Pauline Koh-Banerjee, Nain-Feng Chu, Donna Spiegelman, Bernard Rosner, Graham Colditz, Walter Willett, and Eric Rimm

Transcription

1 Prospective study of the association of changes in dietary intake, physical activity, alcohol consumption, and smoking with 9-y gain in waist circumference among US men 1 4 Pauline Koh-Banerjee, Nain-Feng Chu, Donna Spiegelman, Bernard Rosner, Graham Colditz, Walter Willett, and Eric Rimm ABSTRACT Background: Although it is known that abdominal obesity increases the risk of chronic diseases, prospective data examining the relation between lifestyle factors and the accumulation of abdominal adipose tissue are sparse. Objective: The objective of the study was to determine the associations of changes in diet, physical activity, alcohol consumption, and smoking with 9-y waist gain among US men. Design: A prospective cohort comprised US men aged y at baseline in Data on lifestyle factors were provided periodically with the use of self-reported questionnaires, and participants measured and reported their waist circumference in 1987 and Results: In multivariate analyses, a 2% increment in energy intake from trans fats that were isocalorically substituted for either polyunsaturated fats or carbohydrates was significantly associated with a 0.77-cm waist gain over 9 y (P < for each comparison). An increase of 12 g total fiber/d was associated with a 0.63-cm decrease in waist circumference (P < 0.001), whereas smoking cessation and a 20-h/wk increase in television watching were associated with a 1.98-cm and 0.59-cm waist gain, respectively (P < 0.001). Increases of 25 metabolic equivalent tasks (METs) h/wk in vigorous physical activity and of 0.5 h/wk in weight training were associated with 0.38-cm and 0.91-cm decreases in waist circumference, respectively (P < for each comparison). These associations remained significant after further adjustment for concurrent change in body mass index. Changes in total fat and alcohol consumption and in walking volume were not significantly related to waist gain. Conclusions: Waist gain may be modulated by changes in trans fat and fiber consumption, smoking cessation, and physical activity. Am J Clin Nutr 2003;78: KEY WORDS Dietary fat, trans fats, fiber, physical activity, smoking, waist gain, obesity INTRODUCTION Android obesity is characterized by the localization of body fat in the upper truncal region. This phenotype is more common among males, in contrast with the gynoid phenotype that is more common among females, who have the tendency for fat to accumulate in the hips and thighs. Android obesity is associated with an atherogenic profile (1, 2) and is a risk factor for the development of type 2 diabetes, stroke, coronary heart disease, and total mortality, independent of and additive to total obesity (3 8). Whereas the rising prevalence of obesity in the past few years has been attributed to changes in lifestyle associated with increasing modernization, few studies have prospectively examined the relation between lifestyle factors and the accumulation of abdominal fat. The modifiable factors that were associated with changes in android obesity include generalized obesity (9 11), physical activity (12, 13), and cigarette smoking (14). The effect of diet on central fat stores is more controversial, and the effects of macronutrient composition, alcohol consumption, and dietary fiber are not clearly established. In addition, the findings from various observational and intervention studies are difficult to compare because various anthropometric indicators of abdominal obesity were used. Whereas the waist-to-hip ratio is often used as an indicator of abdominal fat mass, this ratio is difficult to interpret biologically because the waist and hip circumference measures are reflective of different anatomical entities (15). The waist circumference measures both visceral and subcutaneous fat, whereas the hip circumference includes fat mass, lean muscle mass, and skeletal frame (15). Furthermore, the waist circumference contributes less error than does the waist-to-hip ratio because the former is a single measurement (8) and has been independently associated with increased serum indexes of cardiovascular disease risk (16, 17) and with an increased risk of diabetes (8), cardiovascular disease (6), and total mortality (18, 19). Therefore, we prospectively 1 From the Departments of Nutrition (PK-B, WW, and ER), Epidemiology (DS, GC, WW, and ER), and Biostatistics (DS and BR), Harvard School of Public Health, Boston; the Departments of Public Health and Community Medicine, Tri-Service General Hospital, National Defense Medical Center, Taipei, Taiwan, Republic of China (N-FC); and the Channing Laboratory, Department of Medicine, Brigham and Women s Hospital, and Harvard Medical School, Boston (BR, GC, WW, and ER). 2 Supported by research grants CA55075 and HL35464 from the NIH and grant DK46200 from the Boston Obesity and Nutrition Research Center. 3 Address reprint requests to E Rimm, Department of Nutrition, Harvard School of Public Health, 665 Huntington Avenue, Boston, MA erimm@hsph.harvard.edu. 4 Address correspondence to P Koh-Banerjee, Department of Preventive Medicine, University of Tennessee Health Science Center, 66 North Pauline, Suite 633, Memphis, TN Received February 6, Accepted for publication April 25, Am J Clin Nutr 2003;78: Printed in USA American Society for Clinical Nutrition 719

2 720 KOH-BANERJEE ET AL examined the association of changes in dietary intake, physical activity, alcohol consumption, and smoking with a 9-y gain in waist circumference among a cohort of men. SUBJECTS AND METHODS Study population The Health Professionals Follow-up Study is a prospective investigation of male health professionals aged y at baseline in This cohort includes dentists, veterinarians, 4185 pharmacists, 3745 optometrists, 2218 osteopathic physicians, and 1600 podiatrists. In 1986 participants completed a detailed questionnaire regarding medical history, diet, and physical activity. The participants self-reported their age, current height (in inches), weight (in pounds), current smoking and smoking history, marital status, and family history of coronary heart disease and cancer. On a biennial basis thereafter, participants were followed with mailed questionnaires on which they provided updated information on exposures and on any diseases diagnosed since the last questionnaire. Body mass index (BMI; in kg/m 2 ) at each follow-up cycle was calculated with the use of each subject s self-reported weight and height. In addition, in a separate mailing in1987 and along with the biennial questionnaire in 1996, we sent the men a tape measure to assist them in selfreporting their waist and hip circumferences. Men were asked to take measurements while standing and to avoid measuring over bulky clothing. They were instructed to take their waist measurement at the umbilicus and to take their hip measurement at the largest circumference between the waist and thighs; illustrations were included with the directions. Because the 1987 questionnaire was not part of the usual biennial mailings, we did not use our typical extensive follow-up procedures, and thus our followup rate was 65% (6). We excluded from the analysis men who either died (n = 1751) or developed cardiovascular disease, cancer, or diabetes (n = ) before 1996 because the development of those diseases may alter weight and waist measures, dietary intake, and physical activity level. Furthermore, we excluded men who failed to report waist circumference measures, body weights, or dietary data. Our analysis is therefore based on healthy men for whom we have a complete set of predictor and outcome information for the study period of 1986 to The Institutional Review Board of the Harvard School of Public Health approved the protocol for this study. Outcome assessment We evaluated the reproducibility and validity of the selfreported measures of waist circumference and weight by comparing them with technician-assessed measurements taken 6 mo apart in a subset of the cohort participants (20). The self-reported measurements and the average of 2 technician measurements were highly correlated (weight: r = 0.97; waist circumference: r = 0.95). Furthermore, there were no significant linear trends in accuracy of reported waist circumference across quartiles of either age or BMI (20). The validity of self-reported waist measurements was further examined by using Bland-Altman plots with ANALYSE-IT for Excel software (version 1.67; Analyse-It Software, Leeds, United Kingdom; 21). The differences between the self-reported measurements and the average of the 2 technician measurements were normally distributed, and the degree of bias was 0.14 cm in (95% FIGURE 1. Difference between self-reported and technician-assessed waist measures plotted against the mean for all methods (in cm). CI: 0.40, 0.69). Bias did not appear to increase or decrease with the underlying true value. On the basis of these observations and of the observation that 2.3% of the data fell outside the 95% limits of agreement ( 6.02 to 6.30 cm), the two methods were deemed interchangeable (Figure 1). The validity of self-reported height was not evaluated because it was previously reported as highly valid (22). Variables of interest Detailed dietary information was obtained in 1986, 1990, and 1994 through the use of a semi-quantitative food-frequency questionnaire (FFQ) developed by Willett et al (23, 24). The FFQ is used to assess typical food intake over the previous year and currently includes 131 items. For each food, a commonly used unit or portion size is specified, and the participant is asked how often, on average, he had consumed that amount during the previous year. Nine responses are possible, ranging from never to 6 times/d. Alcohol consumption per day was calculated as the sum of the amount of alcohol in any wine, beer, and liquor consumed, multiplied by the average number of servings per day. The amount of alcohol contained within each beverage is 12.8 g/355-ml (12-oz) can or bottle of beer, 11 g/118-ml (4-oz) glass of wine, and 14 g/44-ml (1.5-oz) shot of liquor (25). The FFQ was validated among a subset of the study participants (23). In the validation study, 2 FFQs were administered 1 y apart, with two 1-wk diet records administered 6 mo apart during this 1-y period. The Pearson correlation coefficients between the FFQs and the average of the two 1-wk dietary records, after correction for variation in the 1-wk dietary records, ranged from 0.44 for protein to 0.92 for vitamin C (23). The correlation for total alcohol was 0.86 and that for alcohol-containing beverages ranged from 0.70 for wine to 0.86 for liquor (26). The level of physical activity was ascertained in 1986 and biennially thereafter. Participants were queried regarding the average time spent per week over the past year in specific activities including walking or hiking outdoors, jogging [< 9.6 km/h (< 6 mph)], running [ 9.6 km/h ( 6 mph)], bicycling, swimming, tennis, squash, racquetball, rowing, and calisthenics. After 1986, questions were added regarding the average time spent per week in heavy outdoor work, weight training, and television watching. Walking pace, categorized as casual [ 3.2 km/h ( 2 mph)], normal [ km/h (2 2.9 mph)], brisk [ km/h (3 3.9 mph)], or striding ( 6.4 km/h ( 4 mph)], was also recorded. The time spent at each

3 PREDICTORS OF WAIST GAIN 721 activity in hours per week was multiplied by its typical energy expenditure, expressed in metabolic equivalent tasks (METs), then summed over all activities, to yield a MET hour score (27). One MET, the energy expended by sitting quietly, is equivalent to 3.5 ml of oxygen uptake per kilogram of body weight per minute. Vigorous activities were defined as those requiring 6 METs or more: jogging, running, bicycling, swimming, tennis, squash, racquetball, and rowing. The validity and reproducibility of the physical activity questionnaire were assessed in 1991 when 280 participants in the HPFS completed a 1-wk activity diary at 4 time periods corresponding to different seasons of the year (28). The correlations between scores of physical activity from the diaries and from the questionnaires were 0.65 for total physical activity, 0.28 for nonvigorous activity, and 0.58 for vigorous activity. The correlation between questionnaire-derived vigorous activity and resting pulse was 0.45; after a self-administered step test, the correlation of the same measurements was 0.41 (28). Statistical analysis The mean baseline characteristics for the different age groups were first compared by using analysis of variance with the generalized linear model procedure. We further performed post hoc multiple comparison tests by using the Tukey method to examine all pairwise comparisons at the overall experiment rate of P < Using multivariate linear regression, we examined how changes in lifestyle factors ( for dietary exposures, and for all other exposures) were associated with the nonrepeated dependent variable, namely the change in waist circumference (in cm) in the same period ( ). We used the robust variance estimate (29) to avoid needing assumptions of normality for the linear regression to obtain valid inference. In all analyses, there was one observation per participant. The exposures were modeled as differences between the baseline and most recent follow-up measure when tests for nonlinearity using spline regression were not statistically significant; otherwise, the exposures were categorized. We calculated the age-adjusted regression coefficients for each of the lifestyle factors and waist gain, and associations were estimated from the regression models as the change in waist circumference (in cm) over the 9-y period per unit of change in the lifestyle factor. To control for potential confounding, we adjusted the models for baseline age (continuous variable), baseline waist circumference (quartiles), baseline BMI (quartiles), baseline and changes in total calories (continuous variables), baseline and changes in alcohol consumption (continuous variables), baseline (continuous variable) and changes in (quintiles) total physical activity, and changes in smoking. Smoking was included as a categorical variable, and men were classified according to their change in smoking status between 1986 and Men who were nonsmokers at both time periods were classified as nonsmokers, whereas men who were smokers at both time periods were categorized as habitual smokers. Men who reported a change in their smoking status from smoking in 1986 to nonsmoking in 1996 were classified as quitters, and men who reported never smoking in 1986 and smoking in 1996 were classified as new smokers. All nutrient values (except those for alcohol) were energyadjusted by using the residual method to examine the nutrient composition of the diet rather than the effect of absolute intakes (24). To determine the relative effects of the dietary fat subtypes, the multivariate regression model simultaneously adjusted for intakes of energy, total fat, saturated fat, monounsaturated fat, and trans fatty acid (as percentages of energy). To examine the effect of isoenergetically replacing carbohydrates with each fat subtype, the percentages of energy derived from protein and all fat subtypes were simultaneously included. To identify the lifestyle factors that predicted increases in waist circumference independent of weight gain, we further adjusted for changes in BMI in this period. All statistical analyses were conducted with the use of SAS software (version 8.2; SAS Institute Inc, Cary, NC). To evaluate the influence of measurement error on the association between changes in exposures and waist gain, we used a subset of participants from a separate but similar study (30) for whom repeated dietary records and FFQs were available in both 1980 and Using the regression calibration approach (31) and results from our validation study together with data from our main study, we estimated regression coefficients adjusted for measurement error (see Appendix A). RESULTS Over the 9-y follow-up period, the mean (± SD) waist circumference increased 3.3 ± 6.2 cm, from 93.8 ± 8.5 cm in 1987 to 97.2 ± 9.9 cm in The greatest mean increase (3.9 ± 6.1 cm) was observed among the men aged y at baseline (Table 1). In contrast, the mean waist gain among men aged y was 3.2 ± 5.9 cm, and that among men aged 60 y was 2.3 ± 6.6 cm. At baseline, men aged y had the highest BMI (25.2 ± 2.8). However, the mean BMI change was greatest among men aged y (1.1 ± 1.7); the change among men aged y was 0.7 ± 1.5, and that among men aged 60 y was 0.1 ± 1.5. Over time, men of all age groups decreased their total fat consumption to 30.0 ± 6.8% of total caloric intake, increased their carbohydrate consumption to 50.8 ± 8.8% of total calories, and increased their consumption of dietary fiber to an overall average of 22.4 ± 7.4 g/d. The consumption of trans fats increased slightly over the follow-up period and was 1.3 ± 0.6% of total calories. Alcohol consumption remained fairly stable over time, at an overall average of 11.5 ± 14.9 g/d. Men aged y reported the least time spent watching television both at baseline and at follow-up ( 9.1 ± 7.7 h/wk in 1996). The younger men were more physically active at baseline than were the older men, but the older men increased their levels of vigorous physical activity over time. At follow-up in 1996, men aged y reported 16.8 ± 28.2 MET h/wk of total vigorous activity, whereas those aged y reported 14.0 ± 24.7 MET h/wk, and men aged 60 y reported 11.8 ± 21.6 MET h/wk. Multivariate models In age-adjusted analyses, a higher total fat intake was significantly related to waist gain (Table 2). This association was only modestly attenuated after adjustment for confounding by changes in other lifestyle factors including smoking, alcohol consumption, and physical activity. Because waist gain was only modestly correlated with changes in BMI (r = 0.43), we examined the effects of changes in behaviors on waist gain independent of changes in total obesity. The association between total fat and waist gain virtually disappeared after further adjustment for concurrent change in BMI, which suggested that increases in dietary fat did not alter abdominal adipose tissue independent of overall obesity. In contrast, greater intakes of trans fats were consistently

4 722 KOH-BANERJEE ET AL TABLE 1 Selected characteristics by age category for men in the Health Professionals Follow-Up Study 1 Age group in y old (n = 7577) y old (n = 5314) y old (n = 3696) Age, 1986 (y) 43.8 ± ± ± 3.7 Waist (cm) ± 8.5 3, ± ± ± , ± ± 9.6 Waist change, (cm) 3.9 ± 6.1 3,4 3.2 ± ± 6.6 BMI (kg/m 2 ) ± ± ± ± ± ± 2.9 BMI change, (kg/m 2 ) 1.1 ± 1.7 3,4 0.7 ± ± 1.5 Hypercholesterolemia, 1986 (%) ,4, Hypertension, 1986 (%) , Nonsmokers, 1996 (%) Habitual smokers, 1996 (%) New smokers, 1996 (%) Quitters, 1996 (%) Total energy (kcal/d) ± , ± ± ± ± ± Energy change, (kcal/d) 1.4 ± , ± ± Total fat (% of energy) ± ± ± ± ± ± 6.8 Total fat change, (% of energy) 2.0 ± ± ± 6.2 trans Fat (% of energy) ± 0.5 3,4 1.3 ± ± ± ± ± 0.6 trans Fat change, (% of energy) 0.1 ± ± ± 0.6 Carbohydrate (% of energy) ± ± ± ± ± ± 8.8 Carbohydrate change, (% of energy) 3.1 ± 7.9 3,4 3.8 ± ± 8.0 Total fiber (g/d) ± 6.4 3, ± ± ± 7.1 3, ± ± 7.6 Fiber change, (g/d) 1.4 ± 6.3 3,4 1.7 ± ± 7.1 Alcohol (g/d) ± , ± ± ± , ± ± 15.2 Alcohol change, (g/d) 0.1 ± ± ± 11.5 Television watching (h/wk) ± 7.5 3, ± ± ± 7.7 3, ± ± 9.3 Television watching change, (h/wk) 0.1 ± ± ± 9.6 Total activity (MET h/wk) ± , ± ± ± , ± ± 37.8 Activity change, (MET h/wk) 11.5 ± , ± ± 38.2 Walking (MET h/wk) ± 8.9 3,4 7.0 ± ± ± , ± ± 18.9 Walking change, (MET h/wk) 2.6 ± ,4 4.6 ± ± 18.4 Weight training (h/wk) ± 1.2 3,4 0.2 ± ± ± 1.2 3,4 0.3 ± ± 1.0 Weight training change, (h/wk) 0.1 ± ± ± 1.2 Vigorous activity (MET h/wk) ± , ± ± ± , ± ± 21.6 Vigorous activity change, (MET h/wk) 0.3 ± ,4 0.9 ± ± MET, metabolic equivalent task. 2 x ± SD. 3 Significantly different from group 2 (50 59 y old), overall rate of P < Significantly different from group 3 (60 75 y old), P < A professional diagnosis was self-reported on the questionnaire. 6 x.

5 PREDICTORS OF WAIST GAIN 723 TABLE 2 Estimated adjusted 9-y waist change among men in the Health Professionals Follow-Up Study per unit change in dietary factors 1 Age-adjusted Multivariate 2 Multivariate 3 Lifestyle factor Waist change P Waist change P Waist change P cm cm cm Increase in total fat intake by 5% of energy 0.30 ± 0.04 < ± 0.05 < ± Replacement of polyunsaturated fats (by 2% of energy) 4 With trans fats 0.74 ± ± 0.21 < ± With saturated fats 0.06 ± ± ± With monounsaturated fats 0.10 ± ± ± Replacement of carbohydrates (by 2% of energy) 5 With trans fats 1.21 ± 0.19 < ± 0.21 < ± With saturated fats 0.32 ± 0.04 < ± 0.07 < ± With monounsaturated fats 0.23 ± 0.04 < ± ± With polyunsaturated fats 0.16 ± ± ± Increase in total fiber intake by 12 g/d 0.64 ± 0.09 < ± 0.10 < ± x ± SE. 2 All multivariate models controlled for baseline age (continuous variable), baseline waist circumference (quartiles), baseline BMI (quartiles), baseline and change in total calories (continuous variables), change in smoking status (categorized as nonsmokers, habitual smokers, new smokers, and quitters), baseline (continuous variable) and change (quintiles) in total physical activity, and baseline and change in alcohol intake (continuous variables). 3 Additionally controlled for change in BMI (quartiles). 4 Model simultaneously controlled for baseline and changes in intakes of total fat, trans fat, saturated fat, and monounsaturated fat (as percentage of energy). 5 Model simultaneously adjusted for baseline and changes in intakes of protein and all fat subtypes (as percentage of energy). related to waist gain. A 2% increment in energy intake from trans fats, isocalorically substituted for either polyunsaturated fats or carbohydrates, was associated with a 0.77-cm waist gain in multivariate analyses (P < for each comparison). After adjustment for changes in BMI, the associations were only modestly reduced (0.52- or 0.53-cm waist gain, respectively; P < 0.01 for each comparison). Because common sources of trans fats such as commercial baked goods are also high in sugars, we simultaneously controlled for changes in glycemic load, a measure of both the quality and quantity of carbohydrates consumed, and the results did not differ appreciably. An increase in dietary fiber consumption of 12 g/d significantly predicted a reduction in waist circumference of 0.63 cm (P < 0.001), and this relation was only slightly reduced after adjustment for changes in BMI ( 0.23 cm; P < 0.01). Further adjustment for changes in all fat subtypes did not considerably alter the association (results not shown). Habitual smokers experienced a loss in waist circumference of 0.68 cm (P = 0.01), whereas those who quit smoking gained waist circumference (1.98 cm; P < 0.001) (Table 3). However, after we controlled for changes in BMI, only smoking cessation remained significantly related to waist circumference (0.77 cm; P < 0.01). No significant associations were observed between changes in total alcohol consumption and 9-y waist gain. An increase in total vigorous activity (by 25 MET h/wk) was significantly related to 9-y waist reduction of 0.38 cm (P < 0.001), and this relation remained significant (P < 0.05) after control for concurrent change in BMI ( 0.19 cm; Table 4). Men who added 0.5 h/wk of weight training experienced a loss in waist circumference of 0.91 cm over 9 y (P < 0.001); when BMI was held constant, waist circumference was reduced by 0.74 cm (P < 0.001). Because walking was the most common type of physical activity reported, we examined changes in this activity in relation to 9-y waist gain. To separate the effects of total walking volume from those of walking pace, we simultaneously controlled for both factors in multivariate analyses. Whereas no significant association between changes in walking volume and waist gain was observed, an increase in walking pace of 1.6 km/h ( 1 mph) was related to a loss in waist circumference of 0.50 cm (P = 0.002). When BMI was held constant, waist circumference was reduced by 0.27 cm (P = 0.05). Independent of changes in physical activity, an increase in television watching (by 20 h/wk) was significantly related to a 0.59-cm waist gain (P < 0.001). After adjustment for change in BMI, television watching was associated with a 0.30-cm waist gain (P = 0.02). Because the 1987 questionnaire was not part of the usual biennial mailings, we did not use extensive follow-up procedures to increase our follow-up rate (32). For this reason, men were excluded from the analysis for failure to report their baseline waist circumference. However, this group of men did not substantially differ in BMI, physical activity, or diet in 1986 (mean characteristics not shown) from the cohort used in the analysis. Additional analyses Also excluded from the population for analysis were men who developed cardiovascular disease, cancer, or diabetes before We chose to exclude these men to avoid biases related to changes in diet or physical activity patterns before those diagnoses. However, these men had greater waist circumferences at baseline, and it is possible that dietary fat or vigorous physical activity may differentially affect this population. To explore this possibility, we included these men and reanalyzed the data after we controlled for the development of disease. The results did not differ significantly from those presented above. Using the results of our validation study, we further adjusted the estimated coefficients for measurement error in the significant predictors. Although measures of trans fatty acids were not available from the diet records used in the validation study, we assumed that the validity of the FFQ in measuring trans fats was comparable to its validity in assessing other types of fats (total fat: r = 0.67; saturated fats, r = 0.75). After error correction, the substitution of trans fats as 2% of energy for polyunsaturated fats was associated with a 2.7-cm increase in waist circumference over 9 y (P < 0.001)

6 724 KOH-BANERJEE ET AL TABLE 3 Estimated adjusted 9-y waist change among men in the Health Professionals Follow-Up Study per unit change in smoking status and alcohol consumption 1 Age-adjusted Multivariate 2 Multivariate 3 Lifestyle factor Waist change P Waist change P Waist change P cm cm cm Change in smoking status Nonsmokers (reference group) Habitual smokers 0.40 ± ± ± New smokers 0.40 ± ± ± Quitters 1.95 ± 0.32 < ± 0.32 < ± Increase in alcohol consumption by 12 g/d 0.06 ± ± ± x ± SE. 2 All multivariate models controlled for baseline age (continuous variable), baseline waist circumference (quartiles), baseline BMI (quartiles), baseline and change in total calories (continuous variables), change in smoking status (categorized as nonsmokers, habitual smokers, new smokers, and quitters), baseline (continuous variable) and change (quintiles) in total physical activity, and baseline and change in alcohol intake (continuous variables). 3 Additionally controlled for change in BMI (quartiles). (as compared with a 0.77-cm waist gain, uncorrected). An increase of 12 g fiber/d (r = 0.68 between FFQ and diet records) was associated with a 2.21-cm reduction in waist circumference after error correction (P < 0.001) (0.63-cm waist gain, uncorrected). Although the change in physical activity was not assessed in our validation study, we assumed that the level of error in the assessment of the change in vigorous physical activity was comparable to that in the assessment of the change in dietary exposures. This seems reasonable, because, in our previous validation studies, the correlation for vigorous physical activity between the activity diaries and our questionnaire was 0.58 (28), which was of similar magnitude to correlations between the dietary records and the FFQs for total fat and saturated fat (r = 0.67 and 0.75, respectively) (23). After correction for measurement error, an increase of 25 MET h/wk of vigorous activity was associated with a 1.33-cm decrease in waist circumference (P < 0.001) (as compared with a 0.38-cm waist gain, uncorrected). DISCUSSION In this population of men, changes in several modifiable lifestyle factors were significantly associated with 9-y waist gain. The substitution of energy from trans fats for that from polyunsaturated fats or carbohydrates was associated with waist gain, whereas an increase in fiber was associated with a reduction in waist circumference. Smoking cessation and television watching were related to increases in central fat, whereas vigorous activity, weight training, and walking pace were negatively associated with waist gain. These lifestyle factors predicted 9-y waist gain even after control for concurrent change in BMI, which suggests that the accumula- TABLE 4 Estimated adjusted 9-y waist change among men in the Health Professionals Follow-Up Study per unit change in types of physical activity 1 Age-adjusted Multivariate 2 Multivariate 3 Lifestyle factor Waist change P Waist change P Waist change P cm cm cm Change in total activity No change (reference group) Decrease by 3 MET h/wk 0.55 ± 0.17 < ± ± Increase by 3 11 MET h/wk 0.31 ± ± ± Increase by MET h/wk 0.52 ± ± 0.17 < ± Increase by 25 MET h/wk 0.74 ± 0.16 < ± 0.16 < ± 0.15 <0.001 Increase in vigorous activity of 25 MET h/wk 0.35 ± 0.10 < ± 0.11 < ± Change in weight training No change Decrease by 0.5 h/wk 0.17 ± ± ± Increase by 0.5 h/wk 1.02 ± 0.15 < ± 0.16 < ± 0.14 <0.001 Increase in walking of 3 MET h/wk 0.03 ± ± ± Change in walking pace No change Decrease 1 mph 0.56 ± 0.14 < ± 0.14 < ± Increase 1 mph 0.49 ± ± ± Increase in television watching of 20 h/wk 0.45 ± ± 0.14 < ± x ± SE. MET, metabolic equivalent task. 2 All multivariate models controlled for baseline age (continuous variable), baseline waist circumference (quartiles), baseline BMI (quartiles), baseline and change in total calories (continuous variables), change in smoking status (categorized as nonsmokers, habitual smokers, new smokers, and quitters), baseline and change in total fiber intake (continuous variables), and baseline and change in alcohol intake (continuous variables). 3 Additionally controlled for change in BMI (quartiles).

7 PREDICTORS OF WAIST GAIN 725 tion of abdominal fat tissue was not explained by increases in total obesity in this cohort. In the few studies that examined the relation between dietary fat and central obesity, no association was observed (33, 34), and data from long-term intervention trials suggested that fat consumption within the range of 18 40% of energy intake has little effect on body fatness (35). However, individual fatty acids may differentially affect abdominal adiposity through their effects on insulin action (36). In the Normative Aging Study (2), saturated fat intake was positively correlated with the abdomen-to-hip ratio (r = 0.13), and in a recent clinical trial, the replacement of dietary saturated fats with polyunsaturated fats resulted in improvements in insulin sensitivity and abdominal fat distribution (37). To our knowledge, the current study is the first prospective study to report the association between changes in trans fatty acid intake and increases in abdominal adiposity. It may be that trans fats, which have been associated with an elevated risk of type 2 diabetes (38), impair insulin sensitivity by eliciting alterations in cell membrane structure (39) and by increasing concentrations of interleukin 6, tumor necrosis factor, and prostaglandins, which may reduce insulin sensitivity (40). The resulting hyperinsulinemia may promote lipid accumulation by expressing lipoprotein lipase activity (41), and insulin s effects may be more highly expressed in abdominal visceral adipocytes than in subcutaneous adipocytes because of the potentially greater cellularity, innervation, and blood flow (41). Fiber may affect abdominal adipose tissue through its effects on insulin sensitivity; in particular, soluble fiber may blunt postprandial glycemic and insulinemic responses in the small intestine (42) that are linked to reductions in the rate of return of hunger and subsequent energy intake (43). In cross-sectional studies, fiber generally has been inversely associated with body weight (44) and body fat (45), and, in a longitudinal investigation, fiber was inversely associated with BMI at all levels of fat intake (46). In the Coronary Artery Risk Development in Young Adults Study, fiber consumption further predicted insulin concentrations and 10-y weight gain (46). Prior studies linked increased central adiposity with smoking cessation. Kahn et al (12) reported that men who quit smoking were twice as likely to report waist gain at 10-y follow-up as were those who remained habitual smokers (12). Similarly, men in the Normative Aging Study who quit smoking experienced increases in central adiposity that were independent of age or initial BMI that were greater than the increases in both never or former smokers and current smokers during 15 y of follow-up (14). Sedentary behavior, represented in this study by television watching, was significantly related to increases in abdominal adiposity independent of physical activity. Fung et al (47) reported that television watching was significantly related to plasma biomarkers of obesity among men. The effects of physical inactivity may have been underestimated in our study because time spent watching television may not have represented the total amount of time spent inactively (48). Vigorous physical activity and weight training were significantly inversely associated with waist gain. Whereas an increase in total walking volume was not significantly related to reduced waist circumference, an increase in walking pace was inversely associated with waist gain, which supported the importance of exercise intensity in protection against abdominal obesity. Similarly, in the Cancer Prevention II Study (12), men who engaged in vigorous activities such as running 1 3 h/wk were significantly less likely to experience waist gain than were men who did not FIGURE 2. Possible effect of changes in behaviors on 9-y waist gain in the presence and absence of weight change (corrected for measurement error)., not controlled for weight change;, with weight change held constant. The potential effect on 9-y waist circumference (in cm) was calculated for an increase in dietary fiber of 12 g/d, an increase in weight training by 0.5 h/wk, the replacement of 2% of energy from trans fats with 2% of energy from polyunsaturated fats, and a reduction in television watching of 20 h/wk. engage in any vigorous activities (odds ratio: 0.75; 95% CI: 0.64, 0.88). Walking afforded no significant benefit when performed < 4 h/wk (odds ratio: 1.08; 95% CI: 0.99, 1.17), and nonvigorous activities including gardening provided no significant benefit (12). Physical activity may reduce abdominal obesity through the utilization of more fat from the intraabdominal region than from the gluteal region, which results in the redistribution of adipose tissue (49). Chronic exercise training may also enhance insulin-stimulated glucose uptake through increased activity or expression of key proteins involved in skeletal muscle glucose metabolism (50). To induce the physiologic changes necessary to influence central fat stores, weight training and more vigorous types of activity may be needed. Because our study obtained only 2 waist measurements, we were not able to assess changes in exposures that preceded waist gain, and this inability limited the drawing of causal inferences. However, the consistency of our findings with those of prior studies and the likely effects induced by the exposures on waist gain lend credence to important causal relations. Furthermore, whereas the educational backgrounds, socioeconomic status, and behavioral practices of our participants may not be characteristic of the entire US population, the physiologic effects induced by these lifestyle exposures should not vary substantially by income or education. The prospective collection of data in our study further reduces the potential for bias that is attributable to differential recall by the amount of waist gain. Implemented together, a 2% reduction in energy from trans fats, a 12-g/d increase in fiber, a 3-h/d reduction in television watching, and a 0.5-h/wk increase in weight training may decrease waist circumference by 2.9 cm over 9 y. Whereas the individual contributions of these lifestyle modifications may appear trivial, the clinical implications of even minor reductions in waist circumference are substantial. In our previous report, a 2.5-cm difference in waist circumference was associated with an 20% greater risk for the development of diabetes (8). Furthermore, this risk reduction does not take into account the weight loss that would likely ensue because of the implementation of these behaviors or the favorable changes in insulin sensitivity, serum cholesterol, and cardiorespiratory fitness that may result.

8 726 KOH-BANERJEE ET AL Because trans fatty acid intake, low fiber consumption, and sedentary behavior tend to be more prevalent in the general population, the magnitude of waist reduction that could be achieved would be even greater than that observed in our population of healthy men. Because of measurement errors that are included at both the beginning and end of follow-up, our regression coefficients are also underestimations of the true effect estimates. After accounting for errors in assessing changes in diet and physical activity, these behaviors would predict a difference in waist gain of 10.2 cm (Figure 2). Our results support the greater importance of the type of fat consumed than of the total quantity of fat in the diet, and they add to the growing discussion of the adverse health consequences associated with trans fats. Furthermore, we found that benefits for the prevention of long-term waist gain were associated with particular types and intensity of exercise. A better understanding of the biological mechanisms involved in the various obesity phenotypes may lead to more targeted and effective treatments. Equally as important, simple alterations in diet and physical activity hold promise for reducing age-related increases in abdominal adiposity and the incidence of the metabolic complications associated with this obesity phenotype. The authors are indebted to Lydia Liu, Ellen Hertzmark, and Sydney Atwood for their technical support and assistance. PK-B was responsible for the design of the study, analysis of data, and writing of the manuscript. N-FC was responsible for analysis of data and writing of the manuscript. DS contributed to the design of the study, analysis of data, and writing of the manuscript. BR provided statistical advice and contributed to the writing of the manuscript. GC contributed to the collection of data and the writing of the manuscript. WW contributed to the securing of funding, collection of data, design of the study, analysis of data, and writing of the manuscript. ER contributed to the securing of funding, collection of data, design of the study, analysis of data, and writing of the manuscript. None of the authors had advisory board affiliations or financial interests in organizations sponsoring the research. REFERENCES 1. Seidell JC, Cigolini M, Deslypere JP, Charzewska J, Ellsinger BM, Cruz A. Body fat distribution in relation to serum lipids and blood pressure in 38-year-old European men: the European fat distribution study. Atherosclerosis 1991;86: Ward KD, Sparrow D, Vokonas PS, Willett WC, Landsberg L, Weiss ST. The relationships of abdominal obesity, hyperinsulinemia and saturated fat intake to serum lipid levels: the Normative Aging Study. Int J Obes Relat Metab Disord 1994;18: Lundgren H, Bengtsson C, Blohme G, Lapidus L, Sjostrom L. Adiposity and adipose tissue distribution in relation to incidence of diabetes in women: results from a prospective population study in Gothenburg, Sweden. Int J Obes 1989;13: Manson J, Willett W, Stampfer M, et al. Body weight and mortality among women. N Engl J Med 1995;333: Folsom AR, Stevens J, Scheiner PJ, McGovern PG. Body mass index, waist/hip ratio, and coronary heart disease incidence in African Americans and whites. Atherosclerosis Risk in Communities Study Investigators. Am J Epidemiol 1998;148: Rimm EB, Stampfer MJ, Giovannucci E, et al. Body size and fat distribution as predictors of coronary heart disease among middle-aged and older US men. Am J Epidemiol 1995;141: Lakka HM, Lakka TA, Tuomilehto J, Salonen JT. Abdominal obesity is associated with increased risk of acute coronary events in men. Eur Heart J 2002;23: Chan JM, Rimm EB, Colditz GA, Stampfer MJ, Willett WC. Obesity, fat distribution, and weight gain as risk factors for clinical diabetes in men. Diabetes Care 1994;17: Hodge AM, Dose GK, Gareeboo H, Tuomilehto J, Alberti KG, Zimmet PZ. Incidence, increasing prevalence, and predictors of change in obesity and fat distribution over 5 years in the rapidly developing population of Mauritius. Int J Obes Relat Metab Disord 1996;20: Eck LH, Pascale RW, Klesges RC, White Ray JA, Klesges LM. Predictors of waist circumference change in healthy young adults. Int J Obes Relat Metab Disord 1995;19: Lahti-Koski M., Pietinen P, Mannisto S, Vartiainen E. Trends in waist-to-hip ratio and its determinants in adults in Finland from 1987 to Am J Clin Nutr 2000;72: Kahn HS, Tatham LM, Rodriguez C, Calle EE, Thun MJ, Heath CW Jr. Stable behaviors associated with adults 10-year change in body mass index and likelihood of gain at the waist. Am J Public Health 1997; 87: Despres JP, Tremblay A, Nadeau A, Bouchard C. Physical training and changes in regional adipose tissue distribution. Acta Med Scand Suppl 1988;723: Grinker JA, Tucker K, Vokonas PS, Rush D. Body habitus changes among adult males from the normative aging study: relations to aging, smoking history and alcohol intake. Obes Res 1995;3: Molarius A, Seidell JC. Selection of anthropometric indicators for classification of abdominal fatness a critical review. Int J Obes Relat Metab Disord 1998;22: Grinker JA, Tucker KL, Vokonas PS, Rush D. Changes in patterns of fatness in adult men in relation to serum indices of cardiovascular risk: the Normative Aging Study. Int J Obes Relat Metab Disord 2000;24: Ho S, Chen Y, Woo J, Leung S, Lam T, Janus E. Association between simple anthropometric indices and cardiovascular risk factors. Int J Obes Relat Metab Disord 2001;25: Baik I, Ascherio A, Rimm E, et al. Adiposity and mortality in men. Am J Epidemiol 2000;152: Visscher TLS, Seidell J, Molarius A, van der Kuip D, Hofman A, Witteman J. A comparison of body mass index, waist-hip ratio and waist circumference as predictors of all-cause mortality among the elderly: the Rotterdam study. Int J Obes Relat Metab Disord 2001; 25: Rimm EB, Stampfer MJ, Colditz GA, Chute CG, Litin LB, Willett WC. Validity of self-reported waist and hip circumferences in men and women. Epidemiology 1990;1: Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986;1: Palta M, Prineas RJ, Berman R, et al. Comparison of self-reported and measured height and weight. Am J Epidemiol 1982;115: Rimm EB, Giovannucci EL, Stampfer MJ, Colditz GA, Litin LB, Willett WC. Reproducibility and validity of an expanded selfadministered semiquantitative food frequency questionnaire among male health professionals. Am J Epidemiol 1992;135: Willett WC. Nutritional epidemiology. 2nd edition. New York: Oxford University Press, US Department of Agriculture. Composition of foods raw, processed, and prepared, Washington, DC: Department of Agriculture, Government Printing Office, (Agricultural handbook no. 8 series.) 26. Giovannucci E, Colditz G, Stampfer MJ, et al. The assessment of alcohol consumption by a simple self-administered questionnaire. Am J Epidemiol 1991;133: Ainsworth BE, Haskell WL, Leon AS, Jacobs DR Jr, Montoye HJ, Sallis JF. Compendium of physical activities: classification of energy costs of human physical activity. Med Sci Sports Exerc 1993; 25:71 80.

9 PREDICTORS OF WAIST GAIN Chasan-Taber S, Rimm EB, Stampfer MJ, et al. Reproducibility and validity of a self-administered physical activity questionnaire for male health professionals. Epidemiology 1996;7: White H. A heteroskedasticity-consistent covariance matrix estimator and a direct test for heteroskedasticity. Econometrica 1980;48: Willett WC, Sampson L, Stampfer M, et al. Reproducibility and validity of a semiquantitative food frequency questionnaire. Am J Epidemiol 1985;122: Rosner B, Willett WC, Spiegelman D. Correction of logistic regression relative risk estimates and confidence intervals for systematic within-person measurement error. Stat Med 1989;8: Rimm EB, Stampfer MJ, Colditz GA, Giovannucci E, Willett WC. Effectiveness of various mailing strategies among nonrespondents in a prospective cohort study. Am J Epidemiol 1990; 131: Larson D, Hunter G, Williams M, Kekes-Szabo T, Nyikos I, Goran M. Dietary fat in relation to body fat and intra-abdominal adipose tissue: a cross-sectional analysis. Am J Clin Nutr 1996;64: Samaras K, Fleury A, Spector T, Kelly P, Campbell L. Is there a relationship between current diet and adiposity in post-menopausal women? Proc Australas Soc Study Obes 1996;4:19 (abstr). 35. Willett WC. Is dietary fat a major determinant of body fat? Am J Clin Nutr 1998;67(suppl):556S 62S. 36. Storlien LH, Kriketos AD, Jenkins AB, et al. Does dietary fat influence insulin action? Ann N Y Acad Sci 1997;827: Summers LKM, Fielding BA, Bradshaw HA, et al. Substituting dietary saturated fat with polyunsaturated fat changes abdominal fat distribution and improves insulin sensitivity. Diabetologia 2002;45: Salmeron J, Manson JE, Stampfer MJ, Colditz GA, Rimm EB, Willett WC. Dietary fat, intake and risk of type 2 diabetes mellitus in women. Am J Clin Nutr 2001;73: Hu FB, van Dam RM, Liu S. Diet and risk of type II diabetes: the role of types of fat and carbohydrate. Diabetologia 2001;44: Prentice AM. Overeating: the health risks. Obes Res 2001;9: 234S 8S. 41. Bjorntorp P. The regulation of adipose tissue distribution in humans. Int J Obes Relat Metab Disord 1996;20: Pereira MA, Ludwig DS. Dietary fiber and body-weight regulation. Pediatr Clin North Am 2001;48: Roberts SB. High-glycemic index foods, hunger and obesity: is there a connection? Nutr Rev 2000;58: Alfieri M, Pomerleau J, Grace DM, et al. Fiber intake of normal weight, moderately obese and severely obese subjects. Obes Res 1995;3: Nelson LH, Tucker LA. Diet composition related to body fat in a multivariate study of 293 men. J Am Diet Assoc 1996;96: Ludwig DS, Pereira MA, Kroenke CH, et al. Dietary fiber, weight gain, and cardiovascular disease risk factors in young adults. JAMA 1999;282: Fung TT, Hu FB, Yu J, et al. Leisure-time physical activity, television watching, and plasma biomarkers of obesity and cardiovascular disease risk. Am J Epidemiol 2000;152: Coakley, EH, Rimm EB, Colditz G, Kawachi I, Willett W. Predictors of weight change in men: results from the Health Professionals Follow-up Study. Int J Obes Relat Metab Disord 1998;22: Krotkiewski M. Can body fat patterning be changed? Acta Med Scand Suppl 1998;723: Ryder JW, Gilbert M, Zierath JR. Skeletal muscle and insulin sensitivity: pathophysiological alterations. Front Biosci 2001;6:D APPENDIX A. Let: X 1 represent true dietary intake (diet record) in 1980; X 2 represent true dietary intake (diet record) in 1986; Z 1 represent dietary intake measured by a surrogate (food-frequency questionnaire) in 1980; and Z 2 represent dietary intake measured by a surrogate (food-frequency questionnaire) in Using the regression calibration approach, the change in true dietary intakes (X 2 X 1 ) is estimated as a function of the change in surrogate intakes (Z 2 Z 1 ) derived from the validation study data: (X 2 X 1 ) = + ( ) (Z 2 Z 1 ) + (A1) In the case of univariate linear regression for which the exposure is a difference in 2 linear terms measured with error, the corrected point estimate for the exposure measure is: * = / (A2) where is the estimated linear regression coefficient from the main study, and is the estimated regression slope of X on Z from the validation study.