Comparison of the competing risks model for SGA with the
stillbirth-specific logistic regression model
Prediction of stillbirth, expressed as AUROC and detection rate at 10%
FPR, in screening by maternal risk factors and combinations with EFW and
UtA-PI for all stillbirths and the subgroups of antepartum stillbirths
and those that were related to placental dysfunction by the two models
of screening are summarised in Table 2. At 10% FPR, the competing risks
model predicted 58.6% (52.6 - 64.6), 66.2% (59.9 - 72.6), 70.8% (63.8
- 77.8) of placental dysfunction
related stillbirths, at any gestation, at <37 weeks’ gestation
and <32 weeks, respectively, which were similar to the
respective figures of 62.3% (57.2-67.4), 69.8% (65.0-74.6), and 72.5%
(67.8-77.2) achieved by the application of the stillbirth-specific
logistic regression model.
The ROC curves for prediction of all antepartum stillbirths and
placental dysfunction related stillbirths, at any gestation, at
<37 weeks’ gestation and at <32 weeks,
by the competing risks model for
SGA fetuses, are shown in Figure 2. The detection rates at 1%, 3%, 5%
and 10% FPR in screening by the competing risks model utilising
maternal risk factors and combinations with EFW and UtA-PI are shown in
Table S1. Reducing the FPR from 10% to 3% results in a relatively mild
reduction in the detection rate, an observation that might be useful in
balancing effective prediction and availability of resources.