Differential Performance of Machine Learning Models in Prediction of Procedure-Specific Outcomes