The societal need for better public healthcare calls for granular, continuous, nationwide instrumentation and data fusion technologies. However, the current trend of centralised (database) health analytics gives rise to data privacy issues. This paper proposes sensor data mining algorithms that help infer health/well-being related lifestyle patterns and anomalous (or privacy-sensitive) events. Such algorithms enable a user-centric context awareness at the network edge, which can be used for decentralised eHealth decision making and privacy protection by design. The main hypothesis of this work involves the detection of atypical behaviours from a given stream of energy consumption data recorded at eight houses over a period of a year for cooking, microwave, and TV activities. Our initial exploratory results suggest that in the case of an unemployed single resident, the dayby-day variability of TV or microwave operation, in conjunction with the variability of the absence of other cooking activity, is more significant as compared with the variability of other combinations of activities. The proposed methodology brings together appliance monitoring, privacy, and anomaly detection within a healthcare context, which is readily scalable to include other health-related sensor streams.