Predictive modeling for placenta accreta diagnosis among pre-operatively suspected cases
Abstract Number: OP1-4
Abstract Type: Original Research
Background: Diagnosis of placenta accreta prior to surgery is tentative, limiting anesthesia planning. Predictive modeling techniques were employed to improve the assessment of likelihood of the real risk of placenta accreta.
Methods: Consecutively over a 9 year period, women with predictors for placenta accreta were categorized prior to cesarean delivery into high or low susipcion. Predictors included ultrasound signs, placenta previa, previous cesarean delivery. The uncertainty with this categorization meant that women without accreta diagnosed at surgery were undergoing preparations such as general anesthesia and large intravenous access unnecessarily. A more accurate model with higher discriminatory power was sought. A new model, determined by area under the curve of the receiver operating characteristic curve was chosen. The new risk score was derived from the model, optimal cut-points were chosen to classify subjects into high/low risk for accreta, hysterectomy and massive transfusion.
Results: Of ninety-two women with suspected placenta accreta, 52 (56.5%) had accreta diagnosis confirmed at surgery. Classification to low, 25 (29.3%) and high, 65 (70.1%) suspicion predicted definite accreta among high suspicion cases with sensitivity 90% and specificity 55%. General anesthesia was performed for 50/52 (96.2%) patients with accreta diagnosed at surgery and 29/40 (72.5%) without, p=0.001. The new model with highest discriminatory power to differentiate between women found to have/not have accreta at surgery used ultrasound signs of accreta, previous cesarean delivery and placenta previa, with odds ratios (95%CI); 7.8 (2.66-22.68), p=0.0002; 1.77 (1.15-2.71), p=0.0091; 3.9 (1.08-13.98), p=0.0381, respectively. The risk score for accreta derived from the model has an area under curve of 0.84. Using an optimal cut-off value on the risk score, a sensitivity of 94% and specificity of 52% was obtained for predicting the likelihood of accreta, figure 1. Sensitivity and specificity for prediction of massive blood transfusion was 96% and 63% respectively.
Conclusion: A risk score derived from a prospectively assessed cohort of parturients using pre-delivery predictors of placenta previa, ultrasound signs and previous multiple cesarean deliveries can improve accuracy of predicting placenta accreta. This enables preparation for the likelihood of massive transfusion among patients likely to have accreta.