Supplementary Figure 4.

Decision trees to predict the risk of miscarriage. Only the initial blood sample taken at first presentation was used to generate this decision tree. The risk of miscarriage (between 0 and 1) is presented within each node (rectangles with rounded corners coloured in either green (starting), blue (decision) or red (terminal). The proportion of all women who would reach that node is presented above each node. At the starting node (green), the risk of miscarriage is 0.28. The arrows can then be followed according to the level of unadjusted βhCG (IU/L), kisspeptin levels (pmol/L) and gestational age by last menstrual period (days). The final miscarriage risk is indicated once the terminal nodes (red) have been reached. For example, a pregnant woman that has a βhCG level <14,000 iU/L will have a 0.81 risk of miscarriage and this risk will be present in 17% of the total study population (Decision Tree A). The left panel (Decision Tree A) has an accuracy for correctly classifying women of 87.4% (95% CI: 83.4% - 90.7%), but this complex model could be susceptible to overfitting (i.e. may perform with less accuracy when used in other datasets). The right panel (Decision Tree B) is a parsimonious (simplified) model with an 83.1% accuracy (95% CI: 78.6% - 86.7%) that is more likely to maintain similar performance when applied to other datasets. βhCG, beta chorionic gonadotropin (iU/L); KP, Kisspeptin (pmol/L).