Prediction of ectopic pregnancy in women with a pregnancy of unknown location.

OBJECTIVE

We have previously published on the use of mathematical Model M1 to predict ectopic pregnancy in women with no signs of intra- or extrauterine pregnancy. The aim of this study was to improve on the performance of this model for the detection of developing ectopic pregnancies in women with pregnancies of unknown location (PULs). We therefore generated and evaluated a new logistic regression model from simple hormonal data and compared it with Model M1.

METHODS

Data were collected prospectively from women classified as having a PUL. These women were followed until the diagnosis was established as: a failing PUL, an intrauterine pregnancy (IUP) or an ectopic pregnancy. A multinomial logistic regression model, Model M4, was developed on 201 training cases and it was tested prospectively on another 175 women with a PUL. M4 performance was evaluated using receiver-operating characteristics (ROC) curves and compared with Model M1 based on the human chorionic gonadotropin (hCG) ratio alone.

RESULTS

A total of 376 women with a PUL were recruited into this study: 201 in the training set (109 (54.2%) with a failing PUL, 76 (37.8%) with an IUP and 12 (6.0%) with an ectopic pregnancy; four with a persisting PUL were excluded from analysis) and 175 in the test set (94 (53.7%) with a failing PUL, 64 (36.6%) with an IUP and 15 (8.6%) with an ectopic pregnancy; two with a persisting PUL were excluded from analysis). The log serum hCG average ((hCG 0 h + hCG 48 h)/2) and the hCG ratio (hCG 48 h/hCG 0 h) were encoded as variables following multivariate analysis of the basic data. The new Model M4 contained the log of the hCG average, the hCG ratio and its quadratic effect. In the prediction of ectopic pregnancy, M4 gave an area under the ROC curve (AUC) of 0.900 and M1 gave an AUC of 0.842 (P = 0.0303).

CONCLUSIONS

Although Model M4 is superior to Model M1 when comparing the AUCs for prediction of developing ectopic pregnancies in a PUL population, in real terms this model did not result in substantially more pregnancies being classified correctly as developing ectopic pregnancies. Prospective multicenter studies are needed to assess the diagnostic performance of such models in different populations.