Prediction of preeclampsia throughout gestation with maternal characteristics and biophysical and biochemical markers: a longitudinal study





Background


The current approach to predict preeclampsia combines maternal risk factors and evidence from biophysical markers (mean arterial pressure, Doppler velocimetry of the uterine arteries) and maternal blood proteins (placental growth factor, soluble vascular endothelial growth factor receptor-1, pregnancy-associated plasma protein A). Such models require the transformation of biomarker data into multiples of the mean values by using population- and site-specific models. Previous studies have focused on a narrow window in gestation and have not included the maternal blood concentration of soluble endoglin, an important antiangiogenic factor up-regulated in preeclampsia.


Objective


This study aimed (1) to develop models for the calculation of multiples of the mean values for mean arterial pressure and biochemical markers; (2) to build and assess the predictive models for preeclampsia based on maternal risk factors, the biophysical (mean arterial pressure) and biochemical (placental growth factor, soluble vascular endothelial growth factor receptor-1, and soluble endoglin) markers collected throughout pregnancy; and (3) to evaluate how prediction accuracy is affected by the presence of chronic hypertension and gestational age.


Study Design


This longitudinal case-cohort study included 1150 pregnant women: women without preeclampsia with (n=49) and without chronic hypertension (n=871) and those who developed preeclampsia (n=166) or superimposed preeclampsia (n=64). Mean arterial pressure and immunoassay-based maternal plasma placental growth factor, soluble vascular endothelial growth factor receptor-1, and soluble endoglin concentrations were available throughout pregnancy (median of 5 observations per patient). A prior-risk model for preeclampsia was established by using Poisson regression based on maternal characteristics and obstetrical history. Next, multiple regression was used to fit biophysical and biochemical marker data as a function of maternal characteristics by using data collected at 8 to 15 +6 , 16 to 19 +6 , 20 to 23 +6 , 24 to 27 +6 , 28 to 31 +6 , and 32 to 36 +6 week intervals, and observed values were converted into multiples of the mean values. Then, multivariable prediction models for preeclampsia were fit based on the biomarker multiples of the mean data and prior-risk estimates. Separate models were derived for overall, preterm, and term preeclampsia, which were evaluated by receiver operating characteristic curves and sensitivity at fixed false-positive rates.


Results


(1) The inclusion of soluble endoglin in prediction models for all preeclampsia, together with the prior-risk estimates, mean arterial pressure, placental growth factor, and soluble vascular endothelial growth factor receptor-1, increased the sensitivity (at a fixed false-positive rate of 10%) for early prediction of superimposed preeclampsia, with the largest increase (from 44% to 54%) noted at 20 to 23 +6 weeks (McNemar test, P <.05); (2) combined evidence from prior-risk estimates and biomarkers predicted preterm preeclampsia with a sensitivity (false-positive rate, 10%) of 55%, 48%, 62%, 72%, and 84% at 8 to 15 +6 , 16 to 19 +6 , 20 to 23 +6 , 24 to 27 +6 , and 28 to 31 +6 week intervals, respectively; (3) the sensitivity for term preeclampsia (false-positive rate, 10%) was 36%, 36%, 41%, 43%, 39%, and 51% at 8 to 15 +6 , 16 to 19 +6 , 20 to 23 +6 , 24 to 27 +6 , 28 to 31 +6 , and 32 to 36 +6 week intervals, respectively; (4) the detection rate for superimposed preeclampsia among women with chronic hypertension was similar to that in women without chronic hypertension, especially earlier in pregnancy, reaching at most 54% at 20 to 23 +6 weeks (false-positive rate, 10%); and (5) prediction models performed comparably to the Fetal Medicine Foundation calculators when the same maternal risk factors and biomarkers (mean arterial pressure, placental growth factor, and soluble vascular endothelial growth factor receptor-1 multiples of the mean values) were used as input.


Conclusion


We introduced prediction models for preeclampsia throughout pregnancy. These models can be useful to identify women at risk during the first trimester who could benefit from aspirin treatment or later in pregnancy to inform patient management. Relative to prediction performance at 8 to 15 +6 weeks, there was a substantial improvement in the detection rate for preterm and term preeclampsia by using data collected after 20 and 32 weeks’ gestation, respectively. The inclusion of plasma soluble endoglin improves the early prediction of superimposed preeclampsia, which may be valuable when Doppler velocimetry of the uterine arteries is not available.


Introduction


Preeclampsia is a complex obstetrical syndrome responsible for maternal and infant morbidity and mortality worldwide. , A growing body of evidence indicates several clinical subtypes (eg, early or late, mild or severe, with or without hemolysis, elevated liver enzymes, and low platelet count syndrome) distinguishable at a molecular level involve different pathways, which lead to the clinical and laboratory manifestations that define the syndrome. Among these, an imbalance in angiogenic (placental growth factor [PlGF]) and antiangiogenic (soluble vascular endothelial growth factor receptor-1 [sVEGFR-1] and soluble endoglin [sEng]) factors is considered central to the pathophysiology of the terminal pathway of preeclampsia, especially in cases necessitating indicated preterm delivery. Therefore, these biomarkers have been proposed for the first-, second-, and third-trimester predictions of preeclampsia. , , , The prognostic value of such models , , , , , , for identifying patients at risk of preterm preeclampsia was proven superior to historic-based methods , , , and useful in clinical practice to guide decision-making regarding the administration of low-dose aspirin in the first trimester, steroids for fetal lung maturity, magnesium for seizure prophylaxis, and timing of delivery. , , ,



AJOG at a Glance


Why was this study conducted?


This study aimed to generate models and calculators for the prediction of preeclampsia based on maternal risk factors and multiples of the mean of biophysical and biochemical markers collected longitudinally.


Key findings


Relative to the results based on data collected at 8 to 15 +6 weeks, the sensitivity for preterm and term preeclampsia improved after 20 and 32 weeks, respectively. The inclusion of plasma soluble endoglin (sEng) improved early prediction of disease. Models performed similarly to the Fetal Medicine Foundation calculators when the same biomarker data were used as input, suggesting a modest impact of differences in the analytical approaches.


What does this add to what is known?


Models and calculators to predict preeclampsia throughout gestation were developed, and prediction performance increased with advancing gestational age. The inclusion of sEng increased accuracy early in gestation for preterm and superimposed preeclampsia, which may be valuable when Doppler velocimetry of the uterine arteries is unavailable.



State-of-the-art prediction models for preeclampsia were proposed in a series of papers published by the Fetal Medicine Foundation (FMF). These models involve a combination of maternal risk factors and multiples of the mean (MoM) values of mean arterial pressure (MAP), uterine artery pulsatility index (UtA-PI), and blood concentrations of PlGF, sVEGFR-1 (also known as sFLT-1), and pregnancy-associated plasma protein A (PAPP-A). The use of such prediction models in practice requires not only the availability of biomarker measurements but also standards to define their expected values, given maternal characteristics, gestational age at measurement, and the type of assay used to measure biochemical markers. Such customized biomarker standards (referred to herein as MoM models) are then used to determine how abnormal the observed biomarker values are by calculating the MoM values. Additional customization of risk cutoff values may be required depending on ethnicity and clinical site. For example, a pooled analysis of 3 prospective nonintervention screening studies (61,174 women with a singleton pregnancy; 1770 cases of preeclampsia) found that first-trimester screening in white women detected 70% of preterm preeclampsia (<37 weeks’ gestation) and 40% of term preeclampsia (≥37 weeks’ gestation) at a 10% false-positive rate (FPR). The same risk cutoff values applied in women of Afro-Caribbean racial origin led to detection rates of 92% for preterm preeclampsia and 75% for term preeclampsia, yet the FPR was 3-fold higher than that of white women.


Therefore, based on a previously described retrospective longitudinal study design, we aimed first to develop MoM models for biophysical (MAP) and biochemical (PlGF, sVEGFR-1, and sEng) markers in our majority African American population attending the Detroit Medical Center that could determine MoM values to be used in existing or novel prediction models for preeclampsia. A second goal was to assess the predictive performance of risk models based on maternal characteristics and obstetrical history and biophysical and biochemical marker MoM values throughout pregnancy. Finally, we evaluated how the prediction performance of the novel models was affected by the presence of chronic hypertension, the timing of delivery, and the method used to combine prior-risk and biomarker-based evidence. This latter aspect is important because state-of-the-art approaches for the prediction of preeclampsia give equal weight to both sources of evidence.


Materials and Methods


Study design


This was a retrospective analysis of data from 1150 pregnancies, previously described as part of a case-cohort of 1499 pregnancies on which we reported the prediction of early delivery of placentas presenting lesions of maternal vascular underperfusion.


The original multiple disease case-cohort (n=1499), from which this preeclampsia case-cohort study (n=1150) was drawn, was designed in 2 stages to include 1000 randomly selected women and all remaining major obstetrical complications (ie, preeclampsia, preterm labor, preterm prelabor rupture of the membranes, and small-for-gestational-age gestation [<5th percentile]) from a cohort of 4006 women with a singleton pregnancy, enrolled at 6 to 22 weeks’ gestation, in a longitudinal biomarker study. The preeclampsia case-cohort retained for this study included all women from the random sample of 1000 pregnancies and all additional preeclampsia cases from the cohort of 4006. Therefore, the resulting preeclampsia case-cohort included women without preeclampsia or chronic hypertension (normotensive controls, n=871), women with chronic hypertension without preeclampsia (hypertensive controls, n=49), women who developed preeclampsia (n=166), or women with preeclampsia superimposed on chronic hypertension (n=64). Of the 230 cases of preeclampsia, 83 were preterm preeclampsia (delivery at <37 weeks’ gestation).


Blood samples in the original case-cohort were collected in 3 to 5 of 6 predefined intervals of gestation (8–15 +6 , 16–19 +6 , 20–23 +6 , 24–27 +6 , 28–31 +6 , and 32–36 +6 weeks). Maternal plasma PlGF, sEng, and sVEGFR-1 were determined by enzyme-linked immunosorbent assays. In total, 6035 samples across multiple gestational intervals were included in the current analysis (median number of samples, 5; interquartile range [IQR], 4–6).


The use of clinical data and biologic specimens was approved by the Institutional Review Boards of Wayne State University and the Eunice Kennedy Shriver National Institute of Child Health and Human Development, National Institutes of Health, United States Department of Health and Human Services. All patients provided written informed consent before the collection of samples.


Sample collection and immunoassays


Samples were collected by venipuncture into tubes containing ethylenediaminetetraacetic acid, centrifuged, and stored at −70°C. The most centrally located venipuncture sample within each of the 6 intervals of gestational age for each patient was used for analysis. The inter- and intra-assay coefficients of variation of the assays were 1.4% and 3.9% for sVEGFR-1, 2.3% and 4.6% for sEng, and 6.02% and 4.8% for PlGF, respectively. The sensitivity of each assay was 16.97 pg/mL for sVEGFR-1, 0.08 ng/mL for sEng, and 9.52 pg/mL for PlGF. Sample collection methods, biospecimen processing, and validation of the assays used were previously reported in greater detail.


Clinical definitions and outcomes


The primary outcome was the development of preeclampsia. Secondary outcomes were preterm (<37 weeks’ gestation) preeclampsia, term preeclampsia, and superimposed preeclampsia. We defined uncomplicated pregnancy as a pregnancy with no major obstetrical, medical, or surgical complications, where women delivered a term neonate. Preeclampsia was defined as new-onset proteinuria and hypertension—blood pressure of ≥140/90 mm Hg on 2 occasions at least 4 hours apart or ≥160/110 mm Hg within a shorter interval (minutes)—at ≥20 weeks’ gestation. Proteinuria was defined as a urine protein of ≥300 mg in a 24-hour urine collection, or 2 random urine specimens, obtained 4 hours to 1 week apart, showing ≥1+by dipstick. Hypertension was defined as a systolic blood pressure of ≥140 and/or a diastolic blood pressure of ≥90 mm Hg, measured at least on 2 occasions, 4 hours to 1 week apart. Chronic hypertension was defined in women with hypertension at <20 weeks’ gestation or in those who reported a history of hypertension. For women with chronic hypertension, superimposed preeclampsia was diagnosed by a new onset of proteinuria (either 300 mg/24 hours or ≥2+by dipstick) or thrombocytopenia (platelet count of <100×10 3 /mm 3 ), elevated liver enzymes (aspartate aminotransferase or alanine aminotransferase of >70 IU/L), or pulmonary edema.


Statistical analysis


Demographic data analysis


Demographic and clinical characteristics were compared between the groups by using Fisher exact tests for categorical data and 2-tailed t tests for continuous variables, respectively. P <.05 was considered a statistically significant result.


Calculation of the multiples of the mean for biomarkers


The plasma concentrations of PlGF, sEng, and sVEGFR-1 and MAP were first logarithmically transformed to improve the normality of distribution and to stabilize variance. The expected values of these variables among controls (with and without chronic hypertension) were estimated as a function of maternal characteristics and gestational age in each interval of gestation by using linear regression. The following maternal characteristics were considered for inclusion in the regression models and retained if they contributed to decreasing the Akaike information criterion by using stepwise backward elimination: gestational age at sample collection, maternal age, nulliparity, history of preeclampsia, weight, smoking status, and interaction terms between these covariates and chronic hypertension. MoM values of biomarkers were then calculated as the ratio between the observed biomarker value and expected values for gestational age and maternal characteristics for all samples. MoM values were further log 10 transformed. Box-and-whisker plots (median, IQR, and range) of the biomarker log 10 MoM values for each outcome group were created. Two-tailed t tests were used to determine the significance of differences in the MoM values between groups.


Prior-risk model for preeclampsia by using maternal risk factors and obstetrical history


Poisson regression with sandwich estimation of variance was used to estimate the prior (anterior) risk of preeclampsia based on maternal characteristics and obstetrical history. Model selection was based on the findings from a logistic regression model with backward stepwise elimination of variables, starting with maternal age, weight, height, nulliparity, smoking status, history of preeclampsia, and interaction terms between these covariates and chronic hypertension. Once variables were selected, the Poisson regression model was fit, setting the weight of cases to 1.0, whereas the weights of the noncases were set so that the ratio of the total weight of cases to controls was the same as in the parent cohort of 4006 pregnancies.


Developing prediction models for preeclampsia by using prior risk and multiples of the mean data of biophysical and biochemical markers


Multivariable Poisson regression models were fit by using data collected in each interval of gestation (8–15 +6 , 16–19 +6 , 20–23 +6 , 24–27 +6 , 28–31 +6 , and 32–36 +6 weeks). The models included as predictors the log 10 MoM of MAP, PlGF, sVEGFR-1, sEng, and all pairwise interaction terms among these variables. The predicted risk (log thereof) based on the prior-risk model was included as an input in the biomarker-based model; hence, the 2 types of evidence were automatically weighted to maximize model fit. As in the prior-risk model, the weight of cases was set to 1.0 whereas the weights of controls were set so that the ratio of the total weight of cases to controls was the same as in the parent cohort of 4006 pregnancies. Cutoff values on the predicted combined-risk scores were identified so that the FPRs were either 10% or 20%. Performance indices (area under the receiver operating characteristic [ROC] curve [AUC], sensitivity, specificity, and positive [+] and negative [−] predictive likelihood ratios) were calculated for each interval of gestation. A risk assessment calculator is available from the authors’ website ( https://bioinformaticsprb.med.wayne.edu/software/ ). Analysis was performed using caret, ROCR, and pROC packages for the R statistical language and environment ( www.r-project.org ).


Results


Maternal characteristics, obstetrical history, and the prior risk of preeclampsia


Demographic and clinical characteristics of the study population according to pregnancy outcome are presented in Table 1 . The Poisson regression model used to calculate the prior risk of preeclampsia based on maternal characteristics and obstetrical history is presented in Supplemental Table 1 . Risk factors included in the model were chronic hypertension, maternal weight, nulliparity, and history of preeclampsia, and all were associated with an increased risk of disease. Based on these variables, the prior-risk model predicted overall preeclampsia, preterm preeclampsia, and term preeclampsia with an AUC of 0.7 (0.66–0.74), 0.67 (0.6–0.74), and 0.71 (0.67–0.76), respectively ( Figure 1 ).



Table 1

Demographic characteristics of the study population



















































































Clinical features Controls without chronic hypertension n=871 a Controls with chronic hypertension n=49 a Preeclampsia n=166 (56 a ) Superimposed preeclampsia n=64 (24 a )
Age in y, median (IQR) 23 (20–27) 26 (22–31) b 21 (19–26) 27 (22–32.2) b
Racial origin
African American, n (%) 850 (92.4) 47 (95.9) 159 (95.8) c 60 (93.8) b
White, n (%) 23 (2.5) 0 (0) 3 (1.8) b 1 (1.6) b
Other d , n (%) 47 (5.1) 2 (4.1) c 4 (2.4) b 3 (4.7) b
Nulliparity, n (%) 354 (38.5) 15 (30.6) b 87 (52.4) 19 (29.7) b
History of preeclampsia, n (%) 29 (3.2) 4 (8.2) 11 (6.6) 10 (15.6)
Smoking, n (%) 188 (20.4) 12 (24.5) b 28 (16.9) 22 (34.4) b
Weight (kg), median (IQR) 70 (59–86) 99 (73–115) b 73 (61–87.8) 91 (78–109) b
Height (cm), median (IQR) 162.6 (157.5–167.6) 165.1 (157.5–170.2) 162.6 (157.5–167.6) 165.1 (159.4–170.2)
Birthweight (g), median (IQR) 3185 (2820–3485) 3060 (2637–3265) c 2820 (2171.2–3268.8) b 2725 (2087.5–3270) b
Gestational age at delivery in wk, median (IQR) 39.3 (38.1–40.3) 38.4 (37.6–39.3) 37.7 (36.1–39.1) b 37.2 (35.6–38.2) b

Data are expressed as median (IQR) or number (percentage).

IQR , interquartile range.

Tarca et al. Prediction of preeclampsia throughout pregnancy. Am J Obstet Gynecol 2022 .

a Indicates the number of cases part of the random samples of 1000 pregnancies


b P <.001


c P <.05


d Includes women who identified as Hispanic, Asian, or “Other.” Maternal height and weight were recorded in inches and pounds and then converted into cm and kg, respectively, before analysis. Statistically significant differences are reported vs normotensive controls (Fisher exact for categorical variables and 2-sided t tests for continuous variables).




Figure 1


Prediction performance of the preeclampsia prior-risk model and the combined-risk models

ROC curve for prediction of ( A) overall preeclampsia ( B ) preterm preeclampsia, and ( C ) term preeclampsia based on the prior-risk model and the combined evidence (prior risk and biomarkers).

ROC , receiver operating characteristic.

Tarca et al. Prediction of preeclampsia throughout pregnancy. Am J Obstet Gynecol 2022.


Factors affecting the biophysical and biochemical marker data in women without preeclampsia


Concentrations of PlGF, sVEGFR-1, and sEng were measured in 6035 maternal plasma samples (median number of samples, 5; IQR, 4–6) collected throughout gestation from the 1150 women in the study case-cohort. The MAP data were also available at the date of blood sample collection. To create standards for MoM calculation, the biomarker data among women without preeclampsia were fit as a function of maternal characteristics and obstetrical history by using linear regression. The MoM models ( Supplemental Table 2 ) show that biochemical and biophysical markers in women without preeclampsia change with maternal weight, height, parity, smoking status, and history of preeclampsia and that the effect of these covariates is modified by the presence of chronic hypertension.


Distribution of biomarker multiples of the mean data by the presence or absence of preeclampsia, chronic hypertension status, and gestational age at blood sample collection


The biomarker MoM values (log 10 thereof) were displayed and compared among groups based on disease status (ie, preeclampsia) and the presence of chronic hypertension ( Supplemental Figure 1 ). The angiogenic profiles differed by preeclampsia and chronic hypertension status throughout gestation ( Supplemental Figure 1 ). The PlGF MoM values were significantly lower and sEng MoM values were higher in the preeclampsia and/or superimposed preeclampsia groups than in the normotensive control group from the 20 to 23 +6 week interval onward (all P <.05). The sVEGFR-1 MoM values were higher in the preeclampsia and/or superimposed preeclampsia group than in the normotensive control group from the 24 to 27 +6 week interval onward (all P <.05). Of note, after 20 weeks onward, the PlGF MoM values were more abnormal in the preeclampsia group than in the superimposed preeclampsia group, relative to normotensive controls ( Supplemental Figure 1 ). The MAP MoM values were significantly increased, starting with the 8 to 15 +6 week interval in the preeclampsia and/or superimposed preeclampsia group compared to the control group, and the magnitude of the difference increased with advancing gestational age.


At 8 to 15 +6 weeks, the log 10 PlGF MoM values showed a significant correlation with gestational age at delivery with preeclampsia, yet this was not the case for MAP ( Supplemental Figure 2 ). This finding suggests that although abnormally high MAP MoM values at this gestational age are predictive of preeclampsia, they do not predict gestational age at delivery with preeclampsia, which was the case in other reports that used a standardized approach to blood pressure measurement.


Prediction performance for preeclampsia based on combined prior risk and evidence from biomarkers


Prediction models for preeclampsia, which are based on combined prior risk and evidence from biophysical and biochemical marker values throughout gestation, are presented in Supplemental Table 3 . Of note, the combined-risk models displayed different weights for the prior-risk component (coefficient for prior-risk variable in Supplemental Table 3 ) depending on the gestational age at measurements. The weight of the prior-risk component diminished with advancing gestational age at measurement as the biomarker data became more informative. Assigning a fixed weight to the prior-risk component would have resulted in significantly lower sensitivity (FPR, 10%) than allowing it to vary as shown in Supplemental Table 3 (McNemar test, P <.05 at 20–23 +6 , 24–27 +6 , and 28–31 +6 weeks when combining prior risk with PlGF-derived evidence for predicting all preeclampsia).


The ROC curves for the prediction of preeclampsia by the prior-risk model and the combination of prior risk with biomarker-based evidence collected throughout gestation are presented in Figure 1 . The AUC for the prediction of all preeclampsia slightly improved from 0.75 (0.69–0.8), when only prior-risk evidence was considered, to 0.76 (0.71–0.81) when the prior risk was combined with evidence from the biomarker-based model by using data from patients with available biomarker data collected at 8 to 15 +6 weeks ( Table 2 ). As gestational age increased, hence approaching diagnosis, the AUC increased to 0.86 (0.83–0.89) at 32 to 36 +6 weeks. The detection rates (FPR, 10%) for preeclampsia (all cases) were 44%, 36%, 45%, 52%, 55%, and 61% at 8 to 15 +6 , 16 to 19 +6 , 20 to 23 +6 , 24 to 27 +6 , 28 to 31 +6 , and 32 to 36 +6 week intervals, respectively.



Table 2

Screening performance for preeclampsia using the prior-risk model and the combined-risk model (prior risk and biomarkers)












































































































































































































































































Gestational age (wk) Prior-risk AUC Combined-risk AUC Combined-risk FPR of 10% Combined-risk FPR of 20%
Cutoff Sensitivity LR+ LR− Cutoff Sensitivity LR+ LR−
All preeclampsia
8–15 +6 0.75 (0.69–0.8) 0.76 (0.71–0.81) 0.084 0.44 (0.35–0.54) 4.43 (3.14–6.24) 0.62 (0.52–0.73) 0.055 0.55 (0.45–0.64) 2.72 (2.12–3.49) 0.57 (0.46–0.7)
16–19 +6 0.68 (0.63–0.73) 0.71 (0.66–0.76) 0.084 0.36 (0.28–0.45) 3.62 (2.61–5.01) 0.71 (0.63–0.81) 0.058 0.53 (0.44-0.61) 2.62 (2.09–3.28) 0.59 (0.5–0.71)
20–23 +6 0.67 (0.62–0.72) 0.77 (0.72–0.81) 0.087 0.45 (0.37–0.53) 4.51 (3.4–5.98) 0.61 (0.53–0.7) 0.061 0.6 (0.52–0.67) 3.00 (2.46–3.65) 0.50 (0.41–0.61)
24–27 +6 0.70 (0.65–0.74) 0.80 (0.76–0.83) 0.081 0.52 (0.44–0.59) 5.13 (3.97–6.63) 0.54 (0.46–0.63) 0.055 0.65 (0.57–0.72) 3.23 (2.7–3.86) 0.44 (0.36–0.54)
28–31 +6 0.69 (0.64–0.73) 0.80 (0.76–0.84) 0.075 0.55 (0.48–0.63) 5.58 (4.35–7.15) 0.49 (0.42–0.58) 0.051 0.66 (0.59–0.73) 3.32 (2.79–3.95) 0.42 (0.34–0.52)
32–36 +6 0.69 (0.65–0.74) 0.86 (0.83–0.89) 0.084 0.61 (0.54–0.69) 6.15 (4.86–7.79) 0.43 (0.36–0.52) 0.05 0.74 (0.67–0.8) 3.69 (3.14–4.34) 0.33 (0.25–0.42)
Preterm preeclampsia
8–15 +6 0.70 (0.6–0.8) 0.78 (0.7–0.86) 0.045 0.55 (0.39–0.7) 5.45 (3.71–8.02) 0.50 (0.36–0.7) 0.024 0.62 (0.46–0.76) 3.08 (2.29–4.15) 0.48 (0.32–0.7)
16–19 +6 0.65 (0.56–0.74) 0.80 (0.74–0.86) 0.04 0.48 (0.34–0.62) 4.84 (3.35–6.98) 0.58 (0.44–0.75) 0.026 0.63 (0.49–0.76) 3.14 (2.42–4.07) 0.46 (0.33–0.66)
20–23 +6 0.63 (0.54–0.72) 0.88 (0.83–0.92) 0.034 0.62 (0.48–0.75) 6.15 (4.53–8.36) 0.42 (0.3–0.59) 0.018 0.76 (0.63–0.87) 3.83 (3.1–4.73) 0.30 (0.18–0.48)
24–27 +6 0.65 (0.56–0.73) 0.91 (0.88–0.95) 0.027 0.72 (0.6–0.83) 7.18 (5.53–9.32) 0.31 (0.21–0.46) 0.016 0.82 (0.7–0.9) 4.08 (3.39–4.9) 0.23 (0.14–0.39)
28–31 +6 0.64 (0.55–0.73) 0.94 (0.91–0.98) 0.016 0.84 (0.71–0.92) 8.41 (6.61–10.68) 0.18 (0.1–0.33) 0.009 0.87 (0.76–0.95) 4.36 (3.67–5.17) 0.16 (0.08–0.32)
32–36 +6 0.65 (0.56–0.73) 0.97 (0.95–0.99) 0.016 0.92 (0.81–0.98) 9.26 (7.44–11.53) 0.09 (0.03–0.22) 0.006 0.96 (0.87–1) 4.8 (4.14–5.55) 0.05 (0.01–0.19)
Term preeclampsia
8–15 +6 0.78 (0.72–0.84) 0.78 (0.72–0.84) 0.056 0.36 (0.25–0.49) 3.62 (2.38–5.5) 0.71 (0.59–0.85) 0.034 0.64 (0.51–0.75) 3.17 (2.45–4.1) 0.46 (0.33–0.63)
16–19 +6 0.70 (0.64–0.76) 0.71 (0.65–0.77) 0.047 0.36 (0.26–0.48) 3.67 (2.53–5.3) 0.71 (0.6–0.83) 0.036 0.49 (0.38–0.6) 2.46 (1.88–3.22) 0.63 (0.51–0.78)
20–23 +6 0.69 (0.64–0.75) 0.73 (0.68–0.79) 0.06 0.41 (0.32–0.51) 4.12 (2.99–5.68) 0.65 (0.55–0.77) 0.041 0.54 (0.44–0.64) 2.70 (2.14–3.41) 0.58 (0.47–0.71)
24–27 +6 0.73 (0.68–0.78) 0.77 (0.72–0.82) 0.053 0.43 (0.34–0.53) 4.31 (3.19–5.81) 0.63 (0.53–0.74) 0.038 0.58 (0.48–0.67) 2.88 (2.32–3.56) 0.53 (0.43–0.66)
28–31 +6 0.71 (0.66–0.76) 0.75 (0.7–0.8) 0.054 0.39 (0.3–0.48) 3.92 (2.88–5.33) 0.68 (0.59–0.78) 0.036 0.59 (0.5–0.68) 2.96 (2.42–3.63) 0.51 (0.41–0.63)
32–36 +6 0.71 (0.66–0.76) 0.82 (0.78–0.86) 0.061 0.51 (0.42–0.6) 5.14 (3.93–6.72) 0.54 (0.45–0.65) 0.04 0.68 (0.59–0.76) 3.38 (2.82–4.06) 0.40 (0.31–0.52)

The sensitivity at fixed 10% and 20% FPR for the gestational-age interval–specific models predicting preeclampsia. Cutoffs were chosen so that FPR is 10% or 20%. AUC CIs were calculated using DeLong method. Note that slight variations in prediction performance of the prior-risk model are caused by differences in the sets of cases and controls with an available sample in each interval.

AUC , area under the receiver operating characteristic curve; CI , confidence interval; FPR , false-positive rate; LR , likelihood ratio.

Tarca et al. Prediction of preeclampsia throughout pregnancy. Am J Obstet Gynecol 2022 .


The AUC for prediction of term preeclampsia by prior-risk and biomarkers was 0.78 (0.72–0.84) at 8 to 15 +6 weeks and did not improve with advancing gestational age at measurements until the 32 to 36 +6 week interval when it reached 0.82 (0.78–0.86). The detection rates (FPR, 10%) for term preeclampsia were 36%, 36%, 41%, 43%, 39%, and 51% at 8 to 15 +6 , 16 to 19 +6 , 20 to 23 +6 , 24 to 27 +6 , 28 to 31 +6 , and 32 to 36 +6 week intervals, respectively.


In agreement with previous reports, the prediction of preterm preeclampsia was more accurate than that of term preeclampsia. The AUC for preterm preeclampsia improved from 0.7 (0.6–0.8), when only prior-risk evidence was considered, to 0.78 (0.7–0.86) when the prior-risk evidence was combined with evidence derived from biomarkers at 8 to 15 +6 weeks, and it reached 0.94 (0.91–0.98) at the 28 to 31 +6 week interval. The detection rates (FPR, 10%) for preterm preeclampsia were 55%, 48%, 62%, 72%, and 84% at 8 to 15 +6 , 16 to 19 +6 , 20 to 23 +6 , 24 to 27 +6 , and 28 to 31 +6 week intervals, respectively.


When the prediction models for all preeclampsia were evaluated in the subset of women with chronic hypertension (comparison of superimposed preeclampsia with chronic hypertension without superimposed preeclampsia), the detection rates (FPR, 10%) were 40%, 38% 54%, 48%, 38%, and 43% at 8 to 15 +6 , 16 to 19 +6 , 20 to 23 +6 , 24 to 27 +6 , 28 to 31 +6 , and 32 to 36 +6 week intervals, respectively ( Figure 2 , Table 3 ). Of note, the risk cutoff values required to reach the same FPR when screening patients with chronic hypertension ( Table 3 ) were higher than the cutoff values used when screening the entire study population ( Table 2 ). This finding can be explained by the higher prior risk of disease among women with chronic hypertension than in the overall population.




Figure 2


Prediction performance of the preeclampsia combined-risk models for patients with and without chronic hypertension

ROC curve for prediction of preeclampsia are based on the all preeclampsia risk models in Supplemental Table 3 . Separate ROC curves are drawn for patients with and without chronic hypertension.

ROC , receiver operating characteristic.

Tarca et al. Prediction of preeclampsia throughout pregnancy. Am J Obstet Gynecol 2022.


Table 3

Prediction of preeclampsia in women with chronic hypertension






































































































GA (wk) Prior-risk AUC Combined-risk AUC Combined-risk FPR of 10% Combined-risk FPR of 20%
Cutoff Sensitivity LR+ LR− Cutoff Sensitivity LR+ LR−
All preeclampsia
8–15 +6 0.49 (0.34–0.63) 0.67 (0.54–0.8) 0.325 0.40 (0.24–0.58) 4.13 (1.31–13.05) 0.66 (0.5–0.89) 0.307 0.40 (0.24–0.58) 2.07 (0.91–4.72) 0.74 (0.54–1.03)
16–19 +6 0.52 (0.39–0.66) 0.64 (0.51–0.77) 0.263 0.38 (0.22–0.55) 3.50 (1.27–9.64) 0.70 (0.53–0.92) 0.236 0.43 (0.27–0.61) 2.29 (1.07–4.9) 0.70 (0.51–0.97)
20–23 +6 0.56 (0.44–0.68) 0.69 (0.59–0.8) 0.284 0.54 (0.39–0.69) 4.98 (2.09–11.86) 0.51 (0.37–0.71) 0.214 0.60 (0.45–0.74) 3.09 (1.65–5.79) 0.49 (0.34–0.72)
24–27 +6 0.51 (0.39–0.63) 0.65 (0.53–0.76) 0.243 0.48 (0.34–0.62) 4.81 (1.82–12.7) 0.58 (0.44–0.76) 0.224 0.52 (0.38–0.66) 2.60 (1.32–5.09) 0.60 (0.44–0.83)
28–31 +6 0.55 (0.43–0.67) 0.70 (0.59–0.81) 0.312 0.38 (0.25–0.54) 3.93 (1.45–10.66) 0.68 (0.53–0.88) 0.202 0.55 (0.4–0.7) 2.84 (1.45–5.56) 0.56 (0.39–0.79)
32–36 +6 0.48 (0.36–0.6) 0.75 (0.65–0.85) 0.288 0.43 (0.29–0.59) 4.09 (1.68–9.97) 0.63 (0.48–0.83) 0.219 0.59 (0.43–0.73) 3.07 (1.62–5.79) 0.51 (0.35–0.74)

Only gold members can continue reading. Log In or Register to continue

Stay updated, free articles. Join our Telegram channel

Feb 23, 2022 | Posted by in OBSTETRICS | Comments Off on Prediction of preeclampsia throughout gestation with maternal characteristics and biophysical and biochemical markers: a longitudinal study

Full access? Get Clinical Tree

Get Clinical Tree app for offline access