Towards Quantification of Bias in Machine Learning for Healthcare: A Case Study of Renal Failure Prediction