External validation of models and simple scoring systems to predict miscarriage in intrauterine pregnancies of uncertain viability

Does a logistic regression model and scoring system to predict viability of an intrauterine pregnancy of uncertain viability (PUV) perform as well in an independent patient group as the original patient group? The model and scoring system showed good performance on external validation confirming their value for the prediction of miscarriage/viability in PUV patients up to 1114 weeks of gestation. Several individual ultrasound and demographic factors have been described as predictors for miscarriage. A logistic regression model and simple scoring system using basic clinical and ultrasound features, such as maternal age, bleeding score, mean gestational sac diameter (MSD) and presence or absence of yolk sac, have been developed to allow patient-specific prediction of viability of PUV beyond the first trimester. Prospective observational external validation cohort study in two inner city early pregnancy assessment units over a period of 18 months. All consecutive women with a PUV were recruited. Ultrasound (mean sac diameter and presence of yolk sac) and demographic variables (maternal age, bleeding score and gestational age) were noted. The outcome measure was first trimester (1114 week) viability. Women with unknown first trimester outcome were excluded. Receiver operating characteristic (ROC) curves and calibration plots were constructed. Test performance was compared with the original development data set. A new model and scoring system, which did not include gestational age, was built and evaluated. Of the 575 women who were recruited, first trimester outcome was known for 89.2 (n 513). The model could only be validated in 400 patients, due to missing values in model variables and outcome. The model predicted viability with an area under the ROC curve (AUC) of 0.845 [95 confidence interval (CI), 0.8060.884] compared with 0.774 (95 CI, 0.7010.848) in the original study. The AUC for the scoring system was 0.832 (95 CI, 0.7920.872) compared with 0.771 (95 CI, 0.6980.844) from the original study data set. The new model and the scoring system, excluding gestational age, could be evaluated on 503 patients and resulted in an AUC of 0.801 (95 CI, 0.7650.841) for the model and 0.773 (95 CI, 0.7330.812) for the scoring system. Approximately 22 of patients could not be validated due to missing variables and for 11 of patients the first trimester outcome was unknown. Both the model and the scoring system showed excellent performance on external validation confirming their generalizability and utility in prediction of viability beyond the first trimester in clinical practice. An advantage of the mathematical models original M-o and new M-n and scoring systems original SSo and new SSn is that they can provide women with an individualized probability of the viability of their pregnancy using only demographic information, symptoms and TVS findings. Furthermore, the risk of miscarriage can be given immediately following examination. T.B. is supported by the Imperial Healthcare NHS Trust NIHR Biomedical Research Centre. This research is supported by Research Council KUL GOA MaNet, iMinds 2012, Belgian Federal Science Policy Office IUAP P719. VVB is a postdoctoral fellow of the Research Foundation Flanders (FWO). There are no conflicts of interest.