Comparison of ultrasound and magnetic resonance imaging parameters in predicting survival in isolated left‐sided congenital diaphragmatic hernia

Abstract

To compare test characteristics of ultrasound- and magnetic resonance imaging (MRI)-derived parameters in predicting newborn survival in cases of isolated left-sided congenital diaphragmatic hernia (CDH). This was a retrospective study involving 85 fetuses with an isolated left CDH. All had detailed prenatal evaluation, prenatal care, delivery and postnatal care at a single institution. Ultrasound images were reviewed to allow calculation of the lung-to-head ratio (LHR) and the observed/expected LHR (O/E-LHR), and MRI images were reviewed to determine the observed/expected total lung volume (O/E-TLV) and the percent herniated liver (%HL). Univariable logistic regression was used to evaluate each parameter for its ability to predict survival. Receiver-operating characteristics (ROC) curves were constructed and test characteristics were determined for each parameter as a predictor of survival. The overall survival for all fetuses included was 65%. Pseudo-R(2) values for all parameters were similar and were statistically significant as predictors of survival, with %HL having the highest pseudo-R(2) , of 0.28. ROC curve analysis showed ultrasound-determined parameters (LHR and O/E-LHR) to have a similar area under the curve (AUC), of 0.70, whilst MRI parameters (O/E-TLV and %HL) had AUC values of 0.82 and 0.84, respectively. At ROC-curve-determined cut-off values, MRI parameters had better test characteristics than did ultrasound parameters. At a standardized 5% false-positive rate, %HL performed best, with a sensitivity of 0.54 and a specificity of 0.95. At clinically employed cut-off values, sensitivity was similar for all parameters but MRI parameters provided the best combination of sensitivity and specificity, as evidenced by better likelihood ratios. A variety of measures have been proposed as antenatal predictors of survival in CDH. Ultrasound parameters function at a similar level, whereas MRI-determined parameters appear to offer better predictive value.