Study population

This retrospective multicenter cohort study recruited patients from the First Hospital of Shanxi Medical University (FHSMU) between June 2017 and May 2020 and Shanxi Children’s Hospital and Women Health Center (SCWHC) from September 2019 to May 2020 with a confirmed diagnosis of PE or who developed PE after admission. A study flowchart is shown in Fig. 1. The diagnosis of PE conformed to the 2018 definition of International Society for the Study of Hypertension in Pregnancy and ACOG 2019 [1, 2]. PE is characterized by new-onset of hypertension (systolic BP ≥ 140 mmHg and/or diastolic BP ≥ 90 mmHg) and exhibits at least one of the following new-onset symptoms during or after 20 weeks of gestation: 1) proteinuria (24-h urinary protein ≥ 300 mg/day or dipstick reading ≥ 2+); 2) other maternal organ dysfunction, such as acute kidney injury (creatinine ≥ 90 μmol/L or 1.0 mg/dL), hepatic dysfunction (alanine aminotransferase or aspartate aminotransferase > 40 IU/L, with or without epigastric abdominal or right upper quadrant pain), neurological dysfunction (such as eclampsia and change in mental status), or hematologic complications (blood platelet count < 150,000/μL, disseminated intravascular coagulation, or hematolysis); and 3) uteroplacental complications (abnormal Doppler waveform of the umbilical artery, fetal growth restriction, or stillbirth). Patients who were diagnosed with PE superimposed upon chronic hypertension, pre-pregnancy hypertension, or hypertension that occurred within the first 20 weeks of pregnancy were not included in the study.

Fig. 1
figure 1

Study flowchart outlining the composition of final PE cohort using the datasets from the First Hospital of Shanxi Medical University (FHSMU) and Shanxi Children’s Hospital and Women Health Center (SCWHC)

The study flowchart and patient information are presented in Fig. 1. Between June 2017 and May 2020, a total of 722 PE patients who were admitted to the FHSMU and ended their pregnancies were included, excluding 16 without follow-up and 16 without postpartum BP measurement data. Thus, 690 of the 722 PE patients treated at the FHSMU were included; the other 32 were excluded. Of the 690 patients from the FHSMU, the BP data of 94 (13.62%) remained higher at 3 months postpartum. Among patients treated at the SCWHC between September 2019 and May 2020, 328 with PE ended their pregnancies, excluding 16 patients without follow-up and two without postpartum BP measurement data; the remaining 310 were included in the study. Among the 310 patients in the SCWHC group, the BP values did not return to normal by 3 months postpartum in 40 (12.90%). Thus, 1,000 (690 + 310) PE patients from the two hospitals were included, while 50 were excluded due to lack of postpartum BP data or phone interview failure.

Defining persistent PHTN

The recruited PE patients were followed up by telephone interviews to confirm persistent PHTN at 3 months postpartum. BP was measured twice a day by trained nurses in community clinics and/or by trained family members at home using mercurial or electronic arm cuff BP meters after a 5-min rest. BP measurements were repeated three times within 10 min, with a 1-min interval between measurements. Mean systolic and diastolic BP values were recorded. In the present study, patients with persistent PHTN were defined as those who experienced PE and still showed hypertension (average systolic BP ≥ 140 mmHg; average diastolic BP ≥ 90 mmHg) within 3 months postpartum and the requirement for cardiovascular consults for further investigation and medication. The end date of follow-up was October 2020. Patients with BP that did not return to normal by 3 months postpartum were included as outcome events. Patients who could not be reached by telephone for follow-up and whose BP was not monitored within 3 months postpartum were excluded from the study cohort.

Data collection

The data and diagnoses of the enrolled patients were collected, including maternal demographic characteristics and relevant clinical laboratory tests performed within 7 days prior to the end of pregnancy. If the index visit involved multiple tests, the worst value was selected. A total of 35 candidate risk factors, including laboratory test results, were entered from the literature reviews [1, 2, 20, 25,26,27]. Laboratory indicators were converted from continuous to categorical variables based on whether they were outside the normal range (Table 1).

Table 1 Description of the eight indicator variables for LCCA and the baseline characteristics of the 1,000 enrolled PE patients

Latent class cluster analysis

The 35 candidate risk factors were categorized into eight important indicator variables, including maternal delivery age, mean arterial pressure (MAP = diastolic BP + 1/3 pulse pressure difference; maximum MAP levels during pregnancy were used in the study), drug use during pregnancy, medical history, adverse pregnancy outcomes, blood cell and coagulation tests performed within 7 days before delivery, altered liver and renal functions within 7 days before delivery, and elevated blood myocardial enzymes and electrolyte disbalance within 7 days before delivery. Among the 35 risk factors, no categorical variable data were missing. Some continuous variable data were missing, including up to 2.5% of those for serum creatine kinase and serum lactate dehydrogenase, while data for prothrombin activity (%), international normalized ratio, albumin, serum creatinine, serum urea nitrogen, and serum potassium were missing for fewer than five cases; instead, mean values were used. Each of the eight indicator variables except maternal age and MAP included multiple risk factors. These variables were aggregated and assessed as total risk factor scores, with a dimensionality reduction of 0-N (Table 1) [28]. All eight indicators were considered continuous variables and standardized by the LCCA.

LCCA, a model-based clustering approach, was conducted to analyze the eight indicator variables using R 3.6.1 software. It assumes that heterogeneous populations are a mixture of populations; that is, a latent class is used to classify populations. This method classifies the population by probability; that is, the individual belongs to a cluster with a certain probability, and the individual is ultimately assigned to the cluster with the highest posterior probability [29]. LCCA for categorical indicator variables is called latent class analysis, while that for continuous indicator variables is called latent profile analysis (LPA). The eight indicators we studied were continuous variables, and the basic principle of LPA was to suppose that the probability density function of the P-dimensional continuous manifest variable vector yi can be expressed as the following equation 1:



Here ηk denotes the latent class probabilities and K is the number of clusters (= 1, 2, …, K); yi is the score of object i on a set of manifest variables, assuming that within the cluster k, yi came from an independent multivariate normal distribution; μis the mean vector; and ∑k is the variance-covariance matrix. After model establishment using Bayesian theory, the posterior probability of assigning patients to class k was calculated using the following equation 2:



LPA with the mclust package was used to define clusters of participants with similar clinical profiles. We used mclustBIC to observe the Bayesian Information Criterion (BIC) for different profiles and the integrated completed likelihood (ICL) to penalize the model’s instability to stabilize the number of obtained models. Finally, PE patients were classified into different latent classes.

Statistical analysis

Continuous variables in the baseline information are expressed as median and quartile [M (P25, P75)], and comparisons between PE patients with versus without persistent PHTN were made using the Mann-Whitney U test. Categorical variables are expressed as count and percentage, and the chi-squared test was used to compare PE patients with versus without persistent PHTN. Standardized characteristics of clusters are expressed as mean ± standard deviation (SD), while cluster comparisons were performed using analysis of variance. The logistic regression analysis was performed to explore the association between exposure clusters and persistent PHTN. The statistical analysis was performed using SPSS 22.0, and statistical significance was set at P < 0.05. Bonferroni correction was used to adjust the P values for multiple tests.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit The Creative Commons Public Domain Dedication waiver ( applies to the data made available in this article, unless otherwise stated in a credit line to the data.


This article is autogenerated using RSS feeds and has not been created or edited by OA JF.

Click here for Source link (