Study design and participants

This study was conducted at the Jiangxi Maternal and Child Health Hospital, a leading institution specializing in maternal and child healthcare in Jiangxi Province, with an annual delivery volume of 22,000 and over 2000 PE patients treated each year. It encompassed data from 12,699 pregnant women who received prenatal care at the hospital between January 2019 and June 2022, including all singleton pregnancies screened with NT ultrasound or Down syndrome screening between 11 + 0 and 13 + 6 weeks of gestation. The consistency and accuracy of data collection were ensured through adherence to the hospital’s routine prenatal care and assessment protocols.

Inclusion criteria were primarily single pregnancies with a viable fetus between 11 + 0 and 13 + 6 weeks, ensuring consistency and validity in comparative analyses. All risk factors included in the analysis were obtained before the diagnosis of preeclampsia, ensuring that the predictive model is based solely on pre-diagnostic data. Exclusion criteria were designed to remove factors that might affect the accuracy of the results, including multiple gestations, embryonic developmental arrest or severe structural anomalies, incomplete medical records, transfers after delivery, and mental health issues in pregnant women. Additionally, cases terminated due to medical or social reasons were excluded to minimize potential biases and uncertainties ( Fig. 1 ).

Patient inclusion and exclusion flowchart.

The design of this study carefully considered aspects, including the representativeness of the study population, the accuracy of data collection, and the ethical nature of the research process, ensuring the validity and reliability of the results. Through this investigation of a specific population, we aim to provide new insights and methods for predicting early-onset PE, ultimately contributing to improved health outcomes for pregnant women and their newborns.

Research process and data collection

Data collection encompassed various clinical and biochemical markers randomly retrieved from the obstetric database. This process included discrete data entered directly into clinicians’ or nurses’ electronic medical record systems, structured data, and diagnostic codes inputted by professional coders. The clinical data covered age, blood pressure, pre-pregnancy body mass index (BMI), number of pregnancies and deliveries, conception methods, educational level, dietary and lifestyle habits, and current gestational age. These data provided a comprehensive view for assessing the health status of pregnant women and their risk of PE.

In addition to clinical data, we collected a series of critical biochemical markers crucial for assessing the overall health of pregnant women and their potential risk of PE. This included fundamental blood composition analysis such as white blood cell count, hemoglobin levels, and other significant biochemical markers like liver function tests (alanine transaminase (ALT) and aspartate transaminase (AST)), kidney function tests (blood urea nitrogen (BUN) and serum creatinine), and blood lipid levels (total cholesterol, triglycerides, and lipoprotein levels). The integrated analysis of these data helps us better understand the relationship between pregnant women’s health status and PE risk.

Development of the predictive model

To develop a model capable of accurately predicting early-onset PE, this study analyzed 12,699 valid samples after thorough data cleansing, including 1200 positive cases diagnosed with PE and 11,499 control samples. The samples were randomly assigned into two datasets: 70 % formed the training set for model fitting, while the remaining 30 % served as the test set to evaluate model performance. We maintained a consistent distribution of PE cases in both datasets, approximately 9.45 %.

Regarding model development, we employed both the WRF and Logistic regression models. The RF, an ensemble learning method, makes its final judgment by constructing multiple decision trees and aggregating their predictions. To address the data imbalance in this study, we assigned higher weights to the PE category—specifically, 9.5 times the weight of the non-PE cases. This weighting approach helps reduce the model’s bias toward the majority class (non-PE cases) and enhances the detection of the minority class (PE cases).

To optimize the construction of decision trees, we utilized both the Gini index and entropy as measures. These metrics help assess the purity of data splits, allowing for the most effective selection of features that reduce uncertainty. This method improves the accuracy and generalizability of the decision trees on unseen data.

Additionally, we employed grid search techniques to identify the optimal hyperparameter settings, systematically exploring various parameter combinations to find those that yield the best model performance. We also ranked the importance of various features within the predictive model, providing valuable insights for the clinical assessment of PE risk.

The entire model development and evaluation process was conducted in a Python 3.7 environment using the scikit-learn 1.0.1 library. The application of these models, combined with data preprocessing and optimization strategies, aims to provide medical professionals with a reliable tool for effectively identifying and managing high-risk pregnancies, thereby making strides in the early prevention of PE.

Model evaluation metrics

To comprehensively evaluate the performance of our model, we employed vital indicators, including accuracy, sensitivity, specificity, and AUC score in the receiver operating characteristic curve (ROC). Accuracy measures the model’s ability to make correct predictions across all cases. Sensitivity assesses the model’s capability to correctly identify positive cases of PE, while specificity indicates the efficiency with which the model identifies non-PE cases. The AUC score provides a holistic assessment of the model’s performance across all possible classification thresholds. A higher AUC value, approaching 1, signifies the model’s more robust predictive capability.

Statistical analysis

Initially, all collected data underwent preliminary processing, which included descriptive statistical analysis for quantitative data (expressed as mean ± standard deviation or median with interquartile range) and percentage representation for categorical data. Quantitative differences between groups were assessed using either the Student’s t-test or the Kruskal–Wallis test, while differences in proportions of categorical data were evaluated using the chi-square test. Additionally, univariate binary Logistic regression analysis was used to identify variables significantly associated with PE (P < 0.05), further validated for their independent predictive value through multivariable binary Logistic regression analysis.

All statistical analyses were conducted using SPSS software version 24.0. The development and evaluation of the traditional Logistic regression and the WRF models were performed in a Python 3.7 environment, utilizing the scikit-learn 1.0.1 library. The performance of these models was assessed based on metrics such as ROC-AUC, accuracy, sensitivity, specificity, precision, and concordance. Any model with an AUC value greater than 0.5 was considered predictive, and results with a P-value less than 0.05 were deemed statistically significant.

Research ethics and approval

Before commencing this study on the predictive model for early-onset PE, all procedures and methods adhered to strict ethical guidelines to protect participant rights and received corresponding ethical approval.

The research protocol was submitted for detailed evaluation to the Jiangxi Provincial Medical Ethics Committee and was approved on June 23, 2022 (Approval Number: C-KT-202221). This approval was a critical prerequisite for initiating the study, ensuring its legality and ethical compliance while safeguarding participants’ health and privacy.

Throughout the study, we strictly adhered to the standards and treatment policies defined in the “Hypertension in Pregnancy (2020) Clinical Guidelines,” ensuring that all medical practices conformed to current medical knowledge and guidelines. Decision-making for the reception of PE patients was conducted through a tiered obstetric emergency area, assisted by experienced clinical personnel, further ensuring the professionalism and correctness of practical operations.

To ensure the accuracy and reliability of data, the research facility was equipped with a comprehensive electronic medical record registration system. All personnel involved in the study underwent standardized training, ensuring high standards and consistency in data collection, processing, and analysis. These measures not only helped to enhance the quality of the research but also safeguarded participant information and the accuracy of the research outcomes.

Analysis of clinical markers and their importance in early detection of PE risk

In this study, we tracked and analyzed 12,699 pregnant women to develop a pre-pregnancy predictive model for PE ( Fig. 2 ). Clinical markers for the subjects obtained during the first trimester are detailed in Table 1 . Of our study cohort, 1200 women (9.4 %) were diagnosed with PE during the follow-up period. This finding highlights the relative prevalence of PE among pregnant women and underscores the importance of early identification and prevention of this complication.

Predictive model for PE.

A comparison of clinical markers between the two groups of women (those with PE and those without) revealed that women who developed PE were typically older and more likely to be first-time mothers, suggesting that age and first pregnancy are potential risk factors for developing PE. Notably, although the gestational weight gain was similar between the two groups, women who developed PE had significantly higher BMI, mean systolic blood pressure (SBP(m)), mean diastolic blood pressure (DBP(m)), maximum systolic pressure (SBP-max), and maximum diastolic pressure (DBP-max) compared to those who did not develop PE. These findings highlight the potential role of blood pressure control in preventing the development of PE.

In addition to physiological parameters, our observations indicated that women who developed PE had a significantly higher prevalence of underlying conditions, such as cardiovascular diseases, renal disorders, diabetes, and autoimmune diseases, compared to those who did not develop PE. Furthermore, these women were more likely to have a family history of PE or a prior diagnosis of the condition in previous pregnancies. These findings underscore the critical importance of personal and family medical histories in assessing the risk of PE.

Regarding lifestyle habits and health behaviors, significant differences were noted between women with and without PE in aspects such as mode of conception, urban residency, educational level, hormone use, hyperemesis gravidarum, smoking/drinking habits, and early pregnancy folic acid supplementation. These results provide valuable insights for future preventive strategies and recommendations for prenatal care.

In summary, this study highlights the significance of various clinical markers in predicting the development of PE. By identifying these risk factors, we aim to furnish healthcare professionals with enhanced information for risk assessment and intervention in early pregnancy, ultimately aiming to reduce the incidence of PE and improve health outcomes for pregnant women.

Significant differences in biochemical markers for early prediction of PE

In this study, we compared a series of laboratory markers between pregnant women diagnosed with PE and those without to explore their potential roles in predicting the condition. The analysis revealed significant differences across multiple laboratory markers between the two groups, providing crucial insights into the risk factors for PE ( Table 2 ).

Table 2

Laboratory parameters between the preeclampsia group and the control group.

Variable	preeclampsia group (n = 1200)	Non-preeclampsia group (n = 11,499)	P Value
Coagulation function indicators
Prothrombin time (s)	10.03 ± 0.66	10.20 ± 0.57	<0.001∗
Partial prothrombin time (s)	27.25 ± 2.46	26.84 ± 2.37	<0.001∗
Hepatobiliary function index
ALT (U/L)	11 (8, 15)	9 (7, 12)	<0.001
AST (U/L)	20 (16, 25)	18 (15, 21)	<0.001
Seroglobulin (g/L)	27.65 ± 3.81	27.56 ± 3.73	0.395∗
Total protein (g/L)	60.15 ± 6.5	61.65 ± 5.82	<0.001∗
Total bilirubin (μmol/L)	10.92 ± 3.74	11.63 ± 4.04	<0.001∗
Renal function index
Serum creatinine (μmol/L)	53 (45.25, 63)	46 (40, 54)	<0.001
BUN (mmol/L)	3.8 (3.0, 4.7)	3.3 (2.7, 4.0)	<0.001
Uric acid (μmol/L)	390 (321, 462.75)	328 (280, 384)	<0.001
Serum lipid parameters
Total cholesterol (mmol/L)	5.99 (5.2, 6.84)	5.77 (5.03, 6.59)	<0.001
Triglycerides (mmol/L)	3.53 (2.69, 4.87)	3.12 (2.42, 4.12)	<0.001
LDL (mmol/L)	3.25 (2.59, 3.98)	3.26 (2.58, 3.99)	0.774
HDL (mmol/L)	1.77 (1.45, 2.22)	1.97 (1.63, 2.32)	<0.001
Lipoprotein (a) (g/L)	0.15 (0.07, 0.31)	0.13 (0.06, 0.27)	0.001
ApoA1 (g/L)	2.22 (1.88, 2.5)	2.12 (1.76, 2.45)	<0.001
Blood cell count
Leukocytes (10 9 /L)	9.71 (8.24, 11.94)	9.53 (7.89, 11.61)	0.001
Neutrophils (10 9 /L)	7.31 (5.94, 9.40)	7.16 (5.73, 9.05)	0.001
Lymphocytes (10 9 /L)	1.55 (1.25, 1.96)	1.51 (1.22, 1.83)	<0.001
Platelets (10 9 /L)	200 (165, 241)	198 (165, 235)	0.141
Hemoglobin (g/L)	114 (103, 124)	111 (102, 120)	<0.001
Urine routine index, n (%)
Urine protein, n (%)	833 (69.4)	5175 (45)	<0.001

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

Share this:

• X
• Facebook

Related

Related posts:

1. Incidences and risk factors of postoperative urinary retention after mid-urethral sling placement with and without pelvic reconstructive surgery
2. Differences in sexually transmitted infection-associated cervical infections in pelvic inflammatory disease patients between adolescents and adults
3. Prevalence of premenstrual syndrome and its relationship to depression among Korean adolescents: A nationwide cross-sectional study
4. A novel technique with cool-tip radiofrequency ablation for selective fetal reduction in complicated monochorionic twin

Stay updated, free articles. Join our Telegram channel

Join