Complex and heterogeneous personal health time-series data present significant challenges for existing modeling approaches, especially concerning the accuracy and robustness of reasoning processes. Traditional methods often fail to explicitly capture the inherent logical structures necessary for accurate health predictions, leading to suboptimal performance in complex real-world scenarios. This paper proposes HealthTimeLLM-R1, a novel time-series reasoning enhancement method leveraging reinforcement learning (RL)-enhanced large language models (LLMs). We first construct a high-quality healthcare reasoning dataset containing detailed reasoning pathways by employing automated annotation techniques using advanced LLMs. We then introduce a two-stage optimization strategy combining chain-of-thought (CoT) supervised fine-tuning with Group Relative Policy Optimization (GRPO)-based reinforcement learning, systematically enhancing the model’s reasoning capability in healthcare tasks. Experimental results on multiple health-related time-series tasks show that our method substantially improves reasoning accuracy compared to existing benchmarks, highlighting the potential of RL-enhanced LLMs for robust and reliable healthcare applications.