Prospective Investigation of Metabolic Characteristics in Relation to Weight Gain in Older Adults: The Hoorn Study
The objective of this investigation was to determine the relation between baseline glucose, insulin, adiponectin, and leptin levels and subsequent 6-year weight and waist change in older men and women without diabetes in a prospective cohort study. Participants were 1,198 Dutch men and women without diabetes who were aged 50–77 years when baseline metabolic and anthropometric measurements were evaluated (1989–1991). Approximately 6 years later, body weight and waist circumference were re-measured at a follow-up examination (1996–1998). Metabolic variables (fasting plasma glucose, 2-h postchallenge plasma glucose, homeostasis model assessment of insulin resistance (HOMA-IR), adiponectin, and leptin) were evaluated as predictors of changes in weight and waist circumference. Postchallenge plasma glucose (mmol/l) significantly predicted less gain in both weight and waist circumference (β = −0.28 kg, s.e. = 0.11; β = −0.31 cm, s.e. = 0.14, respectively) during follow-up. Leptin (µg/l) significantly predicted greater increases in weight (β = 0.29 kg, s.e. = 0.07) and waist (β = 0.16 cm, s.e. = 0.08) among men and in waist among women (β = 0.06 cm, s.e. = 0.02). Fasting plasma glucose (mmol/l) predicted an increase in waist among women (β = 1.59 cm, s.e. = 0.63), but not in men (β = −0.74 cm, s.e. = 0.55). Adiponectin and insulin did not predict weight or waist change. The authors conclude that lower postchallenge plasma glucose and higher fasting leptin levels significantly predicted long-term increases in weight and waist circumference. In contrast, measures of insulin resistance and adiponectin were not associated with weight change in this cohort of older persons without diabetes.
Rates of obesity are soaring in the United States and worldwide (1,2), and more insight into the etiology of excess weight gain is urgently needed. The physiologic mechanisms by which energy balance is maintained in humans involve a complex web of feedback loops and networks under neuroendocrine regulation (3). With respect to regulation of long-term weight change, the role of glucose homeostasis is presently unresolved. Few studies have examined glucose levels after an oral glucose tolerance test in relation to weight change, and recent evidence suggests that higher postchallenge glucose response is associated with less weight gain (4,5). Increased postchallenge glucose may inhibit energy intake by delaying gastric emptying, thereby prolonging satiety and increasing the time between meals (6,7). Leptin is an adipokine which has been extensively examined in animal experiments and human observational studies to determine the relationship with appetite regulation and long-term energy balance (8). Findings with respect to long-term weight change in humans have been mixed with higher leptin levels being associated with more weight gain in some studies (9,10,11,12,13,14,15,16) and less weight gain in others (17,18,19,20). Adiponectin is another adipokine suspected to have involvement in energy homeostasis via peripheral and central nervous system actions. In animal models, administration of adiponectin increased fatty acid oxidation in muscle tissue and reduced body weight gain when mice were fed a high-fat/sucrose diet (21). However, adiponectin administered under fasting conditions appeared to increase adenosine monophosphate-activated protein kinase in the hypothalamus stimulating food intake (22). In humans, few studies on adiponectin and long-term weight change have been conducted (23,24). We therefore conducted a prospective study to examine fasting and postload glucose, insulin, leptin, and adiponectin concentrations in relation to weight and waist changes in an older Dutch population. Waist change was included as an outcome in this analysis as waist may be a more sensitive measure of body composition change in older persons because this measure is less affected by age-related changes in body composition (i.e., shrinkage and loss of lean body mass).
Methods and Procedures
Study population and design
The Hoorn Study design and population has previously been described in detail (25). Briefly, a population-based cohort study was conducted in Hoorn, Netherlands in older Dutch men and women from October 1989 to December 1991. Adults invited to participate were from a random sample of inhabitants listed in the municipal registry aged ≥50 years. Approximately 71% persons participated, leaving an original cohort size of 2,484. A follow-up examination was conducted from January 1996 to December 1998 at which 1,513 participants who were alive, living locally, and willing and able to participate were re-examined. The average length of follow-up was ∼6.4 years (interquartile range 6.1–6.7). The Ethical Review Committee of VU Medical Center approved the Hoorn Study, and all participants gave their written informed consent.
For this prospective analysis, there were 1,487 persons with anthropometric measurements taken at both baseline and follow-up examinations. Of these, we excluded subjects with known (n = 40) or newly diagnosed (n = 90) diabetes at baseline. We also excluded subjects developing diabetes over the follow-up period (n = 132) because weight loss is one of the classic symptoms of diabetes, and diabetes medication as well as dietary advice after diabetes diagnosis may affect weight and thus bias the results. An additional 27 subjects were excluded with missing data for potential confounders, leaving 1,198 participants (535 men and 663 women) for the present analysis.
Comparing baseline characteristics from the original cohort of 2,484 adults to this study population (n = 1,198), there were no appreciable differences between these samples for age, anthropometric measurements, alcohol consumption, cigarette smoking status, or physical activity. Participants in the initial sample had higher levels of glucose, insulin, more hypertension, and cardiovascular disease than our analytical sample. However, when we restricted the original cohort to only nondiabetic persons, we found that there were no appreciable differences between populations for any of these variables with the exception of slightly higher prevalence of cardiovascular disease and hypertension.
Blood samples were collected after an overnight fast of at least 10 h before and at 2 h after a 75-g oral glucose tolerance test was administered. Fasting and 2-h glucose levels were measured by a glucose dehydrogenase method (Merck, Darmstadt, Germany). Insulin was measured by double-antibody radioimmunoassay (Linco Research, St. Louis, MO). Adiponectin was measured by a latex particle-enhanced turbidimetric immunoassay (Otsuka Pharmaceutical, Osaka, Japan). Leptin levels were measured by radioimmunoassay previously described in detail (26). Insulin resistance was estimated using the homeostasis model assessment for insulin resistance (HOMA-IR: (fasting insulin (µU/ml)) × fasting glucose (mmol/l))/22.5) (ref. 27).
Weight and height were measured with participants wearing light clothing and no shoes. BMI was calculated as weight (kg) divided by height squared (m2). Waist circumference was measured in centimeters at the midway point between the lowest rib and the iliac crest. Weight change was determined by subtracting the participant's weight (kg) at baseline from weight at the follow-up visit. Waist change (cm) was calculated similarly.
Participants completed a questionnaire at the baseline visit as described previously (25). Cigarette smoking and alcohol consumption were assessed; alcohol consumption was reported in glasses per week and converted to grams (g) per day. Physical activity was assessed with a modified questionnaire used in the Zutphen Elderly Study (28), and a score calculated from 0 to 9, representing the range of activity from low to high intensity, respectively. Information on prevalent cardiovascular disease was determined by a questionnaire based on a translated version of the Rose questionnaire (29). Hypertension was defined as having an elevated blood pressure (systolic blood pressure ≥160 mm Hg or diastolic blood pressure ≥95 mm Hg) or using antihypertensive medications (30).
Cross-sectional Pearson correlation coefficients (ρ) were calculated for BMI and waist with baseline metabolic measurements (i.e., fasting plasma glucose, 2-h glucose, HOMA-IR, adiponectin, leptin), adjusted for age.
For prospective analysis, general linear models were used to evaluate weight or waist change regressed on each of the metabolic markers. The first model adjusted for age and baseline BMI (or waist in the model using waist change as the dependent variable). The second model additionally adjusted for confounding by physical activity score (range 0–9), categories of alcohol consumption (none, ≤10, 10–30, or ≥30 g/day), and cigarette smoking (never, past, or current). A final model was evaluated entering all metabolic variables into the model simultaneously, controlling for the aforementioned covariates. Continuous variables were examined for normality by Shapiro–Wilk test; no variables required transformation in the present analyses. Categorical measurements were entered into the model as indicator variables to avoid assumptions of a linear trend. The lowest level of the categorical variable was used as the reference category. Effect modification by covariates was evaluated by entering multiplicative interaction terms into the multivariable models. All regression models were initially stratified by sex, but significant interactions were only present for leptin and fasting glucose. Therefore, results are presented for men and women combined for the other analyses. We also evaluated weight change according to quintiles of the metabolic variables to investigate potential nonlinear trends. Outliers, defined in our analysis as ±3 times the interquartile range, were determined for each of the metabolic variables. In a sensitivity analysis where outlying observations were excluded, regression coefficients changed <10% and we therefore only presented results without exclusion of outliers.
All presented P values are two-sided. Statistical significance was defined at an alpha level of 0.05, including the assessment of significant interaction terms. The Statistical Analysis System version 9.1 was used for all analyses (SAS Institute, Cary, NC).
Cross-sectional associations at baseline
The mean age for this cohort was 60 years (range 50–77 years). Fifty-five percent of the participants were women, among whom 87% was postmenopausal. As shown in Table 1, there were no appreciable differences between men and women for any of the metabolic biomarkers, with the exception of leptin. At baseline, leptin levels were at least fourfold higher in women as compared with men. About one-third of the population reported current smoking, with more men being current smokers compared to women.
Table 1. Characteristics of study population of nondiabetic men and women (Hoorn Study: baseline visit 1989–1991, follow-up 1996–1998)
All metabolic variables, except adiponectin, were significantly positively correlated with baseline BMI and baseline waist circumference in the cross-sectional analysis (Table 2). Adiponectin was significantly inversely correlated with baseline BMI and baseline waist circumference. Generally, waist circumference was more strongly correlated the metabolic variables than BMI in this cohort of older men and women.
Table 2. Cross-sectional age-adjusted correlation coefficients for baseline BMI and waist circumference with baseline metabolic measurements (Hoorn Study: baseline visit 1989–1991)
As shown in Table 1, the mean weight change during follow-up was ∼1 kg gain (interquartile range: −1.5 to 3.8 kg), and the mean waist change ∼3 cm increase (interquartile range: −0.6 to 6.8 cm). Women had a significantly greater increase in waist circumference over follow-up compared to men (difference 0.95 cm, P < 0.01). Significant interactions with sex were found for leptin and fasting plasma glucose in relation to weight change; thus, results are presented for men and women separately for these two variables.
Table 3 shows the regression coefficients for the association between baseline metabolic measurements and subsequent weight change. We first adjusted for age and baseline BMI, and found that only leptin among men was a significant predictor for weight change during follow-up. Except for a slight attenuation of the regression coefficient for leptin, results did not substantially change in multivariable-adjusted models. In a final model, all metabolic predictors were simultaneously included in the multivariable model. Two-hour glucose significantly predicted less gain in weight during follow-up. Leptin predicted weight gain in men, but this association was not significant in women. There was no substantial association with weight change for fasting glucose, HOMA-IR, and adiponectin.
Table 3. Regression coefficients (β) for weight change (kg) regressed on each baseline metabolic measurement (Hoorn Study: baseline 1989–1991, follow-up 1996–1998)
Table 4 shows the regression coefficients for the association between metabolic measurements and change in waist circumference. In all models, fasting glucose predicted an increase in waist circumference for women, but not for men. Similar to the results for weight change, in the final multivariable model, higher 2-h glucose predicted less gain in waist circumference. Leptin significantly predicted an increase in waist circumference in both men and women. HOMA-IR and adiponectin did not predict a change in waist circumference during follow-up.
Table 4. Regression coefficients (β) for waist change (cm) regressed on each baseline metabolic measurement (Hoorn Study: baseline 1989–1991, follow-up 1996–1998)
Fasting insulin was highly correlated to HOMA-IR (ρ = 0.98) and, similar to HOMA-IR, was not appreciably associated with changes in weight or waist circumference (data not shown). Postload insulin concentration was also evaluated in a subsample for which we had measurements (n = 221). We found no significant associations with either weight or waist change, possibly due to limited power of a small sample size (data not shown). When we evaluated weight and waist change according to quintiles of the metabolic variables, there were no trends observed to suggest nonlinear associations.
We conducted a sensitivity analysis to determine the effect of major weight loss due to poor health. We removed persons losing ≥5 kg during follow-up (n = 93) from multivariable-adjusted models. To address potential selection bias, we performed a sensitivity analysis where persons who developed diabetes by the time of the follow-up examination (n = 132) were included in the analysis. Finally, we explored the magnitude of not adjusting for baseline BMI or waist. For each of these sensitivity analyses, results were not materially different as regression coefficients were of similar magnitude and in no instance changed direction.
High postchallenge glucose predicted less gain in body weight and waist circumference over 6 years in this large population-based cohort of older men and women without diabetes. In contrast, leptin concentrations predicted greater weight gain. Adiponectin and measures of insulin resistance did not predict weight or waist change. This study addresses the lack of adequate evidence from prospective studies that simultaneously consider several metabolic predictors to help elucidate the complex role of glucose in regulating long-term body weight (8).
Our findings for 2-h glucose and weight change are consistent with recent results from two prospective studies (4,5). In a study of 253 Pima Indian normal glucose tolerant adults aged 18–44 years, higher glucose response marked by area under the curve during an oral glucose tolerance test was associated with less weight gain over an ∼7 years of follow-up (4). Similarly, in an analysis of 258 participants from the Quebec Family Study aged 20–65 years, 2-h postload glucose concentrations were significantly inversely correlated with subsequent weight change (5). The San Antonio Heart Study found that 2-h glucose was significantly associated with less weight gain over 8 years, but this association was no longer significant when baseline BMI was included in the model (31). As proposed by Jean Mayer in the 1950s, the classical glucostatic theory postulates that the glucose level in the brain acts as a satiety signal and that decreasing levels can trigger meal initiation (32,33). A sustained level of postprandial hyperglycemia may thus inhibit feeding through prolonged satiety (32,33,34). In addition to central nervous system appetite signaling, gastrointestinal motility may also be affected by postload glucose levels. Increased levels are associated with delayed gastric emptying accompanied by gastric distension and with an increased perception of gut stimuli (7). Consistent with our observations, three previous prospective studies did not find a significant association between fasting glucose and weight change (4,31,35). Although we found that fasting plasma glucose predicted an increase in waist circumference among women, this finding requires confirmation from further studies.
Leptin is an adipokine implicated in the relationship between appetite signaling and long-term weight regulation (8,36). Our findings for leptin are consistent with previous studies which have reported an association between higher leptin levels and weight gain (9,10,11,12,13,14,15,16). However, four studies found that increased leptin levels were associated with weight loss (17,18,19,20), and five studies found no substantial association (23,37,38,39,40). It has been hypothesized that an increasing leptin level is a marker for leptin resistance (41). Depending on how chronic the positive energy balance is, or what the degree of adipose tissue excess is, leptin may have orexogenic or anorexogenic actions (8). For example, in overweight persons, increased leptin levels may predict further weight gain as fat continues to accumulate (8). In contrast, in persons with chronic obesity, increased leptin levels may predict less weight gain marking resistance to leptin action (8). Adiponectin, also an adipokine, has been observed in fasting conditions to increase food intake and reduce energy expenditure in animal experiments via stimulation of adenosine monophosphate-activated protein kinase in the hypothalamus (22). However, adiponectin did not predict weight change in this cohort of older adult men and women or in other prospective observational cohort studies (23,24).
Insulin has anabolic peripheral effects in skeletal and adipose tissues as well as catabolic effects in the central nervous system (42,43). While it has been suggested that hyperinsulinemia promotes fat storage through peripheral actions, concurrent catabolic actions of insulin in the central nervous system may decrease hunger and thereby limit food intake (42). To date, there are 16 prospective investigations of measures of insulin resistance and weight change with insulin resistance predicting less weight gain (31,35,44,45,46,47), more weight gain (48,49), or having no clear association with weight change (35,50,51,52,53,54,55,56). Insulin resistance as assessed by HOMA-IR or fasting insulin did not predict weight or waist change in our study. Discrepant findings could be due to genuine differences between study populations or methodological differences such as the assessment of insulin resistance and duration of follow-up.
Our study had several limitations. First, the potential for selection bias has to be considered. When we compared baseline characteristics from the original cohort to the current study population we found a slightly higher prevalence of cardiovascular disease and hypertension in the original cohort. However, in all multivariable models, both heart disease and hypertension were evaluated as potential confounders. Selection bias could also be present as we excluded participants with diabetes that occurred during follow-up from the analysis. However, in a sensitivity analysis where we did not exclude persons with incident diabetes, results did not materially change. Second, we only had crude markers for insulin resistance and only one postchallenge glucose concentration. Future prospective studies of predictors of weight gain including more detailed measures of glucose homeostasis such as the hyperinsulinemic euglycemic clamp would be of interest. Third, we considered potential confounders in detail, but residual confounding by unknown or imperfectly measured factors cannot be completely excluded.
This study contributes to the current body of knowledge on the complex role of glucose in regulating long-term weight change. Our results suggest that high-leptin concentrations and low-postchallenge glucose concentrations are important predictors of long-term weight gain in the general population. Further studies on the role of these metabolic factors as determinants of energy balance may provide clues for interventions to reduce weight gain.
The Hoorn Study has been made possible by VU University Amsterdam, the VU University Medical Center, the Dutch Diabetes Research Foundation, the Dutch Organization for Scientific Research, the Netherlands Heart Foundation, and the Health Research and Development Council of the Netherlands.
The authors declared no conflict of interest.