Objectives: To evaluate the prognostic significance of weight change in frail elderly patients.

Design: Prospective study.

Setting: The general medical or surgical wards of a university‐affiliated Department of Veterans Affairs hospital.

Participants: Six hundred sixty randomly selected subjects aged 65 or older.

Measurements: At admission and discharge, each subject completed a standardized diagnostic evaluation, including demographic variables, and basic medical, functional, and nutritional assessments. Weights were recorded from hospital or clinic visits in the year before admission, during the current hospitalization, and after discharge for a median of 5.6 years. Average yearly intraindividual weight change was calculated using least‐squares regression. Associations between intraindividual weight change and mortality were identified using Cox proportional hazards regression.

Results: During the study, 314 subjects (48%) died. A U‐shaped association between weight change and mortality was observed. Those who were relatively weight stable (±1 kg/y) had the lowest mortality (28%). Compared with this group, the adjusted relative risks (ARR) of death for those who lost 1 or more to less than 3 kg/y and 3 or more kg/y were 2.14 (95% confidence interval (CI)=1.52–3.00) and 3.59 (95% CI=2.58–4.99), respectively. The ARR associated with a weight gain of 1 or more to less than 3 kg/y and 3 or more kg/y was 1.38 (95% CI=0.91–2.10) and 3.73 (95% CI=2.34–5.94), respectively. The amount of bidirectional weight fluctuation (estimated using coefficient of variance) was not significantly associated with mortality.

Conclusion: For reasons that are not clear, elderly patients who gain 3 or more kg/y after hospital discharge are at nearly the same risk of mortality as those who lose this amount of weight.