Body Mass and Weight Change in Relation to Mortality Risk
Body Mass and Weight Change in Relation to Mortality Risk
The National Institutes of Health-AARP Diet and Health Study is composed of men and women living in 6 US states and 2 metropolitan areas. A self-administered baseline questionnaire was mailed to 3.5 million members of AARP (formerly the American Association of Retired Persons) aged 50–71 years in 1995–1996, with an initial cohort size of 566,402. The baseline questionnaire included current height and weight. Six months later, a second questionnaire seeking information on height and weight at earlier ages and on other variables was sent to persons who responded to the baseline questionnaire. From a starting population of 334,908 men and women who completed both questionnaires, we restricted the data set to the 118,823 who reported never smoking. We excluded subjects whose questionnaires were completed by a surrogate respondent (n = 3,373) and subjects who reported extremely high values for alcohol consumption (more than 3 standard deviations above the 75th percentile (n = 766)) or provided no weight or height data (n = 509). Finally, to ensure that no individual had an age interval greater than 20 years for the analysis of weight change between age 50 years and study baseline, we excluded subjects aged 70 years or more (n = 4,228). The remaining 109,947 participants (53,126 men and 56,821 women) comprised our analytical cohort.
Cohort members were followed for vital status from the return date of the second questionnaire in 1996–1997 through December 31, 2009. Vital status was ascertained by annual linkage of the cohort to the Social Security Administration Death Master File, which contains information on all deaths occurring in the United States. The design and maintenance of this cohort have been described previously. All study participants provided written informed consent, and the National Institutes of Health-AARP Diet and Health Study was approved by the Special Studies Institutional Review Board of the US National Cancer Institute.
Cause of death was obtained through linkage to the National Death Index Plus, using International Classification of Diseases, Ninth Revision (ICD-9), and International Classification of Diseases, Tenth Revision (ICD-10), codes for death certificate underlying cause of death. The codes were initially grouped using the National Center for Health Statistics' "113 selected causes of death" and then further consolidated to create broad categories, as follows: cancer (ICD-9 codes 140–239; ICD-10 codes C00–C97 and D00–D48), cardiovascular disease (CVD) (ICD-9 codes 390–398, 401–404, 410–429, and 440–448; ICD-10 codes I00–I13, I20–I51, and I70–I78), and stroke (ICD-9 codes 430–438; ICD-10 codes I60–I69). All other codes were considered together.
Data on current height and weight, health status, smoking habits, race/ethnicity, physical activity, alcohol consumption, and recalled height at age 18 years and weight at ages 18, 35, and 50 years were collected by questionnaire. We created BMI variables for 3 distinct ages (18, 35, and 50 years) and for the weighted average of BMI at ages 18, 35, and 50 years. BMI categories were nested within broad, generally accepted weight classifications, including underweight (<18.5), 2 categories of normal weight (18.5–22.4 (referent) and 22.5–24.9), 2 categories of overweight (25.0–27.4 and 27.5–29.9), and 2 categories of obesity (30.0–32.4 and ≥32.5). The BMI categories were constructed to: 1) reflect the weight distributions of the study population at ages 18, 35, and 50 years; 2) create large, statistically stable groups; and 3) incorporate established BMI cutpoints. For example, at age 18 years, most subjects had a BMI in the normal or underweight category and relatively few (1.4%) had a BMI in the obese range. In contrast, at age 50 years, sufficient numbers of subjects had an obese BMI to allow the creation of 2 obesity categories. The leaner category in the normal BMI range was chosen as the referent BMI category. For an alternative analysis of BMI at age 18 years, we calculated BMI percentiles using the Centers for Disease Control and Prevention (CDC) sex-specific BMI-for-age growth charts for children and categorized these values according to the CDC-recommended cutpoints.
Weight-change categories were created for 3 age intervals spanning adulthood: ages 18–35 years, ages 35–50 years, and ages 50–69 years (i.e., self-reported weight from the baseline questionnaire). In contrast to the 2 younger age intervals, the width of the 50- to 69-year interval differed according to the subject's age at entry (maximum age at baseline questionnaire = 69.9 years; average age = 61.9 years (standard deviation, 5.2)). As described above, we truncated the study population age limit slightly to ensure that the weight-change interval at older ages was no more than 20 years in duration, to provide greater comparability with the younger age intervals. A weight change of 0.2 kg/year or less during an age interval was defined as stable weight, and this category served as the referent group. Health status was evaluated using subjects' responses to the question, "Would you say your health in general is: ——?," with 5 response categories ranging from poor to excellent.
Multivariate hazard ratios for mortality were estimated using Cox regression analysis, with age at return of the second questionnaire used as the underlying time metric. On average, the time between return of the baseline questionnaire and return of the second questionnaire was 0.57 years (standard deviation, 0.19). Age-adjusted mortality rates were calculated using the Mantel-Haenszel method. We estimated associations for BMI at ages 18, 35, and 50 years, average BMI from early adulthood to middle adulthood (as a measure of cumulative body mass), and weight change over 3 consecutive age intervals of 18–35 years, 35–50 years, and 50 years to study entry (ages 50–69 years). Subjects who died but had no available information on underlying cause of death were excluded from cause-specific analyses (n = 1,706). Subjects who were missing body weight data were excluded on an analysis-specific basis. To examine the potential influence of illness on body weight, we stratified analyses of weight change from age 50 years to age 69 years according to self-reported health status. All models adjusted for race/ethnicity, education, physical activity, and alcohol consumption at study entry. Analyses combining men and women included adjustment for gender. Weight-change models additionally adjusted for initial BMI (i.e., BMI at the beginning of the weight-change age interval) and recalled height at age 18 years. Our final analysis considered the influence of long-term excess weight by estimating risks based on the youngest age at which participants reported having a BMI of 25.0 or higher, compared with a referent group that maintained a BMI below 25.0 in all 4 age periods (this analysis included BMI at study entry).
Materials and Methods
Study Population
The National Institutes of Health-AARP Diet and Health Study is composed of men and women living in 6 US states and 2 metropolitan areas. A self-administered baseline questionnaire was mailed to 3.5 million members of AARP (formerly the American Association of Retired Persons) aged 50–71 years in 1995–1996, with an initial cohort size of 566,402. The baseline questionnaire included current height and weight. Six months later, a second questionnaire seeking information on height and weight at earlier ages and on other variables was sent to persons who responded to the baseline questionnaire. From a starting population of 334,908 men and women who completed both questionnaires, we restricted the data set to the 118,823 who reported never smoking. We excluded subjects whose questionnaires were completed by a surrogate respondent (n = 3,373) and subjects who reported extremely high values for alcohol consumption (more than 3 standard deviations above the 75th percentile (n = 766)) or provided no weight or height data (n = 509). Finally, to ensure that no individual had an age interval greater than 20 years for the analysis of weight change between age 50 years and study baseline, we excluded subjects aged 70 years or more (n = 4,228). The remaining 109,947 participants (53,126 men and 56,821 women) comprised our analytical cohort.
Cohort Follow-up
Cohort members were followed for vital status from the return date of the second questionnaire in 1996–1997 through December 31, 2009. Vital status was ascertained by annual linkage of the cohort to the Social Security Administration Death Master File, which contains information on all deaths occurring in the United States. The design and maintenance of this cohort have been described previously. All study participants provided written informed consent, and the National Institutes of Health-AARP Diet and Health Study was approved by the Special Studies Institutional Review Board of the US National Cancer Institute.
Causes of Death
Cause of death was obtained through linkage to the National Death Index Plus, using International Classification of Diseases, Ninth Revision (ICD-9), and International Classification of Diseases, Tenth Revision (ICD-10), codes for death certificate underlying cause of death. The codes were initially grouped using the National Center for Health Statistics' "113 selected causes of death" and then further consolidated to create broad categories, as follows: cancer (ICD-9 codes 140–239; ICD-10 codes C00–C97 and D00–D48), cardiovascular disease (CVD) (ICD-9 codes 390–398, 401–404, 410–429, and 440–448; ICD-10 codes I00–I13, I20–I51, and I70–I78), and stroke (ICD-9 codes 430–438; ICD-10 codes I60–I69). All other codes were considered together.
Assessment of BMI and Weight Change
Data on current height and weight, health status, smoking habits, race/ethnicity, physical activity, alcohol consumption, and recalled height at age 18 years and weight at ages 18, 35, and 50 years were collected by questionnaire. We created BMI variables for 3 distinct ages (18, 35, and 50 years) and for the weighted average of BMI at ages 18, 35, and 50 years. BMI categories were nested within broad, generally accepted weight classifications, including underweight (<18.5), 2 categories of normal weight (18.5–22.4 (referent) and 22.5–24.9), 2 categories of overweight (25.0–27.4 and 27.5–29.9), and 2 categories of obesity (30.0–32.4 and ≥32.5). The BMI categories were constructed to: 1) reflect the weight distributions of the study population at ages 18, 35, and 50 years; 2) create large, statistically stable groups; and 3) incorporate established BMI cutpoints. For example, at age 18 years, most subjects had a BMI in the normal or underweight category and relatively few (1.4%) had a BMI in the obese range. In contrast, at age 50 years, sufficient numbers of subjects had an obese BMI to allow the creation of 2 obesity categories. The leaner category in the normal BMI range was chosen as the referent BMI category. For an alternative analysis of BMI at age 18 years, we calculated BMI percentiles using the Centers for Disease Control and Prevention (CDC) sex-specific BMI-for-age growth charts for children and categorized these values according to the CDC-recommended cutpoints.
Weight-change categories were created for 3 age intervals spanning adulthood: ages 18–35 years, ages 35–50 years, and ages 50–69 years (i.e., self-reported weight from the baseline questionnaire). In contrast to the 2 younger age intervals, the width of the 50- to 69-year interval differed according to the subject's age at entry (maximum age at baseline questionnaire = 69.9 years; average age = 61.9 years (standard deviation, 5.2)). As described above, we truncated the study population age limit slightly to ensure that the weight-change interval at older ages was no more than 20 years in duration, to provide greater comparability with the younger age intervals. A weight change of 0.2 kg/year or less during an age interval was defined as stable weight, and this category served as the referent group. Health status was evaluated using subjects' responses to the question, "Would you say your health in general is: ——?," with 5 response categories ranging from poor to excellent.
Data Analysis
Multivariate hazard ratios for mortality were estimated using Cox regression analysis, with age at return of the second questionnaire used as the underlying time metric. On average, the time between return of the baseline questionnaire and return of the second questionnaire was 0.57 years (standard deviation, 0.19). Age-adjusted mortality rates were calculated using the Mantel-Haenszel method. We estimated associations for BMI at ages 18, 35, and 50 years, average BMI from early adulthood to middle adulthood (as a measure of cumulative body mass), and weight change over 3 consecutive age intervals of 18–35 years, 35–50 years, and 50 years to study entry (ages 50–69 years). Subjects who died but had no available information on underlying cause of death were excluded from cause-specific analyses (n = 1,706). Subjects who were missing body weight data were excluded on an analysis-specific basis. To examine the potential influence of illness on body weight, we stratified analyses of weight change from age 50 years to age 69 years according to self-reported health status. All models adjusted for race/ethnicity, education, physical activity, and alcohol consumption at study entry. Analyses combining men and women included adjustment for gender. Weight-change models additionally adjusted for initial BMI (i.e., BMI at the beginning of the weight-change age interval) and recalled height at age 18 years. Our final analysis considered the influence of long-term excess weight by estimating risks based on the youngest age at which participants reported having a BMI of 25.0 or higher, compared with a referent group that maintained a BMI below 25.0 in all 4 age periods (this analysis included BMI at study entry).