This study analyzed the data of a birth cohort, which were collected from 2013 to 2017 at Tongzhou Maternal and Child health hospital in Beijing, China. There are 17,901 pregnant women enrolled in the current study. The inclusion criteria of this cohort study were: (1) age of delivery above 18 years, (2) single gestation, (3) _T1GWG available at the first trimester (13 ± 1 weeks). The exclusive criteria were: (1) diagnosis of chronic hypertension and history of HDP, (2) HEELP or metabolic syndrome, (3) blood pressure missing at 13 ± 1 and 20 ± 1 weeks, (4) preterm birth, (5) postpartum hypertension, (6) extreme values for height, weight, and blood pressure (height < 1 m or height > 2 m, weight < 30 kg or weight > 150 kg, SBP < 70 mmHg or SBP > 270 mmHg, DBP < 50 mmHg or DBP > 140 mmHg, _T1GWG < −5.9 kg or _T1GWG > 11 kg). The flow chart was presented as Figure 1.

The sociodemographic data of pregnant women were collected from the first antenatal clinical record, such as race, age, level of education, employment condition, elder gestation, gestational season, elder gestation, and pre-pregnant weight and height. The pre-pregnant weight was obtained from the pregnant women’s self-reports. The gestational weight and blood pressure were measured by trained nurses at regular antenatal clinic. The GWG was defined as the difference between gestational weight and pre-pregnant weight. The MAP equaled to SBP plus 2/3 DBP. The pre-pregnant BMI was calculated using formula: BMI = (weight in kg) divided by (square of height in m). The pre-pregnant BMI class was identified according to the BMI criteria for Asian women (the BMI criteria of underweight, normal weight, overweight, and obesity were defined as < 18.5, 18.5–23, 23–27.5, ≥27.5) (12). De novo HDP that includes GH and de novo PE was the primary outcome of this study. The diagnosis of GH or de novo PE was made by obstetricians according to the latest Chinese clinical practice guideline which was consistent with the diagnostic guidelines in the developed countries (13). International Classification of Diseases 10 (ICD-10) codes were used to define GH and de novo PE (such as O13.01, O13.02 for GH; O11.01, O14.001, O14.101, O14.102, O14.901 for de novo PE). The text containing information related to GH and de novo PE in medical records was extracted to double confirm the disease and identify the week of disease diagnosis. The normotensive group was defined as free from any diagnosis of GH, PE, and without a history of hypertensive disorders.

The study was approved by the Institutional Review Board of Peking University Health Science Center (No. IRB00001052-21023).

Missing values of demographic variables, we imputed the missing data by the k-nearest neighbor (KNN) algorithm [Beretta and Santaniello (14)]. The Shapiro–Wilk test was applied to check the normality of the distribution for each variable. Normally distributed variables were presented as means with SD and analyzed with t-test for group comparisons between HDP and Normotensive groups. Non-normally distributed variables were shown as medians with interquartile range and analyzed with chi-square test or Mann–Whitney U-test.

An early GWG category (EwtGCat) in terms of the risk of HDP was constructed. The risk ratios (RR) for HDP were calculated using the multivariable Poisson regression model for each _T1GWG interval within the particular pre-pregnant BMI class vs. all other women within that BMI class, borrowing the LifeCycle Project method (6). Considering the _T1GWG was in a relatively small scale, we choose the 25th, 50th, 75th percentiles as the cut-off points for the category. The Class 1 of the EwtGCat included the women with underweight and normal weight pre-pregnant BMI. The women with overweight or obese during pregnancy were further grouped into class 2, 3, and 4. Class 2 was defined as the weight gain with significant negative association (RR < 1). Class 3 was the weight gain interval without statistical difference. Class 4 was the weight gain interval that showed remarkable positive association (RR > 1). The continuous _T1GWG was also transformed into categorical variables using quartile method and NAM criteria. Multivariable Cox hazards regression model was utilized (15) to access the association between _T1GWG (continuous, quartile, EwtGCat, and NAM criteria) and HDP after adjusting the covariables. The covariables include race, age, level of education, employment condition, elder gestation, and gestational season.

We constructed a serial mediation model to assess the effect of _T1GWG on HDP with MAP at 13th and 20th weeks (MAP_{13 week} and MAP_{20 week}) as mediators by adjusting covariables. The direct effect (DE) defined as _T1GWG to HDP. The indirect effects (IE) included three aspects: (1) IE₁ was defined as the pathway from _T1GWG to MAP_{13 week} to HDP, (2) IE₂ was the pathway from _T1GWG to MAP_{20 week} to HDP, (3) IE₃ was the pathway from _T1GWG to MAP_{13 week} to MAP_{14 week} to HDP. All the estimated effects were validated by bootstrapping method for 5,000 loops. The serial mediation model pattern was presented as Figure 2. The bruceR package (version 0.7.0) was adopted in this part of analyses. The stratified analyses were conducted to explore the effect of EwtGCat on the risk of HDP across different subgroups, such as parity (unipara vs. multipara), maternal age (>35 age vs. ≤35 age), conceptional season (spring/winter vs. summer/autumn), education level (high vs. low). The sensitivity analyses were further performed to assess the association between _T1GWG and onset time of HDP.

The adjusted logistic regression model was adopted to assess the prediction performance for HDP with 1,000 bootstrapping replications (16). Three different models were constructed and compared with each other by the Area Under Receiver Operator Characteristic Curve (AUC), sensitivity, and specificity. The models were as follow: model_NAM includes NAM criteria and covariables; model_EwtGCat included EwtGCat and covariables; model_EwtGCat&MAP included EwtGCat, MAP_{13 week}, MAP_{20 week} and covariables. All the statistical analyses were performed on the R software (version 4.0.0). The p-value < 0.05 was considered as significant difference.

Table 1 shows the maternal socio-demographical and first-trimester weight gain characteristics between de novo HDP and normotensive groups. The percentage of de novo HDP in our study was 5.31%. The 17,901 pregnant women included in this study had the mean maternal age of 29.0 years (SD = 3.8). Women in the following strata had a significantly higher percentage of HDP diagnosis: educational level with high school or lower, overweight or obesity with pre-pregnant BMI class and unipara. The _T1GWG was significantly higher in HDP group compared with normotensive women (p = 0.001). The detailed characteristics of the participants are listed in Table 1.

The Early GWG Category (EwtGCat) was constructed according to the Supplementary Table 1. The Class 1 of EwtGCat was identified as the women with underweight and normal weight pre-pregnant BMI classes, and there was no association between _T1GWG and HDP. As for underweight and obesity, the significant negative association was obtained in Class 2 which the interval of _T1GWG was below 0 kg. Class 3 was the interval of _T1GWG from 0 to 2.5 kg which we failed to observe any statistical association. Class 4 was the interval of _T1GWG above 2.5 kg which presented a significant positive association between _T1GWG and HDP.

As the Table 2 shown, _T1GWG was significantly higher in the HDP pregnancies compared with normotensive women (HR = 1.07, 95% CI: 1.04–1.09, p < 0.001). In comparison with women in the lowest quartile of _T1GWG, the hazards of HDP elevated in the women with Q3, Q4 of _T1GWG (HR = 1.29, 95% CI: 1.08–1.55, p = 0.005; HR = 1.43, 95% CI: 1.21–1.69, p < 0.001). There were significant associations between _T1GWG and HDP in the Class 2, 3, 4 compared with Class 1 under EGC criteria (HR = 2.42, 95% CI: 2.11–2.77, p < 0.001; HR = 5.27, 95% CI: 4.05–6.86, p < 0.001; HR = 5.62, 95% CI: 4.05–7.79, p < 0.001). However, there is no significant hazard difference between normal and abnormal groups under NAM criteria (HR = 1.04, 95% CI: 0.91–1.20, p = 0.551).

The Table 3 indicated that MAP_{13 week} and MAP_{20 week} totally mediated 37.7% of association between _T1GWG and HDP in all participants (IE_total = 1.001 × 10^–3, p < 0.001; DE = 2.654 × 10^–3, p < 0.001). The indirect mediation effect was consisted in 3 pathways, such as (1) _T1GWG to MAP_{13 week} to HDP (IE₁ = 3.330 × 10^–4, mediate proportion: 12.5%, p < 0.001); (2) _T1GWG to MAP_{20 week} to HDP (IE₂ = 3.970 × 10^–4, mediate proportion: 15.0%, p < 0.001); (3) _T1GWG to MAP_{13 week} to MAP_{20 week} to HDP (IE₃ = 2.710 × 10^–4, mediate proportion: 10.2%, p < 0.001). We reran the serial mediation model to test the IE of MAP in overweight/obesity women, which is similar to the results of all participants. The MAP_{13 week} and MAP_{14 week} mediated 26.7% of total effect of _T1GWG on HDP. IE₁, IE₂, and IE₃ explained 6.1, 13.7, and 6.9%, respectively, of the total effect through 3 mediated pathways (all p < 0.001). However, we could not detect meaningful mediation effect in the underweight and normal weight women.

The stratified analyses were conducted to assess the effect of EwtGCat on HDP under different variables. The Class 1 and Class 2 of EwtGCat were defined as low EwtGCat and the Class 3 and Class 4 were defined as high EwtGCat. The Figure 3 showed that high EwtGCat was elevated the risk of HDP especially in the women with multipara, maternal age above 35, spring/winter conception and low education level (HR = 3.338, 95% CI: 1.607–6.936; HR = 3.004, 95% CI: 1.119–8.064; HR = 2.313, 95% CI: 1.447–3.696; HR = 2.704, 95% CI: 1.431–5.016, respectively). In the sensitivity analyses, we failed to observe a significant association between _T1GWG and onset time of HDP (p = 0.677).

We assessed the predictive capacity for HDP by model_NAM, modele_EwtGCat, and model_EwtGCat&MAP. As shown in Figure 4, the AUC, sensitivity, and specificity of the model_NAM, which included NAM criteria and covariables, were 0.587 (95% CI: 0.561–0.611), 0.544 (95% CI: 0.463–0.606), and 0.578 (95% CI: 0.492–0.667). In terms of model_EwtGCat, the AUC was improved to 0.668 (95% CI: 0.649–0.688). The model_EwtGCat&MAP demonstrated the best predictive performance for HDP comparing to the other two model (AUC = 0.760, 95% CI: 0.739–0.777; sensitivity = 0.703, 95% CI: 0.677–0.719; specificity = 0.686, 95% CI: 0.635–0.726). The detail information of the predictive models was listed in Supplementary Table 2.