Efficient management of healthcare resources is a central issue for modern hospitals and public health systems. Patient flows, scheduling of medical examinations, and treatment planning require robust, interpretable, and resource-efficient optimization methods. In this paper, we propose Q2OPT-MED, a reinforcement learning (RL) method guided by the 2-opt heuristic that combines the simplicity of tabular Q-Learning (QL) with a reward function enriched with clinical knowledge. Unlike cumbersome, poorly interpretable neural approaches, this approach is lightweight, runs on a standard central processing unit (CPU), and is well-suited for constrained hospital environments. Experiments conducted on real-world hospital scheduling datasets demonstrate that Q2OPT-MED improves patient throughput, reduces waiting times, and accelerates convergence compared to classical methods. These results confirm the relevance of hybrid approaches (AI + heuristics) for optimizing patient care pathways.