Abstract Body: Introduction: Identifying cardiovascular disease (CVD) risk in adults under age 50 remains an unmet need. Pregnancy offers a unique opportunity to identify CVD risk given high health care engagement and frequent medical follow-up. We aimed to develop a prediction model for long-term CVD risk using clinical factors that can be ascertained at prenatal care initiation.
Methods: We conducted a retrospective study in an electronic health record-based pregnancy cohort from a single institution. We included singleton pregnancies from individuals aged ≥18 years who delivered between 1998 and 2016, had ≥3 months of postpartum follow-up data, and were free of CVD before pregnancy. We split the cohort into training (60%) and validation (40%) sets. We used Cox regression to model time from delivery to diagnosis of CVD, with censoring at the last non-obstetric encounter, the next delivery, or 12/31/2024. We identified CVD using diagnostic codes for atherosclerotic CVD and heart failure. Predictors were adapted from the PREVENT equations and included universally available factors at initiation of prenatal care: age (years), pre-pregnancy diabetes, pre-pregnancy hypertension (HTN), pre-pregnancy smoking, 1^st trimester body mass index (kg/m²) and systolic blood pressure (mmHg), and the social deprivation index. We used time-varying predictors to account for multiple pregnancies per individual. We assessed model performance using the C-statistic, smoothed calibration curves and the integrated calibration index (an index = 0 indicates perfect calibration) at 10 years.
Results: We included 46,392 pregnancies from 32,600 individuals (Table 1). Participants had a mean (SD) age at delivery of 31.6 (5.6) years. Over a median (IQR) of 7.3 (2.4, 13.6) years, 2.1% of individuals overall (n=561 [training], n=400 [validation]) experienced CVD. Hazard ratios for CVD for each of the predictors are given in Table 2. In the validation dataset, the model had a C-statistic of 0.668, indicating moderate discrimination. The model demonstrated excellent calibration, with an integrated calibration index of 0.00197 (Figure 1).
Conclusions: A prediction model based on clinical factors routinely available at the initiation of prenatal care demonstrated moderate discrimination and excellent calibration in predicting 10-year CVD risk.