Behavior change is associated with important decrease of cognitive and physical capacities among elderly people. Therefore, a proactive detection of long-term behavior changes in early stages of their evolution is a keystone to improve elderly healthcare services. In fact, nowadays’ geriatric methods mainly rely on scales and questionnaires, and are inconvenient to investigate long-term changes on a daily basis. Therefore, our proposed approach for behavior change detection analyzes elderly people behavior over long periods via ambient technologies. In fact, employed technologies are unobtrusive, do not interfere with the natural behavior of elderly people and do not affect their privacy. Furthermore, our long-term behavior analysis is based on the identification of significant behavior change indicators (e.g., mobility, memory, nutrition and social life indicators significantly correlate with cognitive and physical diseases), and the application of efficient statistical techniques that differentiate long-term and short-term changes in analyzed behavior. In addition, our two-year deployment validates our objective technological observations through real correlations with medical observations of nursing-home team.