Uncovering Interpretable Potential Confounders in Electronic Medical Records