Enabling Scalable Evaluation of Bias Patterns in Medical LLMs