Pregnancy of unknown location: external validation of the hCG-based M6NP and M4 prediction models in an emergency gynaecology unit

Abstract

Objective To investigate if M6NP predicting ectopic pregnancy (EP) among women with pregnancy of unknown location (PUL) is valid in an emergency gynaecology setting and comparing it with its predecessor M4.Design Retrospective cohort study.Setting University Hospital.Participants Women with PUL.Methods All consecutive women with a PUL during a study period of 3 years were screened for inclusion. Risk prediction of an EP was based on two serum human chorionic gonadotropin (hCG) levels taken at least 24 hours and no longer than 72 hours apart.Main outcome measures The area under the ROC curve (AUC) expressed the ability of a model to distinguish an EP from a non-EP (discrimination). Calibration assessed the agreement between the predicted risk of an EP and the true risk (proportion) of EP. The proportion of EPs and non-EPs classified as high risk assessed the model’s sensitivity and false positive rate (FPR). The proportion of non-EPs among women classified as low risk was the model’s negative predictive value (NPV). The clinical utility of a model was evaluated with decision curve analysis.Results 1061 women were included in the study, of which 238 (22%) had a final diagnosis of EP. The AUC for EP was 0.85 for M6NP and 0.81 for M4. M6NP made accurate risk predictions of EP up to predictions of 20% but thereafter risks were underestimated. M4 was poorly calibrated up to risk predictions of 40%. With a 5% threshold for high risk classification the sensitivity for EP was 95% for M6NP, the FPR 50% and NPV 97%. M6NP had higher sensitivity and NPV than M4 but also a higher FPR. M6NP had utility at all thresholds as opposed to M4 that had no utility at thresholds≤5%.Conclusions M6NP had better predictive performance than M4 and is valid in women with PUL attending an emergency gynaecology unit. Our results can encourage implementation of M6NP in related yet untested clinical settings to effectively support clinical decision-making.