Metabolic Predictors of 5-Year Change in Weight and Waist Circumference in a Triethnic Population

Transcription

1 American Journal of Epidemiology Copyright 2003 by the Johns Hopkins Bloomberg School of Public Health All rights reserved Vol. 157, No. 7 Printed in U.S.A. DOI: /aje/kwg022 Metabolic Predictors of 5-Year Change in Weight and Waist Circumference in a Triethnic Population The Insulin Resistance Atherosclerosis Study Elizabeth J. Mayer-Davis 1, Gregory J. Kirkner 1, Andrew J. Karter 2, and Daniel J. Zaccaro 3 1 Department of Epidemiology and Biostatistics, School of Public Health, University of South Carolina, Columbia, SC. 2 Northern California Region, Division of Research, Kaiser Permanente, Oakland, CA. 3 Department of Public Health Sciences, Bowman Gray School of Medicine, Wake Forest University, Winston-Salem, NC. Received for publication May 13, 2002; accepted for publication October 17, Insulin resistance, insulin secretion, and glucose tolerance may predict weight change. A total of 1,194 adults aged years at baseline (46% with normal glucose tolerance according to World Health Organization criteria, 23% with impaired glucose tolerance, and 31% with type 2 diabetes mellitus who were not taking insulin) were evaluated at baseline ( ) and after 5 years. Baseline insulin sensitivity (S I ) was measured by means of a 12-sample, insulin-enhanced, frequently sampled intravenous glucose tolerance test. Insulin secretion was assessed in terms of acute insulin response and disposition index, both obtained from the frequently sampled intravenous glucose tolerance test. At follow-up, 25% of subjects had lost more than 2.27 kg (>5 pounds), 38% weighed within 2.27 kg of their baseline weight, and 37% had gained more than 2.27 kg. In separate models, greater weight loss occurred among those with type 2 diabetes than among those with either impaired glucose tolerance or normal glucose tolerance (p < 0.001); baseline acute insulin response and disposition index were positively associated and baseline fasting insulin level was inversely associated with 5- year weight change (p < 0.05 for each; data were adjusted for baseline body mass index and demographic and behavior change variables). Upon simultaneous inclusion of metabolic variables within glucose tolerance status groups, none was a significant predictor of weight loss. Apart from glucose tolerance status itself, measures of insulin metabolism appear to have little effect on weight change over 5 years. body composition; body weight changes; diabetes mellitus; insulin; insulin resistance; metabolism; weight loss Abbreviations: IRAS, Insulin Resistance Atherosclerosis Study; S I, insulin sensitivity. In 1991, it was reported that insulin resistance was associated with reduced rates of weight gain among Pima Indians (1). Subsequently, some studies reported similar findings (albeit using insulin concentration as a surrogate measure of insulin resistance (2 4)), while others found that insulin resistance predicted increased weight gain (5, 6). Still other investigators found no association (7 9) or found differential effects of insulin level on weight change according to age, race/ethnicity, gender, or type of modeling procedure used for longitudinal data (10, 11). It has also been postulated that increased insulin secretion is a determinant of reduced weight gain through the direct effect of insulin on the central nervous system in inducing satiety and reducing food intake over time in response to the fed state (12). Among Pima Indians with normal glucose tolerance, insulin secretion was inversely related to 3-year rate of weight gain independently of initial body weight and insulin sensitivity (S I ) (13). However, Sigal et al. (14) observed the greatest weight gain over 16 years among persons with both high S I and a high acute insulin response. Among Caucasian women (but not men), Gould et al. (7) recently reported an inverse association between insulin secretion and increased weight gain and a positive association of fasting insulin concentration with increased waist:hip ratio over 4 years. Fasting insulin was not related to weight gain in either men or women. Correspondence to Dr. Elizabeth J. Mayer-Davis, Department of Epidemiology and Biostatistics, Normal J. Arnold School of Public Health, University of South Carolina, 800 Sumter Street, Columbia, SC ( mayer@gwm.sc.edu). 592

2 Metabolic Predictors of Weight Change 593 In the largest multiethnic cohort to be studied to date, the hypothesis that increased insulin resistance and increased insulin concentration independently predict reduced rates of weight gain was evaluated among 1,194 adults aged years at baseline who participated in the 5-year follow-up examination of the Insulin Resistance Atherosclerosis Study (IRAS). Glucose tolerance was also considered as a potential determinant of weight change on the basis of the common clinical observation of weight loss in the face of uncontrolled diabetes or in response to medical nutrition therapy designed to encourage weight loss. Change in waist circumference was considered as a second outcome variable. The analyses considered potential differential effects according to ethnicity, gender, and glucose tolerance (normal glucose tolerance, impaired glucose tolerance, and non-insulin-using type 2 diabetes mellitus). MATERIALS AND METHODS The IRAS was a multicenter study of the relation of insulin and insulin resistance to atherosclerosis and its risk factors among Hispanic, non-hispanic White, and African-American men and women with normal glucose tolerance, impaired glucose tolerance, or non-insulin-using type 2 diabetes. Participants, who were aged years at baseline, were enrolled from clinical centers located in Oakland, California; Los Angeles, California; San Antonio, Texas; and the San Luis Valley, Colorado. Further details on the IRAS study design have been published previously (15). The study protocols were approved by the institutional review board of each clinical center, and participants provided informed consent. The baseline examination occurred in ; the follow-up examination was conducted approximately 5 years later. Sample selection Of the original cohort of 1,624 IRAS participants, 1,313 (80.1 percent) were seen at the follow-up examination. Of these, 1,195 had complete data for the present analyses and one individual was excluded as an outlier (normal glucose tolerance at baseline, weight loss of 73 kg), for a final sample size of 1,194. Some analyses varied slightly in sample size (1,192 1,194) because of occasional missing data. At baseline, on the basis of World Health Organization criteria for a 2-hour, 75-g oral glucose tolerance test (16) or use of oral hypoglycemic agents, 46 percent of subjects had normal glucose tolerance, 23 percent had impaired glucose tolerance, and 31 percent had type 2 diabetes and were not taking insulin. Variable measurement The fasting plasma sample drawn at the time of the oral glucose tolerance test was used to determine fasting insulin concentration (µu/ml). Between 1 week and 1 month after the oral glucose tolerance test, a 12-sample, insulinenhanced, frequently sampled intravenous glucose tolerance test was used to assess S I and acute insulin response. Methods used have been described in detail previously (17). Frequently sampled intravenous glucose tolerance tests were performed with glucose (0.3 g/kg body weight) and insulin (0.03 units/kg) injections at 0 and 20 minutes, respectively. Blood samples were collected at 0, 2, 4, 8, 19, 22, 30, 40, 50, 70, 100, and 180 minutes for centralized determination of glucose and insulin levels (laboratory of R. Bergman, University of Southern California, Los Angeles). S I was calculated with minimal model analyses. Fasting insulin, which is inversely correlated with S I (18), was also used as a surrogate marker for S I. Acute insulin response was calculated on the basis of insulin levels through the 8-minute blood sample (prior to insulin infusion). Recent studies have also measured insulin secretion as the disposition index. The disposition index is based on the hyperbolic relation between S I and insulin secretion and quantifies pancreatic functionality as the product of S I and insulin secretion (19). This measure can be interpreted as pancreatic beta cell compensation for the degree of insulin resistance present, and it has been strongly related to diabetes incidence over 5 years in the IRAS cohort (20). Anthropometric measures were taken with the participant in lightweight clothing with shoes removed. Height and weight were measured in duplicate, with the participant in lightweight clothing and no shoes, and recorded to the nearest 0.5 cm and 0.1 kg, respectively. Body mass index was calculated as weight (kg)/height (m) 2. Minimum waist circumference was measured using a flexible steel tape measure at the natural indentation or at a level midway between the iliac crest and the lower edge of the rib cage if no natural indentation was visible. Waist circumference was recorded to the nearest 0.5 cm, and the mean of two measures within 1 cm of each other was used. Standardized interviews were used to ascertain self-report of being on a weight loss diet at the time of either the baseline examination or the 5-year follow-up examination. Subjects were asked to report their overall perceived health status as excellent, good, fair, or poor. At each examination, dietary intake was assessed by means of a validated food frequency interview (21), and frequency of participation in vigorous activity was assessed through a validated question with a response set that ranged from rare to never to five or more times per week (22). Smoking status was assessed by standardized interview at both examinations. Statistical analyses Age-adjusted partial correlation coefficients were obtained from linear regression. Because of right-skewness of the data, data for S I, acute insulin response, disposition index, and fasting insulin were log-transformed in these analyses. Constants were added to S I (constant = 1) and acute insulin response (constant = 360) as appropriate in order to avoid negative or zero values prior to log transformation. Multiple linear regression was used to test hypotheses for the main effects (α = 0.05) of S I, acute insulin response, disposition index, fasting insulin, and glucose tolerance on the primary outcome, 5-year weight change. Because the overall model R 2 varied little based on whether the log-transformed or untransformed variables were used for S I, acute insulin response, disposition index, and fasting insulin as independent variables, the untransformed variables were

3 594 Mayer-Davis et al. TABLE 1. Characteristics of participants in the Insulin Resistance Atherosclerosis Study (n = 1,194) according to 5-year weight change, Characteristic * SD, standard deviation; S I, insulin sensitivity. Weight (kg)/height (m) 2. 5-year follow-up examination minus baseline examination. 5-year follow-up examination minus baseline examination, adjusted for body mass index. Weight change Loss of 2.27 kg (n = 299) Stable within 2.27 kg (n = 447) Gain of 2.27 kg (n = 448) Mean or no. SD* or % Mean or no. SD or % Mean or no. SD or % Weight (kg) at baseline examination Body mass index at baseline examination Weight change (kg) Waist circumference (cm) at baseline examination Waist change (cm) S I * ( 10 4 minutes 1 /µu/ml) at baseline examination Acute insulin response (µu/ml minutes) at baseline examination Disposition index (acute insulin response S I ) at baseline examination Fasting insulin level (µu/ml) at baseline examination Diabetes status at baseline examination (no.) Normal % % % Impaired glucose tolerance % % % Diabetes mellitus % % % Age (years) at baseline examination Ethnicity (no.) Non-Hispanic White % % % Hispanic % % % African-American % % % Gender (no.) Female % % % Male % % % On a weight loss diet (no.) Yes, at baseline examination only % % % Yes, at 5-year follow-up examination only % % % Yes, at both examinations 3 1.0% 4 0.9% 6 1.4% No % % % Perceived health at baseline examination (no.) Excellent % % % Good % % % Fair % % % Poor % 6 1.3% 9 2.0% retained for simplicity in model interpretation. To account for regression to the mean, these models included baseline body mass index. The models also included adjustment for potential confounding by demographic variables (age, gender, ethnicity, clinical site), baseline perceived health status, and use of oral hypoglycemic agents. In addition, the data were also adjusted for 5-year change in behavioral variables that could have an impact on energy balance: caloric intake, physical activity, alcohol intake, and smoking. In exploratory analyses of predictors of 5-year change in waist circumference, the series of models was repeated with adjustment for baseline waist circumference and, by regression, change in body mass index. Thus, in these analyses, waist change can be interpreted as change in central adiposity, independent of change in overall obesity. In order to maximize the observable range of variables in the data set and to avoid unintentional introduction of bias, we pooled all participants across glucose tolerance groups in

4 Metabolic Predictors of Weight Change 595 TABLE 2. Characteristics of participants in the Insulin Resistance Atherosclerosis Study (n = 1,194) according to baseline glucose tolerance status, Characteristic Normal glucose tolerance (n = 554) * SD, standard deviation; S I, insulin sensitivity. Weight (kg)/height (m) 2. 5-year follow-up examination minus baseline examination. 5-year follow-up examination minus baseline examination, adjusted for body mass index. Glucose tolerance status Impaired glucose tolerance (n = 274) Diabetes mellitus (n = 366) Mean or no. SD* or % Mean or no. SD or % Mean or no. SD or % Weight (kg) at baseline examination Body mass index at baseline examination Weight change (kg) Waist circumference (cm) at baseline examination Waist change (cm) S I * ( 10 4 minutes 1 /µu/ml) at baseline examination Acute insulin response (µu/ml minutes) at baseline examination Disposition index (acute insulin response S I ) at baseline examination Fasting insulin level (µu/ml) at baseline examination On a weight loss diet (no.) Yes, at baseline examination only % % % Yes, at 5-year follow-up examination only % 9 3.3% % Yes, at both examinations 1 0.2% 4 1.5% 8 2.2% No % % % the initial analyses. We added interaction terms one at a time to evaluate potential differential effects of the main effect variables on the outcomes, according to glucose tolerance, ethnicity, and gender (α = 0.05). Finally, regardless of the statistical significance of the glucose tolerance interaction term, analyses were repeated stratified according to glucose tolerance at baseline because of the a priori interest in the impact of glucose tolerance itself on weight change. RESULTS At baseline, average body mass index was The prevalences of normal weight (body mass index <25), overweight (body mass index ), and obesity (body mass index >30) were 21 percent, 44 percent, and 35 percent, respectively. At the 5-year follow-up, 25 percent of subjects had lost more than 2.27 kg (>5 pounds), 38 percent weighed within 2.27 kg of their baseline weight, and 37 percent had gained more than 2.27 kg. Additional descriptive characteristics of the participants are shown in table 1, according to category of 5-year weight change. Table 2 shows subjects characteristics according to baseline glucose tolerance status. Mean weight changes among persons with normal and impaired glucose tolerance were 1.98 kg and 1.30 kg, respectively, compared with 1.21 kg among those with diabetes. Age-adjusted Pearson correlation coefficients for correlations between the metabolic variables at baseline (including S I, acute insulin response, disposition index, and fasting insulin) and measures of baseline and follow-up obesity are given in table 3, according to glucose tolerance at baseline. As expected, among persons with normal glucose tolerance, baseline body mass index, weight, and waist circumference were strongly correlated with lower baseline S I, higher acute insulin response, lower disposition index, and higher concentrations of fasting insulin (all p s < 0.001). Crosssectional associations were similar among persons with impaired glucose tolerance. Presumably because of the restricted range of S I and acute insulin response among persons with diabetes, associations with measures of obesity were generally similar but attenuated in magnitude for this group. With regard to correlations of the metabolic variables with 5-year change in weight and waist circumference, results for persons with normal or impaired glucose tolerance were nonsignificant, with the exception of a significant inverse correlation between fasting insulin and change in waist circumference among persons with impaired glucose tolerance (r = 0.13, p < 0.05). Among persons with diabetes, S I was positively correlated (r = 0.11, p < 0.05), fasting insulin was inversely correlated (r = 0.20, p < 0.001), and disposition index was positively correlated (r = 0.11, p < 0.05) with 5-year weight change. Results were similar for 5-year change in waist circumference. We conducted the next series of analyses with the glucose tolerance groups combined in order to allow consideration of the full range of variability present in the sample. Age-

5 596 Mayer-Davis et al. TABLE 3. Age-adjusted partial correlation coefficients (r) for measures of baseline insulin resistance and insulin secretion in relation to baseline body mass index, weight, waist circumference, and change in weight and waist circumference, according to glucose tolerance status (n = 1,194), Insulin Resistance Atherosclerosis Study, Baseline metabolic measure Body mass index Baseline Weight (kg) Waist circumference (cm) Weight (kg) 5-year change Waist circumference (cm) Normal glucose tolerance S I, 0.49*** 0.37*** 0.49*** Acute insulin response 0.24*** 0.16*** 0.24*** Disposition index 0.27*** 0.23*** 0.27*** Fasting insulin level 0.48*** 0.38*** 0.46*** Impaired glucose tolerance S I 0.43*** 0.40*** 0.45*** Acute insulin response 0.14* 0.12* 0.17** Disposition index 0.24*** 0.23*** 0.23*** Fasting insulin level 0.44*** 0.37*** 0.43*** * * p < 0.05; ** p < 0.01; *** p < Weight (kg)/height (m) 2. S I, insulin sensitivity. Data for this variable were log-transformed. Type 2 diabetes mellitus S I 0.23*** 0.20*** 0.27*** 0.11* 0.12* Acute insulin response Disposition index 0.20*** 0.15*** 0.21*** 0.11* 0.12* Fasting insulin level 0.40*** 0.39*** 0.43*** 0.20*** 0.18*** adjusted correlation coefficients for S I, acute insulin response, disposition index, and fasting insulin in relation to weight change were 0.14, 0.13, 0.18, and 0.16, respectively (all p s < 0.001), suggesting increased weight gain with increased baseline S I, increased insulin secretion, and increased pancreatic functionality. For waist change, results were 0.16, 0.09, 0.16, and 0.16, respectively (all p s < 0.01). Table 4 shows results for each of the four baseline metabolic variables and glucose tolerance itself, considered separately in relation to 5-year change in weight and waist circumference, with adjustment for baseline body mass index, potential demographic confounders, and 5-year behavior change variables. Insulin resistance as measured by S I was not associated with 5-year weight change (p = 0.96). In contrast, both measures of insulin secretion (acute insulin response and disposition index) were significantly, positively related to weight change (p < 0.01), and fasting insulin was inversely related to weight change (p = 0.05). Significantly greater weight loss occurred among persons with diabetes than among those with either impaired glucose tolerance or normal glucose tolerance (p < 0.001). None of the metabolic variables were associated with waist change (independent of change in overall obesity). Before proceeding to models that considered the metabolic variables simultaneously, we added interaction terms to the models that predicted 5-year weight change (shown in table 4) in order to determine whether any of the observed effects differed significantly according to ethnicity, gender, or glucose tolerance. Of the 12 interaction terms tested, one reached statistical significance: The effect of S I on 5-year weight change appeared to differ according to glucose tolerance (p = 0.03 for the interaction term). Consistent with the unadjusted results shown in table 3, no significant relation between S I and weight change was observed among either persons with normal glucose tolerance or persons with impaired glucose tolerance; however, among persons with diabetes at baseline, higher S I was significantly associated with subsequent weight gain. Table 5 shows predicted 5-year change in weight for all metabolic variables considered simultaneously, with data stratified according to baseline diabetes status and adjusted for baseline body mass index, potential demographic confounders, and 5-year behavior change variables. None of the variables (S I, acute insulin response, fasting insulin) were significantly associated with 5-year weight change within subgroups of normal glucose tolerance (model 1), impaired glucose tolerance (model 2), or type 2 diabetes mellitus (model 3) (p > 0.15 for each).

6 Metabolic Predictors of Weight Change 597 TABLE 4. Predicted 5-year change in weight and waist circumference for each metabolic variable considered separately, adjusted for potentially confounding factors (n = 1,194), Insulin Resistance Atherosclerosis Study, Model 1 Model Change in weight (kg) per 1-SD increase Weight change 95% CI Waist circumference change Change in waist circumference (cm) per 1-SD increase 95% CI S I , , 0.17 Model 2 Acute insulin response , 0.95** , 0.36 Model 3 Disposition index , 1.24** , 0.20 Model 4 Fasting insulin level , 0.002* , 0.27 Model 5 Glucose tolerance status Normal glucose tolerance 1.05, , 0.63 Impaired glucose tolerance , 1.27*** , 0.78 Type 2 diabetes mellitus * p < 0.05; ** p < 0.01; *** p < Potentially confounding factors included baseline body mass index, age, gender, ethnicity, clinical site, perceived health, use of oral hypoglycemic agents, and behavior change (i.e., difference between baseline examination and follow-up examination for alcohol, smoking, physical activity, and dietary calories). Change in waist circumference was adjusted, by regression, for change in body mass index. The models also included baseline waist circumference. SD, standard deviation; CI, confidence interval; S I, insulin sensitivity. Reference category. TABLE 5. Predicted 5-year change in weight for all metabolic variables considered simultaneously, with data adjusted for potentially confounding factors* and stratified according to baseline glucose tolerance status (n = 1,194), Insulin Resistance Atherosclerosis Study, Model Model 1: normal glucose tolerance (n = 554) Change in weight (kg) per 1-SD increase 95% CI S I , 0.24 Acute insulin response , 0.69 Fasting insulin level , 0.67 Model 2: impaired glucose tolerance (n = 274) S I , 0.93 Acute insulin response , 1.06 Fasting insulin level , 0.16 Model 3: type 2 diabetes mellitus (n = 366) S I , 2.4 Acute insulin response , 2.72 Fasting insulin level , 0.82 * Potentially confounding factors included baseline body mass index, age, gender, ethnicity, clinical site, perceived health, and behavior change (i.e., alcohol, smoking, physical activity, and dietary calories). For the model including persons with diabetes, confounding factors included use of oral hypoglycemic agents. SD, standard deviation; CI, confidence interval; S I, insulin sensitivity.

7 598 Mayer-Davis et al. FIGURE 1. Five-year weight change according to diabetes status, before and after adjustment for metabolic and behavior change variables, Insulin Resistance Atherosclerosis Study, NGT, normal glucose tolerance; IGT, impaired glucose tolerance; DMs, diabetes mellitus. Bars, 95% confidence interval. To further address the issue of intentional weight loss beyond adjustment for reported 5-year change in behaviors likely to have an impact on weight, we repeated analyses in the subset of study participants who, by self-report, were not on a weight-loss diet at the time of either examination 1 or examination 2 (n = 1,085). The results were essentially unchanged. We further explored the impact of diabetes status. Figure 1 shows 5-year weight change according to diabetes status both before and after inclusion of the metabolic variables and behavior change variables considered in the present study. Any potential effect of these covariates on weight loss was not sufficient to explain the observed impact of diabetes status on weight change over the 5-year follow-up period. However, as evidenced by the increased width of the confidence intervals, precision was lost, presumably because of the inclusion of metabolic variables that are associated with each other both biologically and statistically. Next, persons with diabetes at baseline were stratified by severity of disease according to baseline fasting glucose level (categories: <140, , , and 250 mg/dl). Figure 2 shows a pattern of incremental weight loss across the first three categories, with weight gain being observed among those in the highest category of baseline fasting glucose even after adjustment for metabolic variables and 5-year behavior change variables (sample sizes in the categories ranged from 49 to 134). Because insulin therapy or use of most oral hypoglycemic agents can promote weight gain, the final model excluded persons who initiated use of insulin or oral hypoglycemic agents during the follow-up period (sample sizes in the categories ranged from 25 to 96). This essentially explained the weight gain in the most severe disease category (baseline fasting glucose level 250 mg/dl). DISCUSSION These findings show that the diabetic state, characterized by relatively low S I, low acute insulin response, low disposition index, and (for persons with type 2 diabetes who were not taking insulin) high fasting insulin concentrations, was by far the strongest predictor of 5-year weight loss in this cohort. None of the metabolic variables related to glucose and insulin metabolism explained the impact of having diabetes on subsequent weight loss. Furthermore, among persons with either normal glucose tolerance or impaired glucose tolerance, none of the measures of insulin resistance, insulin secretion, pancreatic function as assessed by disposition index, or fasting insulin concentration were predictive of 5-year weight change. Because the pathways that determine observed levels of these metabolic measures at any given point in time are highly dependent on one another as a function of biologic feedback systems, it is not surprising that various studies reported one or another of these variables to be associated with weight change, somewhat inconsistently. The present analysis differed from results previously published in the direct consideration of glucose tolerance itself. It is possible that the variables under study in previous reports (e.g., fasting insulin level) were actually reflecting risk for development of the diabetic state, rather than any single mechanism related to glucose or insulin metabolism. It is clear that increasing levels of obesity are strongly related to increased risk of diabetes, and evidence is accumulating that sustained, intentional weight loss may reduce diabetes risk (23 25). Recent findings from the Diabetes Prevention Program confirm the efficacy of moderate weight loss through lifestyle change in reducing risk for develop-

8 Metabolic Predictors of Weight Change 599 FIGURE 2. Five-year weight change according to diabetes status, before and after adjustment for metabolic and behavior change variables, excluding persons taking insulin at visit 2, Insulin Resistance Atherosclerosis Study, NGT, normal glucose tolerance; IGT, impaired glucose tolerance; DMs, diabetes mellitus; FPG, fasting plasma glucose (mg/dl); OHGA, oral hypoglycemic agents. Bars, 95% confidence interval. ment of type 2 diabetes (26). However, this does not preclude the potential for a natural history of diabetes in which weight loss ensues as a result of the impaired fat and carbohydrate metabolism that is central to the disease. In 1981, Knowler et al. (27) reported similar prospective data from the Pima Indian cohort in which weight loss was observed over the 2 years following diabetes diagnosis, despite the observation of weight gain in the 2 4 years prior to diagnosis among those aged years. It is possible that simply feeling unwell near the time of clinical diagnosis of diabetes may contribute to weight loss. Perhaps more likely, in addition to inefficient fuel storage, weight loss may occur with diabetes because of the loss of calories that may occur with glucosuria. Figure 2 is consistent with this notion in that increasing baseline fasting glucose concentration was associated with greater subsequent 5-year weight loss. The exception to this observation occurred among persons with a fasting glucose level 250 mg/dl, in whom weight gain was observed. However, initiation of insulin therapy or oral hypoglycemic agents explained the observed weight gain. Regarding the question of whether increased insulin secretion may lead to reduced weight gain through central nervous system signaling of the fed state (i.e., via feelings of satiety), the inverse association of fasting insulin concentration with weight change in table 4 might suggest that this may indeed be the case. However, results for acute insulin response and disposition index suggest just the opposite, namely, an association of increased insulin secretion with increased weight gain. Fasting insulin concentration (analyzed in the absence of other metabolic variables) may reflect insulin resistance more than insulin secretion. Again, final results as shown in table 5 and figures 1 and 2 suggest that any direct impact of individually measured aspects of this complex biologic system may be relatively unimportant in comparison with the confluence of the underlying processes when diabetes develops, leading to otherwise unexplained weight loss. Individuals with a diagnosis of type 2 diabetes are commonly advised to lose weight, which raised a concern that at least some of the observed weight loss might have been due to intentional weight loss. We addressed this through two analytical approaches. First, because increased caloric intake, decreased physical activity, change in smoking habits, or change in alcohol intake over the 5-year follow-up period would be expected to contribute to weight change, we calculated change in these variables and included it in the regression analyses. Second, in recognition of the imprecision inherent in assessment of these behavioral variables despite use of validated instruments, we restricted analyses to persons who reported that they were not on a weight loss diet, either at the first examination or at the second examination, and results were unchanged. Although neither of these two approaches can entirely rule out mixing of intentional and unintentional weight loss in the final results, it seems unlikely that all of the observed effects would be due to intentional weight loss. The question of metabolic predictors of change in waist circumference was motivated by previously published IRAS data that showed a positive relation between waist circumference and measures of insulin secretion (28), independent of body mass index. However, as indicated above in the Materials and Methods section, analyses of metabolic

9 600 Mayer-Davis et al. predictors of change in waist circumference should be viewed as exploratory. This is because the IRAS protocol did not include measures of central adiposity beyond simple anthropometric measurements and because previous work focused on energy balance, not on fat distribution. Initial analyses (table 3) showed quite similar findings for weight and waist circumference, presumably because waist circumference provides information about overall obesity as well as visceral fat (29). In the present study, the age-adjusted partial correlation coefficient for baseline body mass index and baseline waist circumference was 0.78 for those with normal glucose tolerance or impaired glucose tolerance and 0.81 for those with diabetes. In order to distinguish between associations with overall adiposity and associations with central adiposity, we used residual values from regression analysis as a measure of central adiposity (i.e., waist circumference, adjusted for body mass index) in subsequent modeling. Overall, these adjusted results (table 4) suggested no association between any of the metabolic variables (including glucose tolerance status) and 5-year change in body mass index-adjusted waist circumference. The residual method has commonly been used with variables that are strongly correlated with each other (e.g., dietary fat intake adjusted for total calorie intake by regression); however, the method is subject to fairly stringent distributional requirements, and results can be difficult to interpret from a clinical perspective (30). Because waist circumference explains variance in visceral fat measured by magnetic resonance imaging to a greater degree than does body mass index (29), future studies may be warranted to further explore the hypothesis of an effect of metabolic predictors on change in central adiposity, using both anthropometric measures and more precise measures such as magnetic resonance imaging. Overall, given the results of the published studies to date, it is possible that insulin resistance or insulin secretion may play a subtle role in determination of energy balance over time in nondiabetic persons. However, other influences (such as socially or culturally determined dietary choices and physical activity) are likely to be far more important to long-term weight change, unless alterations in glucose and insulin metabolism are sufficiently severe that diabetes ensues. Further study is warranted to determine the natural history of weight change and its attendant metabolic status among persons with diagnosed diabetes, including both persons who intentionally lose weight and those who do not. ACKNOWLEDGMENTS This study was supported by grants UO1-HL47887, UO1- HL47889, UO1-HL47890, UO1-HL47892, UO , and DK29867 from the National Heart, Lung, and Blood Institute. REFERENCES 1. Swinburn BA, Nyomba BL, Saad MF, et al. Insulin resistance associated with lower rates of weight gain in Pima Indians. J Clin Invest 1991;88: Hoag S, Marshall JA, Jones RH, et al. High fasting insulin levels associated with lower rates of weight gain in persons with normal glucose tolerance: The San Luis Valley Diabetes Study. Int J Obes Relat Metab Disord 1995;19: Valdez R, Mitchell BD, Haffner SM, et al. Predictors of weight change in a bi-ethnic population: The San Antonio Heart Study. Int J Obes Relat Metab Disord 1994;18: Wedick NM, Mayer-Davis EJ, Wingard DL, et al. Insulin resistance precedes weight loss in adults without diabetes: The Rancho Bernardo Study. Am J Epidemiol 2001;153: Hodge AM, Dowse GK, Alberti KG, et al. Relationship of insulin resistance to weight gain in nondiabetic Asian Indian, Creole, and Chinese Mauritians. Metabolism 1996;45: Odeleye OE, de Courten M, Pettitt DJ, et al. Fasting hyperinsulinemia is a predictor of increased body weight gain and obesity in Pima Indian children. Diabetes 1997;46: Gould A, Williams D, Byrne C, et al. Prospective cohort study of the relationship of markers of insulin resistance and secretion with weight gain and changes in regional adiposity. Int J Obes Relat Metab Disord 1999;23: Zavaroni I, Zuccarelli A, Gasparini P, et al. Can weight gain in healthy, nonobese volunteers be predicted by differences in baseline plasma insulin concentration? J Clin Endocrinol Metab 1998;83: McLaughlin T, Abbasi F, Carantoni M, et al. Differences in insulin resistance do not predict weight loss in response to hypocaloric diets in healthy obese women. J Clin Endocrinol Metab 1999;84: Folsom AR, Vitelli LL, Lewis CE, et al. Is fasting insulin concentration inversely associated with rate of weight gain? Contrasting findings from the CARDIA and ARIC study cohorts. Int J Obes 1998;22: Lazarus R, Sparrow D, Weiss S. Temporal relations between obesity and insulin: longitudinal data from the Normative Aging Study. Am J Epidemiol 1998;147: Porte DJ, Seeley R, Woods S, et al. Obesity, diabetes, and the central nervous system. Diabetologia 1998;41: Schwartz MW, Boyko EJ, Kahn SE, et al. Reduced insulin secretion: an independent predictor of body weight gain. J Clin Endocrinol Metab 1995;80: Sigal RJ, el-hashimy M, Martin BC, et al. Acute postchallenge hyperinsulinemia predicts weight gain: a prospective study. Diabetes 1997;46: Wagenknecht LE, Mayer EJ, Rewers M, et al. The Insulin Resistance Atherosclerosis Study (IRAS): objectives, design, and recruitment results. Ann Epidemiol 1995;5: World Health Organization. Diabetes mellitus: report of a WHO study group. (WHO Technical Report Series, no. 727). Geneva, Switzerland: World Health Organization, Saad MF, Anderson RL, Laws A, et al. A comparison between the minimal model and the glucose clamp in the assessment of insulin sensitivity across the spectrum of glucose tolerance. Diabetes 1994;43: Anderson RL, Hamman RF, Savage PJ, et al. Exploration of simple insulin sensitivity measures derived from frequently sampled intravenous glucose tolerance (FSIGT) tests: The Insulin Resistance Atherosclerosis Study. Am J Epidemiol 1995; 142: Elbein SC, Wegner K, Kahn SE. Reduced β-cell compensation to the insulin resistance associated with obesity in members of Caucasian familial type 2 diabetic kindred. Diabetes Care 2000; 23: Mayer-Davis EJ, Bell RA, D Agostino RJ. Moderate alcohol consumption in relation to higher insulin sensitivity: The Insulin Resistance and Atherosclerosis Study (IRAS). Diabetes

10 Metabolic Predictors of Weight Change ;47(suppl 1):A Mayer-Davis EJ, Vitolins MZ, Carmichael S, et al. Validity and reproducibility of a food frequency interview in a multi-cultural epidemiologic study. Ann Epidemiol 1999;9: Mayer-Davis EJ, D Agostino RJ, Karter AJ, et al. Intensity and amount of physical activity in relation to insulin sensitivity. JAMA 1998;279: Pan X, Li G, Hu Y, et al. Effects of diet and exercise in preventing NIDDM in people with impaired glucose tolerance. Diabetes Care 1997;20: Moore L, Visioni A, Wilson P, et al. Can sustained weight loss in overweight individuals reduce the risk of diabetes mellitus? Epidemiology 2000;3: Tuomilehto J, Lindstrom J, Eriksson JG, et al. Prevention of type 2 diabetes mellitus by changes in lifestyle among subjects with impaired glucose tolerance. N Engl J Med 2001;344: Knowler WC, Barrett-Connor E, Fowler SE, et al. Reduction in incidence of type 2 diabetes with lifestyle intervention or metformin. Diabetes Prevention Program Research Group. N Engl J Med 2002;346: Knowler WC, Pettitt DJ, Savage PJ, et al. Diabetes incidence in Pima Indians: contributions of obesity and parental diabetes. Am J Epidemiol 1981;113: Mayer-Davis EJ, Levin S, Bergman RN, et al. Insulin secretion, obesity, and potential behavioral influences: results from the Insulin Resistance Atherosclerosis Study (IRAS). Diabetes Metab Res Rev 2001;17: Janssen I, Heymsfield SB, Allison DB, et al. Body mass index and waist circumference independently contribute to the prediction of nonabdominal, abdominal subcutaneous, and visceral fat. Am J Clin Nutr 2002;75: Willet W, Stampfer M. Implications of total energy intake for epidemiologic analyses. In: Nutritional epidemiology. 2nd ed. New York, NY: Oxford University Press, 1998:

11 American Journal of Epidemiology Copyright 2003 by the Johns Hopkins Bloomberg School of Public Health All rights reserved Vol. 157, No. 7 Printed in U.S.A. DOI: /aje/kwg022 Metabolic Predictors of 5-Year Change in Weight and Waist Circumference in a Triethnic Population The Insulin Resistance Atherosclerosis Study Elizabeth J. Mayer-Davis 1, Gregory J. Kirkner 1, Andrew J. Karter 2, and Daniel J. Zaccaro 3 1 Department of Epidemiology and Biostatistics, School of Public Health, University of South Carolina, Columbia, SC. 2 Northern California Region, Division of Research, Kaiser Permanente, Oakland, CA. 3 Department of Public Health Sciences, Bowman Gray School of Medicine, Wake Forest University, Winston-Salem, NC. Received for publication May 13, 2002; accepted for publication October 17, Insulin resistance, insulin secretion, and glucose tolerance may predict weight change. A total of 1,194 adults aged years at baseline (46% with normal glucose tolerance according to World Health Organization criteria, 23% with impaired glucose tolerance, and 31% with type 2 diabetes mellitus who were not taking insulin) were evaluated at baseline ( ) and after 5 years. Baseline insulin sensitivity (S I ) was measured by means of a 12-sample, insulin-enhanced, frequently sampled intravenous glucose tolerance test. Insulin secretion was assessed in terms of acute insulin response and disposition index, both obtained from the frequently sampled intravenous glucose tolerance test. At follow-up, 25% of subjects had lost more than 2.27 kg (>5 pounds), 38% weighed within 2.27 kg of their baseline weight, and 37% had gained more than 2.27 kg. In separate models, greater weight loss occurred among those with type 2 diabetes than among those with either impaired glucose tolerance or normal glucose tolerance (p < 0.001); baseline acute insulin response and disposition index were positively associated and baseline fasting insulin level was inversely associated with 5- year weight change (p < 0.05 for each; data were adjusted for baseline body mass index and demographic and behavior change variables). Upon simultaneous inclusion of metabolic variables within glucose tolerance status groups, none was a significant predictor of weight loss. Apart from glucose tolerance status itself, measures of insulin metabolism appear to have little effect on weight change over 5 years. body composition; body weight changes; diabetes mellitus; insulin; insulin resistance; metabolism; weight loss Abbreviations: IRAS, Insulin Resistance Atherosclerosis Study; S I, insulin sensitivity. In 1991, it was reported that insulin resistance was associated with reduced rates of weight gain among Pima Indians (1). Subsequently, some studies reported similar findings (albeit using insulin concentration as a surrogate measure of insulin resistance (2 4)), while others found that insulin resistance predicted increased weight gain (5, 6). Still other investigators found no association (7 9) or found differential effects of insulin level on weight change according to age, race/ethnicity, gender, or type of modeling procedure used for longitudinal data (10, 11). It has also been postulated that increased insulin secretion is a determinant of reduced weight gain through the direct effect of insulin on the central nervous system in inducing satiety and reducing food intake over time in response to the fed state (12). Among Pima Indians with normal glucose tolerance, insulin secretion was inversely related to 3-year rate of weight gain independently of initial body weight and insulin sensitivity (S I ) (13). However, Sigal et al. (14) observed the greatest weight gain over 16 years among persons with both high S I and a high acute insulin response. Among Caucasian women (but not men), Gould et al. (7) recently reported an inverse association between insulin secretion and increased weight gain and a positive association of fasting insulin concentration with increased waist:hip ratio over 4 years. Fasting insulin was not related to weight gain in either men or women. Correspondence to Dr. Elizabeth J. Mayer-Davis, Department of Epidemiology and Biostatistics, Normal J. Arnold School of Public Health, University of South Carolina, 800 Sumter Street, Columbia, SC ( mayer@gwm.sc.edu). 592

12 Metabolic Predictors of Weight Change 593 In the largest multiethnic cohort to be studied to date, the hypothesis that increased insulin resistance and increased insulin concentration independently predict reduced rates of weight gain was evaluated among 1,194 adults aged years at baseline who participated in the 5-year follow-up examination of the Insulin Resistance Atherosclerosis Study (IRAS). Glucose tolerance was also considered as a potential determinant of weight change on the basis of the common clinical observation of weight loss in the face of uncontrolled diabetes or in response to medical nutrition therapy designed to encourage weight loss. Change in waist circumference was considered as a second outcome variable. The analyses considered potential differential effects according to ethnicity, gender, and glucose tolerance (normal glucose tolerance, impaired glucose tolerance, and non-insulin-using type 2 diabetes mellitus). MATERIALS AND METHODS The IRAS was a multicenter study of the relation of insulin and insulin resistance to atherosclerosis and its risk factors among Hispanic, non-hispanic White, and African-American men and women with normal glucose tolerance, impaired glucose tolerance, or non-insulin-using type 2 diabetes. Participants, who were aged years at baseline, were enrolled from clinical centers located in Oakland, California; Los Angeles, California; San Antonio, Texas; and the San Luis Valley, Colorado. Further details on the IRAS study design have been published previously (15). The study protocols were approved by the institutional review board of each clinical center, and participants provided informed consent. The baseline examination occurred in ; the follow-up examination was conducted approximately 5 years later. Sample selection Of the original cohort of 1,624 IRAS participants, 1,313 (80.1 percent) were seen at the follow-up examination. Of these, 1,195 had complete data for the present analyses and one individual was excluded as an outlier (normal glucose tolerance at baseline, weight loss of 73 kg), for a final sample size of 1,194. Some analyses varied slightly in sample size (1,192 1,194) because of occasional missing data. At baseline, on the basis of World Health Organization criteria for a 2-hour, 75-g oral glucose tolerance test (16) or use of oral hypoglycemic agents, 46 percent of subjects had normal glucose tolerance, 23 percent had impaired glucose tolerance, and 31 percent had type 2 diabetes and were not taking insulin. Variable measurement The fasting plasma sample drawn at the time of the oral glucose tolerance test was used to determine fasting insulin concentration (µu/ml). Between 1 week and 1 month after the oral glucose tolerance test, a 12-sample, insulinenhanced, frequently sampled intravenous glucose tolerance test was used to assess S I and acute insulin response. Methods used have been described in detail previously (17). Frequently sampled intravenous glucose tolerance tests were performed with glucose (0.3 g/kg body weight) and insulin (0.03 units/kg) injections at 0 and 20 minutes, respectively. Blood samples were collected at 0, 2, 4, 8, 19, 22, 30, 40, 50, 70, 100, and 180 minutes for centralized determination of glucose and insulin levels (laboratory of R. Bergman, University of Southern California, Los Angeles). S I was calculated with minimal model analyses. Fasting insulin, which is inversely correlated with S I (18), was also used as a surrogate marker for S I. Acute insulin response was calculated on the basis of insulin levels through the 8-minute blood sample (prior to insulin infusion). Recent studies have also measured insulin secretion as the disposition index. The disposition index is based on the hyperbolic relation between S I and insulin secretion and quantifies pancreatic functionality as the product of S I and insulin secretion (19). This measure can be interpreted as pancreatic beta cell compensation for the degree of insulin resistance present, and it has been strongly related to diabetes incidence over 5 years in the IRAS cohort (20). Anthropometric measures were taken with the participant in lightweight clothing with shoes removed. Height and weight were measured in duplicate, with the participant in lightweight clothing and no shoes, and recorded to the nearest 0.5 cm and 0.1 kg, respectively. Body mass index was calculated as weight (kg)/height (m) 2. Minimum waist circumference was measured using a flexible steel tape measure at the natural indentation or at a level midway between the iliac crest and the lower edge of the rib cage if no natural indentation was visible. Waist circumference was recorded to the nearest 0.5 cm, and the mean of two measures within 1 cm of each other was used. Standardized interviews were used to ascertain self-report of being on a weight loss diet at the time of either the baseline examination or the 5-year follow-up examination. Subjects were asked to report their overall perceived health status as excellent, good, fair, or poor. At each examination, dietary intake was assessed by means of a validated food frequency interview (21), and frequency of participation in vigorous activity was assessed through a validated question with a response set that ranged from rare to never to five or more times per week (22). Smoking status was assessed by standardized interview at both examinations. Statistical analyses Age-adjusted partial correlation coefficients were obtained from linear regression. Because of right-skewness of the data, data for S I, acute insulin response, disposition index, and fasting insulin were log-transformed in these analyses. Constants were added to S I (constant = 1) and acute insulin response (constant = 360) as appropriate in order to avoid negative or zero values prior to log transformation. Multiple linear regression was used to test hypotheses for the main effects (α = 0.05) of S I, acute insulin response, disposition index, fasting insulin, and glucose tolerance on the primary outcome, 5-year weight change. Because the overall model R 2 varied little based on whether the log-transformed or untransformed variables were used for S I, acute insulin response, disposition index, and fasting insulin as independent variables, the untransformed variables were

13 594 Mayer-Davis et al. TABLE 1. Characteristics of participants in the Insulin Resistance Atherosclerosis Study (n = 1,194) according to 5-year weight change, Characteristic * SD, standard deviation; S I, insulin sensitivity. Weight (kg)/height (m) 2. 5-year follow-up examination minus baseline examination. 5-year follow-up examination minus baseline examination, adjusted for body mass index. Weight change Loss of 2.27 kg (n = 299) Stable within 2.27 kg (n = 447) Gain of 2.27 kg (n = 448) Mean or no. SD* or % Mean or no. SD or % Mean or no. SD or % Weight (kg) at baseline examination Body mass index at baseline examination Weight change (kg) Waist circumference (cm) at baseline examination Waist change (cm) S I * ( 10 4 minutes 1 /µu/ml) at baseline examination Acute insulin response (µu/ml minutes) at baseline examination Disposition index (acute insulin response S I ) at baseline examination Fasting insulin level (µu/ml) at baseline examination Diabetes status at baseline examination (no.) Normal % % % Impaired glucose tolerance % % % Diabetes mellitus % % % Age (years) at baseline examination Ethnicity (no.) Non-Hispanic White % % % Hispanic % % % African-American % % % Gender (no.) Female % % % Male % % % On a weight loss diet (no.) Yes, at baseline examination only % % % Yes, at 5-year follow-up examination only % % % Yes, at both examinations 3 1.0% 4 0.9% 6 1.4% No % % % Perceived health at baseline examination (no.) Excellent % % % Good % % % Fair % % % Poor % 6 1.3% 9 2.0% retained for simplicity in model interpretation. To account for regression to the mean, these models included baseline body mass index. The models also included adjustment for potential confounding by demographic variables (age, gender, ethnicity, clinical site), baseline perceived health status, and use of oral hypoglycemic agents. In addition, the data were also adjusted for 5-year change in behavioral variables that could have an impact on energy balance: caloric intake, physical activity, alcohol intake, and smoking. In exploratory analyses of predictors of 5-year change in waist circumference, the series of models was repeated with adjustment for baseline waist circumference and, by regression, change in body mass index. Thus, in these analyses, waist change can be interpreted as change in central adiposity, independent of change in overall obesity. In order to maximize the observable range of variables in the data set and to avoid unintentional introduction of bias, we pooled all participants across glucose tolerance groups in

14 Metabolic Predictors of Weight Change 595 TABLE 2. Characteristics of participants in the Insulin Resistance Atherosclerosis Study (n = 1,194) according to baseline glucose tolerance status, Characteristic Normal glucose tolerance (n = 554) * SD, standard deviation; S I, insulin sensitivity. Weight (kg)/height (m) 2. 5-year follow-up examination minus baseline examination. 5-year follow-up examination minus baseline examination, adjusted for body mass index. Glucose tolerance status Impaired glucose tolerance (n = 274) Diabetes mellitus (n = 366) Mean or no. SD* or % Mean or no. SD or % Mean or no. SD or % Weight (kg) at baseline examination Body mass index at baseline examination Weight change (kg) Waist circumference (cm) at baseline examination Waist change (cm) S I * ( 10 4 minutes 1 /µu/ml) at baseline examination Acute insulin response (µu/ml minutes) at baseline examination Disposition index (acute insulin response S I ) at baseline examination Fasting insulin level (µu/ml) at baseline examination On a weight loss diet (no.) Yes, at baseline examination only % % % Yes, at 5-year follow-up examination only % 9 3.3% % Yes, at both examinations 1 0.2% 4 1.5% 8 2.2% No % % % the initial analyses. We added interaction terms one at a time to evaluate potential differential effects of the main effect variables on the outcomes, according to glucose tolerance, ethnicity, and gender (α = 0.05). Finally, regardless of the statistical significance of the glucose tolerance interaction term, analyses were repeated stratified according to glucose tolerance at baseline because of the a priori interest in the impact of glucose tolerance itself on weight change. RESULTS At baseline, average body mass index was The prevalences of normal weight (body mass index <25), overweight (body mass index ), and obesity (body mass index >30) were 21 percent, 44 percent, and 35 percent, respectively. At the 5-year follow-up, 25 percent of subjects had lost more than 2.27 kg (>5 pounds), 38 percent weighed within 2.27 kg of their baseline weight, and 37 percent had gained more than 2.27 kg. Additional descriptive characteristics of the participants are shown in table 1, according to category of 5-year weight change. Table 2 shows subjects characteristics according to baseline glucose tolerance status. Mean weight changes among persons with normal and impaired glucose tolerance were 1.98 kg and 1.30 kg, respectively, compared with 1.21 kg among those with diabetes. Age-adjusted Pearson correlation coefficients for correlations between the metabolic variables at baseline (including S I, acute insulin response, disposition index, and fasting insulin) and measures of baseline and follow-up obesity are given in table 3, according to glucose tolerance at baseline. As expected, among persons with normal glucose tolerance, baseline body mass index, weight, and waist circumference were strongly correlated with lower baseline S I, higher acute insulin response, lower disposition index, and higher concentrations of fasting insulin (all p s < 0.001). Crosssectional associations were similar among persons with impaired glucose tolerance. Presumably because of the restricted range of S I and acute insulin response among persons with diabetes, associations with measures of obesity were generally similar but attenuated in magnitude for this group. With regard to correlations of the metabolic variables with 5-year change in weight and waist circumference, results for persons with normal or impaired glucose tolerance were nonsignificant, with the exception of a significant inverse correlation between fasting insulin and change in waist circumference among persons with impaired glucose tolerance (r = 0.13, p < 0.05). Among persons with diabetes, S I was positively correlated (r = 0.11, p < 0.05), fasting insulin was inversely correlated (r = 0.20, p < 0.001), and disposition index was positively correlated (r = 0.11, p < 0.05) with 5-year weight change. Results were similar for 5-year change in waist circumference. We conducted the next series of analyses with the glucose tolerance groups combined in order to allow consideration of the full range of variability present in the sample. Age-

15 596 Mayer-Davis et al. TABLE 3. Age-adjusted partial correlation coefficients (r) for measures of baseline insulin resistance and insulin secretion in relation to baseline body mass index, weight, waist circumference, and change in weight and waist circumference, according to glucose tolerance status (n = 1,194), Insulin Resistance Atherosclerosis Study, Baseline metabolic measure Body mass index Baseline Weight (kg) Waist circumference (cm) Weight (kg) 5-year change Waist circumference (cm) Normal glucose tolerance S I, 0.49*** 0.37*** 0.49*** Acute insulin response 0.24*** 0.16*** 0.24*** Disposition index 0.27*** 0.23*** 0.27*** Fasting insulin level 0.48*** 0.38*** 0.46*** Impaired glucose tolerance S I 0.43*** 0.40*** 0.45*** Acute insulin response 0.14* 0.12* 0.17** Disposition index 0.24*** 0.23*** 0.23*** Fasting insulin level 0.44*** 0.37*** 0.43*** * * p < 0.05; ** p < 0.01; *** p < Weight (kg)/height (m) 2. S I, insulin sensitivity. Data for this variable were log-transformed. Type 2 diabetes mellitus S I 0.23*** 0.20*** 0.27*** 0.11* 0.12* Acute insulin response Disposition index 0.20*** 0.15*** 0.21*** 0.11* 0.12* Fasting insulin level 0.40*** 0.39*** 0.43*** 0.20*** 0.18*** adjusted correlation coefficients for S I, acute insulin response, disposition index, and fasting insulin in relation to weight change were 0.14, 0.13, 0.18, and 0.16, respectively (all p s < 0.001), suggesting increased weight gain with increased baseline S I, increased insulin secretion, and increased pancreatic functionality. For waist change, results were 0.16, 0.09, 0.16, and 0.16, respectively (all p s < 0.01). Table 4 shows results for each of the four baseline metabolic variables and glucose tolerance itself, considered separately in relation to 5-year change in weight and waist circumference, with adjustment for baseline body mass index, potential demographic confounders, and 5-year behavior change variables. Insulin resistance as measured by S I was not associated with 5-year weight change (p = 0.96). In contrast, both measures of insulin secretion (acute insulin response and disposition index) were significantly, positively related to weight change (p < 0.01), and fasting insulin was inversely related to weight change (p = 0.05). Significantly greater weight loss occurred among persons with diabetes than among those with either impaired glucose tolerance or normal glucose tolerance (p < 0.001). None of the metabolic variables were associated with waist change (independent of change in overall obesity). Before proceeding to models that considered the metabolic variables simultaneously, we added interaction terms to the models that predicted 5-year weight change (shown in table 4) in order to determine whether any of the observed effects differed significantly according to ethnicity, gender, or glucose tolerance. Of the 12 interaction terms tested, one reached statistical significance: The effect of S I on 5-year weight change appeared to differ according to glucose tolerance (p = 0.03 for the interaction term). Consistent with the unadjusted results shown in table 3, no significant relation between S I and weight change was observed among either persons with normal glucose tolerance or persons with impaired glucose tolerance; however, among persons with diabetes at baseline, higher S I was significantly associated with subsequent weight gain. Table 5 shows predicted 5-year change in weight for all metabolic variables considered simultaneously, with data stratified according to baseline diabetes status and adjusted for baseline body mass index, potential demographic confounders, and 5-year behavior change variables. None of the variables (S I, acute insulin response, fasting insulin) were significantly associated with 5-year weight change within subgroups of normal glucose tolerance (model 1), impaired glucose tolerance (model 2), or type 2 diabetes mellitus (model 3) (p > 0.15 for each).

16 Metabolic Predictors of Weight Change 597 TABLE 4. Predicted 5-year change in weight and waist circumference for each metabolic variable considered separately, adjusted for potentially confounding factors (n = 1,194), Insulin Resistance Atherosclerosis Study, Model 1 Model Change in weight (kg) per 1-SD increase Weight change 95% CI Waist circumference change Change in waist circumference (cm) per 1-SD increase 95% CI S I , , 0.17 Model 2 Acute insulin response , 0.95** , 0.36 Model 3 Disposition index , 1.24** , 0.20 Model 4 Fasting insulin level , 0.002* , 0.27 Model 5 Glucose tolerance status Normal glucose tolerance 1.05, , 0.63 Impaired glucose tolerance , 1.27*** , 0.78 Type 2 diabetes mellitus * p < 0.05; ** p < 0.01; *** p < Potentially confounding factors included baseline body mass index, age, gender, ethnicity, clinical site, perceived health, use of oral hypoglycemic agents, and behavior change (i.e., difference between baseline examination and follow-up examination for alcohol, smoking, physical activity, and dietary calories). Change in waist circumference was adjusted, by regression, for change in body mass index. The models also included baseline waist circumference. SD, standard deviation; CI, confidence interval; S I, insulin sensitivity. Reference category. TABLE 5. Predicted 5-year change in weight for all metabolic variables considered simultaneously, with data adjusted for potentially confounding factors* and stratified according to baseline glucose tolerance status (n = 1,194), Insulin Resistance Atherosclerosis Study, Model Model 1: normal glucose tolerance (n = 554) Change in weight (kg) per 1-SD increase 95% CI S I , 0.24 Acute insulin response , 0.69 Fasting insulin level , 0.67 Model 2: impaired glucose tolerance (n = 274) S I , 0.93 Acute insulin response , 1.06 Fasting insulin level , 0.16 Model 3: type 2 diabetes mellitus (n = 366) S I , 2.4 Acute insulin response , 2.72 Fasting insulin level , 0.82 * Potentially confounding factors included baseline body mass index, age, gender, ethnicity, clinical site, perceived health, and behavior change (i.e., alcohol, smoking, physical activity, and dietary calories). For the model including persons with diabetes, confounding factors included use of oral hypoglycemic agents. SD, standard deviation; CI, confidence interval; S I, insulin sensitivity.

17 598 Mayer-Davis et al. FIGURE 1. Five-year weight change according to diabetes status, before and after adjustment for metabolic and behavior change variables, Insulin Resistance Atherosclerosis Study, NGT, normal glucose tolerance; IGT, impaired glucose tolerance; DMs, diabetes mellitus. Bars, 95% confidence interval. To further address the issue of intentional weight loss beyond adjustment for reported 5-year change in behaviors likely to have an impact on weight, we repeated analyses in the subset of study participants who, by self-report, were not on a weight-loss diet at the time of either examination 1 or examination 2 (n = 1,085). The results were essentially unchanged. We further explored the impact of diabetes status. Figure 1 shows 5-year weight change according to diabetes status both before and after inclusion of the metabolic variables and behavior change variables considered in the present study. Any potential effect of these covariates on weight loss was not sufficient to explain the observed impact of diabetes status on weight change over the 5-year follow-up period. However, as evidenced by the increased width of the confidence intervals, precision was lost, presumably because of the inclusion of metabolic variables that are associated with each other both biologically and statistically. Next, persons with diabetes at baseline were stratified by severity of disease according to baseline fasting glucose level (categories: <140, , , and 250 mg/dl). Figure 2 shows a pattern of incremental weight loss across the first three categories, with weight gain being observed among those in the highest category of baseline fasting glucose even after adjustment for metabolic variables and 5-year behavior change variables (sample sizes in the categories ranged from 49 to 134). Because insulin therapy or use of most oral hypoglycemic agents can promote weight gain, the final model excluded persons who initiated use of insulin or oral hypoglycemic agents during the follow-up period (sample sizes in the categories ranged from 25 to 96). This essentially explained the weight gain in the most severe disease category (baseline fasting glucose level 250 mg/dl). DISCUSSION These findings show that the diabetic state, characterized by relatively low S I, low acute insulin response, low disposition index, and (for persons with type 2 diabetes who were not taking insulin) high fasting insulin concentrations, was by far the strongest predictor of 5-year weight loss in this cohort. None of the metabolic variables related to glucose and insulin metabolism explained the impact of having diabetes on subsequent weight loss. Furthermore, among persons with either normal glucose tolerance or impaired glucose tolerance, none of the measures of insulin resistance, insulin secretion, pancreatic function as assessed by disposition index, or fasting insulin concentration were predictive of 5-year weight change. Because the pathways that determine observed levels of these metabolic measures at any given point in time are highly dependent on one another as a function of biologic feedback systems, it is not surprising that various studies reported one or another of these variables to be associated with weight change, somewhat inconsistently. The present analysis differed from results previously published in the direct consideration of glucose tolerance itself. It is possible that the variables under study in previous reports (e.g., fasting insulin level) were actually reflecting risk for development of the diabetic state, rather than any single mechanism related to glucose or insulin metabolism. It is clear that increasing levels of obesity are strongly related to increased risk of diabetes, and evidence is accumulating that sustained, intentional weight loss may reduce diabetes risk (23 25). Recent findings from the Diabetes Prevention Program confirm the efficacy of moderate weight loss through lifestyle change in reducing risk for develop-

18 Metabolic Predictors of Weight Change 599 FIGURE 2. Five-year weight change according to diabetes status, before and after adjustment for metabolic and behavior change variables, excluding persons taking insulin at visit 2, Insulin Resistance Atherosclerosis Study, NGT, normal glucose tolerance; IGT, impaired glucose tolerance; DMs, diabetes mellitus; FPG, fasting plasma glucose (mg/dl); OHGA, oral hypoglycemic agents. Bars, 95% confidence interval. ment of type 2 diabetes (26). However, this does not preclude the potential for a natural history of diabetes in which weight loss ensues as a result of the impaired fat and carbohydrate metabolism that is central to the disease. In 1981, Knowler et al. (27) reported similar prospective data from the Pima Indian cohort in which weight loss was observed over the 2 years following diabetes diagnosis, despite the observation of weight gain in the 2 4 years prior to diagnosis among those aged years. It is possible that simply feeling unwell near the time of clinical diagnosis of diabetes may contribute to weight loss. Perhaps more likely, in addition to inefficient fuel storage, weight loss may occur with diabetes because of the loss of calories that may occur with glucosuria. Figure 2 is consistent with this notion in that increasing baseline fasting glucose concentration was associated with greater subsequent 5-year weight loss. The exception to this observation occurred among persons with a fasting glucose level 250 mg/dl, in whom weight gain was observed. However, initiation of insulin therapy or oral hypoglycemic agents explained the observed weight gain. Regarding the question of whether increased insulin secretion may lead to reduced weight gain through central nervous system signaling of the fed state (i.e., via feelings of satiety), the inverse association of fasting insulin concentration with weight change in table 4 might suggest that this may indeed be the case. However, results for acute insulin response and disposition index suggest just the opposite, namely, an association of increased insulin secretion with increased weight gain. Fasting insulin concentration (analyzed in the absence of other metabolic variables) may reflect insulin resistance more than insulin secretion. Again, final results as shown in table 5 and figures 1 and 2 suggest that any direct impact of individually measured aspects of this complex biologic system may be relatively unimportant in comparison with the confluence of the underlying processes when diabetes develops, leading to otherwise unexplained weight loss. Individuals with a diagnosis of type 2 diabetes are commonly advised to lose weight, which raised a concern that at least some of the observed weight loss might have been due to intentional weight loss. We addressed this through two analytical approaches. First, because increased caloric intake, decreased physical activity, change in smoking habits, or change in alcohol intake over the 5-year follow-up period would be expected to contribute to weight change, we calculated change in these variables and included it in the regression analyses. Second, in recognition of the imprecision inherent in assessment of these behavioral variables despite use of validated instruments, we restricted analyses to persons who reported that they were not on a weight loss diet, either at the first examination or at the second examination, and results were unchanged. Although neither of these two approaches can entirely rule out mixing of intentional and unintentional weight loss in the final results, it seems unlikely that all of the observed effects would be due to intentional weight loss. The question of metabolic predictors of change in waist circumference was motivated by previously published IRAS data that showed a positive relation between waist circumference and measures of insulin secretion (28), independent of body mass index. However, as indicated above in the Materials and Methods section, analyses of metabolic

19 600 Mayer-Davis et al. predictors of change in waist circumference should be viewed as exploratory. This is because the IRAS protocol did not include measures of central adiposity beyond simple anthropometric measurements and because previous work focused on energy balance, not on fat distribution. Initial analyses (table 3) showed quite similar findings for weight and waist circumference, presumably because waist circumference provides information about overall obesity as well as visceral fat (29). In the present study, the age-adjusted partial correlation coefficient for baseline body mass index and baseline waist circumference was 0.78 for those with normal glucose tolerance or impaired glucose tolerance and 0.81 for those with diabetes. In order to distinguish between associations with overall adiposity and associations with central adiposity, we used residual values from regression analysis as a measure of central adiposity (i.e., waist circumference, adjusted for body mass index) in subsequent modeling. Overall, these adjusted results (table 4) suggested no association between any of the metabolic variables (including glucose tolerance status) and 5-year change in body mass index-adjusted waist circumference. The residual method has commonly been used with variables that are strongly correlated with each other (e.g., dietary fat intake adjusted for total calorie intake by regression); however, the method is subject to fairly stringent distributional requirements, and results can be difficult to interpret from a clinical perspective (30). Because waist circumference explains variance in visceral fat measured by magnetic resonance imaging to a greater degree than does body mass index (29), future studies may be warranted to further explore the hypothesis of an effect of metabolic predictors on change in central adiposity, using both anthropometric measures and more precise measures such as magnetic resonance imaging. Overall, given the results of the published studies to date, it is possible that insulin resistance or insulin secretion may play a subtle role in determination of energy balance over time in nondiabetic persons. However, other influences (such as socially or culturally determined dietary choices and physical activity) are likely to be far more important to long-term weight change, unless alterations in glucose and insulin metabolism are sufficiently severe that diabetes ensues. Further study is warranted to determine the natural history of weight change and its attendant metabolic status among persons with diagnosed diabetes, including both persons who intentionally lose weight and those who do not. ACKNOWLEDGMENTS This study was supported by grants UO1-HL47887, UO1- HL47889, UO1-HL47890, UO1-HL47892, UO , and DK29867 from the National Heart, Lung, and Blood Institute. REFERENCES 1. Swinburn BA, Nyomba BL, Saad MF, et al. Insulin resistance associated with lower rates of weight gain in Pima Indians. J Clin Invest 1991;88: Hoag S, Marshall JA, Jones RH, et al. High fasting insulin levels associated with lower rates of weight gain in persons with normal glucose tolerance: The San Luis Valley Diabetes Study. Int J Obes Relat Metab Disord 1995;19: Valdez R, Mitchell BD, Haffner SM, et al. Predictors of weight change in a bi-ethnic population: The San Antonio Heart Study. Int J Obes Relat Metab Disord 1994;18: Wedick NM, Mayer-Davis EJ, Wingard DL, et al. Insulin resistance precedes weight loss in adults without diabetes: The Rancho Bernardo Study. Am J Epidemiol 2001;153: Hodge AM, Dowse GK, Alberti KG, et al. Relationship of insulin resistance to weight gain in nondiabetic Asian Indian, Creole, and Chinese Mauritians. Metabolism 1996;45: Odeleye OE, de Courten M, Pettitt DJ, et al. Fasting hyperinsulinemia is a predictor of increased body weight gain and obesity in Pima Indian children. Diabetes 1997;46: Gould A, Williams D, Byrne C, et al. Prospective cohort study of the relationship of markers of insulin resistance and secretion with weight gain and changes in regional adiposity. Int J Obes Relat Metab Disord 1999;23: Zavaroni I, Zuccarelli A, Gasparini P, et al. Can weight gain in healthy, nonobese volunteers be predicted by differences in baseline plasma insulin concentration? J Clin Endocrinol Metab 1998;83: McLaughlin T, Abbasi F, Carantoni M, et al. Differences in insulin resistance do not predict weight loss in response to hypocaloric diets in healthy obese women. J Clin Endocrinol Metab 1999;84: Folsom AR, Vitelli LL, Lewis CE, et al. Is fasting insulin concentration inversely associated with rate of weight gain? Contrasting findings from the CARDIA and ARIC study cohorts. Int J Obes 1998;22: Lazarus R, Sparrow D, Weiss S. Temporal relations between obesity and insulin: longitudinal data from the Normative Aging Study. Am J Epidemiol 1998;147: Porte DJ, Seeley R, Woods S, et al. Obesity, diabetes, and the central nervous system. Diabetologia 1998;41: Schwartz MW, Boyko EJ, Kahn SE, et al. Reduced insulin secretion: an independent predictor of body weight gain. J Clin Endocrinol Metab 1995;80: Sigal RJ, el-hashimy M, Martin BC, et al. Acute postchallenge hyperinsulinemia predicts weight gain: a prospective study. Diabetes 1997;46: Wagenknecht LE, Mayer EJ, Rewers M, et al. The Insulin Resistance Atherosclerosis Study (IRAS): objectives, design, and recruitment results. Ann Epidemiol 1995;5: World Health Organization. Diabetes mellitus: report of a WHO study group. (WHO Technical Report Series, no. 727). Geneva, Switzerland: World Health Organization, Saad MF, Anderson RL, Laws A, et al. A comparison between the minimal model and the glucose clamp in the assessment of insulin sensitivity across the spectrum of glucose tolerance. Diabetes 1994;43: Anderson RL, Hamman RF, Savage PJ, et al. Exploration of simple insulin sensitivity measures derived from frequently sampled intravenous glucose tolerance (FSIGT) tests: The Insulin Resistance Atherosclerosis Study. Am J Epidemiol 1995; 142: Elbein SC, Wegner K, Kahn SE. Reduced β-cell compensation to the insulin resistance associated with obesity in members of Caucasian familial type 2 diabetic kindred. Diabetes Care 2000; 23: Mayer-Davis EJ, Bell RA, D Agostino RJ. Moderate alcohol consumption in relation to higher insulin sensitivity: The Insulin Resistance and Atherosclerosis Study (IRAS). Diabetes

20 Metabolic Predictors of Weight Change ;47(suppl 1):A Mayer-Davis EJ, Vitolins MZ, Carmichael S, et al. Validity and reproducibility of a food frequency interview in a multi-cultural epidemiologic study. Ann Epidemiol 1999;9: Mayer-Davis EJ, D Agostino RJ, Karter AJ, et al. Intensity and amount of physical activity in relation to insulin sensitivity. JAMA 1998;279: Pan X, Li G, Hu Y, et al. Effects of diet and exercise in preventing NIDDM in people with impaired glucose tolerance. Diabetes Care 1997;20: Moore L, Visioni A, Wilson P, et al. Can sustained weight loss in overweight individuals reduce the risk of diabetes mellitus? Epidemiology 2000;3: Tuomilehto J, Lindstrom J, Eriksson JG, et al. Prevention of type 2 diabetes mellitus by changes in lifestyle among subjects with impaired glucose tolerance. N Engl J Med 2001;344: Knowler WC, Barrett-Connor E, Fowler SE, et al. Reduction in incidence of type 2 diabetes with lifestyle intervention or metformin. Diabetes Prevention Program Research Group. N Engl J Med 2002;346: Knowler WC, Pettitt DJ, Savage PJ, et al. Diabetes incidence in Pima Indians: contributions of obesity and parental diabetes. Am J Epidemiol 1981;113: Mayer-Davis EJ, Levin S, Bergman RN, et al. Insulin secretion, obesity, and potential behavioral influences: results from the Insulin Resistance Atherosclerosis Study (IRAS). Diabetes Metab Res Rev 2001;17: Janssen I, Heymsfield SB, Allison DB, et al. Body mass index and waist circumference independently contribute to the prediction of nonabdominal, abdominal subcutaneous, and visceral fat. Am J Clin Nutr 2002;75: Willet W, Stampfer M. Implications of total energy intake for epidemiologic analyses. In: Nutritional epidemiology. 2nd ed. New York, NY: Oxford University Press, 1998: