OC31.08: What is the optimal approach to accurately classifying failing pregnancies of unknown location (PULs)?

Abstract

Background: Banerjee et al. (2001) have previously developed a logistic regression model to predict failing pregnancies of unknown location (PULs). In the same study, this model did not outperform serum progesterone < 20 nmol/L taken at presentation. We compare these diagnostic tests to the human chorionic gonadotrophin (hCG) ratio for the prediction of failing PULs. Methods: Retrospective observational study. We compared the performance of three models for the prediction of failing PULs. 1) Logistic regression model (Banerjee et al., incorporating vaginal bleeding, endometrial thickness; 2) serum progesterone level and serum hCG level), 2) Serum progesterone <20 nmol/L at 0 hour (hr); and 3) hCG ratio (hCG 48 hr/hCG 0 hr) < 0.8. The performance of these models was evaluated using receiver operating characteristic curves (ROC) curves. P-values for comparison of AUC curves calculated using DeLong et al.’s method. Results: 4698 consecutive women were scanned and 370 were classified as PULs. For the prediction of failing PULs, the area under the ROC (AUC) for the logistic regression model was 0.944; the AUC for serum progesterone < 20 nmol/L at 0 hr was 0.963; and the AUC for hCG ratio < 0.8 was 0.972. P-values for comparison of AUC curves (See Table 1): Conclusions: The hCG ratio is the optimal test for the prediction of failing PULs.