE delivering at <32, 32 +0 to 36 +6 , 37 +0 to 39 +6 , and ≥40 weeks. The unaffected group contained all pregnancies without PE or PIH.
Statistical analyses
Performance of screening was assessed as follows: firstly, by examining the empirical results in 7748 pregnancies with complete data on UTPI, MAP, PLGF, and SFLT; secondly, by examining the empirical results using all available data for each biomarker; and thirdly, by modeling, whereby values on biomarkers were simulated for all 123,406 cases with available data on maternal factors. In selecting the second option, we wanted to have the maximum possible data for developing the models and examining performance of the various biomarkers; for example, in examining UTPI we could use data from 67,605 pregnancies, rather than just 7748. However, the distribution of maternal factors was not identical in each subset used for assessment of each biomarker or their combinations; consequently, there were differences between the datasets in the maternal factor–related performance of screening and it was therefore difficult to compare meaningfully the additional contribution to performance between biomarkers and their combinations over and above that of maternal factors alone. To overcome this problem we used modeling by imputing values for all biomarkers in the large dataset of 123,406 pregnancies.
Competing risks model
This model assumes that if the pregnancy were to continue indefinitely all women would develop PE, and whether they do so or not before a specified gestational age depends on competition between delivery before or after development of PE. The effect of maternal factors is to modify the mean of the distribution of gestational age at delivery with PE so that in pregnancies at low risk for PE the gestational age distribution is shifted to the right, with the implication that in most pregnancies delivery will actually occur for other reasons before development of PE. In high-risk pregnancies the distribution is shifted to the left; and the smaller the mean gestational age, the higher is the risk for PE. The distribution of biomarkers is specified conditionally on the gestational age at delivery with PE. For any women with specific maternal factors and biomarker multiple of the normal median (MoM) values, the posterior distribution of the time to delivery with PE, assuming there is no other cause of delivery, is obtained from the application of Bayes’ theorem.
Gestational age at delivery with PE was defined by 2 components: firstly, the prior distribution based on maternal factors, and secondly, the conditional distribution of MoM biomarker values, given the gestational age, with PE and maternal factors. Values of UTPI, MAP, PLGF, and SFLT were expressed as MoMs adjusting for those characteristics found to provide a substantive contribution to their values, including the maternal factors in the prior model. In the PE group, the mean log 10 MoM was assumed to depend linearly with gestational age at delivery and this linear relationship was assumed to continue until the mean log 10 MoM of zero, beyond which the mean was taken as zero; this assumption was confirmed by the empirical results shown in Figure 1 . Multivariable Gaussian distributions were fitted to the log 10 MoM values of the biomarkers and a common covariance matrix was assumed for these distributions. Analysis of residuals was used to check the adequacy of the model and assess the effects of maternal factors on log 10 -transformed MoM values in pregnancies with PE.
Empirical performance of screening
Empirical performance of screening was carried out for all available data and for the subset of 7748 pregnancies with complete data on UTPI, MAP, PLGF, and SFLT. Five-fold cross validation was used to assess the empirical performance of screening for PE by maternal factors and the combination of maternal factors with biomarkers. The data were divided into 5 equal subgroups; the model was then fitted 5 times to different combinations of 4 of the 5 subgroups and used to predict risk of PE in the remaining fifth of the data. In each case, the maternal factor model, the regression models, and the covariance matrix were fitted to the training dataset comprising four fifths of the data and used to produce risks for the hold-out sample comprising the remaining fifth of the data.
Model-based estimates of screening performance
To provide model-based estimates of screening performance, the following procedure was adopted. First, we obtained the dataset of 123,406 singleton pregnancies, including 2748 (2.2%) with PE, that was previously used to develop a model for PE based on maternal demographic characteristics and medical history. Second, for each case of PE (n = 2748) and pregnancies unaffected by PE or PIH (n = 117,710), the biophysical and biochemical MoM values were simulated from the fitted multivariate Gaussian distribution for log-transformed MoM values. Third, risks were obtained using the competing risk model from the simulated MoM values and the pregnancy characteristics. These 3 steps were applied to the pregnancies within the unaffected group with no restriction on the time of delivery. Fourth, for a given FPR, risks from the unaffected group were used to define a risk cutoff. The proportion of PE risks was then used to obtain an estimate of the associated DR. The area under the receiver operating characteristic curve (AUROC) was also calculated. The simulations were repeated 100 times to reduce variability due to the simulation process and provide suitably precise model-based estimates of performance.
The statistical software package R was used for data analyses. The survival package was used for fitting the maternal factors model and the package pROC was used for the receiver operating characteristic (ROC) curve analysis.
Results
The characteristics of the total population of 123,406 singleton pregnancies are given in Table 1 and those of the subset of 7748 pregnancies with complete data on UTPI, MAP, PLGF, and SFLT are given in Supplemental Table 1 ( Appendix ).
| Variable | Unaffected (n = 117,710) | PE <37 w (n = 790) | PE ≥37 w (n = 1958) | PIH (n = 2948) |
|---|---|---|---|---|
| Maternal age in years, median (IQR) | 31.3 (26.7, 35.1) | 31.8 (26.9, 36.5) a | 31.3 (26.5, 35.8) | 31.8 (27.2, 35.5) a |
| Maternal weight in kg, median (IQR) | 69.8 (62.4, 79.9) | 74.0 (65.0, 88.0) a | 77.4 (67.8, 91.9) a | 76.0 (67.0, 88.0) a |
| Maternal height in cm, median (IQR) | 164 (160, 169) | 163 (158, 167) a | 164 (160, 168) a | 165 (160, 169) |
| Body mass index, median (IQR) | 25.8 (23.2, 29.4) | 28.4 (24.6, 32.8) a | 28.8 (25.4, 33.7) a | 28.1 (25.0, 32.4) a |
| Gestational age in weeks, median (IQR) | 22.1 (21.1, 22.7) | 22.2 (21.2, 22.8) a | 22.2 (21.4, 22.7) a | 22.2 (21.4, 22.7) a |
| Racial origin | a | a | a | |
| White, n (%) | 87,373 (74.2) | 420 (53.2) | 1165 (59.5) | 2010 (68.2) |
| Afro-Caribbean, n (%) | 18,313 (15.6) | 293 (37.1) | 614 (31.4) | 668 (22.7) |
| South Asian, n (%) | 6120 (5.2) | 51 (6.5) | 102 (5.2) | 148 (5.0) |
| East Asian, n (%) | 3106 (2.6) | 10 (1.3) | 37 (1.9) | 53 (1.8) |
| Mixed, n (%) | 2798 (2.4) | 16 (2.0) | 40 (2.0) | 69 (2.3) |
| Medical history | ||||
| Chronic hypertension, n (%) | 1198 (1.0) | 102 (12.9) a | 186 (9.5) a | 0 (0.0) a |
| Diabetes mellitus, n (%) | 893 (0.8) | 30 (3.8) a | 31 (1.6) a | 35 (1.2) a |
| SLE/APS, n (%) | 207 (0.2) | 9 (1.1) a | 7 (0.4) | 9 (0.3) |
| Conception | a | a | ||
| Natural, n (%) | 113,530 (96.5) | 727 (92.0) | 1868 (95.4) | 2823 (95.8) |
| In vitro fertilization, n (%) | 2632 (2.2) | 43 (5.4) | 68 (3.5) | 83 (2.8) |
| Ovulation induction drugs, n (%) | 1548 (1.3) | 20 (2.5) | 22 (1.1) | 42 (1.4) |
| Family history of preeclampsia, n (%) | 4243 (3.6) | 67 (8.5) a | 134 (6.8) a | 220 (7.5) a |
| Parity | ||||
| Nulliparous, n (%) | 57,720 (49.0) | 468 (59.2) a | 1,250 (63.8) a | 1,888 (64.0) a |
| Parous with no previous PE, n (%) | 56,848 (48.3) | 196 (24.8) a | 476 (24.3) a | 765 (26.0) a |
| Parous with previous PE, n (%) | 3142 (2.7) | 126 (16.0) a | 232 (11.9) a | 295 (10.0) a |
| Inter-pregnancy interval in years, median (IQR) | 2.9 (1.9, 4.8) | 4.2 (2.4, 7.3) a | 3.7 (2.3, 6.7) a | 3.4 (2.0, 5.7) a |
Distribution of biomarkers
The distributions of log 10 MoM values of the biomarkers in unaffected pregnancies and in those that developed PE are shown in Supplemental Tables 2 and 3 ( Appendix ). In the unaffected group, the median MoM value is 1.0 and on the log scale the distribution of MoM values is very well approximated by a Gaussian distribution with mean zero. The MoM values in the PE group and the fitted regression relationships with gestational age at delivery are shown in Figure 1 . All markers showed more separation at earlier than later gestations and this is reflected in their superior performance at detection of early vs late PE.
Performance of screening for preeclampsia
Empirical and model-based performance of screening for PE by maternal factors and combinations of biomarkers are shown in Tables 2 and 3 , Supplemental Tables 4-7 ( Appendix ), and Figures 2 and 3 . The empirical performance of screening for PE at <37 and ≥37 weeks in the 7748 pregnancies with complete data is shown in Table 2 ; the DRs at 5% and 10% FPR were compatible with the model-based rates. The AUROC curves for prediction of PE at <32, <37, and ≥37 weeks based on empirical results from all available data are shown in Table 3 and these were compatible with the model-based results. Empirical performance of screening for PE with delivery at <37, ≥37, <32, 32 +0 to 36 +6 , 37 +0 to 39 +6 , and ≥40 weeks’ gestation is shown in Supplemental Tables 4-6 ( Appendix ); the number of cases for each biomarker and combinations of biomarkers varied, with inevitable differences in composition of the populations and, consequently, differences in performance of screening by maternal factors alone. The model-based performance of screening for PE with delivery at <37, ≥37, <32, 32 +0 to 36 +6 , 37 +0 to 39 +6 , and ≥40 weeks’ gestation is shown in Supplemental Table 7 ( Appendix ). Figure 2 shows the ROC curves for model-based prediction of PE at <32, <37, and ≥37 weeks’ gestation by maternal factors, combination of maternal factors with each biomarker, and combination of maternal factors with MAP, UTPI, and PLGF. Figure 3 shows the empirical performance of screening for PE at <37 and ≥37 weeks, by combination of maternal factors with all available data on MAP, UTPI, and PLGF; the empirical results were compatible with the model-based results.
| Method of screening | Preeclampsia at <37 weeks | Preeclampsia at ≥37 weeks | ||||||
|---|---|---|---|---|---|---|---|---|
| FPR 5% | FPR 10% | FPR 5% | FPR 10% | |||||
| n/N | % (95% CI) a | n/N | % (95% CI) a | n/N | % (95% CI) a | n/N | % (95% CI) a | |
| History | 21/62 | 34 (22, 47); 34 | 29/62 | 47 (34, 60); 47 | 55/206 | 27 (21, 33); 26 | 75/206 | 36 (30, 43); 37 |
| MAP | 30/62 | 48 (35, 61); 47 | 37/62 | 60 (46, 72); 60 | 55/206 | 27 (21, 33); 30 | 90/206 | 44 (37, 51); 43 |
| UTPI | 37/62 | 60 (46, 72); 57 | 47/62 | 76 (63, 86); 70 | 52/206 | 25 (19, 32); 28 | 78/206 | 38 (31, 45); 40 |
| PLGF | 34/62 | 55 (42, 68); 64 | 44/62 | 71 (58, 82); 73 | 55/206 | 27 (21, 33); 27 | 75/206 | 36 (30, 43); 37 |
| SFLT | 20/62 | 32 (21, 45); 38 | 33/62 | 53 (40, 66); 50 | 55/206 | 27 (21, 33); 26 | 75/206 | 36 (30, 43); 37 |
| MAP, UTPI | 49/62 | 79 (67, 88); 67 | 50/62 | 81 (69, 90); 78 | 59/206 | 29 (23, 35); 33 | 90/206 | 44 (37, 51); 46 |
| MAP, PLGF | 38/62 | 61 (48, 73); 69 | 45/62 | 73 (60, 83); 78 | 55/206 | 27 (21, 33); 30 | 89/206 | 43 (36, 50); 43 |
| MAP, SFLT | 31/62 | 50 (37, 63); 49 | 38/62 | 61 (48, 73); 62 | 55/206 | 27 (21, 33); 30 | 90/206 | 44 (37, 51); 42 |
| UTPI, PLGF | 43/62 | 69 (56, 80); 72 | 50/62 | 81 (69, 90); 81 | 53/206 | 26 (20, 32); 28 | 75/206 | 36 (30, 43); 40 |
| UTPI, SFLT | 41/62 | 66 (53, 78); 61 | 45/62 | 73 (60, 83); 72 | 54/206 | 26 (20, 33); 28 | 78/206 | 38 (31, 45); 40 |
| PLGF, SFLT | 35/62 | 56 (43, 69); 65 | 44/62 | 71 (58, 82); 75 | 55/206 | 27 (21, 33); 27 | 75/206 | 36 (30, 43); 37 |
| MAP, UTPI, PLGF | 45/62 | 73 (60, 83); 77 | 52/62 | 84 (72, 92); 85 | 58/206 | 28 (22, 35); 33 | 90/206 | 44 (37, 51); 46 |
| MAP, UTPI, SFLT | 46/62 | 74 (62, 84); 69 | 50/62 | 81 (69, 90); 79 | 57/206 | 28 (22, 35); 33 | 92/206 | 45 (38, 52); 46 |
| MAP, PLGF, SFLT | 37/62 | 60 (46, 72); 69 | 45/62 | 73 (60, 83); 79 | 56/206 | 27 (21, 34); 33 | 89/206 | 43 (36, 50); 46 |
| UTPI, PLGF, SFLT | 41/62 | 66 (53, 78); 74 | 50/62 | 81 (69, 90); 82 | 54/206 | 26 (20, 33); 28 | 74/206 | 36 (29, 43); 40 |
| MAP, UTPI, PLGF, SFLT | 46/62 | 74 (62, 84); 78 | 53/62 | 85 (74, 93); 86 | 56/206 | 27 (21, 34); 33 | 91/206 | 44 (37, 51); 46 |
a The last numbers in each cell are the values obtained from modeling.
| Screening | Areas under the receiver operating characteristic curve | |||||
|---|---|---|---|---|---|---|
| PE <32 w | PE <37 w | PE ≥37 w | ||||
| Empirical (95% CI) | Model | Empirical (95% CI) | Model | Empirical (95% CI) | Model | |
| History | 0.820 (0.791, 0.848) | 0.827 | 0.789 (0.773, 0.804) | 0.796 | 0.748 (0.737, 0.759) | 0.752 |
| MAP | 0.902 (0.862, 0.942) | 0.906 | 0.849 (0.824, 0.874) | 0.860 | 0.787 (0.769, 0.805) | 0.784 |
| UTPI | 0.949 (0.931, 0.968) | 0.957 | 0.898 (0.883, 0.912) | 0.895 | 0.766 (0.753, 0.779) | 0.771 |
| PLGF | 0.962 (0.914, 0.999) | 0.989 | 0.887 (0.849, 0.926) | 0.905 | 0.732 (0.701, 0.763) | 0.752 |
| SFLT | 0.906 (0.820, 0.993) | 0.875 | 0.820 (0.771, 0.869) | 0.810 | 0.733 (0.700, 0.766) | 0.752 |
| MAP, UTPI | 0.969 (0.940, 0.997) | 0.975 | 0.918 (0.895, 0.941) | 0.924 | 0.801 (0.784, 0.819) | 0.801 |
| MAP, PLGF | 0.981 (0.957, 0.999) | 0.992 | 0.909 (0.875, 0.943) | 0.924 | 0.766 (0.738, 0.795) | 0.784 |
| MAP, SFLT | 0.941 (0.892, 0.990) | 0.924 | 0.858 (0.811, 0.906) | 0.865 | 0.769 (0.738, 0.801) | 0.784 |
| UTPI, PLGF | 0.976 (0.947, 0.999) | 0.995 | 0.926 (0.895, 0.956) | 0.934 | 0.736 (0.705, 0.768) | 0.771 |
| UTPI, SFLT | 0.973 (0.941, 0.999) | 0.973 | 0.909 (0.875, 0.944) | 0.903 | 0.741 (0.707, 0.775) | 0.772 |
| PLGF, SFLT | 0.957 (0.896, 0.999) | 0.993 | 0.878 (0.836, 0.921) | 0.910 | 0.734 (0.701, 0.768) | 0.752 |
| MAP, UTPI, PLGF | 0.979 (0.949, 0.999) | 0.996 | 0.932 (0.899, 0.965) | 0.948 | 0.772 (0.742, 0.801) | 0.801 |
| MAP, UTPI, SFLT | 0.994 (0.989, 0.999) | 0.983 | 0.915 (0.872, 0.958) | 0.927 | 0.780 (0.749, 0.812) | 0.801 |
| MAP, PLGF, SFLT | 0.980 (0.952, 0.999) | 0.983 | 0.899 (0.859, 0.940) | 0.927 | 0.768 (0.737, 0.800) | 0.801 |
| UTPI, PLGF, SFLT | 0.984 (0.959, 0.999) | 0.998 | 0.926 (0.894, 0.957) | 0.939 | 0.739 (0.706, 0.773) | 0.772 |
| MAP, UTPI, PLGF, SFLT | 0.995 (0.990, 0.999) | 0.998 | 0.930 (0.892, 0.968) | 0.951 | 0.773 (0.741, 0.805) | 0.801 |
Empirical performance for early, preterm, and term preeclampsia
On the basis of all available data, the empirical performance of screening for early PE by maternal factors (AUROC, 0.820; 95% CI, 0.791, 0.848) was improved by the addition of MAP (AUROC, 0.902; 95% CI, 0.862, 0.942) or PLGF (AUROC, 0.962; 95% CI, 0.914, 0.999) and the performance of maternal factors and MAP was improved by the addition of PLGF (AUROC, 0.981; 95% CI, 0.957, 0.999), UTPI and PLGF (AUROC, 0.979; 95% CI, 0.949, 0.999), UTPI and SFLT (AUROC, 0.994; 95% CI, 0.989, 0.999), and PLGF and SFLT (AUROC, 0.980; 95% CI, 0.952, 0.999); addition of SFLT to the combination of maternal factors, MAP, UTPI, and PLGF provided a small nonsignificant improvement in performance of screening (AUROC, 0.995; 95% CI, 0.990, 0.999) ( Table 3 , Figure 2 ).
The performance of screening for preterm PE by maternal factors (AUROC, 0.789; 95% CI, 0.773, 0.804) was improved by the addition of MAP (AUROC, 0.849; 95% CI, 0.824, 0.874), UTPI (AUROC, 0.898; 95% CI, 0.883, 0.912), or PLGF (AUROC, 0.887; 95% CI, 0.849, 0.926) and the performance of maternal factors and MAP was improved by the addition of either UTPI (AUROC, 0.918; 95% CI, 0.895, 0.941), PLGF (AUROC, 0.909; 95% CI, 0.875, 0.943), or both UTPI and PLGF (AUROC, 0.932; 95% CI, 0.899, 0.965); SFLT did not provide significant improvement to any combination of biomarkers ( Table 3 , Figure 2 ).
The performance of screening for term PE by maternal factors (AUROC, 0.748; 95% CI, 0.737, 0.759) was improved by the addition of MAP (AUROC, 0.787; 95% CI, 0.769, 0.805) and both MAP and UTPI (AUROC, 0.801; 95% CI, 0.784, 0.819); serum PLGF and SFLT, either on their own or in combination, did not improve the prediction provided by maternal factors alone ( Table 3 , Figure 2 ).
Performance of screening in subgroups of racial origin and obstetric history
In the dataset of 123,406 pregnancies, 61,326 women (49.7%) were nulliparous and 62,080 (50.3%) were parous, including 3795 (6.1%) with history of PE in a previous pregnancy and 58,285 (93.9%) without history of PE. The contribution of parous women to PE was 37.5% (1030/2748), including 34.8% (358/1030) from parous women with PE in a previous pregnancy and 65.2% (672/1030) from parous women without a history of PE.
The model-based performance of screening by a combination of maternal factors, MAP, UTPI, and PLGF in the prediction of preterm PE and term PE for nulliparous and parous women of Afro-Caribbean and white racial origin are given in Table 4 . In these calculations a risk cutoff was selected to achieve a screen positive rate of about 10%. At a risk cutoff of 1 in 100 for preterm PE and 1 in 15 for term PE, the FPR and DR were higher in parous women with vs without PE in a previous pregnancy and in those of Afro-Caribbean vs white racial origin. In all groups, the risk of being affected given a screen positive result was considerably higher than the prevalence of the disease, whereas in those with a screen negative result the risk was considerably reduced.
| Group | Prevalence (%) | Screen positive (%) | False positive (%) | DR (%) | Risk of being affected given result: | |
|---|---|---|---|---|---|---|
| Screen positive (%) a | Screen negative (%) b | |||||
| Preeclampsia <37 w | ||||||
| All pregnancies | 0.64 | 11.4 | 10.4 | 85 | 4.77 | 0.11 |
| Nulliparous | 0.76 | 14.7 | 13.7 | 84 | 4.37 | 0.14 |
| Parous | 0.52 | 8.0 | 7.2 | 85 | 5.50 | 0.08 |
| No previous PE | 0.34 | 5.9 | 5.4 | 78 | 4.45 | 0.08 |
| Previous PE | 3.32 | 41.6 | 37.6 | 97 | 7.76 | 0.16 |
| Afro-Caribbean | 1.47 | 23.3 | 21.1 | 91 | 5.78 | 0.17 |
| Nulliparous | 1.64 | 30.0 | 27.8 | 92 | 5.03 | 0.20 |
| Parous | 1.36 | 18.8 | 16.8 | 91 | 6.58 | 0.15 |
| No previous PE | 0.93 | 15.4 | 14.1 | 86 | 5.20 | 0.15 |
| Previous PE | 6.83 | 62.6 | 57.1 | 100 | 10.87 | 0.07 |
| White | 0.46 | 8.8 | 8.2 | 80 | 4.20 | 0.10 |
| Nulliparous | 0.62 | 12.1 | 11.4 | 81 | 4.12 | 0.13 |
| Parous | 0.29 | 5.2 | 4.7 | 78 | 4.41 | 0.07 |
| No previous PE | 0.19 | 3.4 | 3.2 | 66 | 3.65 | 0.07 |
| Previous PE | 2.01 | 34.1 | 31.5 | 95 | 5.61 | 0.14 |
| Preeclampsia ≥37 w | ||||||
| All pregnancies | 1.59 | 9.9 | 9.3 | 44 | 7.09 | 0.98 |
| Nulliparous | 2.04 | 13 | 12.4 | 41 | 6.47 | 1.38 |
| Parous | 1.14 | 6.9 | 6.4 | 50 | 8.24 | 0.61 |
| No previous PE | 0.82 | 4 | 3.8 | 33 | 6.66 | 0.57 |
| Previous PE | 6.11 | 54.8 | 52.5 | 85 | 9.53 | 1.98 |
| Afro-Caribbean | 3.09 | 28 | 26.5 | 70 | 7.68 | 1.31 |
| Nulliparous | 3.96 | 41.2 | 39.8 | 74 | 7.07 | 1.77 |
| Parous | 2.51 | 19.2 | 18 | 65 | 8.52 | 1.08 |
| No previous PE | 1.91 | 14.6 | 13.8 | 52 | 6.85 | 1.07 |
| Previous PE | 10.13 | 84 | 82.4 | 96 | 11.58 | 2.5 |
| White | 1.28 | 6.2 | 5.8 | 32 | 6.62 | 0.93 |
| Nulliparous | 1.74 | 8.4 | 8 | 30 | 6.16 | 1.34 |
| Parous | 0.79 | 3.8 | 3.5 | 37 | 7.69 | 0.51 |
| No previous PE | 0.55 | 1.4 | 1.3 | 16 | 6.53 | 0.46 |
| Previous PE | 4.63 | 44.8 | 43.1 | 76 | 7.86 | 2.02 |
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
