Background The relevance of body mass index (BMI) to cause-specific mortality in old age is uncertain.

Objectives To examine cause-specific 5 year mortality in old age by BMI in old age and middle age (40–69 years).

Methods Cox proportional hazards for mortality rates among 4862 former male civil servants in relation to quartiles of BMI measured when screened in 1968–70 and when resurveyed in 1997–98 (median age 76 years).

Results The association between all-cause mortality after resurvey and BMI in old age was U-shaped with hazard ratios (HRs) of 1.3 (95% CI 1.1–1.5) for the lightest and heaviest categories relative to the middle two. Among ‘healthy’ men the lightest (<22.7 kg/m²) had greatest all-cause mortality. The heaviest men (>26.6 kg/m²) had increased risk of cardiovascular disease (CVD) mortality in the first two years or for the whole period if never-smokers. Respiratory mortality was inversely associated with BMI in old age [adjusted HR for trend per BMI category increase 0.6 (0.5–0.7)] but cancer mortality lacked a clear pattern. Net gain or loss of 10 kg or more between middle and old age was a strong predictor of all-cause and CVD mortality.

Conclusions The shape of the association between BMI in old age and mortality differs by cause of death. Major weight change over time is a warning signal for higher CVD mortality. Having BMI <22.7 kg/m² in old age is associated with above-average mortality rates even if apparently healthy.