The weighted sample at baseline () was 52.9% female and 83.5% non-Hispanic white. A significant portion of the weighted sample could be characterized as socioeconomically disadvantaged, with 25.6% reporting less than a high school education and 19.2% reporting a household income of less than $10,000 in the previous year in 1986.
Baseline Characteristics of Americans’ Changing Lives Sample (N=3,617), 1986
In terms of health risk behaviors at baseline (), 30.4% reported that they currently smoked, 4.3% were categorized as heavy drinkers, and 14.4% had a BMI in the obese range. Previous research using data from ACL Wave 1 has shown that the four health risk variables under study were significantly more prevalent in low-education and low-income populations, even after adjusting for differences in the age distribution across education and income strata (blinded for review). For example, after adjusting for age, the prevalence of smoking and overweight in the lowest income category was 37.7% and 24.4% respectively, compared with 27.4% and 14.0% in the highest income category (results not shown).
During the 19 year follow-up period, 1,409 study subjects died (26.0% of the weighted sample) (). The mean age at death was 73.9 years, with significant differences across sociodemographic groups. The mean age at death was 70.8 years for men versus 77.1 for women (p<.001), and 75.0 for non-Hispanic whites versus 68.1 for others (p<.001). Of the 1,409 people who died between 1986 and 2005, 4.9% were <45 years of age, 19.1% were between 45 and 65 years, 21.3% were between 65 and 74 years, and 54.6% were 75 years or older at the time of death.
The results of Cox proportional hazards models including demographic and socioeconomic characteristics as covariates revealed that mortality was significantly associated with age, gender, race, and place of residence. As expected, the hazard rate ratio (HRR) of mortality increased dramatically with age, and was lower for females compared to males (, Model 1). Non-Hispanic whites had a lower rate of mortality after controlling for other sociodemographic characteristics (HRR=0.86, 95% C.I. 0.71-1.05), but this was not statistically significant in any of the models. In addition, the mortality rate for those living in central cities was significantly higher than those residing in suburban and rural areas, controlling for sociodemographic characteristics, health risk behaviors, and health status (, Models1-3).
Mortality Hazard Rate Ratios (95% Confidence Intervals) for Americans’ Changing Lives Full Sample (n=3,617), 1986-2005
Controlling for demographic factors, income was significantly related to mortality (, Model 1). Compared to the highest income group, the HRR for the middle income group was 1.50 (95% C.I.=1.11-1.99) and 2.12 (95% C.I.=1.60-2.81) for the lowest income group. Education was significantly associated with mortality in Model 1, with a HRR of 1.40 (95% C.I.=1.05–1.85) for those with less than a high school education controlling for income and other sociodemographic characteristics. When health risk variables were added to the model, the education effect attenuated to non-significance (, Model 3). Supplemental analyses revealed that when income was omitted from the model, the education effect was strong and significant across the models. This finding suggests that education indirectly influences mortality through its strong association with income.
Income remained a strong and significant predictor of 19-year mortality after controlling for health risk behaviors (, Models 1 and 2), although there was a slight attenuation in the HRR for lower income groups when controlling for risk behaviors. Controlling for demographic characteristics and the 4 risk behaviors, the mortality HRR was 1.76 (95% C.I.=1.28-2.41) for the lowest income group and 1.42 (95% C.I.= 1.05-1.92) for the middle income group (, Model 2). The mortality disadvantage for the lower income group remained significant in a model controlling for sociodemographics, health risk behaviors and health status (, Model 3). These results suggest that even though there is indeed a higher prevalence of major health risk behaviors among people in the lowest income group, this does not account for the majority of the relationship between income and mortality.
Many of the health risk factors under study were predictive of mortality in the ACL sample. Those subjects who were current or former smokers, non-drinkers, severely underweight (BMI<18.5), and in the lowest quintile for physical activity were all at a significantly higher risk of mortality during the 19-year study period (, Model 2). These relationships all remained significant when baseline health status was added to the model (, Model 3).
Those categorized as obese (BMI=30+) did not have an elevated mortality risk; in fact their risk was significantly lower than those with normal weight in a model controlling for other health risk behaviors and health status (HRR=0.79; 95% C.I.=0.64-0.97). Analyses (not shown) focusing on those with morbid obesity (BMI > 35) also did not reveal a significantly elevated mortality rate, although the sample size for this group was small (baseline N=202, average age=48.4 years). Physical activity, however, was associated with lower mortality risk, with those in the lowest activity group (i.e., those with a sedentary lifestyle) having a significantly elevated rate of mortality (HRR=1.58; 95% 1.20-2.07) controlling for other health risk behaviors and health status (, Model 3).
Analyses stratifying the ACL sample into those <55 and those 55+ years at baseline revealed similarities to and differences from the results for the full sample (). The female mortality advantage was significant in the two age groups; and education was not significantly related to mortality in either group.
Mortality Hazard Rate Ratios (95% Confidence Intervals) for Americans’ Changing Lives Sample Stratified by Age, 1986-2005
Notable differences between the two age groups included results related to income, smoking and BMI. In regard to income, an elevated risk of mortality among those with low incomes was only observed for those age 55+ at baseline when controlling for health risk behaviors and health status (, Models 1b and 2b). The mortality risks for current smoker were larger among those in the young age group (<55 at baseline). This could be the result of a cohort effect, or because competing mortality risks for smokers increase as they age.
Obesity (BMI=30+) was not predictive of mortality in either age group (). However, being overweight or obese was significantly protective against mortality for those ages 55 and older in 1986. Specifically, those 55 and older who were obese had a 27% lower rate of mortality in a model controlling for other health risk behaviors (, Model 2a) and a 32% lower rate of mortality when controlling for health risk behaviors and health status (, Model 2b). In addition, those respondents age 55+ who were overweight (but not obese) also experienced a reduced risk of dying. These results suggest that being overweight or obese conferred a protection or benefit in terms of mortality for those who were age 55 and older in 1986, when considering sociodemographic characteristics, health behaviors and health status indicators simultaneously.
In the older age group, those in the lowest quintile of physical activity had a significantly elevated risk of mortality (HRR=1.88, 95% C.I.=1.41-2.50; , Model 2b). These results demonstrated that, for those Americans age 55 and older, a lack of physical activity was significantly associated with increased the risk of mortality over the next 19 years while obesity itself was not a risk factor. These results also suggest that any amount of physical activity, relative to the most sedentary group, had a positive benefit in terms of mortality.
BMI and Physical Activity
We conducted additional analyses to investigate the impact of BMI and physical activity on mortality without the other variable in the model, since these two variables are correlated. The results (not shown) for physical activity were robust to whether or not a control for BMI (measured both categorically and continuously) was in the model. For BMI, the HRR for obesity in a full model including physical activity was 0.79 (95% CI=0.64-0.97), although this protective effect attenuated to insignificance when physical activity was excluded from the model (HRR=0.88, 95% CI=0.70-1.10). Thus, for the full sample, the significant protective effect of obesity is not observed in a model excluding physical activity. This is because the relationship between obesity on mortality is upwardly biased when physical activity is not included in the model.
In addition, age-stratified analyses reveal similar effects of BMI with and without physical activity in the model for study subjects who were age 55 or older at baseline. As shown in (models 2a and 2b), being obese was significantly protective against mortality for those in the older subsample both with physical activity included in (HRR=0.68, 95% CI=0.55-0.84) and excluded from the analysis (HRR=0.79, 95% CI=0.64-0.98). The results of all of these analyses suggest that the effects of overweight/obesity are upwardly biased when physical activity is not considered simultaneously.
Mortality Hazard Rate Ratios (95% Confidence Intervals) for Body Weight and Physical Activity, Americans’ Changing Lives Sample Stratified by Age, 1986-2005
Other Sensitivity Analysis
We conducted additional analysis to make sure that our results were not sensitive to how categorical independent variables were operationalized. This included: a) using 6 categories for education and income; b) including a control for number of people dependent upon the household income; c) using a variety of different cut-points for alcohol consumption, including different cut-points for “heavy drinking” and for males versus females; d) using different categorizations of race. In summary, the results reported above are robust to different measurement approaches for the independent variables (myriad results not shown). The main results and conclusions hold across different specifications without exception.