Adult pregnant women who underwent physical examinations and delivered at Liyang People's Hospital were continuously recruited for the study. The exclusion criteria included: (1) absence of baseline or follow-up blood pressure assessments; (2) pre-existing hypertension (chronic hypertension concurrent with pregnancy); (3) secondary hypertension; (4) severe cardiac, liver, or kidney dysfunction; and (5) autoimmune diseases. Ultimately, 446 participants without pre-existing hypertension were enrolled in the study. These participants were initially registered during their first prenatal visit (approximately at the 12th week of pregnancy) and were then monitored every four weeks until delivery.

Maternal clinical parameters and serum biomarkers were collected after an overnight fast between the 16th and 20th weeks of gestation. These clinical parameters included age, height, baseline weight, weight change, baseline systolic blood pressure (SBP), baseline diastolic blood pressure (DBP), gravidity, and parity. Weight change was calculated as the difference between baseline weight and weight measured during the final examination. The body mass index (BMI) was calculated by weight(kg)/[height(m)]^2. SBP and DBP were measured thrice using an automatic blood pressure monitor following at least 30 min of rest, with the mean blood pressure being utilised for subsequent analyses. The fasting serum biomarkers included haemoglobin, leukocyte, and platelet counts.

According to the 2018 ISSHP guideline, HDP is categorised into two types: (1) hypertension known before pregnancy or present in the first 20 weeks of gestation, and (2) hypertension arising de novo at or after 20 weeks (10). Following the 2018 ISSHP guidelines, our study defined new-onset hypertension as hypertension arising de novo at or after 20 weeks, encompassing both gestational hypertension and preeclampsia. The SBP ≥ 140 mmHg or DBP ≥ 90 mmHg was set a cut-off for hypertension. Participants who developed HDP were classified as the disease group, while those who did not develop HDP were designated as the control group.

The statistical analysis was conducted following the guidelines of the Scientific Publication Committee of the American Heart Association (15). The multivariate multiple imputation method was performed to fill in missing variates, which could minimise selection bias and improve statistical efficiency (16, 17). The "Multivariate Imputation by Chained Equations" package was applied and the default parameters were applied. Continuous variables adhering to a normal distribution (determined by the Kolmogorov-Smirnov test) were articulated as the mean ± standard deviation, whereas those with a skewed distribution were presented as a median along with the interquartile range. The representation of categorical variables was done through frequencies paired with percentages. For the comparison between the HDP group and the control group, we applied one-way ANOVA test (for normally distributed variables), the Kruskal-Wallis test (for skewed distribution), or chi-square test (for categorical variables) as appreciated.

Logistic regression was employed to evaluate the impact of the maternal clinical parameters and serum biomarkers on the development of HDP. Risk factors that achieved statistical significance (P < 0.05) in univariate logistic regression analyses were subsequently included in a multivariate logistic regression model. The effect sizes were presented using odd ratios (ORs) with 95% confidence intervals (CIs). Moreover, we illustrated the influence of waist circumference on new-onset hypertension via a restricted cubic spline (RCS) with 4 knots located at the 5th, 35th, 65th, and 95th percentiles. The 65th knots were set as the reference unless otherwise stated.

Moreover, to establish the prediction model for new-onset hypertension, we first screened the risk factors using an automatic backwards selection algorithm based on the Classification and Regression Training (caret) package (version 6.0–94) in R (18). Then, the identified risk factors were used to establish the prediction model using multiple machine learning methods, including the generalised linear model, multivariate adaptive regression splines, random forest, and k-nearest neighbours. The model performance was evaluated using the area under the curve. All participants were randomly allocated to either a training set or an internal validation set in an 8:2 ratio. The training set facilitated the selection of features and the training of the prediction model, while the internal validation set was employed to evaluate the model's performance. This cross-validation strategy ensures that the model is both trained and tested on independent subsets of the dataset, enhancing the generalizability and robustness of the predictive model.

A P-value < 0.05 was considered as statistical significance. All statistical analyses were conducted using R software (version 4.3.0).

Table 1 summarises the characteristics of the participants in the HDP and control groups. Of the 446 pregnant participants without baseline hypertension, 153 developed new-onset HDP before delivery. We observed significant disparities in several baseline and physiological attributes, mainly favouring the development of HDP. Baseline weight, BMI, weight change during pregnancy, baseline SBP, and baseline DBP values were notably higher in the HDP group. Regarding pregnancy history, the HDP group contained a greater proportion of first-time pregnancies (44.4%) and participants without previous childbirth (60.1%) compared to the control group.

Moreover, haematological parameters also showed a significant difference between the two groups. The HDP group exhibited increased haemoglobin and platelet levels, while leukocyte counts were marginally lower than those in the control group. These observations indicate an intricate relationship between HDP and specific haematological parameters, adding another dimension to understanding HDP risk factors. However, the study found no significant differences in age and height between the HDP and control groups.

Table 2 outlines the relationship between various risk factors and the incidence of new-onset HDP, as analysed by univariate and multivariate logistic regression. In univariate analysis, baseline weight (OR = 1.04, 95% CI: 1.02–1.06, P < 0.001), baseline BMI (OR = 1.13, 95% CI: 1.07–1.19, P < 0.001), weight change (OR = 1.05, 95% CI: 1.02–1.09, P = 0.006), baseline SBP (OR = 1.09, 95% CI: 1.06–1.11, P < 0.001), baseline DBP (OR = 1.13, 95% CI: 1.10–1.17, P < 0.001), hemoglobin (OR = 0.89, 95% CI: 0.81–0.97, P = 0.009), leukocyte (OR = 1.04, 95% CI: 1.02–1.06, P < 0.001), and platelets (OR = 1.01, 95% CI: 1.00–1.01, P < 0.001) were found to have a significant effect on the onset of HDP.

Next, we input the risk factors with a P of <0.05 into the multivariate regression model (Table 2). In multivariate analysis, baseline BMI (OR = 1.10, 95% CI: 1.03–1.17, P = 0.007), weight change (OR = 1.10, 95% CI: 1.05–1.16, P < 0.001), baseline SBP (OR = 1.04, 95% CI: 1.01–1.08, P = 0.008), baseline DBP (OR = 1.10, 95% CI: 1.05–1.14, P < 0.001), hemoglobin (OR = 0.80, 95% CI: 0.71–0.9, P < 0.001), leukocyte (OR = 1.02, 95% CI: 1–1.04, P = 0.03), and platelets (OR = 1.01, 95% CI: 1–1.01, P < 0.001) remained significantly correlated to the risk of HDP. The baseline weight was not input into the multivariate analysis since the BMI was calculated based on body weight. However, other factors such as age, height, and number of pregnancies did not show any significant association. Furthermore, we illustrated the dose-dependent association between maternal clinical parameters and HDP risk (Figure 1). The RCS plots showed the significant effect of baseline BMI, weight change, baseline SBP, and baseline DBP on HDP (all P < 0.001). The linear dose-dependent relationship was observed in baseline BMI and baseline hypertension, whereas the U-shape relationship was observed between weight change and HDP and the J-shape curve was observed in baseline DBP. These results underscore the importance of monitoring physiological parameters such as BMI, weight change, baseline blood pressure, and blood cell count during pregnancy as potential predictive factors for new-onset HDP.

The backwards selection algorithm of the caret package identified several features, including age, weight change, baseline SBP, baseline DBP, gravidity, parity, haemoglobin, leukocyte, and platelets. We established four distinct predictive models based on different machine learning methods (generalised linear model, multivariate adaptive regression splines, random forest, and k-nearest neighbours). In the training set, the area under the curve (AUCs) for the generalised linear model, multivariate adaptive regression splines, random forest, and k-nearest neighbours models were 0.87 (0.83–0.91), 0.91 (0.87–0.95), 0,99 (0.98–0.99), and 0.88 (0.84–0.92), respectively.

In the internal validation set, the AUCs for the generalised linear model, multivariate adaptive regression splines, random forest, and k-nearest neighbours models were 0.76 (0.65–0.87), 0.85 (0.76–0.94), 0.85 (0.76–0.94), and 0.76 (0.65–0.87), respectively. The ROC curves of the proposed four models are given in Figure 2. The prediction models based on multivariate adaptive regression splines and random forest showed high performance with the area under the curve above 0.8 in the training set.