OBJECTIVE To examine the stability of patients' choices for life-sustaining treatments. METHODS A longitudinal cohort study. METHODS Primary care practices in central North Carolina. METHODS Medicare recipients (n = 2536). METHODS Participants were asked about demographic characteristics, health status, well-being, depression, social support, use of a living will, and desire for life-sustaining treatment if they were to become terminally ill. These questions were repeated 2 years later (n = 2073, 82% follow-up). RESULTS The population tended to choose to forego one more treatment at follow-up than they did at baseline. A choice to forego treatment was twice as stable as a choice to receive treatment. Patients with a living will were less likely to change their wishes (14%) than those without a living will (41%). Persons were more likely to want increased treatment at a later time if they had been hospitalized (23% compared with 18%), had had an accident (29% compared with 19%), had become more immobile (23% compared with 19%), had become more depressed (25% compared with 15%), or had less social support (25% compared with 14%). CONCLUSIONS Most patients (85%) who had chosen to forego life-sustaining treatments did not change their choices. Nonetheless, these data suggest that it is important to review patients' preferences for life-sustaining treatments rather than to assume the stability of their choices.