///2019 Abstract Details
2019 Abstract Details2019-07-13T07:45:15-05:00

External validation of a multivariable prediction model for the diagnosis of placenta accreta in women with suspicion of placenta accreta spectrum

Abstract Number: F3H-451
Abstract Type: Original Research

Shubhangi Singh MBBS1 ; Thomas D Shipp MD2; Daniela A. Carusi MD3; Carolyn F. Weiniger MD4; Kara G Fields MS5; Michaela K Farber MD MS6

Background: Placenta accreta spectrum (PAS) is associated with major postpartum hemorrhage and increased morbidity. Anticipatory recognition is critical to minimize life-threatening hemorrhage from unexpected PAS. A previously published multivariable logistic regression model for the antenatal diagnosis of women with suspected PAS is based on 3 risk factors: prior cesarean delivery (CD), placenta previa, and ultrasound suspicion of PAS (1). We aimed to validate this prediction model.

Methods: Women who delivered at a single tertiary center between January 2007 and December 2017 with above risk factors were included in this analysis. Specifically, women with placenta previa and prior CD, and women with prior CD and/or placenta previa and ultrasound suspicion for PA. Women with prior CD or previa with a reassuring ultrasound were excluded. Risk scores for PA were calculated for each patient with the previously published prediction model. The ability of the model to discriminate between patients with and without PA was assessed by calculating the area under the receiver operating characteristic (ROC) curve (AUC). Sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) were calculated for the two probability cut-offs identified previously: 0.174 (corresponding to 100% sensitivity in model development sample) and 0.208 (corresponding to the optimal sensitivity/specificity tradeoff).

Results: Of 307 identified cases of suspected PAS, 140 (46%) were confirmed at the time of delivery. The model AUC when applied to our sample was 0.683 (95% CI: 0.615, 0.751) (Figure). The sensitivity, specificity, PPV and NPV for the probability cut-off of 0.174 in our sample were 75.7%, 10.8%, 41.6%, 34.6% respectively. The sensitivity, specificity, PPV and NPV for the probability cut-off of 0.208 were 70.7%, 64.1%, 62.3% and 72.3% respectively.

Conclusions: The current study offers validation of the prediction model for PAS described by Weiniger et al (1). As expected, the prediction model had a weaker ability to discriminate between patients with and without PAS in our external sample (AUC 0.683) than in the model development sample (AUC 0.846). Interestingly, specificity using the optimal cutoff was higher in our validation (64.1%) than in the original study (52.5%). Model discrimination may be improved by updating model coefficients and/or adding additional model predictors.


1. Int J Obstet Anesth 2013;22:273-9.

SOAP 2019