Development and Validation of Case-Ascertainment Algorithms for Hypertensive Disorders of Pregnancy Using Longitudinal Electronic Health Records Data.

Objectives: Hypertensive disorders of pregnancy (HDP), including chronic hypertension, gestational hypertension, and preeclampsia/eclampsia, is a leading cause of maternal and perinatal morbidity. Accurate identification of individual HDP subtypes in electronic health records (EHRs) is critical for research and surveillance but remains a challenge. We aimed to develop and validate EHR-based case-ascertainment algorithms for individual HDP conditions using medical chart review.

Study design and setting: We conducted a validation study within the Blood Pressure in Pregnancy, Obesity, Diabetes and Perinatal Outcomes (BIPOD) cohort at Kaiser Permanente Northern California, comprising 441,147 singleton pregnancies from 2011 to 2021. Using a stratified sampling approach, we selected 980 pregnancies for medical chart review: 200 chronic hypertension, 280 gestational hypertension, 300 preeclampsia/eclampsia, and 200 normotensive pregnancies. Following the American College of Obstetricians and Gynecologists diagnosis criteria, we developed HDP case-ascertainment algorithms incorporating clinician diagnosis codes, antihypertensive medications, systolic/diastolic blood pressure, and laboratory test results. Normotension was defined as not meeting HDP definitions throughout pregnancy. Positive predictive value (PPV), negative predictive value (NPV), sensitivity, and specificity were calculated, with weighting to account for sampling design. Minimum validity thresholds were set as 80% PPV, 90% NPV, 80% sensitivity, and 90% specificity.

Results: Algorithms for chronic and gestational hypertension demonstrated high diagnostic validity across all definitions, with all performance statistics exceeding the minimum thresholds. All definitions were retained in the final algorithms for chronic hypertension [weighted PPV: 87.0% (95% CI 86.4%-87.6%), NPV: 99.5% (99.5%-99.5%); sensitivity: 84.9% (84.2%-85.5%); specificity 99.6% (99.6%-99.6%)] and gestational hypertension [weighted PPV 91.4% (91.1%-91.7%); NPV: 99.5% (99.5%-99.5%); sensitivity: 94.0% (93.7%-94.2%); specificity: 99.3% (99.2%-99.3%)]. For preeclampsia/eclampsia, only the definition using inpatient diagnosis had acceptable validity (PPV: 94.9%), while definitions using outpatient diagnoses or laboratory results had poor PPV (0.0%-8.0%). Weighted performance for the preeclampsia/eclampsia final algorithm using inpatient diagnosis was high: PPV 94.9% (94.5%-95.2%); NPV 99.5% (99.5%-99.5%); sensitivity 88.8% (88.3%-89.3%); specificity 99.8% (99.8%-99.8%). Similarly, normotensive had high validation performance: PPV 99.5% (99.5%-99.5%); NPV 91.4% (91.2%-91.7%); sensitivity 98.6% (98.6%-98.7%); specificity 96.7% (96.5%-96.8%).

Conclusion: EHR-based case-ascertainment algorithms for HDP demonstrated high validity in a large, diverse population. These algorithms can facilitate accurate HDP phenotyping in population-based studies.