Body Mass Trajectories and Mortality Among Older Adults
Body Mass Trajectories and Mortality Among Older Adults
Table 1 summarizes key characteristics of the sample. At baseline, the sample consisted of about 48% men and 52% women; the respondents were about 56 years old on average. The initial average BMI was about 27 for both men and women, increasing slightly to a mean BMI of 28 among survivors at the end of the study. Few respondents were underweight (0.4% of men and 1.7% of women) at baseline and only 21% of men and about 24% of women were obese; most had BMIs in the overweight or normal weight categories. About 23% of men and 15% of women died during the 16 years of follow-up.
Figure 2 shows the estimated sample trajectories from a 3-class model for men (solid lines) and a random sample of observed BMI trajectories (dotted lines). The plots visually corroborate that the linear specification fit the data well: The direction of change and linearity of the individual observed trajectories follow the estimated trajectories neatly although there is variability in the overall BMI level in each class. Results for women were similar to those for men. (The figure was omitted for parsimony but is available on request.)
(Enlarge Image)
Figure 2.
Three classes of BMI trajectories for men–estimated trajectory for each class and a random sample of observed individual trajectories. Panel A: Stable overweight class (92.9%); Panel B: Obese gaining class (2.8%); Panel C: Obese losing class (4.3%). Note: Results for women are visually nearly indistinguishable. The estimated sample trajectories and sample proportions for both genders are summarized in Figure 3.
Figure 3 summarizes the estimated BMI trajectories from the 3-class GMM for men (Panel A) and women (Panel B). Although these models were estimated independently, the resulting trajectories and the distributions of the samples into each class were remarkably similar for the two genders. The majority of respondents, 93% of men and almost 90% of women, had a BMI trajectory that could be characterized as being stable overweight. This group experienced only a slight increase in BMI over the duration of follow-up, and their initial BMI was distributed around the low-overweight range. The remaining 7%–10% of the sample had BMIs primarily in the obese range. About 3% of men and 6% of women had a BMI trajectory described as obese gaining, which was characterized by a steady weight gain over time. The mean BMI in this class increased from just above 30 for men or 35 for women to BMIs well above 40. The last group, obese losing, comprised 4%–5% of the sample. This group was characterized by decreases of body weight, from BMIs centered around the 35–40 range to the high-overweight range among men and low-obese range among women.
(Enlarge Image)
Figure 3.
Estimated BMI trajectories from unadjusted growth mixture models, by sex. Panel A: Men; Panel B. Women. Note: The plot shows the estimated and observed sample trajectories and the proportion of the total sample for each class estimated by the model. Data source: Health and Retirement Survey 1992–2008 (N = 4,579 men and N = 5,124 women).
We then used the joint GMM-DTSA model to estimate mortality differences across these three trajectory classes. Table 2 shows the odds ratios for the risk of dying, and associated p values, by gender. Relative to the stable overweight group, men and women in the obese gaining groups had about a 50% higher odds of dying—although among men, this difference was not statistically significant (for women, OR = 1.54, p < .001). Men and women in the obese losing groups were about 3–4 times more likely to die during follow-up, compared with those in the reference stable overweight class (for men, OR = 2.8, p < .001; for women, OR = 3.7, p < .001).
Finally, we estimated growth mixture models adjusted for several baseline characteristics—year of birth, dichotomized SRH, and smoking status—in order to gain insight into the reasons for the high mortality of the obese gaining and especially obese losing groups. Table 3 shows the results for men and women. In both genders, poor/fair baseline health was strongly associated with membership in the obese gaining groups relative to the stable overweight group (for men, OR = 2.4, p < .001; for women, OR = 3.2, p < .001). The effects were even stronger for the obese losing groups: men in this group had 4.6 times the odds of reporting fair/poor health (p < .001) compared with the stable overweight group; for women, the odds were 6.5 (p < .001). In terms of smoking, men and women in the obese losing class were less likely to be current smokers than those in the stable overweight class (for men, OR = 0.6, p = .03; for women, OR = 0.4, p = .002). In addition, women in the obese gaining class were more likely to be former smokers than those in the stable overweight class (OR = 1.7, p = .002).
Results
Table 1 summarizes key characteristics of the sample. At baseline, the sample consisted of about 48% men and 52% women; the respondents were about 56 years old on average. The initial average BMI was about 27 for both men and women, increasing slightly to a mean BMI of 28 among survivors at the end of the study. Few respondents were underweight (0.4% of men and 1.7% of women) at baseline and only 21% of men and about 24% of women were obese; most had BMIs in the overweight or normal weight categories. About 23% of men and 15% of women died during the 16 years of follow-up.
Figure 2 shows the estimated sample trajectories from a 3-class model for men (solid lines) and a random sample of observed BMI trajectories (dotted lines). The plots visually corroborate that the linear specification fit the data well: The direction of change and linearity of the individual observed trajectories follow the estimated trajectories neatly although there is variability in the overall BMI level in each class. Results for women were similar to those for men. (The figure was omitted for parsimony but is available on request.)
(Enlarge Image)
Figure 2.
Three classes of BMI trajectories for men–estimated trajectory for each class and a random sample of observed individual trajectories. Panel A: Stable overweight class (92.9%); Panel B: Obese gaining class (2.8%); Panel C: Obese losing class (4.3%). Note: Results for women are visually nearly indistinguishable. The estimated sample trajectories and sample proportions for both genders are summarized in Figure 3.
Figure 3 summarizes the estimated BMI trajectories from the 3-class GMM for men (Panel A) and women (Panel B). Although these models were estimated independently, the resulting trajectories and the distributions of the samples into each class were remarkably similar for the two genders. The majority of respondents, 93% of men and almost 90% of women, had a BMI trajectory that could be characterized as being stable overweight. This group experienced only a slight increase in BMI over the duration of follow-up, and their initial BMI was distributed around the low-overweight range. The remaining 7%–10% of the sample had BMIs primarily in the obese range. About 3% of men and 6% of women had a BMI trajectory described as obese gaining, which was characterized by a steady weight gain over time. The mean BMI in this class increased from just above 30 for men or 35 for women to BMIs well above 40. The last group, obese losing, comprised 4%–5% of the sample. This group was characterized by decreases of body weight, from BMIs centered around the 35–40 range to the high-overweight range among men and low-obese range among women.
(Enlarge Image)
Figure 3.
Estimated BMI trajectories from unadjusted growth mixture models, by sex. Panel A: Men; Panel B. Women. Note: The plot shows the estimated and observed sample trajectories and the proportion of the total sample for each class estimated by the model. Data source: Health and Retirement Survey 1992–2008 (N = 4,579 men and N = 5,124 women).
We then used the joint GMM-DTSA model to estimate mortality differences across these three trajectory classes. Table 2 shows the odds ratios for the risk of dying, and associated p values, by gender. Relative to the stable overweight group, men and women in the obese gaining groups had about a 50% higher odds of dying—although among men, this difference was not statistically significant (for women, OR = 1.54, p < .001). Men and women in the obese losing groups were about 3–4 times more likely to die during follow-up, compared with those in the reference stable overweight class (for men, OR = 2.8, p < .001; for women, OR = 3.7, p < .001).
Finally, we estimated growth mixture models adjusted for several baseline characteristics—year of birth, dichotomized SRH, and smoking status—in order to gain insight into the reasons for the high mortality of the obese gaining and especially obese losing groups. Table 3 shows the results for men and women. In both genders, poor/fair baseline health was strongly associated with membership in the obese gaining groups relative to the stable overweight group (for men, OR = 2.4, p < .001; for women, OR = 3.2, p < .001). The effects were even stronger for the obese losing groups: men in this group had 4.6 times the odds of reporting fair/poor health (p < .001) compared with the stable overweight group; for women, the odds were 6.5 (p < .001). In terms of smoking, men and women in the obese losing class were less likely to be current smokers than those in the stable overweight class (for men, OR = 0.6, p = .03; for women, OR = 0.4, p = .002). In addition, women in the obese gaining class were more likely to be former smokers than those in the stable overweight class (OR = 1.7, p = .002).