We propose a new framework combining dynamic sampling rates for healthcare sensors driven by user behavior, and an adaptive MAC scheduling scheme applied to the time-slotted channel hopping (TSCH) protocol in IEEE 802.15.4, which provides high throughput and reliable communications. First, we introduce a system software architecture for machine-learning-assisted healthcare monitoring that detects the user's behavior using edge computing and adjusts the sampling rates of the healthcare sensors accordingly. Second, we propose an adaptive MAC scheduling scheme for TSCH based on a state machine model that reacts to the dynamic traffic generated by the healthcare sensors. In case of an urgent state, the MAC scheduler automatically allocates extra timeslots to the appropriate sensors so as to enable the reliable transfer of high-resolution sensor data for further analysis. Experimental results from our testbed, implemented in Contiki-OS on the OpenMote-CC2538 platform, show that the proposed adaptive scheduling scheme can respond quickly to changes in user behavior and ensure the reliable transfer of sensor data in emergency situations.