Continuous Training and Fine-tuning for Domain-Specific Language Models in Medical Question Answering