<p>Wearable medical devices play an important role in real-time health monitoring, but their adoption is limited by high energy consumption, especially in wireless data transmission. This paper addresses the challenge of reducing power demand while maintaining clinical data fidelity. The study is based on 2000 intraoperative cases drawn from the VitalDB repository, encompassing heart rate, body temperature, and systolic and diastolic blood pressure recordings. Three transmission optimization approaches were evaluated: linear dimensionality reduction through principal component analysis (PCA), threshold-based scheduling derived from clinical criteria, and nonlinear compression using autoencoder (AE) models, including convolutional, recurrent, and temporal convolutional network (TCN) variants. A hybrid framework integrating threshold scheduling with a Masked-TCN autoencoder was subsequently developed. PCA provided modest compression (1.33×) with minimal computational overhead. In contrast, threshold scheduling achieved significant transmission reduction (41.4×) and approximately 97.6% energy savings, though it achieved only partial clinical event recall (76.2%). Autoencoder-based compression, particularly with the Masked-TCN, produced strong reconstruction accuracy, maintaining temporal consistency with PRD at or below 4%, albeit with moderate transmission savings (around 10–20%). The hybrid framework successfully balanced these trade-offs, reducing communication load by 74% compared with naive transmission while preserving high signal fidelity and reliable event detection. These findings indicate that integrating clinically guided threshold rules with deep learning-based compression provides a practical and efficient pathway for extending the operational lifetime of wearable health-monitoring systems without compromising patient safety.</p>