Effect of Age on the Association between Body Mass Index and All-Cause Mortality: The Ohsaki Cohort Study

Background To clarify the effect of age on the association between body mass index (BMI) and all-cause mortality. Methods We followed 43 972 Japanese participants aged 40 to 79 years for 12 years. Cox proportional hazards regression analysis was used to estimate hazard ratios (HRs), using the following BMI categories: <18.5 (underweight), 18.5–20.9, 21.0–22.9, 23.0–24.9 (reference), 25.0–27.4, 27.5–29.9, and ≥30.0 kg/m2 (obese). Analyses were stratified by age group: middle-aged (40–64 years) vs elderly (65–79 years). Results We observed a significantly increased risk of mortality in underweight elderly men: the multivariate HR was 1.26 (0.92–1.73) in middle-aged men and 1.49 (1.26–1.76) in elderly men. In addition, we observed a significantly increased risk of mortality in obese middle-aged men: the multivariate HR was 1.71 (1.17–2.50) in middle-aged men and 1.25 (0.87–1.80) in elderly men. In women, there was an increased risk of mortality irrespective of age group in the underweight: the multivariate HR was 1.46 (0.96–2.22) in middle-aged women and 1.47 (1.19–1.82) in elderly women. There was no excess risk of mortality with age in obese women: the multivariate HR was 1.47 (0.94–2.27) in middle-aged women and 1.26 (0.95–1.68) in elderly women. Conclusions As compared with the reference category, obesity was associated with a high mortality risk in middle-aged men, whereas underweight, rather than obesity, was associated with a high mortality risk in elderly men. In women, obesity was associated with a high mortality risk during middle age; underweight was associated with a high mortality risk irrespective of age. The mortality risk due to underweight and obesity may be related to sex and age.

To further examine the effect of age on the association between BMI and all-cause mortality, we conducted a cohort study among middle-aged and elderly Japanese who were recruited from the general population. We obtained information about their medical history, smoking status, and other possible confounders. In addition, our study overcomes problems in previous studies because we adjusted for several confounders after excluding participants with subclinical disease. We believe that by clarifying the effect of age on the association between BMI and all-cause mortality, it might be possible to improve public health measures by targeting body weight control according to life stage.

Study cohort
The details of the Ohsaki National Health Insurance (NHI) Cohort Study have been described previously. 20 We excluded 776 participants who withdrew from the NHI before 1 January 1995, when we started prospective collection of data on NHI withdrawals. Thus, the study cohort comprised the remaining 51 253 participants. The study protocol was approved by the Ethics Committee of Tohoku University School of Medicine. We considered the return of the selfadministered questionnaires signed by the participants to imply their consent to participate in the study.
For the current analysis, we also excluded 1767 participants with a history of cancer, 1384 participants with a history of myocardial infarction, and 997 participants with a history of stroke, because the presence of these diseases at baseline could have affected their BMI. In addition, we excluded 3133 participants who did not provide information about body weight or height. As a result, a total of 43 972 adults (21 038 men and 22 934 women) participated. After 12 years of follow-up, there were 5707 deaths (3685 men and 2022 women).

Body mass index
The self-administered questionnaire included questions on weight and height. BMI was calculated as weight in kilograms divided by the square of height in meters (kg/m 2 ). We used BMI as a measure of total adiposity and divided the participants into groups according to the following BMI categories: <18.5 (underweight), 18.5-20.9, 21.0-22.9, 23.0-24.9, 25.0-27.4, 27.5-29.9, and ≥30.0 kg/m 2 (obese). These weight categories correspond to the cut-off points proposed by the World Health Organization (WHO), ie, normal BMI range (18.5-24.9 kg/m 2 ), grade 1 overweight (25.0-29.9 kg/m 2 ), grade 2 overweight (30.0-39.9 kg/m 2 ), and grade 3 overweight (≥40.0 kg/m 2 ). 23 We previously evaluated the validity of self-reported weight and height. 22 Briefly, the weight and height of 14 883 participants, who were a subsample of the cohort, were measured during health examinations in 1995. The Pearson correlation coefficient (r) and weighted kappa (κ) for the selfreported values and measured values were r = 0.96 (P < 0.01) for weight, r = 0.93 (P < 0.01) for height, and r = 0.88 (P < 0.01) and κ = 0.72 for BMI. Thus, the self-reported heights and weights in the baseline questionnaire were considered sufficiently valid.

Follow-up
We followed the participants from 1 January 1995 through 31 December 2006 and recorded any mortality or migration by reviewing data on NHI withdrawals. When a participant withdrew from the NHI system because of death, emigration, or employment, the date of and reason for withdrawal were coded in the NHI withdrawal history files. Because we were unable to obtain subsequent information on participants who withdrew from the NHI because of emigration or employment, we discontinued follow-up of these participants.
The end point was all-cause mortality. Data on the death of participants were based on the death certificates filed at Ohsaki Public Health Center.
The person-years of follow-up were counted for each participant, until either the date of death, withdrawal from the NHI, or the end of the study period, whichever occurred first. The total number of person-years accrued was 440 175.

Statistical analysis
We used Cox proportional hazards regression analysis to calculate the hazard ratios (HRs) and 95% confidence intervals (CIs) for all-cause mortality according to BMI category, and to adjust for potential confounding factors, using the SAS version 9.1 statistical software package. 24 To enable detailed examination of the association of BMI and all-cause mortality by WHO categories, the normal weight and overweight categories were divided into 3 and 2 categories, respectively. The BMI category 23.0-24.9 kg/m 2 was selected as the reference because it is the median of the 7 categories.
Stratified analyses were conducted using 2 age groups: middle-aged participants (40-64 years) and elderly participants (65-79 years). The classification of elderly participants was based on a report by the WHO. 25 All P values were 2-tailed, and a P value of <0.05 was considered statistically significant.
The following variables were selected as potential confounding factors: 5-year age group, weight change since age 20 years (loss of ≥10.0 kg, loss of 5.0-9.9 kg, change of less than 5.0 kg, gain of 5.0-9.9 kg, or gain of ≥10.0 kg), education (junior high school or less, high school, or college/ university or higher), marital status (married or unmarried), cigarette smoking (never smoker, past smoker, current smoker consuming 1-19 cigarettes per day, or current smoker consuming at least 20 cigarettes per day), alcohol consumption (never drinker, past drinker, or current drinker), time spent walking per day (less than 1 hour or 1 hour or longer), sports and physical exercise time per week (less than 1 hour, 1-2 hours, 3-4 hours, or 5 hours or longer), history of kidney disease (yes or no), and history of liver disease (yes or no). We further adjusted for hypertension and diabetes mellitus in multivariate model 2. Before including the above potential confounders into the multivariate models, we examined interactions between all-cause mortality and all potential confounders through the addition of cross-product terms to the multivariate model. Based on the results of these analyses (data not shown), we included all the above variables into the multivariate models. In addition, we repeated the analyses after excluding the 739 participants who died within 2 years of baseline.

Baseline characteristics by BMI category
The baseline characteristics of the study participants according to the 7 BMI categories are shown for middleaged men (Table 1), elderly men (Table 2), middle-aged women (Table 3), and elderly women (Table 4). Among middle-aged men and women, 2.3% and 2.9%, respectively, were underweight, about 50% of each had a BMI from 21.0 to 24.9 kg/m 2 ; 25.7% and 28.5% had a BMI from 25.0 to 29.9 kg/m 2 , and 2.3% and 3.4% were obese, respectively. Among elderly men and women, 5.8% and 5.9%, respectively, were underweight, about half of each had a BMI from 21.0 to 24.9 kg/m 2 ; 19.2% and 27.9% had a BMI from 25.0 to 29.9 kg/m 2 , and 1.4% and 4.0% were obese, respectively.
In men, mean age decreased linearly with an increase in BMI category. In women, middle-aged women with a BMI from 25.0 to 27.4 kg/m 2 and elderly women who were underweight were oldest. The proportions of men and women who had lost ≥5 kg of body weight since age 20 years decreased with increasing BMI category. Participants with the highest level of education were middle-aged men with a BMI from 25.0 to 27.4 kg/m 2 , middle-aged women with a BMI from 18.5 to 20.9 kg/m 2 , and underweight elderly men and women. The proportions of unmarried men and women were higher among those who were underweight and obese. The proportions of men and women who were current smokers decreased with increasing BMI. The proportions of men and women who had never drunk alcohol were highest in the underweight, with the exception of middle-aged women. Underweight and obese men and women were less likely to walk 1 hour or longer per day and to participate in <1 hour of sports or physical exercise per week. The proportions of men and women who had histories of hypertension and diabetes increased with an increase in BMI category. The proportions of middle-aged men and elderly women who had histories of kidney disease and liver disease did not significantly differ across BMI categories. The proportions of participants with histories of liver disease and kidney disease were highest among elderly obese men and underweight middle-aged women, respectively.
All-cause mortality by BMI category

DISCUSSION
The present results indicate that the mortality risk associated with underweight and obesity might be dependent upon sex and age group. We noted significant increased risks of mortality only in middle-aged obese men and elderly underweight men. In women, there was no significant excess risk of mortality with age in the obese, and no significant increased risk of mortality, irrespective of age group, in the underweight. We considered several important confounding factors: cigarette smoking, alcohol consumption, and physical activity are major confounding factors associated with both BMI and mortality. [1][2][3][4][5][6][7][8][9][10][11][12][13][14][15]17,18 We also considered education level and marital status as potential confounding factors, as in past studies. 1,3,4,6,7,9,11,17 Furthermore, the presence of subclinical disease or a history of illness could induce weight loss and increase the risk of death. [1][2][3][4]8,9,11,14,15,17,18 To eliminate any effect of medical history, we excluded participants with a history of cancer, myocardial infarction, or stroke, and adjusted for weight change since age 20 years, history of kidney disease, and history of liver disease, in multivariate analysis.
Multivariate adjustment attenuated the HR estimates associated with a BMI of 27.5-29.9 or ≥30.0 kg/m 2 in women, but not in men. No single covariate resulted in significant attenuation, although an increase in body weight of 5 kg or more since age 20 years, current drinking, and ≥1 hour physical activity per week attenuated hazard ratios. In contrast, a decrease in body weight of 5 kg or less since age 20 years, past drinking, being unmarried, <1 hour spent walking per day, and histories of kidney disease and liver disease significantly increased HRs in men. Almost all previous studies agree that the excess risk of mortality due to obesity decreases with age, [1][2][3][4][5][6][7][8][9][10][11][12][13][14]17,18 and our results accord with this. In underweight adults, the results of past studies have been inconsistent. [1][2][3][4][5][6][7][8][9][10][11][12][13][14][15] Our results are in agreement with 2 of 14 studies of men, 5,13 and 4 of 13 studies of women. 3,4,10,14 In Japan, Matsuo et al reported the effect of age on the association between BMI and all-cause mortality. 10 Their findings agree with ours, except for underweight men. They adjusted only for age, alcohol intake, and smoking status in multivariate analysis; however, physical activity and socioeconomic status have also been identified as confounding factors for the risk of all-cause mortality. [1][2][3][4][5][6][7][8][9][10][11][12][13][14][15]17,18 Although their result differ from ours for underweight men, our study was more careful in adjusting for physical activity, socioeconomic status, weight change since age 20 years, marital status, and histories of kidney disease and liver disease.  Development of measures to address underweight has been slower than for obesity. However, Grabowski et al and Sergi et al showed that a low BMI in elderly adults was a predictor of mortality. 26,27 Okoro et al found that underweight was associated with subsequent disability in elderly adults. 28 Our study also found that underweight is associated with a high mortality risk in elderly men and women, irrespective of age group.
A major strength of the present study was that the participants were recruited from the general Japanese population. According to the Global Database on Body Mass Index of the WHO, the prevalence of underweight participants is higher in Japan (10%-20%) than in Western populations (0%-5%). Therefore, the Japanese population is one of the best in which to examine the excess risk of mortality due to underweight.
Several limitations of our study should be considered. First, although BMI has been accepted as satisfactory index of underweight and obesity, it cannot be used to identify distributions of fat and muscle tissue. Second, we used self- reported BMI at baseline. Niedhammer et al showed that there is a systematic bias in self-reported weight and height. 29 However, we previously evaluated the validity of selfreported BMI, and demonstrated a high correlation and appropriate agreement between self-reported BMI and measured BMI in a subsample of 14 883 participants (r = 0.88, κ = 0.72). We consider this bias to be a nondifferential misclassification that is not dependent upon all-cause death. This misclassification weakens the true association toward the null. Third, as a result of stratification by age group, there was a possibility of beta error because of inadequate numbers of participants and events. Finally, there is a possibility of residual confounding by physical activity. In summary, obesity increases mortality risk in middle-aged men, whereas underweight, rather than obesity, is associated with high mortality risk in elderly men. In women, obesity increases mortality risk in middle age, and underweight increases mortality risk irrespective of age. Although there was no significant interaction by age group or sex, the mortality risks associated with underweight and obesity may nevertheless be dependent on sex and age group.