Body Mass Index and Survival in Men and Women Aged 70 to 75

Leon Flicker, PhD; Kieran A. McCaul, PhD; Graeme J. Hankey, MD; Konrad Jamrozik, PhD; Wendy J. Brown, PhD; Julie E. Byles, PhD; Osvaldo P. Almeida, PhD


J Am Geriatr Soc. 2010;58(2):234-241. 

In This Article


Ethical Approval

The human research ethics committee of the University of Western Australia approved the protocol for the Health in Men Study (HIMS), and ethics committees at the University of Newcastle and the University of Queensland approved that for the Australian Longitudinal Study on Women's Health (ALSWH).

Data Sources

Data were used from two population-based longitudinal studies that began in 1996: the HIMS and the older cohort from the ALSWH. The detailed methods for both studies are published elsewhere.[12,13]

The men who were screened for abdominal aortic aneurysm in a randomized controlled trial conducted in Perth, Western Australia, in 1996 formed the HIMS cohort. In this trial, eligible men were aged 65 to 79, resident in Perth (the capital of Western Australia), and not in long-stay institutional accommodation. A list of all potentially eligible men was drawn from an electronic copy of the electoral roll in 1996 (voting is compulsory for adult Australians); after excluding 8,801 who were no longer resident in Perth and 2,296 who had died before being contacted, the remaining men were randomized into the screening group (n=19,352) or control group (n=19,352). Of those invited to be screened, 1,846 were ineligible, 5,303 did not respond or refused, and 12,203 were screened. These 12,203 screened men formed the HIMS cohort and have been followed since their recruitment. The sample size was based on the number of men required to be screened to demonstrate 50% lower mortality from abdominal aortic aneurysm than in the control group.

In the ALSWH study, three cohorts of women were randomly selected in 1996 from electronic records of Australia's universal health insurance scheme, Medicare, which covers all citizens and permanent residents. In the oldest cohort, aged 70 to 75, 39,000 women were invited to participate; of these 1,100 were not contactable, and 2,366 were ineligible. Of those women remaining (35,534), 12,614 responded. Sample size was based on the available funding for the three cohorts.

For this analysis to achieve comparability between the cohorts, only the men aged 70 to 75 at baseline (n=4,931) and the women resident in metropolitan and urban areas (n=5,042) were included. The overall response rate was 69.7% for the men and 35.5% for the women. In both cohorts, survival was better than in the populations from which they were recruited. At 10 years, survival of HIMS respondents was 16% higher than observed in the general population of this age and survival of ALSHW respondents was 8% higher.

The HIMS and ALSWH surveys collected self-reported measures of height and weight, which were used to calculate BMI. In addition, a variety of demographic (e.g., age, education, marital status), lifestyle (e.g., smoking status, alcohol consumption, exercise), and health status characteristics (e.g. self-reported history of hypertension, diabetes mellitus) were ascertained. Current alcohol use was categorized into three levels using the National Health and Medical Research Council of Australia guidelines,[14] which recommend no more than two standard drinks (each containing 10 g alcohol) per day for women and four standard drinks per day for men and, for both sexes, at least 2 alcohol-free days per week. Subjects were classified as nondrinkers, as drinking within recommended levels, or as having alcohol consumption exceeding recommended levels. Subjects answered questions relating to participation in a usual week in vigorous exercise, (e.g., jogging) and nonvigorous exercise (e.g., walking). Subjects were categorized as sedentary if they reported no time in either of these activities in a usual week.

Participants were followed for a decade or until death if sooner. Date of death and multiple causes of death were obtained from the Australian Bureau of Statistics, which allowed 100% ascertainment of cause-specific mortality status until the end of 2005. Causes of death were coded according to the International Classification of Diseases, Ninth Revision (ICD-9) or Tenth Revision (ICD-10). In this analysis, cause-specific mortality was determined using all causes of death documented on the death certificate rather than just the single underlying cause of death, which resulted in some individuals being counted in more than one category of cause-specific mortality. Deaths were grouped into three major categories: cardiovascular disease (ICD9: 390–434, 436–448; ICD10: I00–I78), cancer (ICD9: 140–208, ICD10: C00–C97), and chronic respiratory disease (ICD9: 490–494, 496, ICD10: J40–J47).

Statistical Analysis

Cox proportional hazards regression was used to model survival to death from all causes. Survival time was calculated in days from the date of entry into the relevant study to the date of death or the end of follow-up (December 31, 2005), whichever came first. Individuals still alive at the end of follow-up were censored and tied survival times were broken using Efron's method. The proportional hazards assumption was tested by examining the relationship between the scaled Schoenfeld residuals and survival time. The overall fit of the regression models was assessed by examining the Cox-Snell residuals.

To investigate the functional form of the association between BMI and mortality risk, BMI was modeled as a continuous variable using a restricted cubic spline with three knots. From this analysis, the BMI associated with the minimum mortality risk was predicted for men and women, and their associated 95% confidence intervals were estimated using bootstrapping. To ensure that the choice of knots was optimal, a series of sensitivity analyses was conducted, varying the number of knots used (3, 4, 5, or 6 knots) and their location. Results from these analyses were examined graphically and compared using the Bayesian information criterion (BIC).

Potential confounders and effect modifiers of the relationship between BMI (categorized according to WHO criteria) and mortality were investigated in the following way. Initially, the cohorts were combined, and each covariate, apart from sex, was modeled separately for its relationship with mortality risk. Subsequently, two separate models were fitted for each covariate. The covariate was modeled with BMI and an interaction term between the covariate and BMI. Then each covariate was modeled with sex and an interaction term between the covariate and sex. These analyses were performed to detect any interactions between sex and any covariate. Any interaction that achieved nominal significance at the .05 level (two-tailed) was retained for further modeling. Final modeling was an iterative process of adding covariates (and their relevant interaction terms) to a base model of sex, BMI, and an interaction between sex and BMI. Additional covariates were added to the baseline model if they altered any of the effect estimates in the baseline model by 10% or more.

A potential source of bias arises if, for some participants, illness has caused weight loss and this illness also increases the risk of mortality. To determine whether the presence of preexisting illness modified the relationship with BMI, men and women were categorized as healthy if they reported no prior history of diabetes mellitus, heart disease, stroke, hypertension, or chronic respiratory illness and if they were not current smokers. Regression models were also fitted conditional on 1-, 2-, and 3-year survival. This removed the influence of early mortality from the hazard ratio estimates, and these were compared with those obtained from the full cohort.


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.
Post as: