Abstract
Purpose
We conduct this study to develop and validate a predictive nomogram for preeclampsia (PE) to inform the development of early intervention strategies in clinical practice.
Methods
In this analysis, we collected data from women with medium or high risk for PE who underwent placental growth factor (PlGF)-based testing between December 20, 2021 and December 31, 2022. The gestational age at the time of taking the PlGF-based test for the PE and non-PE groups was 20.0 weeks (range 16.1–26.1 weeks) and 22.2 weeks (range 16.2–27.3 weeks), respectively. The independent risk factors for PE were identified through both univariate and multivariate analyses. Based on these independent risk factors, a logistic regression model for risk prediction was developed. The model was validated using five-fold cross-validation. Moreover, the efficacy of the model was appraised using the area under the receiver operating characteristic curve (AUROC), while the calibration of the model was assessed through calibration curves. Additionally, decision curves and clinical impact curves were leveraged to evaluate the clinical applicability of the model.
Results
In total, 2063 women were included. Of these, 108 had PE. Body mass index, mean arterial pressure, a ratio of soluble fms-like tyrosine kinase-1/PlGF, history of adverse pregnancy, family history of PE, previous history of PE, chronic hypertension, autoimmune disease, and polycystic ovary syndrome were independent risk factors for PE. The model constructed based on independent risk factors demonstrated that the AUROC in the training set was 0.883 (95% confidence interval [CI] 0.838–0.928), with a sensitivity of 0.827 and specificity of 0.816. In the validation set, the AUROC was 0.862 (95% CI 0.774–0.951), with a sensitivity of 0.815 and specificity of 0.772. The decision curve revealed that the model had a large probability interval for the net benefit threshold.
Conclusion
The predictive nomogram for PE constructed based on common interpretable features has desirable efficacy, which informs the development of specialized preventive protocols in clinical practice.
Supplementary Information
The online version contains supplementary material available at 10.1007/s00404-025-08076-6.
Keywords: Preeclampsia, Nomogram, Placental growth factor, Soluble fms-like tyrosine kinase-1
What does this study add to the clinical work
The model showed ideal predictive performance in the validation set. This can be used in clinical practice to identify high-risk groups early and help develop closer monitoring and timely preventive measures. After patients understand the risk of diseases, the compliance of patients can be increased, maximizing the effectiveness of specific preventive measures and reducing the occurrence of adverse pregnancy.
Introduction
Preeclampsia (PE) is a critical disorder affecting multiple systems. It complicates about 2–5% of pregnancies, contributing to 76,000 maternal deaths and 500,000 deaths among babies every year [1, 2]. After PE onset, the only proven method to stop it is the delivery of the fetus and placenta [3, 4]. Therefore, accurately predicting this disorder is essential as it would enable the subsequent application of preventive clinical management strategies [3, 4].
Identifying risk factors based on the mother’s medical history and demographic profile is a common clinical method for predicting PE [5]. Nevertheless, this strategy exhibits inferior performance in prediction [6–8]. An alternative method involves using maternal serum biomarkers alone, including placental growth factor (PlGF), soluble fms-like tyrosine kinase-1 (sFlt-1), and a sFlt-1/PlGF ratio. It has been shown that PlGF alone can predict early PE onset with a detection rate of about 55 at a 10% false positive rate [9, 10]. However, there is still debate on the clinical application value of PlGF and the sFlt-1/PlGF ratio in predicting PE [11–13]. A single-center study has confirmed that quantification of a panel of cfRNA levels in maternal serum could be used to effectively detect PE with a sensitivity of 1.0000 in the early stages of pregnancy and throughout the entire spectrum of pregnancy. However, multicenter large-scale clinical studies are warranted [14]. A meta-analysis revealed that the uterine arteries pulsatility index measured by Doppler ultrasound was useful and effective for predicting PE with high specificity for predicting preeclampsia (0.879) [15]. Nevertheless, this study demonstrated a relatively low sensitivity of 0.586 [15]. Co-screening is currently trending toward the commencement of PE prediction like an algorithm created by the Fetal Medicine Foundation (FMF). Numerous studies have assessed the performance of the FMF algorithm and concluded that the detection rate of preterm PE was approximately 75% and term PE 43% at a screen-positive rate of 10% [6, 8, 16, 17].
While the FMF algorithm offers superior predictive performance compared to using only maternal risk factors or serum biomarkers, the screen-positive rate for both approaches remains low. Furthermore, the predictive capability of individual risk factors for PE is noticeably limited, particularly when dealing with binary variables. In clinical practice, there are numerous independent risk factors tied to PE. However, identifying high-risk populations for PE based solely on these independent risk factors presents a considerable challenge. This difficulty arises from the complexity of quantifying the link between each risk factor and the risk of PE. We hypothesize that constructing a multivariable prediction model may ideally address these challenges. Therefore, we develop and validate a predictive model of PE to inform the development of specific preventive protocols in clinical practice.
Methods
Participants and study design
We conducted a retrospective collection of electronic medical records for all women with medium- or high-risk factors who voluntarily underwent a PlGF-based test at Linyi People’s Hospital between December 20, 2021, and December 31, 2022, utilizing the hospital information system. As the hospital is a tertiary institution and many pregnant women are initially examined at local hospitals in their first trimester, those with risk factors often arrive at the hospital for prenatal examinations at a later gestational age (GA). PlGF measurement could be performed on all pregnant women with risk factors who have not been diagnosed with PE. The medium- and high-risk factors are as follows: maternal age (> 35 years), high body mass index (BMI) (> 30 kg/m2), comorbidities (including chronic hypertension, type 1 or type 2 diabetes, renal disease, and autoimmune), history of PE, history of other pregnancy hypertensive disorder, family history of PE (mother or sister), assisted reproduction, nulliparity, multifetal gestation, more than 10-year pregnancy interval, and obstetric history (low birthweight, small gestational age [GA], or previous adverse pregnancy outcome) [2, 18]. The following patients were excluded: (i) women who could not be followed until the onset of PE, including those who underwent artificial termination of pregnancy not attributable to PE, those whose fetuses died before 24 weeks of gestation, or those who delivered at other hospitals; (ii) women diagnosed with PE, gestational hypertension, or PE superimposed on chronic hypertension. All women who voluntarily consented to participate in the PlGF-based testing had provided written consent.
The study protocol was approved by the Science Research Ethics Committee, Linyi People’s Hospital (YX200683). We reported the study in line with the TRIPOD recommendations [19].
Outcome
The outcome of the study was the incidence of PE, which is clinically defined as the new onset of hypertension accompanied by proteinuria or end-organ failure that occurs after 20 weeks of gestation in women who were normotensive prior to this period, or both [2, 18, 20, 21]. Preeclampsia superimposed on chronic hypertension is also included. Hypertension is defined as either a systolic blood pressure (SBP) of at least 140 mmHg or a diastolic blood pressure (DBP) of at least 90 mmHg, or both, recorded on two different occasions that are at least four hours apart. Proteinuria is characterized by a 24 h urinary protein level that is greater than 300 mg, or by a routine urine test with a dipstick reading of 2 + or above [2, 18, 20, 21].
Predictor variables
We collected the following information from the patients: maternal age, height (cm), admission weight (kg), maternal weight gain (kg), pre-pregnancy weight (kg), BMI, SBP (mmHg), DBP (mmHg), mean arterial pressure (MAP), GA at screening PlGF (wk. + d), PlGF (pg./mL), sFlt-1(pg./mL), sFlt-1/PlGF ratio, nulliparity, type 1 or type 2 diabetes, pregnancy with hypothyroidism, pregnancy with hyperthyroidism, history of adverse pregnancy, multifetal gestation, conception by assisted reproductive technique, family history of PE, history of PE, renal disease, chronic hypertension, autoimmune disease, prior pregnancy with placental abruption (PPWPA), prior pregnancy with fetal growth restriction, aspirin, previous history of stillbirth, and polycystic ovary syndrome (PCOS) [2, 18, 20, 21]. MAP was calculated according to the algorithm of (SBP + 2* DBP)/3. Maternal serum concentrations of biomarkers were measured by a multi-channel dry fluorescence immunoassay platform (AFS2000A analyzer; Lambert Diagnostics, Guangzhou, China). A series of quality control systems were strictly implemented to ensure consistency of measurement of biomarkers. Due to the absence of routine ultrasound monitoring for uterine artery and umbilical cord blood flow at our institution, Doppler analysis on the uterine artery was not incorporated into this study.
Missing data
In the sample included in this study, some variables exhibited a small amount of missing values. This issue was addressed utilizing a multiple imputation method.
Statistical analysis
For continuous variables that followed a normal distribution, data were reported as mean ± standard deviation, and an independent samples t test was conducted to compare intergroup differences. When a skewed distribution was presented, the Mann–Whitney U test was used to compare differences between groups. Measurement data were expressed as n (%), with the chi-square test employed to analyze the differences between groups.
The differences in each parameter were compared between PE and non-PE cohorts. Variables with P < 0.1 were included in a multifactorial stepwise logistic regression to analyze independent risk factors for PE. Additionally, a prediction model based on logistic regression was developed using independent risk factors. The robustness of the model was assessed utilizing five-fold cross-validation. The best model in different validation sets was selected. The efficacy of the model was evaluated utilizing the area under the receiver operating characteristic curve (AUROC). According to the optimal Youden index identified in the training set, a high-risk probability threshold was established to differentiate between high-risk and low-risk PE. The calibration of the model was assessed through calibration curves. Additionally, decision curves and clinical impact curves were leveraged to evaluate the clinical applicability of the model. The study was conducted in R4.4.1.
Results
Study population and baseline characteristics
In total, 9263 women at medium or high risk of developing PE voluntarily underwent a PlGF-based test in clinical routine. After two stages of exclusion, the study finally included 2063 women. Of these, 108 patients were diagnosed with PE (Fig. 1).
Fig. 1.
Flow chart of women included in the study
The characteristics of the women in our study cohort are presented in Table 1. The GA at the time of taking the PlGF-based test for the PE and non-PE groups was 20.0 weeks (range 16.1–26.1 weeks) and 22.2 weeks (range 16.2–27.3 weeks) respectively, showing no notable difference (P = 0.404). Additionally, the sFlt-1 levels between the two groups exhibited no remarkable difference (P = 0.048). However, noticeable differences were observed in BMI, PlGF levels, MAP, and sFlt-1/PlGF ratio between the two groups (P < 0.05).
Table 1.
Characteristics of women in the study population
| Variables | Non-preeclampsia | Preeclampsia | P value |
|---|---|---|---|
| N = 1955 | N = 108 | ||
| Maternal age | 32 (27, 34) | 33 (27, 36.8) | 0.063 |
| Height (cm) | 160 (163, 166) | 162 (160, 165) | 0.246 |
| Admission weight (kg) | 75 (68.5, 83) | 84.5 (74, 97) | < 0.001 |
| Maternal weight gain (kg) | 15 (11, 19) | 14 (10, 19.8) | 0.315 |
| Pre-pregnancy weight (kg) | 60 (54, 67.5) | 71 (59.5, 81) | < 0.001 |
| BMI | 22.58 (20.37, 25.16) | 27.00 (22.84, 31.20) | < 0.001 |
| SBP (mmHg) | 121 (120, 122) | 122 (120, 130) | < 0.001 |
| DBP (mmHg) | 70 (66, 75) | 75 (68, 84.7) | < 0.001 |
| MAP | 86.67 (86.47, 93.33) | 92.67 (85.33, 98.00) | < 0.001 |
| GA at screening PlGF (wk. + d) | 20.0 (16.1, 26.1) | 22.2 (16.2, 27.3) | 0.404 |
| PlGF (pg./mL) | 103 (74.2, 157) | 76.5 (52.5, 112) | < 0.001 |
| sFlt-1 (pg./mL) | 1464 (1240, 1750) | 1602 (1278, 1847) | 0.048 |
| sFlt-1/PlGF ratio | 14.1 (9.38, 20.0) | 22.8 (13.0, 33.0) | < 0.001 |
| Nulliparity (Yes) | 733 (37.5) | 47 (43.5) | 0.209 |
| Type 1 or type 2 diabetes (Yes) | 367 (18.8) | 35 (32.4) | < 0.001 |
| Pregnancy with hypothyroidism (Yes) | 119 (6.09) | 8 (7.41) | 0.578 |
| Pregnancy with hyperthyroidism (Yes) | 5 (0.26) | 0 (0.00) | 1.000 |
| History of adverse pregnancy (Yes) | 102 (5.22) | 19 (17.6) | < 0.001 |
| Multifetal gestation (Yes) | 71 (3.63) | 6 (5.56) | 0.305 |
| Conception by ART (Yes) | 143 (7.31) | 9 (8.33) | 0.693 |
| Family history of PE (Yes) | 6 (0.31) | 9 (8.33) | < 0.001 |
| History of PE (Yes) | 4 (0.20) | 8 (7.41) | < 0.001 |
| Renal disease (Yes) | 2 (0.10) | 7 (6.48) | < 0.001 |
| Chronic hypertension (Yes) | 21 (1.07) | 21 (19.4) | < 0.001 |
| Autoimmune disease (Yes) | 21 (1.07) | 17 (15.7) | < 0.001 |
| PPWPA (Yes) | 2 (0.10) | 0 (0.00) | 1.000 |
| Prior pregnancy with FGR (Yes) | 2 (0.10) | 0 (0.00) | 1.000 |
| Aspirin (Yes) | 28 (1.43) | 15 (13.9) | < 0.001 |
| Previous history of stillbirth (Yes) | 15 (0.77) | 1 (0.93) | 0.578 |
| PCOS (Yes) | 26 (1.33) | 14 (13.0) | < 0.001 |
BMI body mass index, SBP systolic blood pressure, DBP diastolic blood pressure, MAP mean arterial pressure, GA gestational age, PlGF placental growth factor, sFlt-1 soluble fms-like tyrosine kinase-1, ART assisted reproductive technique, FGR fetal growth restriction, PCOS polycystic ovary syndrome, PPWPA prior pregnancy with placental abruption
Feature selection
In PE and non-PE populations, admission weight (kg), pre-pregnancy weight (kg), BMI, SBP (mmHg), DBP (mmHg), MAP, PlGF (pg./mL), sFlt-1(pg./mL), the sFlt-1/PlGF ratio, type 1 or type 2 diabetes, history of adverse pregnancy, family history of PE, history of PE, renal disease, chronic hypertension, autoimmune disease, aspirin, and PCOS exhibited notable differences. Furthermore, the difference in maternal age between the two groups was at a critical threshold. The above indicators were included in a multifactorial stepwise logistic regression. The results indicated that BMI, MAP, the sFlt-1/PlGF ratio, history of adverse pregnancy, family history of PE, history of PE, chronic hypertension, autoimmune disease, and PCOS were independent risk factors for PE (Tables 1 and 2).
Table 2.
Analysis of independent risk factors for preeclampsia
| Factors | β | se | Wald χ2 | P | OR (95%CI) |
|---|---|---|---|---|---|
| BMI | 0.234 | 0.031 | 55.113 | < 0.001 | 1.263 (1.188–1.343) |
| MAP | 0.069 | 0.018 | 14.353 | < 0.001 | 1.071 (1.034–1.110) |
| sFlt-1/PlGF ratio | 0.071 | 0.010 | 47.140 | < 0.001 | 1.074 (1.052–1.095) |
| History of adverse pregnancy (Yes) | 1.193 | 0.358 | 11.092 | 0.001 | 3.297 (1.634–6.652) |
| Family history of PE (Yes) | 2.988 | 0.721 | 17.173 | < 0.001 | 19.852 (4.831–81.584) |
| History of PE (Yes) | 2.531 | 0.835 | 9.186 | 0.002 | 12.564 (2.445–64.555) |
| Chronic hypertension (Yes) | 1.804 | 0.460 | 15.363 | < 0.001 | 6.071 (2.464–14.960) |
| Autoimmune disease (Yes) | 2.765 | 0.465 | 35.356 | < 0.001 | 15.878 (6.382–39.502) |
| PCOS (Yes) | 2.010 | 0.482 | 17.428 | < 0.001 | 7.465 (2.905–19.183) |
| Constant | -16.806 | 1.846 | 82.917 | < 0.001 |
Model building
We developed a predictive model based on logistic regression utilizing independent risk factors and evaluated the performance of the model through five-fold cross-validation. The results of the cross-validation indicated that the AUROC in the training set ranged from 0.881 to 0.890, while the range in the validation set was 0.852 to 0.912 (Fig. 2). In the analysis, the number of PE cases exhibited a notably imbalanced distribution (PE accounted for only 5.23% of the included cases). Therefore, the model with the best sensitivity was selected as the recommended model and a predictive nomogram was constructed (Table 3 and Fig. 3).
Fig. 2.
Receiver operating characteristic curves for five-fold cross-validation in the training and validation sets
Table 3.
Results of five-fold cross-validation
| k-Fold | Preeclampsia size | Training set | Validation set | |||||
|---|---|---|---|---|---|---|---|---|
| Training set | Validation set | c-Index (95% CI) | sen | spe | c-Index (95% CI) | sen | spe | |
| Fold 1 | 94 | 14 | 0.878 (0.835–0.922) | 0.766 | 0.865 | 0.912 (0.827–0.998) | 0.714 | 0.872 |
| Fold 2 | 83 | 25 | 0.882 (0.838–0.928) | 0.795 | 0.840 | 0.875 (0.787–0.963) | 0.760 | 0.832 |
| Fold 3 | 90 | 18 | 0.881 (0.839–0.923) | 0.711 | 0.910 | 0.885 (0.767–1.000) | 0.778 | 0.919 |
| Fold 4 | 81 | 27 | 0.883 (0.838–0.928) | 0.827 | 0.816 | 0.862 (0.774–0.951) | 0.815 | 0.772 |
| Fold 5 | 84 | 24 | 0.890 (0.844–0.936) | 0.750 | 0.927 | 0.852 (0.771–0.933) | 0.500 | 0.941 |
Fig. 3.
Prediction nomogram for preeclampsia based on independent risk factors. The nomogram is composed of scores, predictive factors, total points, and risk probabilities. In the figure, “Points” represents the scores corresponding to different levels of each predictive factor. BMI, MAP, sFlt-1/PlGF ratio [sFlt.1.PlGF], history of adverse pregnancy [History.of.adverse.pregnancy], family history of PE [Family.history.of.PE], history of PE [History.of.PE], chronic hypertension [Chronic.hypertension], autoimmune disease [Autoimmune disease], and PCOS are each independent predictors. “Total Points” is the total number of points for each level of all predictive factors, and “Disease Risk” is the risk probability for each level of each total point.)
Model assessment
Given that the proportion of PE included was only 5.23%, the dataset exhibited a noticeable imbalance. For the evaluation of the best model, higher sensitivity in the validation set is necessary. Consequently, Fold 4 was selected as the best model based on the fivefold cross-validation. In the recommended model, the AUROC for the training set was 0.883 (95% CI 0.838–0.928), with a sensitivity and specificity of 0.827 and 0.816, respectively. The AUROC for the validation set was 0.862 (95% CI 0.774–0.951), with a sensitivity and specificity of 0.815 and 0.772, respectively (Fig. 4). In the ROC curve of the training set, the optimal cutoff value for risk probability was 0.04, corresponding to a total score of 101 in the nomogram. Thus, when an individual factors, including BMI, MAP, the sFlt-1/PlGF ratio, history of adverse pregnancy, family history of PE, history of PE, chronic hypertension, autoimmune disease, and PCOS were incorporated into the nomogram and the calculated total score exceeded 101, the patient was considered at high risk for PE. Preventive strategies should be considered accordingly.
Fig. 4.
Receiver operating characteristic curves of the prediction nomogram in the training and validation sets
The calibration curve indicated a high degree of concordance between the pre-calibration and the resampled calibration curves. However, a notable degree of discrepancy was noted in relation to the standard reference line (Fig. 5). The biomarker incorporated into the model was the sFlt-1/PlGF ratio, demonstrating a modest discriminatory power with an AUROC of 0.692 (95% CI 0.631–0.752). Sensitivity and specificity were 0.630 and 0.733, respectively.
Fig. 5.
Calibration curves of the prediction nomogram in the training and validation sets
In both the training and the validation sets, the decision curves revealed large probability intervals for the net benefit threshold (Fig S1). The clinical impact curve of the model indicated that when the threshold probability exceeded 20%, the prediction of the model highly aligned with actual occurrences, resulting in an effective clinical prediction rate (Fig S2).
Discussion
Main findings
The analysis indicates that risk factors for PE include BMI, MAP, the sFlt-1/PlGF ratio, history of adverse pregnancy, family history of PE, history of PE, chronic hypertension, autoimmune disease, and PCOS. The predictive capability of a singular risk factor for PE is markedly limited, particularly when it comes to binary variables (such as history of adverse pregnancy, family history of PE, and history of PE). Moreover, there are certain deficiencies in the predictive capability of the single continuous indicator (sFlt-1/PlGF). In this study, the AUROC for sFlt-1/PlGF predicting PE is 0.692 (95% CI 0.631–0.752). Therefore, it is essential to develop multivariate models that can integrate various predictive factors. The interpretable predictive nomogram constructed based on independent risk factors demonstrates ideal efficacy. In the validation set, the AUROC is 0.862 (95% CI 0.774–0.951), with a sensitivity and specificity of 0.815 and 0.772, respectively.
Previous studies
Previous studies have also focused on developing predictive models for PE and have published relevant systematic reviews [22–25]. The modeling methods discussed in these reviews primarily utilize Logistic regression. This preference is largely due to the ideal interpretability and recommended applicability of nomograms constructed based on Logistic regression. Furthermore, these reviews do not consider the validity of predictive models for PE in specific countries. Additionally, common predictors in previously constructed models are maternal characteristics (pre-pregnancy BMI, family history of PE, history of PE, chronic disease, and ethnicity) and biomarkers (uterine arterial pulsatility index and pregnancy-associated plasma protein-A) [24]. Notably, only the sFlt-1/PlGF ratio was included in our final model. This finding aligns with previous studies, underscoring the superior predictive performance of the sFlt-1/PlGF ratio compared to PlGF alone [11, 13]. Our findings also indicate that the sFlt-1/PlGF ratio can predict the occurrence of all PE cases, with a detection rate of 63% and a specificity of 73.3%. This performance surpasses previous studies focused on early-onset PE [9, 10, 13]. Our algorithm exhibits superior discrimination performance, achieving a sensitivity of 0.79 at a positive predictive value of 0.221 for all PE, in contrast to the competing risk model that employed the FMF methodology [6, 8, 16, 17]. In comparison to maternal risk factors and other biomarkers, our final model demonstrates superior discrimination in predicting PE. In conclusion, this model not only demonstrates exceptional stratification capability and calibration efficacy, but also shows considerable clinical utility.
Strengths and limitations
This study offers several remarkable strengths. First, our study closely reflects clinical practice by deriving insights from recent real-world cases without intervention and using original data values. This approach allows for the prediction of PE in patients at any GA as long as the factors included in the model are available. Second, this study sheds light on potential correlations between variables. For instance, we examine the well-established relationship between PlGF and sFlt-1, in which sFlt-1 binds to PlGF. This interaction is characterized by a decrease in PlGF concentration and a concomitant increase in sFlt-1 levels, which are crucial in the development of PE. Such insights contribute to a deeper understanding of the pathophysiology and potential predictive markers for PE [13, 26]. This correlation can be understood as collinearity, which has a notable impact on the fitting of the model. To remove this collinearity among predictors, stepwise binomial logistic and LASSO regressions are chosen. Furthermore, binary logistic regression offers several advantages over competing risk models. It is relatively straightforward to comprehend. The interpretability of the model is notably high, allowing for a clear understanding of the influence exerted by various variables on the final results, as evidenced by the weights assigned to each feature.
However, there are several limitations, including the non-multicenter design, a limited number of cases, and the absence of external validation. These factors may adversely affect the generalizability of the model. Moreover, this study excludes GA as the outcome variable is a dichotomous event without a time element. Given the current approaches for predicting PE tend to be based on GA and using a competing risk model that treats PE as a time-dependent event [27–29], we are unable to demonstrate that our prediction model is valid and independent of GA. Due to the limited sample size of cases, subgroup analyses are not conducted. This not only constrains the statistical power of the prediction model but also represents an additional limitation of this study. Compared to the robust screening tool for PE developed by the FMF and recommended by the International Federation of Gynecology and Obstetrics, Doppler velocimetry of the uterine arteries is excluded from the protocol of this study. Our algorithm may be suitable for middle-Socio-demographic Index geographies where Doppler velocimetry of the uterine arteries could not be obtained easily. Additionally, a correct and mandatory validation would be prospectively proofing the nomogram. There is a clear need for our model to undergo further testing and refinement. We expect that future studies, especially external validation studies, will play a crucial role in this process.
Implications for clinical practice
How can we use our results to facilitate clinical practice? One strategy is to integrate the nomogram and the decision curve analysis plot to evaluate whether a PlGF-based test is required. Whenever a pregnant woman attends the outpatient department for prenatal care, regardless of GA, we can calculate the risk of PE using the nomogram with all variables except for the sFlt-1/PlGF ratio. From an economic point of view, this approach may be cost-effective. The other strategy is that we could calculate the total points of a patient using all the variables we could achieve. If total points are above 101 (the cutoff value for the total points of the nomogram), the patient is considered to be at high risk for PE and is prescribed the appropriate preventive measures. For example, if the total score is greater than 101 during early pregnancy, low-dose oral aspirin and reasonable doses of calcium supplements can be given to prevent PE. In mid- to late pregnancy, if the total score exceeds 101, it is too late to use aspirin and determining the optimal delivery time may be premature. High-risk patients can be referred to a high-risk obstetrics specialist for clinical management, allowing for enhanced monitoring and higher-level clinical care to avoid emergency deliveries and fetal death.
Conclusions
Our predictive nomogram for PE constructed based on common interpretable features exhibits desired efficacy, which informs the screening of at-risk populations and the development of specialized preventive protocols in clinical practice.
Supplementary Information
Below is the link to the electronic supplementary material.
Author contributions
Changxiu Wang: Conceptualization, Methodology, Validation, Formal analysis, Data Curation, Writing—Original Draft, Visualization; Tao Zeng: Methodology, Software, Validation, Formal analysis, Data Curation, Visualization; Xiangyu Zhao: Resources, Data Curation; Cuiping You: Writing—Review & Editing, Project administration, Funding acquisition; Yucheng Lu: Investigation; Guanqing Kong: Data Curation; Lingling Hu: Investigation; Jinyan Huang: Conceptualization, Writing—Review & Editing, Supervision; Yanxin Zhang: Conceptualization, Writing-Review & Editing, Supervision, Project administration.
Funding
This study was supported by the Shandong Provincial Natural Science Foundation Innovation and Development Joint Fund (ZR2023LSW014).
Data availability
The data that support this study are available from the corresponding author upon reasonable request.
Declarations
Conflict of interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Ethical approval
The dataset of this retrospective study was from patient’s electronic medical records. Written consent was obtained from all the participants. The study protocol was approved by the Science Research Ethics Committee, Linyi People’s Hospital (YX200683).
Declaration of generative AI in scientific writing
The authors did not use generative AI or AI-assisted technologies in the development of this manuscript.
Footnotes
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Changxiu Wang and Tao Zeng contributed equally to this work and should be considered as co-first authors.
References
- 1.Kassebaum NJ, Barber RM, Bhutta ZA, Dandona L, Gething PW, Hay SI, Kinfu Y, Larson HJ, Liang X, Lim SS, Lopez AD (2016) Global, regional, and national levels of maternal mortality, 1990–2015: a systematic analysis for the global burden of disease study 2015. Lancet 388(10053):1775–812. 10.1016/s0140-6736(16)31470-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Poon LC, Shennan A, Hyett JA, Kapur A, Hadar E, Divakar H et al (2019) The international federation of gynecology and obstetrics (FIGO) initiative on pre-eclampsia: a pragmatic guide for first-trimester screening and prevention. Int J Gynaecol Obstet 145(Suppl 1):1–33. 10.1002/ijgo.12802 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Magee LA, Nicolaides KH, von Dadelszen P (2022) Preeclampsia. N Engl J Med 386(19):1817–32. 10.1056/NEJMra2109523 [DOI] [PubMed] [Google Scholar]
- 4.Chappell LC, Cluver CA, Kingdom J, Tong S (2021) Pre-eclampsia. Lancet 398(10297):341–54. 10.1016/s0140-6736(20)32335-7 [DOI] [PubMed] [Google Scholar]
- 5.Scott G, Gillon TE, Pels A, von Dadelszen P, Magee LA (2022) Guidelines-similarities and dissimilarities: a systematic review of international clinical practice guidelines for pregnancy hypertension. Am J Obstet Gynecol 226(2s):S1222-s36. 10.1016/j.ajog.2020.08.018 [DOI] [PubMed] [Google Scholar]
- 6.Tan MY, Syngelaki A, Poon LC, Rolnik DL, O’Gorman N, Delgado JL et al (2018) Screening for pre-eclampsia by maternal factors and biomarkers at 11–13 weeks’ gestation. Ultrasound Obstet Gynecol 52(2):186–95. 10.1002/uog.19112 [DOI] [PubMed] [Google Scholar]
- 7.Wright D, Syngelaki A, Akolekar R, Poon LC, Nicolaides KH (2015) Competing risks model in screening for preeclampsia by maternal characteristics and medical history. Am J Obstet Gynecol 213(1):62.e1-62.e10. 10.1016/j.ajog.2015.02.018 [DOI] [PubMed] [Google Scholar]
- 8.Tan MY, Wright D, Syngelaki A, Akolekar R, Cicero S, Janga D et al (2018) Comparison of diagnostic accuracy of early screening for pre-eclampsia by NICE guidelines and a method combining maternal factors and biomarkers: results of SPREE. Ultrasound Obstet Gynecol 51(6):743–50. 10.1002/uog.19039 [DOI] [PubMed] [Google Scholar]
- 9.Zhong Y, Zhu F, Ding Y (2015) Serum screening in first trimester to predict pre-eclampsia, small for gestational age and preterm delivery: systematic review and meta-analysis. BMC Pregnancy Childbirth 15:191. 10.1186/s12884-015-0608-y [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Akolekar R, Zaragoza E, Poon LC, Pepes S, Nicolaides KH (2008) Maternal serum placental growth factor at 11 + 0 to 13 + 6 weeks of gestation in the prediction of pre-eclampsia. Ultrasound Obstet Gynecol 32(6):732–9. 10.1002/uog.6244 [DOI] [PubMed] [Google Scholar]
- 11.Chaemsaithong P, Gil MM, Chaiyasit N, Cuenca-Gomez D, Plasencia W, Rolle V et al (2023) Accuracy of placental growth factor alone or in combination with soluble fms-like tyrosine kinase-1 or maternal factors in detecting preeclampsia in asymptomatic women in the second and third trimesters: a systematic review and meta-analysis. Am J Obstet Gynecol 229(3):222–47. 10.1016/j.ajog.2023.03.032 [DOI] [PubMed] [Google Scholar]
- 12.Widmer M, Cuesta C, Khan KS, Conde-Agudelo A, Carroli G, Fusey S et al (2015) Accuracy of angiogenic biomarkers at ⩽20weeks’ gestation in predicting the risk of pre-eclampsia: a WHO multicentre study. Pregnancy Hypertens 5(4):330–8. 10.1016/j.preghy.2015.09.004 [DOI] [PubMed] [Google Scholar]
- 13.Stepan H, Galindo A, Hund M, Schlembach D, Sillman J, Surbek D et al (2023) Clinical utility of sFlt-1 and PlGF in screening, prediction, diagnosis and monitoring of pre-eclampsia and fetal growth restriction. Ultrasound Obstet Gynecol 61(2):168–80. 10.1002/uog.26032 [DOI] [PubMed] [Google Scholar]
- 14.Chen J, Xu X, Xu X, Yang S, Wang X, Ye A et al (2024) Prediction of preeclampsia using maternal circulating mRNAs in early pregnancy. Arch Gynecol Obstet 310(1):327–35. 10.1007/s00404-024-07486-2 [DOI] [PubMed] [Google Scholar]
- 15.Liu Y, Xie Z, Huang Y, Lu X, Yin F (2024) Uterine arteries pulsatility index by Doppler ultrasound in the prediction of preeclampsia: an updated systematic review and meta-analysis. Arch Gynecol Obstet 309(2):427–37. 10.1007/s00404-023-07044-2 [DOI] [PubMed] [Google Scholar]
- 16.O’Gorman N, Wright D, Poon LC, Rolnik DL, Syngelaki A, Wright A et al (2017) Accuracy of competing-risks model in screening for pre-eclampsia by maternal factors and biomarkers at 11–13 weeks’ gestation. Ultrasound Obstet Gynecol 49(6):751–5. 10.1002/uog.17399 [DOI] [PubMed] [Google Scholar]
- 17.Rolnik DL, Wright D, Poon LCY, Syngelaki A, O’Gorman N, de Paco Matallana C et al (2017) ASPRE trial: performance of screening for preterm pre-eclampsia. Ultrasound Obstet Gynecol 50(4):492–5. 10.1002/uog.18816 [DOI] [PubMed] [Google Scholar]
- 18.ACOG Practice Bulletin No (2019) 202: gestational hypertension and preeclampsia. Obstet Gynecol 133(1):1. 10.1097/aog.0000000000003018 [DOI] [PubMed] [Google Scholar]
- 19.Collins GS, Reitsma JB, Altman DG, Moons KG (2015) Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement. Bmj. 350:g7594. 10.1136/bmj.g7594 [DOI] [PubMed] [Google Scholar]
- 20.National Institute for Health and Care Excellence: Guidelines (2019) Hypertension in pregnancy: diagnosis and management. National Institute for Health and Care Excellence (NICE) Copyright © NICE 2019, London.
- 21.Magee LA, Brown MA, Hall DR, Gupte S, Hennessy A, Karumanchi SA et al (2022) The 2021 International Society for the Study of Hypertension in Pregnancy classification, diagnosis & management recommendations for international practice. Pregnancy Hypertens 27:148–69. 10.1016/j.preghy.2021.09.008 [DOI] [PubMed] [Google Scholar]
- 22.Tiruneh SA, Vu TTT, Rolnik DL, Teede HJ, Enticott J (2024) Machine learning algorithms versus classical regression models in pre-eclampsia prediction: a systematic review. Curr Hypertens Rep 26(7):309–23. 10.1007/s11906-024-01297-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Antwi E, Amoakoh-Coleman M, Vieira DL, Madhavaram S, Koram KA, Grobbee DE et al (2020) Systematic review of prediction models for gestational hypertension and preeclampsia. PLoS One 15(4):e0230955. 10.1371/journal.pone.0230955 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Tiruneh SA, Vu TTT, Moran LJ, Callander EJ, Allotey J, Thangaratinam S et al (2024) Externally validated prediction models for pre-eclampsia: systematic review and meta-analysis. Ultrasound Obstet Gynecol 63(5):592–604. 10.1002/uog.27490 [DOI] [PubMed] [Google Scholar]
- 25.Pawlak C, Blois R, Gaillard JM, Richard J (1986) Sleep in Pick disease. Encephale 12(6):327–34 [PubMed] [Google Scholar]
- 26.Benzing T (2016) Hypertension: testing for pre-eclampsia: paving the way to early diagnosis. Nat Rev Nephrol 12(4):200–2. 10.1038/nrneph.2016.21 [DOI] [PubMed] [Google Scholar]
- 27.Tsiakkas A, Saiid Y, Wright A, Wright D, Nicolaides KH (2016) Competing risks model in screening for preeclampsia by maternal factors and biomarkers at 30–34 weeks’ gestation. Am J Obstet Gynecol 215(1):87.e1–81.e17. 10.1016/j.ajog.2016.02.016 [DOI] [PubMed] [Google Scholar]
- 28.Ciobanu A, Wright A, Panaitescu A, Syngelaki A, Wright D, Nicolaides KH (2019) Prediction of imminent preeclampsia at 35–37 weeks gestation. Am J Obstet Gynecol 220(6):584.e1-584.e11. 10.1016/j.ajog.2019.01.235 [DOI] [PubMed] [Google Scholar]
- 29.Mazer Zumaeta A, Wright A, Syngelaki A, Maritsa VA, Da Silva AB, Nicolaides KH (2020) Screening for pre-eclampsia at 11–13 weeks’ gestation: use of pregnancy-associated plasma protein-A, placental growth factor or both. Ultrasound Obstet Gynecol 56(3):400–7. 10.1002/uog.22093 [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The data that support this study are available from the corresponding author upon reasonable request.