An approach to the prediction of neonatal Erb palsy

Abstract

We devised a method to predict risk of neonatal Erb palsy using variables ascertainable before delivery. Multiple logistic regression modeling was used to construct an Erb palsy risk score from a case-control study of 45 consecutive Erb palsy cases and 90 controls. Receiver-operator characteristics identified a range of scores for which likelihood ratios were determined for calculation of predictive values across a range of prior probabilities. In the final model, large estimated fetal weight, gestational diabetes, large maternal body mass index, large weight gain, and black race were associated with 2.3-4.8 times greater odds of Erb palsy. A long second stage had a modest effect on the odds (OR=2.7, P=0.13), unless preceded by a long deceleration phase, which combination increased the odds of Erb palsy to 20.1 (P=0.001). A risk score of 0.72 had a sensitivity of 36% and a specificity of 99%. In a theoretical population with a birth prevalence of brachial plexus palsy of 2.5/1000, application of the risk score would prevent 36% of cases and result in about 14 cesareans for every nerve injury prevented. We conclude that risk scoring using multiple logistic regression coefficients of variables that can be known in time to affect decision-making about mode of delivery has the potential to guide intervention to prevent some Erb palsies.