The authors propose an adaptive frequency hopping (AFH) algorithm, entitled robust adaptive frequency hopping (RAFH), for providing increased reliability of a wireless medical telemetry system (WMTS) under coexistence environment with non-medical devices. The conventional AFH scheme classifies channels into 'good' or 'bad' according to the threshold-based on-off decision by packet error rate (PER) measurement, and only uses good channels with a uniform hop probability. Unlike the conventional AFH scheme, RAFH is a novel technique, which solves a constrained entropy maximisation problem and assigns every channel a different hop probability as a decreasing function of the measured PER. The key novelty of RAFH over existing AFH schemes is that it reflects the relative channel condition by assigning non-uniform hop probabilities. By adopting constrained entropy maximisation, RAFH not only improves the average PER, but also reduces the PER fluctuation over time under a dynamic interference environment, both of which increase the reliability of WMTS. Through extensive simulation, we show that RAFH outperforms basic frequency hopping (FH) and the conventional AFH with respect to the PER under various scenarios of dynamic interference.