Abstract Background Hypertensive disorders of pregnancy (HDP) are associated with an elevated risk of incident hypertension after delivery. Identifying patients at greatest risk of postpartum hypertension could improve targeting of preventive interventions. The use of electronic health record (EHR) data to predict hypertension risk may improve generalizability in real-world settings but may be limited by incomplete data. Purpose To develop a prediction model for incident postpartum hypertension using EHR data in a racially diverse cohort. Methods This cohort study analyzed electronic health record data from a U.S.-based academic health system, including demographics, vital signs, medications, and diagnosis codes for ambulatory and hospital encounters. Patients (aged 15 years) with a delivery between 2012-2020 were included. Patients with pre-pregnancy hypertension, serious medical conditions, or without postpartum data were excluded. Candidate predictors included demographics, health factors, and ambulatory blood pressures (BP) during early pregnancy (0-20 weeks) and postpartum (4-8 weeks). The primary outcome was incident hypertension diagnosed between 6 and 24 months postpartum as defined by a validated algorithm using diagnosis codes, medications and/or BP values (≥140/90). We modeled the risk of incident hypertension using logistic regression models, considering predictors identified a priori based on clinical relevance. Model performance was assessed with area under the receiver operating characteristic curve (AUC). Internal validation was conducted using bootstrap resampling. We imputed missing data for variables with 5% missing. Over 18% of women had missing BP values in their EHRs. To address this, we applied a pseudo-likelihood approach to fit the logistic regression models and estimate predictive accuracy metrics. We further performed sensitivity analysis by conducting standard analyses using complete data records. Results Of 30,744 patients, mean age at delivery was 29.9 years, 43.9% identified as White, 39.9% identified as Black, and 22.5% had HDP (Table 1). Between 6 and 24 months postpartum, 881 (2.9%) developed incident hypertension. The final model included age, body mass index (BMI), HDP, insurance, nulliparity, preterm delivery, pregestational diabetes, BP medication at delivery discharge, BMI by HDP interaction term, and postpartum SBP and DBP (Table 2). The AUC was 0.834. Women with and without missing BP data showed similar distributions for the completely observed variables. The odds ratio parameter estimates and the AUC were similar when analyzing complete data (N=21,733). Conclusion In a real-world cohort of racially diverse patients, postpartum hypertension can be accurately predicted using HDP, BMI, and other widely available clinical factors. These hypertension risk estimates may help guide discussions regarding intensity of cardiovascular preventive measures in the postpartum period.
Lewey et al. (Sat,) studied this question.