Prediction of subsequent miscarriage risk in women who present with a viable pregnancy at the first early pregnancy scan

Objectives: To generate and evaluate a new prediction model for miscarriage in women who present with a viable intrauterine pregnancy (IUP) at the primary early pregnancy scan and to compare this new model to a previously published model.

Materials and methods: Data were collected prospectively from women presenting to the early pregnancy unit with a viable IUP between November 2006 and January 2013. More than 30 historical, clinical and ultrasonographic variables were recorded on a standardised datasheet at the first visit. Women were followed until the final outcome was known at the end of the first trimester: viable IUP or miscarriage. A new multinomial logistic regression model was developed retrospectively on training cases and tested prospectively on test cases. The performance of the new prediction model was evaluated using receiver operating characteristic (ROC) curves and compared to a previously published model. After removing cases with missing values for the model of Oates, the area under the ROC curve (AUC) was also calculated for the new model and the Oates model.

Results: A total of 1115 consecutive first-trimester women presented to the early pregnancy unit. Eight hundred and sixty-two women with a viable IUP at the first scan whose outcome was known at the end of the first trimester were included in the final analysis. Five hundred and sixty-six women were included in the training set and 296 in the test set. 92.1% were viable and 7.9% had miscarried at the end of the first trimester. The most significant independent prognostic variables for the logistic regression model were as follows: maternal age, embryonic heart rate (EHR), logarithm [gestational sac (GS) volume/crown-rump length (CRL)], CRL and the presence or absence of clots per vagina (PV) at presentation. The performance of the new model compared with the Oates model gave an AUC of 0.870 vs 0.847 for the training set and 0.783 vs 0.744 for the test set. After removing cases with missing values for the model of Oates 2013, the performance of the new model compared to the Oates model gave an AUC of 0.887 vs 0.861 for the training set and 0.816 vs 0.734 for the test set (P-value <0.04).

Conclusions: We have developed a new prediction model which indicates the likelihood of miscarriage. In women who present with a viable IUP at the primary scan, advancing maternal age in the presence of clots PV increases the probability of subsequent miscarriage. Whereas, in women with a higher EHR in the presence of an increased GS volume/CRL ratio, the likelihood of subsequent miscarriage is reduced. This new model outperforms the previously published model developed in our unit.