From 1 January 1994 through 3 April 2003, 713 participants were diagnosed with dementia (). Mean age at initial recorded diagnosis was 74.5 years (range 66-82). The mean time to start of ascertainment of dementia was 26.5 years after the multiphasic exam. Those with the diagnosis were more likely to be older, have a grade school education (completed schooling to age 12), and be unmarried in mid-life.
Demographic characteristics of the participants at mid-life by dementia status. Figures are numbers (percentage) of participants unless stated otherwise*
At mid-life, 10% of the cohort were obese, 36% overweight, 53% normal weight, and 1.3% underweight. The prevalence of a subsequent diagnosis of dementia was significantly higher for those who were obese or overweight at mid-life (). Those in the highest fifth for subscapular and tricep skinfold measurements at mid-life were more likely to have dementia than those in the lowest fifth (). Post hoc analyses of the 10 276 participants in the study compared with the 10 407 who were excluded because they were no longer health plan members in 1994 showed no significant differences in any of the mid-life measures of adiposity or covariates by status of health plan membership in 1994. We also conducted post hoc analyses to ensure that there was no selection effect due to age, indeed participants who were obese or overweight were not older at time of ascertainment of dementia than those of normal weight (see table on bmj.com
). We checked the proportionality of hazards for each covariate by entering interaction terms of the covariate by person years into the model. The P values for each were non-significant (P > 0.06), indicating all hazards were proportional.
Mid-life adiposity of participants by dementia status. Figures are numbers (percentage) of participants
Compared with those normal weight at mid-life, obese people had a 74% greater risk of dementia (hazard ratio 1.74, 95% confidence interval 1.34 to 2.26, fully adjusted model, ), while those who were overweight had a 35% greater risk (1.35, 1.14 to 1.60, fully adjusted model). In sex specific models body mass index was associated with dementia more strongly in women (body mass index*sex interaction term P = 0.06). Obese women were twice as likely to have dementia as women of normal weight (2.07, 1.49 to 2.89, fully adjusted model), while obese men had a non-significant 30% increase in risk (1.30, 0.84 to 1.87, fully adjusted model). Overweight women were 55% more likely to have dementia than women of normal weight (1.55, 1.22 to 1.97, fully adjusted model), while overweight men had a non-significant 16% increase in risk compared with men of normal weight (1.16, 0.91 to 1.46, fully adjusted model). Being underweight was not significantly associated with dementia in either sex, but only 0.6% of men and 1.9% of women had a body mass index < 18.5, limiting the power to detect such an association. There were no significant race interactions in the association between body mass index and risk of dementia (P > 0.15 for race*body mass index interaction term).
Cox proportional hazards model of body mass index at mid-life and risk of dementia. Figures are hazard ratios (95% confidence intervals)
Measures of skinfold thickness at mid-life were significantly associated with risk of dementia at a magnitude similar to body mass index (). Those in the highest fifth of subscapular skinfold had a 72% increased risk, while those in the highest fifth of tricep skinfolds had a 59% increased risk compared with those in the lowest fifth (fully adjusted models: 1.72, 1.36 to 2.18, and 1.59, 1.24 to 2.04, respectively). There was a non-significant trend for the effect of high subscapular skinfold thickness on risk of dementia to be stronger among men. Compared with those in the lowest fifth, men in the highest fifth of subscapular skinfold thickness had nearly a twofold increase in risk of dementia, while women had a 50% increase (). Addition of body mass index to the skinfold models did not attenuate the effect (data not shown). Results did not vary by race in the association between skinfolds and dementia risk (P > 0.15, interaction term skinfold measures*race).
Cox proportional hazards model of skinfold thickness (according to fifth of distribution*) at mid-life and risk of dementia. Figures are hazard ratios (95% confidence intervals)