Early Pregnancy Blood Pressure Trajectory Groups Predict Hypertensive Disorders of Pregnancy
Authors: Megan Sheehan, Manyun Zhao, Stacey Alexeeff, Baiyang Sun, Erica P. Gunderson, Lisa D. Levine, Jinbo Chen, Jennifer Lewey
Published: 2026

ABSTRACT
BackgroundHypertensive disorders of pregnancy (HDP) cause significant maternal and fetal morbidity, but clinical tools to identify high-risk patients are lacking. Blood pressure (BP) trajectories in early pregnancy have predicted HDP.ObjectivesIn this study, we aimed to test these early BP trajectories in a distinct health system.MethodsElectronic health record data were used to retrospectively identify patients with a singleton delivery from 2012 to 2020 in a large urban academic center. Patients with pre-existing hypertension or serious medical conditions were excluded. Each patient was assigned 1 of 6 previously identified BP trajectory groups from conception through 20 weeks’ gestation. The primary outcome of HDP diagnosis, including preeclampsia and gestational hypertension, was identified using International Statistical Classification of Diseases codes. Multivariable logistic regression was used to model the relationship between patient factors and HDP diagnosis, and predictive model discrimination was assessed using C-statistics.ResultsAmong 25,879 patients, 5,694 (21.9%) were diagnosed with HDP. After adjustment for age, race/ethnicity, BMI, parity, and pregestational diabetes, BP trajectory groups were independently associated with increasing odds of HDP, with adjusted odds of HDP diagnosis progressively increasing from 1.6 (95% CI: 1.4-1.8) to 6 (95% CI: 5.3-6.9) across trajectories. Inclusion of trajectories into predictive models improved discrimination compared to clinical risk factors alone (C-statistic 0.71 [95% CI: 0.70-0.72] vs 0.66 [95% CI: 0.65-0.67]).ConclusionsBP trajectories, based on early pregnancy BP measurements, improved ability to predict HDP diagnosis in diverse pregnant cohorts. BP trajectories may improve identification of high-risk patients for more intensive BP monitoring during pregnancy.

Hypertensive disorders of pregnancy (HDP), including preeclampsia and gestational hypertension (HTN), affect up to 20% of pregnancies in the United States, are a leading cause of adverse maternal and fetal outcomes, and contribute to long-term maternal cardiovascular disease risk.1, 2, 3 The incidence of HDP has risen globally over recent decades as risk factors like advanced maternal age and obesity become more prevalent.4, 5, 6 Identifying individuals at increased risk for HDP in early pregnancy is crucial for risk stratification, increased maternal and fetal monitoring as pregnancy continues, and timely intervention such as low-dose aspirin in those at high risk.7 The U.S. Preventative Services Task Force recommends screening for HDP with blood pressure (BP) checks at each antenatal visit.8 However, existing HDP risk assessment tools do not use these BP values or account for changes in BP during pregnancy, and instead rely on demographic and clinical history.8, 9, 10

Early pregnancy BP is associated with HDP outcomes and recent studies suggest that the longitudinal trajectories of early pregnancy BP changes add important prognostic information to identify patients at an increased risk of HDP. Using electronic health record (EHR) data, Gunderson et al11 identified 6 distinct longitudinal BP trajectories in the first 20 weeks’ gestation among a community-based diverse cohort of 174,925 patients considered low-to-moderate risk for HDP, who received prenatal care and delivered a singleton live or still birth from 2009 to 2019 within the Kaiser Permanente Northern California health care system. The authors developed and internally validated predictive model performance using social, demographic, and early pregnancy clinical factors to predict patients’ risk of early- and late-onset preeclampsia and gestational HTN. The addition of early pregnancy BP trajectories from 0 to 20 weeks’ gestation to these models significantly improved model risk discrimination beyond clinical risk factors, and demonstrated excellent calibration.12 Models including BP trajectories and other risk factors outperformed those including only a single initial BP measurement, suggesting that longitudinal trends of BP in early pregnancy carry additional prognostic value.12

In this study, we aimed to test the early BP trajectory groups using EHR clinical data obtained from an external health system serving a racially diverse, urban cohort of patients with a delivery between 2012 and 2020. We created new models for HDP diagnosis using our external data set, then aimed to validate the predictive contribution of BP trajectory classification in these models. We hypothesized that the inclusion of BP trajectories with readily available demographic and clinical factors would improve model performance to predict an HDP diagnosis after 20 weeks’ gestation. Establishing early pregnancy BP trajectories as a risk factor for HDP in diverse patient cohorts and data sets could facilitate the incorporation of these BP trajectories in future HDP risk assessment models to improve risk stratification and earlier monitoring for the highest risk patients.

Methods

Data source

EHR data, including demographic variables, obstetric history, and clinical factors were accessed through the Penn Medicine data warehouse. Obstetric variables were captured at the time of delivery hospitalization and included delivery date, gestational age at delivery, mode of delivery, and number of prior pregnancies and deliveries. Clinical data included International Statistical Classification of Diseases (ICD)-9th and -10th revision clinical modification codes and medications for outpatient and inpatient encounters and vital signs (BP, weight, and height) from outpatient encounters. For each patient, data were available 12 months before delivery and through 5 years after delivery.

Study design and population

This retrospective cohort study included patients with a delivery between 2012 and 2020 at 1 of 2 urban, academic Penn Medicine hospitals in Philadelphia. It was approved by the Penn Institutional Review Board.

For each patient, we selected the first delivery occurring after 20 weeks’ gestation during the study period (n = 61,928). We excluded patients <15 or >50 years and patients with prior serious medical conditions such as cancer, kidney disease, or cardiovascular disease, with subsequent delivery or death within 12 months after delivery, or missing all clinical data within 12 months after delivery. Patients diagnosed with chronic HTN before pregnancy or through 20 weeks’ gestational age were also excluded, as were patients missing all BP encounters between the last menstrual period (LMP) and 20 weeks’ gestation. The current analysis is restricted to patients with singleton deliveries, including live and stillbirths. The final analytic cohort consisted of 34,613 patients. See the Supplemental Appendix for exclusion criteria definitions and patient selection diagram.

Outcomes

The primary outcome was a diagnosis of preeclampsia or gestational HTN after 20 weeks’ gestation using ICD codes using a previously validated algorithm13 (Supplemental Appendix). Patients who developed both gestational HTN and preeclampsia were classified as having preeclampsia, the more severe form of HDP. A composite outcome of any HDP (preeclampsia or gestational HTN) was analyzed as secondary outcome to ensure adequate power for subgroup analyses.

Covariates

Demographic covariates included maternal age at delivery, self-reported race and ethnicity (Asian, Hispanic, non-Hispanic Black, non-Hispanic White, or other/unknown), and insurance status at delivery (public, private, or other/unknown). Clinical covariates included pre-pregnancy body mass index (BMI) (kg/m2) calculated by weight closest to the date of LMP and height, mode of delivery (vaginal or cesarean), parity, preterm delivery (delivery prior to 37 weeks’ gestation), and pregestational diabetes. Pre-pregnancy BMI was calculated using the first trimester weight when pre-pregnancy weight was not available. When missing, pre-pregnancy weight was imputed based on Institute of Medicine guidelines of weight gain per trimester and the nearest available weight.14 Race/ethnicity was included to explore effect modification. Race/ethnicity as a covariate indicates differences that may occur as a result of racism or other lived experiences.

BP and BP trajectories

We included all BPs recorded from LMP to 20 weeks’ gestation in the outpatient setting. BPs from inpatient settings or the emergency room were excluded. One BP measurement per encounter day was used. If ≥2 BPs were recorded on the same date, we used the BP with the lower systolic value. We excluded invalid BP measurements (systolic BP ≤60 or ≥300, diastolic ≤30 or ≥180 mm Hg, or diastolic value > systolic). BP trajectories were categorized as 6 groups (ultra-low-declining, low-declining, moderate-fast-decline, low-increasing, moderate-stable, and elevated-stable) using group-based trajectory modeling (Supplemental Figure 1).11 Each patient in the cohort was assigned to a single trajectory group best reflecting the profile of change using maximum probability assignment rule based on the fitted parameter values from the previous group-based trajectory modeling and all available longitudinal BP measurements from LMP to 20 weeks’ gestation. The primary analysis was limited to patients with ≥3 BP values documented between LMP and 20 weeks’ gestation, as prior work demonstrated improved model prediction with this minimum number of BP values12 Sensitivity analyses included patients with 1 to 2 BP values between LMP and 20 weeks’ gestation.

Statistical analysis

For the primary outcome, multinomial regression was used to predict preeclampsia vs gestational HTN vs no HDP as nominal outcomes. Multivariable logistic regression was used to build prediction models for any HDP as a binary outcome. We first built a baseline model including demographic and clinical covariates as risk predictors. Subsequent models included baseline predictors with initial systolic BP included and then baseline predictors with BP trajectory included. Given that a small proportion of patients were identified in the ultra-low-declining BP trajectory group, this group was combined with the low-declining BP trajectory group to serve as the referent group for regression analyses11 We reported adjusted ORs (aORs) and 95% CIs for the association between HDP outcomes and BP trajectories as well as demographic and clinical covariates. Model discrimination was assessed using the mean area under the curve (AUC) from bootstrapping with 200 resamples with 95% CI estimated via the percentile-based bootstrap distribution of the AUC. Multiclass AUC was reported for preeclampsia and gestational HTN. Statistical tests with a 2-sided P value <0.05 were considered statistically significant. Analyses were performed using SAS (version 9.4; SAS Institute) or R (version 4.0.2; R Foundation for Statistical Computing).

Sensitivity analyses

We conducted several sensitivity analyses to evaluate the robustness of our findings. First, we restricted the cohort to nulliparous individuals, so as to exclude patients with a history of HDP in prior pregnancy. Next, we repeated the main analysis among patients with <3 BP available from LMP through 20 weeks’ gestation. Finally, we repeated the main model without race/ethnicity as a covariate. To assess model performance among individuals of diverse racial backgrounds, we conducted a separate analysis restricted to individuals who identify as non-Hispanic White and non-Hispanic Black, as these groups represent the 2 largest racial subgroups in the cohort. The model was configured as a multivariable logistic regression with HDP as a binary outcome. Effect modification by race was evaluated among those who identify as White or Black using an interaction term of race and BP trajectory. This analysis reports aOR and 95% CI for each BP trajectory and racial group relative to a single referent group (Supplemental Appendix).

Results

Of the 61,928 patients with index deliveries from 2012 to 2020, 34,613 (55.9%) met the study inclusion criteria and had ≥1 BP measurement in early pregnancy (Figure 1). Of these, 25,879 (74.8%) had ≥3 BPs from 0 to 20 weeks’ gestation and were included in the primary analysis. The mean maternal age at delivery was 30.8 years (SD: 5.4 years), 19,652 (76%) had private insurance, and 8,314 (32.1%) identified as non-Hispanic Black (Table 1). Of the total cohort, 16,910 (65%) were nulliparous and 593 (2.3%) had pregestational diabetes. Overall, 5,694 (21.9%) patients were diagnosed with HDP, including 2,225 with preeclampsia (8.6%) and 3,449 with gestational HTN (13.3%). There were 101,297 BPs recorded among the cohort within the first 20 weeks of gestation that were used to identify trajectories, with an average of 3.9 (SD: 1.2) BPs per patient.Figure 1CONSORT Diagram∗Patients with incomplete postpartum data were excluded as cohort was primarily developed to study postpartum outcomes for a separate analysis. Patient were excluded if missing all BP and ICD encounters (n = 9,146) or had <1 BP and <2 ICD encounters between delivery and 12 months postpartum (n = 10,738). BP = blood pressure.Table 1Maternal Characteristics Stratified by HDP Status Among Patients With ≥3 Blood Pressures Before 20 Weeks’ GestationAll(N = 25,879)Preeclampsia(n = 2,225, 8.6%)Gestational HTN(n = 3,449 (13.3%)No HDP(n = 20,205, 78.1%)Age (y), mean (SD)30.8 (5.4)30.8 (5.9)30.2 (5.8)30.9 (5.2)Age, y (category) ≤253,601 (13.9)362 (16.3)648 (18.8)2,591 (12.8) 25-≤3516,040 (62)1,270 (57.1)2024 (58.7)12,746 (63.1) >356,238 (24.1)593 (26.7)777 (22.5)4,868 (24.1)Race/ethnicity Asian2,120 (8.2)122 (5.5)147 (4.3)1851 (9.2) Hispanic1,504 (5.8)135 (6.1)168 (4.9)1,201 (5.9) Non-Hispanic Black8,314 (32.1)853 (38.3)1,397 (40.5)6,064 (30.0) Non-Hispanic White12,588 (48.6)1,039 (46.7)1,571 (45.5)9,978 (49.4) Other/missing1,353 (5.2)76 (3.4)166 (4.8)1,111 (5.5)Insurance status Private19,652 (75.9)1,612 (72.4)2,502 (72.5)15,538 (76.9) Public6,066 (23.4)595 (26.7)927 (26.9)4,544 (22.5) Missing161 (0.6)18 (0.8)20 (0.6)123 (0.6)Pre-pregnancy BMI, kg/m2mean (SD)25.9 (6.4)27.8 (7.1)28.2 (7.3)25.3 (5.9)Pre-pregnancy BMI, kg/m2 (category) ≤2513,878 (53.6)899 (40.4)1,335 (38.7)11,644 (57.6) 25-≤305,931 (22.9)602 (27.1)904 (26.2)4,425 (21.9) >305,326 (20.6)654 (29.4)1,118 (32.4)3,554 (17.6) Missing744 (2.9)70 (3.1)92 (2.7)582 (2.9)Delivery method Cesarean7,273 (28.1)930 (41.8)1,011 (29.3)5,332 (26.4) Vaginal16,469 (63.6)1,170 (52.6)2,253 (65.3)13,046 (64.6) Missing2,137 (8.3)125 (5.6)185 (5.4)1827 (9.0)Parity Multiparous8,646 (33.4)508 (22.8)818 (23.7)7,320 (36.2) Nulliparous16,910 (65.3)1,695 (76.2)2,595 (75.2)12,620 (62.5) Missing323 (1.2)22 (1.0)36 (1.0)265 (1.3)Preterm delivery1872 (7.2)459 (20.6)158 (4.6)1,255 (6.2)Diabetes, pregestational593 (2.3)110 (4.9)121 (3.5)362 (1.8)Blood pressure trajectory groups Ultra-low-declining414 (1.6)10 (0.4)14 (0.4)390 (1.9) Low-declining7,387 (28.5)326 (14.7)469 (13.6)6,592 (32.6) Moderate-fast decline2,361 (9.1)156 (7.0)249 (7.2)1956 (9.7) Low-increasing7,151 (27.6)579 (26.0)866 (25.1)5,706 (28.2) Moderate-stable7,152 (27.6)866 (38.9)1,449 (42.0)4,837 (23.9) Elevated-stable1,414 (5.5)288 (12.9)402 (11.7)724 (3.6)BMI = body mass index; HDP = hypertensive disorders of pregnancy; HTN = hypertension.

CONSORT Diagram

∗Patients with incomplete postpartum data were excluded as cohort was primarily developed to study postpartum outcomes for a separate analysis. Patient were excluded if missing all BP and ICD encounters (n = 9,146) or had <1 BP and <2 ICD encounters between delivery and 12 months postpartum (n = 10,738). BP = blood pressure.

Table 1Maternal Characteristics Stratified by HDP Status Among Patients With ≥3 Blood Pressures Before 20 Weeks’ GestationAll(N = 25,879)Preeclampsia(n = 2,225, 8.6%)Gestational HTN(n = 3,449 (13.3%)No HDP(n = 20,205, 78.1%)Age (y), mean (SD)30.8 (5.4)30.8 (5.9)30.2 (5.8)30.9 (5.2)Age, y (category) ≤253,601 (13.9)362 (16.3)648 (18.8)2,591 (12.8) 25-≤3516,040 (62)1,270 (57.1)2024 (58.7)12,746 (63.1) >356,238 (24.1)593 (26.7)777 (22.5)4,868 (24.1)Race/ethnicity Asian2,120 (8.2)122 (5.5)147 (4.3)1851 (9.2) Hispanic1,504 (5.8)135 (6.1)168 (4.9)1,201 (5.9) Non-Hispanic Black8,314 (32.1)853 (38.3)1,397 (40.5)6,064 (30.0) Non-Hispanic White12,588 (48.6)1,039 (46.7)1,571 (45.5)9,978 (49.4) Other/missing1,353 (5.2)76 (3.4)166 (4.8)1,111 (5.5)Insurance status Private19,652 (75.9)1,612 (72.4)2,502 (72.5)15,538 (76.9) Public6,066 (23.4)595 (26.7)927 (26.9)4,544 (22.5) Missing161 (0.6)18 (0.8)20 (0.6)123 (0.6)Pre-pregnancy BMI, kg/m2mean (SD)25.9 (6.4)27.8 (7.1)28.2 (7.3)25.3 (5.9)Pre-pregnancy BMI, kg/m2 (category) ≤2513,878 (53.6)899 (40.4)1,335 (38.7)11,644 (57.6) 25-≤305,931 (22.9)602 (27.1)904 (26.2)4,425 (21.9) >305,326 (20.6)654 (29.4)1,118 (32.4)3,554 (17.6) Missing744 (2.9)70 (3.1)92 (2.7)582 (2.9)Delivery method Cesarean7,273 (28.1)930 (41.8)1,011 (29.3)5,332 (26.4) Vaginal16,469 (63.6)1,170 (52.6)2,253 (65.3)13,046 (64.6) Missing2,137 (8.3)125 (5.6)185 (5.4)1827 (9.0)Parity Multiparous8,646 (33.4)508 (22.8)818 (23.7)7,320 (36.2) Nulliparous16,910 (65.3)1,695 (76.2)2,595 (75.2)12,620 (62.5) Missing323 (1.2)22 (1.0)36 (1.0)265 (1.3)Preterm delivery1872 (7.2)459 (20.6)158 (4.6)1,255 (6.2)Diabetes, pregestational593 (2.3)110 (4.9)121 (3.5)362 (1.8)Blood pressure trajectory groups Ultra-low-declining414 (1.6)10 (0.4)14 (0.4)390 (1.9) Low-declining7,387 (28.5)326 (14.7)469 (13.6)6,592 (32.6) Moderate-fast decline2,361 (9.1)156 (7.0)249 (7.2)1956 (9.7) Low-increasing7,151 (27.6)579 (26.0)866 (25.1)5,706 (28.2) Moderate-stable7,152 (27.6)866 (38.9)1,449 (42.0)4,837 (23.9) Elevated-stable1,414 (5.5)288 (12.9)402 (11.7)724 (3.6)BMI = body mass index; HDP = hypertensive disorders of pregnancy; HTN = hypertension.

Maternal Characteristics Stratified by HDP Status Among Patients With ≥3 Blood Pressures Before 20 Weeks’ Gestation

BMI = body mass index; HDP = hypertensive disorders of pregnancy; HTN = hypertension.

Most patients were classified as having low-declining (28.5%), low-increasing (27.6%), and moderate-stable (27.6%) BP trajectories. (Table 1, Supplemental Figure 2). Fewer patients fell into the extreme trajectories, with only 1.6% assigned to ultra-low-declining and 5.5% to elevated-stable BP groups. Patients in the higher trajectory groups were older, had a higher prevalence of pregestational diabetes, and were more likely to be nulliparous (Supplemental Table 1). Patients who identified as Asian were more frequently assigned ultra-low and low-declining BP trajectories, compared to other racial and ethnic groups. Distribution of HDP across BP trajectories stratified by preeclampsia and gestational HTN is shown in Figure 2. Patients with moderate-stable and elevated-stable trajectories had the highest rates of any HDP (32.4% and 48.8%, respectively). Those with ultra-low-declining BP had the lowest, with 5.8% of patients in this trajectory diagnosed with preeclampsia or gestational HTN during the index delivery.Figure 2HDP Outcome Distribution Among Early Pregnancy Blood Pressure Trajectory GroupsHDP = hypertensive disorders of pregnancy; HTN = hypertension.

HDP Outcome Distribution Among Early Pregnancy Blood Pressure Trajectory Groups

HDP = hypertensive disorders of pregnancy; HTN = hypertension.

After adjustment for age, race/ethnicity, BMI, parity, and pregestational diabetes, BP trajectory groups were independently associated with increasing risk for preeclampsia and gestational HTN in multivariable multinomial regression models (Table 2). Compared to patients with ultra-low or low-declining BP trajectories (referent), the adjusted odds of a preeclampsia diagnosis progressively increased from the moderate-fast decline (aOR: 1.5; 95% CI: 1.2-1.8) to the elevated-stable group (aOR: 6.2; 95% CI: 5.1-7.4). Similar effect sizes were observed for gestational HTN, and the composite outcome of any HDP (Supplemental Table 2). Race/ethnicity, BMI, nulliparity, and pregestational diabetes were independently associated with developing preeclampsia or gestational HTN. Models using the ultra-low-declining BP trajectory alone as the referent group demonstrated similar trends, although the magnitude of the effect size was higher (Supplemental Table 3). In the subgroup of nulliparous patients (N = 16,910), the association between BP trajectories and HDP outcomes was similar to the main analysis (Supplemental Table 4). Absolute risk for each BP trajectory group using median values of the other clinical covariates showed the highest risk among those with elevated-stable trajectory across all race/ethnicity groups (Supplemental Table 5).Table 2Association Between Early Pregnancy BP Trajectory Groups and Specific HDP DiagnosisAdjusted OR (95% CI)PreeclampsiaaGestational HTNaEarly pregnancy BPT groups Ultra-low and low-declining combinedRefRef Moderate-fast decline1.49 (1.22-1.82)1.66 (1.41-1.95) Low-increasing1.83 (1.59-2.11)1.90 (1.68-2.14) Moderate-stable2.92 (2.54-3.36)3.37 (3.00-3.78) Elevated-stable6.17 (5.13-7.42)5.96 (5.07-7.00)Demographic and clinical risk factors Age (1 year)1.02 (1.01-1.03)0.99 (0.98-1.00) Racial and ethnic groups WhiteRefRef Asian0.86 (0.70-1.10)0.71 (0.59-0.85) Hispanic1.31 (1.07-1.60)0.97 (0.81-1.16) Non-Hispanic Black1.61 (1.43-1.81)1.48 (1.34-1.63) Other/unknown0.78 (0.61-0.99)1.10 (0.92-1.31) Pre-pregnancy BMI ≤25RefRef 25-≤301.39 (1.24-1.56)1.41 (1.28-1.55) >301.60 (1.41-1.81)1.83 (1.65-2.02) Nulliparous2.42 (2.16-2.71)2.20 (2.00-2.42) Diabetes, pregestational2.40 (1.91-3.02)1.68 (1.35-2.10)BP = blood pressure; BPT = blood pressure trajectory; other abbreviations as in Table 1.aValues calculated using logistic regression with multinomial outcome (preeclampsia, gestational hypertension, or no HDP).

Table 2Association Between Early Pregnancy BP Trajectory Groups and Specific HDP DiagnosisAdjusted OR (95% CI)PreeclampsiaaGestational HTNaEarly pregnancy BPT groups Ultra-low and low-declining combinedRefRef Moderate-fast decline1.49 (1.22-1.82)1.66 (1.41-1.95) Low-increasing1.83 (1.59-2.11)1.90 (1.68-2.14) Moderate-stable2.92 (2.54-3.36)3.37 (3.00-3.78) Elevated-stable6.17 (5.13-7.42)5.96 (5.07-7.00)Demographic and clinical risk factors Age (1 year)1.02 (1.01-1.03)0.99 (0.98-1.00) Racial and ethnic groups WhiteRefRef Asian0.86 (0.70-1.10)0.71 (0.59-0.85) Hispanic1.31 (1.07-1.60)0.97 (0.81-1.16) Non-Hispanic Black1.61 (1.43-1.81)1.48 (1.34-1.63) Other/unknown0.78 (0.61-0.99)1.10 (0.92-1.31) Pre-pregnancy BMI ≤25RefRef 25-≤301.39 (1.24-1.56)1.41 (1.28-1.55) >301.60 (1.41-1.81)1.83 (1.65-2.02) Nulliparous2.42 (2.16-2.71)2.20 (2.00-2.42) Diabetes, pregestational2.40 (1.91-3.02)1.68 (1.35-2.10)BP = blood pressure; BPT = blood pressure trajectory; other abbreviations as in Table 1.aValues calculated using logistic regression with multinomial outcome (preeclampsia, gestational hypertension, or no HDP).

Association Between Early Pregnancy BP Trajectory Groups and Specific HDP Diagnosis

BP = blood pressure; BPT = blood pressure trajectory; other abbreviations as in Table 1.

Values calculated using logistic regression with multinomial outcome (preeclampsia, gestational hypertension, or no HDP).

The performance of models assessing the association between risk factors, BP, and HDP outcomes, represented by the C-statistics, are displayed in Table 3. The C-statistic for models including risk factors, without BP measurements, was 0.66 (95% CI: 0.65-0.67). The addition of the first systolic BP measurement in early pregnancy improved the C-statistic to 0.69 (95% CI: 0.68-0.70). The addition of BP trajectory in place of a first systolic BP further improved the C-statistic to 0.71 (95% CI: 0.70-0.72). Models performed similarly for both preeclampsia and gestational HTN outcomes individually as well as the composite outcome of any HDP. Decision curve analysis demonstrated greater net benefit for the trajectory-enhanced model compared with model using clinical risk factors alone across a range of threshold probabilities, suggesting improved clinical utility for identifying patients at risk of HDP (Supplemental Figure 3).Table 3Predictive Model Performance of HDP Using Clinical Risk Factors, Initial First-Trimester Systolic Blood Pressure, and Early Pregnancy Blood Pressure Trajectory GroupsCovariatesPerformance, AUC (95% CI)Model NumberEntire Cohort (N = 25,879)PreeclampsiaGestational HTNAny HDP1Age, race/ethnicity, parity, BMI, diabetes0.66 (0.65-0.67)0.66 (0.65-0.67)0.66 (0.65-0.67)2Age, race/ethnicity, parity, BMI, diabetes, initial SBP0.69 (0.68-0.70)0.69 (0.68-0.70)0.69 (0.68-0.70)3Age, race/ethnicity, parity, BMI, diabetes, BPT group0.71 (0.70-0.72)0.71 (0.70-0.72)0.71 (0.70-0.72)Sensitivity analysis: nulliparous patients (N = 16,910) 1aAge, race/ethnicity, BMI, diabetes0.64 (0.63-0.66)0.65 (0.64-0.66)0.65 (0.64-0.65) 2aAge, race/ethnicity, BMI, diabetes, initial SBP0.67 (0.66-0.69)0.68 (0.67-0.69)0.68 (0.67-0.69) 3aAge, race/ethnicity, BMI, diabetes, BPT group0.69 (0.68-0.71)0.70 (0.69-0.71)0.70 (0.69-0.71)Sensitivity analysis: patients with ≤2 BP (N = 8,734) 1bAge, race/ethnicity, parity, BMI, diabetes0.67 (0.65-0.69)0.68 (0.67-0.70)0.68 (0.66-0.69) 2bAge, race/ethnicity, parity, BMI, diabetes, initial SBP0.69 (0.67-0.71)0.70 (0.69-0.72)0.70 (0.68-0.71) 3bAge, race/ethnicity, parity, BMI, diabetes, BPT group0.69 (0.67-0.71)0.70 (0.69-0.72)0.70 (0.68-0.71)Sensitivity analysis: removing race/ethnicity as covariate (N = 25,879) 1cAge, parity, BMI, diabetes0.65 (0.64-0.66)0.65 (0.64-0.66)0.65 (0.64-0.66) 2cAge, parity, BMI, diabetes, initial SBP0.68 (0.67-0.69)0.69 (0.68-0.70)0.68 (0.67-0.69) 3cAge, parity, BMI, diabetes, BPT group0.70 (0.69-0.71)0.71 (0.70-0.72)0.70 (0.70-0.71)AUC = area under the curve; SBP = systolic blood pressure; other abbreviations as in Tables 1 and 2.

Table 3Predictive Model Performance of HDP Using Clinical Risk Factors, Initial First-Trimester Systolic Blood Pressure, and Early Pregnancy Blood Pressure Trajectory GroupsCovariatesPerformance, AUC (95% CI)Model NumberEntire Cohort (N = 25,879)PreeclampsiaGestational HTNAny HDP1Age, race/ethnicity, parity, BMI, diabetes0.66 (0.65-0.67)0.66 (0.65-0.67)0.66 (0.65-0.67)2Age, race/ethnicity, parity, BMI, diabetes, initial SBP0.69 (0.68-0.70)0.69 (0.68-0.70)0.69 (0.68-0.70)3Age, race/ethnicity, parity, BMI, diabetes, BPT group0.71 (0.70-0.72)0.71 (0.70-0.72)0.71 (0.70-0.72)Sensitivity analysis: nulliparous patients (N = 16,910) 1aAge, race/ethnicity, BMI, diabetes0.64 (0.63-0.66)0.65 (0.64-0.66)0.65 (0.64-0.65) 2aAge, race/ethnicity, BMI, diabetes, initial SBP0.67 (0.66-0.69)0.68 (0.67-0.69)0.68 (0.67-0.69) 3aAge, race/ethnicity, BMI, diabetes, BPT group0.69 (0.68-0.71)0.70 (0.69-0.71)0.70 (0.69-0.71)Sensitivity analysis: patients with ≤2 BP (N = 8,734) 1bAge, race/ethnicity, parity, BMI, diabetes0.67 (0.65-0.69)0.68 (0.67-0.70)0.68 (0.66-0.69) 2bAge, race/ethnicity, parity, BMI, diabetes, initial SBP0.69 (0.67-0.71)0.70 (0.69-0.72)0.70 (0.68-0.71) 3bAge, race/ethnicity, parity, BMI, diabetes, BPT group0.69 (0.67-0.71)0.70 (0.69-0.72)0.70 (0.68-0.71)Sensitivity analysis: removing race/ethnicity as covariate (N = 25,879) 1cAge, parity, BMI, diabetes0.65 (0.64-0.66)0.65 (0.64-0.66)0.65 (0.64-0.66) 2cAge, parity, BMI, diabetes, initial SBP0.68 (0.67-0.69)0.69 (0.68-0.70)0.68 (0.67-0.69) 3cAge, parity, BMI, diabetes, BPT group0.70 (0.69-0.71)0.71 (0.70-0.72)0.70 (0.70-0.71)AUC = area under the curve; SBP = systolic blood pressure; other abbreviations as in Tables 1 and 2.

Predictive Model Performance of HDP Using Clinical Risk Factors, Initial First-Trimester Systolic Blood Pressure, and Early Pregnancy Blood Pressure Trajectory Groups

AUC = area under the curve; SBP = systolic blood pressure; other abbreviations as in Tables 1 and 2.

Sensitivity analyses

After restricting the cohort to nulliparous patients (Table 3, models 1a-3a), the model performance C-statistics similarly improved with the addition of BP trajectories (0.70; 95% CI: 0.69-0.71) compared to risk factors alone (0.65; 95% CI: 0.64-0.65) or risk factors with single BP measurement (0.68; 95% CI: 0.67-0.69). To assess model performance among patients with fewer than 3 BP measurements (models 1b-3b), we restricted the cohort to patients with 1 or 2 BP measurements in the first 20 weeks’ gestation (N = 8,734) (Figure 1). The model performance C-statistic with risk factors alone was 0.68 (95% CI: 0.66-0.69). The addition of a single BP measurement improved the C-statistic to 0.70 (95% CI: 0.68-0.69). The addition of BP trajectory in place of a single BP measurement did not further improve the C-statistic. Finally, we assessed model performance without using race/ethnicity as a covariate (models 1c-3c). Model performance C-statistics improved with the addition of BP trajectory and was overall very similar to the models that included race/ethnicity (C-statistic 0.70 without race/ethnicity vs 0.71 with race/ethnicity) (Table 3).

Effect modification by race

To assess the association of BP trajectory with HDP outcomes among different racial groups, we restricted the cohort to patients who identified as non-Hispanic White (N = 12,588) and non-Hispanic Black (N = 8,314). Compared to patients with ultra-low or low-declining BP trajectories (referent), the adjusted odds of the composite outcome of any HDP progressively increased from the moderate-fast decline to the elevated-stable (aOR: 6.2; 95% CI: 5.1-7.4) groups for both White and Black patients (Table 4). Notably, the magnitude of the effect size was greater for White patients compared to Black patients for each BP trajectory group (aOR for the elevated-stable group: 9.0 [95% CI: 7.4-11.0] among White patients vs 3.5 [95% CI: 2.8-4.4] among Black patients). The model performance C-statistic was 0.72 (95% CI: 0.71 – 0.73) in White patients compared to 0.68 (95% CI: 0.67 – 0.69) in Black patients. As shown in Supplemental Table 6, effect modification was present for BP trajectory according to race (P < 0.01 for both preeclampsia and gestational HTN outcomes). Compared to White patients with ultra-low and low-declining BP trajectories (referent), Black patients with ultra-low and low-declining BP trajectories had 2.4 higher odds of preeclampsia (aOR: 2.4; 95% CI: 1.8-3.2), adjusting for other risk factors. Compared to the referent group of ultra-low and low-declining BP trajectory in White patients, the adjusted odds of preeclampsia and gestational progressively increased across BP trajectory groups in both White and Black patients.Table 4HDP Outcome Stratified by Race/EthnicityNon-Hispanic White (N = 12,588)Non-Hispanic Black (N = 8,314)BP TrajectoryHDP, n (%)aORa (95% CI)HDP, n (%)aORa (95% CI)Ultra-low and low-declining246 (7.4)Ref388 (16.9)RefModerate-fast decline194 (16.3)2.2 (1.8-2.7)147 (20.6)1.2 (1.0-1.5)Low-increasing664 (18.2)2.4 (2.1-2.9)585 (25.4)1.6 (1.3-1.8)Moderate-stable1,111 (30.6)4.3 (3.7-5.0)912 (36.0)2.4 (2.1-2.8)Elevated-stable395 (50.4)9.0 (7.4-11.0)218 (46.3)3.5 (2.8-4.4)aOR = adjusted OR; other abbreviation as in Table 1.aAdjusted for age, BMI, parity, and pregestational diabetes.

Table 4HDP Outcome Stratified by Race/EthnicityNon-Hispanic White (N = 12,588)Non-Hispanic Black (N = 8,314)BP TrajectoryHDP, n (%)aORa (95% CI)HDP, n (%)aORa (95% CI)Ultra-low and low-declining246 (7.4)Ref388 (16.9)RefModerate-fast decline194 (16.3)2.2 (1.8-2.7)147 (20.6)1.2 (1.0-1.5)Low-increasing664 (18.2)2.4 (2.1-2.9)585 (25.4)1.6 (1.3-1.8)Moderate-stable1,111 (30.6)4.3 (3.7-5.0)912 (36.0)2.4 (2.1-2.8)Elevated-stable395 (50.4)9.0 (7.4-11.0)218 (46.3)3.5 (2.8-4.4)aOR = adjusted OR; other abbreviation as in Table 1.aAdjusted for age, BMI, parity, and pregestational diabetes.

HDP Outcome Stratified by Race/Ethnicity

aOR = adjusted OR; other abbreviation as in Table 1.

Adjusted for age, BMI, parity, and pregestational diabetes.

Discussion

Summary of findings

We found that inclusion of BP trajectory groups, based on BP measurement from early pregnancy, improved our ability to predict diagnosis of HDP in our racially diverse cohort of 25,879 patients (Central Illustration). Patients in the highest-risk BP trajectory (elevated-stable) had a 6-fold increased risk of HDP compared to those in the lowest risk trajectories, an association which remained true for both preeclampsia and gestational HTN. Models incorporating BP trajectory discriminated risk of HDP better than models that included only clinical risk factors or risk factors along with initial systolic BP, suggesting that BP trajectory captures important prognostic information that is not apparent in a single BP measurement early in pregnancy.Central illustrationBlood Pressure Trajectories Predict Hypertensive Disorders of PregnancyBP = blood pressure; HDP = hypertensive disorders of pregnancy; HTN = hypertension.

Blood Pressure Trajectories Predict Hypertensive Disorders of Pregnancy

BP = blood pressure; HDP = hypertensive disorders of pregnancy; HTN = hypertension.

Comparison to existing literature

Our findings support the utility of discrete BP trajectories previously identified by Gunderson et al11 to improve risk stratification for preeclampsia and gestational HTN in a diverse population in northern California. Our study population is demographically distinct from the original cohort used to develop these trajectories, with a higher proportion of Black patients and patients with public insurance, and a lower proportion of Asian patients. Our cohort had fewer patients in the ultra-low-declining BP trajectory but otherwise demonstrated a similar distribution across BP trajectory groups. Prior development of the BP trajectory groups with internal validation was performed among a low-to-moderate risk community-based sample without pre-existing heart, kidney, or liver disease, and no chronic HTN with an overall prevalence of preeclampsia and gestational HTN of 10%.12 By contrast, our cohort had almost double the rate of HDP, suggesting this risk stratification can be used in populations with varied risk profiles. Our large and demographically diverse patient population further strengthens the external validity of the prognostic value of BP trajectories and supports the generalizability of this risk stratification across heterogeneous populations.

We fit prediction models using similar clinical risk factors (based on covariate availability in our data set) combined with BP trajectory. These models had comparably high discrimination (C-statistic 0.71 for our models, compared to 0.73-0.76 in prior studies).12 Among patients with <3 BPs available in early pregnancy, there was no difference in model performance between inclusion of initial systolic BP and BP trajectory. This suggests that trajectory inclusion may not be applicable for patients who enter prenatal care later than 16 weeks or those with limited access to outpatient appointments in early stages of pregnancy. This is an important limitation, as many patients initiate prenatal care after their first trimester. There are disparities surrounding access to early prenatal care that may limit the ability to apply BP trajectories across all populations.

We also found that BP trajectory improved model discrimination with and without the inclusion of race/ethnicity as a covariate. We chose to include self-reported race/ethnicity as a covariate in primary analysis because of well-described disparities in adverse maternal and perinatal outcomes among Black patients in the United States, particularly among those affected by pregnancy-related HTN.15 Previous work also suggested effect modification by race/ethnicity and BP trajectory associations with HDP.11 However, inclusion of race/ethnicity in prediction models is debated and imperfect, and there is risk of exacerbating inequity.16 Patients may receive different clinical management based on race alone, and focus on individual-level risk through prediction models can decrease attention on higher-level factors that influence disparities in care. Race/ethnicity was an effect modifier of the BP trajectory association for both preeclampsia and gestational HTN; however, risk increased across BP trajectories in a very similar graded manner for both White and Black patients. Among those within the lowest-risk BP trajectory, Black patients remained at elevated risk of preeclampsia or gestational HTN, which may be explained by unmeasured social, environmental, or clinical factors.17 Although BP trajectory groups were associated with a smaller magnitude of HDP risk among Black patients compared to White patients, our findings demonstrate that BP trajectory still provides useful risk stratification among all patients regardless of race.

Several groups have attempted to develop prediction models for preeclampsia and gestational HTN based on BP measurements throughout gestation, although none are widely used in routine clinical practice due to limited efficacy to predict later-onset preeclampsia. Existing models include a wide range of biomarker and Doppler ultrasound variables, as well as known maternal risk factors included in our study. However, many also incorporate variables that are not routinely available; for example, models by Odibo et al18 used clinical history in addition to placental protein 13, pregnancy-associated plasma protein A, and the mean uterine artery pulsatility index to predict preeclampsia-required delivery before 34 weeks’ gestation; other groups’ models also include these blood tests and ultrasound parameters.9,19 These models have a wide performance range (C-statistic 0.7-0.86) for early preterm preeclampsia. The Fetal Medicine Foundation has developed an algorithm with maternal history, biophysical data including BMI and a single mean arterial BP, and biochemical data including blood testing and ultrasound parameters to estimate the risk of preterm preeclampsia.20 This algorithm was evaluated in a cohort of nulliparous women from 11 to 14 weeks gestation in the recent PREDICTION study from Guerby et al. They found a detection rate of 77% for early-onset preeclampsia and 63% for preterm preeclampsia.21 However, these serum markers and Doppler measurements are not routinely collected as part of pregnancy-related care in the first trimester, limiting the clinical utility of this algorithm and other models. Although some of these models do include single BP measurements early in pregnancy, none currently use multiple BPs over time. The American College of Obstetricians and Gynecologists has noted the continued lack of a well-studied and supported model that can be used in routine prenatal care and therefore recommends only utilizing clinical risk factors until a robust and accessible model is developed.22 The additional prognostic information of BP trajectories, generated from readily available data with internal and external validation, may provide substantial support for this.11,12

Strengths and limitations

Our study has several limitations. First, we were unable to externally validate the previously published risk prediction model using BP trajectories due to missing covariates in our EHR data set (eg smoking status, social determinants of health, and prior HDP history). Some of these unmeasured risk factors may be partially captured by trajectory effects and potentially bias our results away from the null. As a result, we had to refit models using risk factors based on available data, which could impact the assessment of BP trajectories. Despite this, we demonstrate that BP trajectories improve model discrimination to a similar degree as prior studies. In addition, we performed several detailed sensitivity analyses to account for limitations in data availability. Second, few patients in our cohort were assigned an ultra-low-declining BP trajectory, which is likely related to the lower number of Asian patients in our cohort. As a result, our models used a different referent group that combined the ultra-low and low BP trajectory groups together. Despite this, we still found similar graded increases in the risk of HDP diagnosis among patients across the 6 early pregnancy BP trajectory groups. Finally, our risk prediction model performed differently in Black individuals. This emphasizes the importance of studying other factors that contribute to differences in HDP outcomes which may be difficult to measure, including environmental exposures and social and structural determinants of health.PerspectivesCOMPETENCY IN PATIENT CARE AND PROCEDURAL SKILLS 1: Future work should evaluate the role of applying BP trajectories into existing HDP risk prediction models. If applied prospectively, BP trajectories could be used to provide anticipatory guidance and reinforce measures to help prevent preeclampsia, such as early aspirin use and potentially more frequent home BP monitoring. Such approaches would need to be studied in a prospective trial. Trajectory group information could help target patients at highest risk who may benefit most from heightened surveillance.COMPETENCY IN PATIENT CARE AND PROCEDURAL SKILLS 2: This study validated the use of previously identified BP trajectories in early pregnancy in an external, demographically distinct cohort to predict diagnosis of HDP. Systolic BP patterns during the first 20 weeks’ gestation provide important prognostic information to identify patients who may benefit from enhanced screening or early interventions during pregnancy.TRANSLATIONAL OUTLOOK: Along with known clinical risk factors, early pregnancy BP trajectory groups can be incorporated into risk prediction for HDP. A risk calculator using readily available data from EHRs could help clinicians identify highest risk patients for monitoring and timely intervention.

COMPETENCY IN PATIENT CARE AND PROCEDURAL SKILLS 1: Future work should evaluate the role of applying BP trajectories into existing HDP risk prediction models. If applied prospectively, BP trajectories could be used to provide anticipatory guidance and reinforce measures to help prevent preeclampsia, such as early aspirin use and potentially more frequent home BP monitoring. Such approaches would need to be studied in a prospective trial. Trajectory group information could help target patients at highest risk who may benefit most from heightened surveillance.

COMPETENCY IN PATIENT CARE AND PROCEDURAL SKILLS 2: This study validated the use of previously identified BP trajectories in early pregnancy in an external, demographically distinct cohort to predict diagnosis of HDP. Systolic BP patterns during the first 20 weeks’ gestation provide important prognostic information to identify patients who may benefit from enhanced screening or early interventions during pregnancy.

TRANSLATIONAL OUTLOOK: Along with known clinical risk factors, early pregnancy BP trajectory groups can be incorporated into risk prediction for HDP. A risk calculator using readily available data from EHRs could help clinicians identify highest risk patients for monitoring and timely intervention.

Funding support and author disclosures

Dr Lewey was supported by 10.13039/100000050NHLBI grant K23 HL153667. The development of the early pregnancy blood pressure trajectory groups was supported by the 10.13039/100000050NHLBI R01 HL145808 (Dr Gunderson). All other authors have reported that they have no relationships relevant to the contents of this paper to disclose.