Objective To assess whether postpartum hemorrhage can be predicted in women with gestational hypertension or mild preeclampsia at term. Design A cohort study in which we used data from our multicentre randomized controlled trial (HYPITAT trial). Setting The study was conducted in 38 hospitals in the Netherlands between 2005 and 2008. Population Women with gestational hypertension or mild preeclampsia at term (n = 1132). Methods An antepartum model (model A) and an antepartum/intrapartum model (model B) were created using logistic regression. The predictive capacity of the models was assessed with receiver operating characteristic analysis and calibration. Main outcome measure Postpartum hemorrhage, defined as blood loss >1000 mL within 24 h after delivery. Results Postpartum hemorrhage occurred in 118 (10.4%) women. Maternal age (odds ratio 1.03), prepregnancy body mass index (odds ratio 0.96), and women with preeclampsia (odds ratio 1.5) were independent antepartum prognostic variables of postpartum hemorrhage. Intrapartum variables incorporated in the model were gestational age at delivery (odds ratio 1.2), duration of dilatation stage (odds ratio 1.1), and episiotomy (odds ratio 1.5). Model A and model B showed moderate discrimination, with areas under the receiver operating characteristic curve of 0.59 (95% confidence interval 0.53-0.64) and 0.64 (95% confidence interval 0.59-0.70), respectively. Calibration was moderate for model A (Hosmer-Lemeshow p = 0.26) but better for model B (Hosmer-Lemeshow p = 0.36). The rates of postpartum hemorrhage ranged from 4% (lowest 10%) to 22% (highest 10%). Conclusion In the assessment of performance of a prediction model, calibration is more important than discriminative capacity. Our prediction model shows that for women with gestational hypertension or mild preeclampsia at term, distinction between low and high risk of developing postpartum hemorrhage is possible when antepartum and intrapartum variables are combined.
- gestational hypertension
- predictive value
- prognostic model
- receiver-operating characteristic curve analysis