Evaluating Bias in Retrieval-Augmented Medical Question-Answering Systems