Evidence from epidemiologic studies that central obesity precedes future metabolic change and does not occur concurrently with the appearance of the blood pressure, glucose, and lipid abnormalities that characterize the metabolic syndrome (MetS) has been lacking. Longitudinal surveys were conducted in Mauritius in 1987, 1992, and 1998, and in Australia in 2000 and 2005 (AusDiab). This analysis included men and women (aged ≥25 years) in three cohorts: AusDiab 2000–2005 (n = 5,039), Mauritius 1987–1992 (n = 2,849), and Mauritius 1987–1998 (n = 1,999). MetS components included waist circumference, systolic blood pressure, fasting and 2-h postload plasma glucose, high-density lipoprotein (HDL) cholesterol, triglycerides, and homeostasis model assessment of insulin sensitivity (HOMA-S) (representing insulin sensitivity). Linear regression was used to determine which baseline components predicted deterioration in other MetS components over 5 years in AusDiab and 5 and 11 years in Mauritius, adjusted for age, sex, and ethnic group. Baseline waist circumference predicted deterioration (P < 0.01) in four of the other six MetS variables tested in AusDiab, five of six in Mauritius 1987–1992, and four of six in Mauritius 1987–1998. In contrast, an increase in waist circumference between baseline and follow-up was only predicted by insulin sensitivity (HOMA-S) at baseline, and only in one of the three cohorts. These results suggest that central obesity plays a central role in the development of the MetS and appears to precede the appearance of the other MetS components.
Substantial evidence exists regarding the important role played by obesity in the development of the metabolic syndrome (MetS), with obesity being included as a required prerequisite in the most recent definition of MetS by the International Diabetes Federation. Indeed, both cross-sectional and longitudinal factor analyses have implicated obesity as a central feature of this multifaceted condition (1,2). However, whether obesity precedes future metabolic change or occurs concurrently with the deterioration in blood pressure, glucose levels, and lipid abnormalities that characterize the MetS is an important question that has remained unclear. The implications of this question for the MetS relate to whether the importance of obesity is merely as one of several components in clinical definitions, or whether it is actually most important before the development of other MetS abnormalities and when the best opportunity for prevention exists.
The few studies that have used longitudinal data to examine the temporal relationships between the components of the MetS have all implicated baseline obesity, the increase in obesity between baseline and follow-up or insulin resistance as being the strongest predictors of deterioration in MetS components or the progression to diabetes (2,3,4,5,6,7,8). None of these studies, however, has examined whether the pre-existing central obesity that so often accumulates over the life course precedes deterioration in the other MetS components.
The theoretical model on which this research is based is that if several of the measured MetS components are similarly related to an unmeasured underlying cause, it would be expected that they would develop concurrently, and that in some people one of these components would appear first, whereas in others another component would be the first abnormality to develop. If, however, one of the MetS components (e.g., central obesity) was itself the underlying cause, it would be expected that this component should predict deterioration in the other components, but not vice versa. By examining the relationships between all MetS components at baseline and change in all other MetS components between baseline and follow-up, an assessment of the validity of these two scenarios is possible.
In this analysis, we use data from two large, national, longitudinal studies (the Australian Diabetes, Obesity and Lifestyle study (AusDiab) and the Mauritius Non-Communicable Diseases Study) to explicitly assess using the theoretical model described, whether central obesity precedes subsequent deterioration in each of the other elements of the MetS.
Methods and Procedures
The study methods and response for both the AusDiab and Mauritius surveys have been described in detail elsewhere (9,10,11,12). In brief, the AusDiab study was a population-based survey of 11,247 adults (5,049 men and 6,198 women), aged ≥25 years in 1999–2000 (response was 55.3% of those completing a household interview). A stratified cluster sample was drawn from 42 randomly selected Census Collector Districts across Australia (six in each of the States and the Northern Territory). Five years later in 2004 and 2005, 6,400 (57%) participants were resurveyed. A comparison of the profile of responders vs. nonresponders to the AusDiab survey has been published previously (13).
In Mauritius, all persons >24 years of age living in selected areas were invited to attend a survey in 1987. The response was 80% (n = 5,083). Of these, 74.2% (n = 3,771) were followed up in 1992 and 55.1% (n = 2,802) were followed up in 1998.
For both the AusDiab and Mauritius surveys, a 75-g oral glucose-tolerance test was performed on all nonpregnant participants, except those taking insulin or oral hypoglycaemic drugs, at all surveys. Biochemical measurements, height, weight, hip and waist circumference, and blood pressure for all surveys were measured as previously described (13,14,15). Elements of the MetS used in this analysis included the five proposed in each of the International Diabetes Federation and National Cholesterol Education Program-Adult Treatment Panel III definitions (waist circumference, triglyceride and high-density lipoprotein (HDL) cholesterol concentration, glucose, and blood pressure) as well as insulin sensitivity due to its close links to the MetS and the possibility that it is an underlying cause. Systolic blood pressure was chosen to represent hypertension (results were similar when diastolic blood pressure was used—data not shown), and both fasting and 2-h postload plasma glucose were included to represent the glucose component. Diabetes was classified according to the World Health Organization criteria and those with diabetes at baseline or who developed diabetes between baseline and follow-up, as well as those on antihypertensive agents, were excluded from all analyses (16). Women who were pregnant at any survey were excluded. In AusDiab only (because this information was not available for Mauritius), participants reporting nonskin cancers during any of the annual follow-up surveys were excluded from all analyses. After exclusions, the numbers available for analysis were Mauritius 1987–1992, n = 2,849; Mauritius 1987–1998, n = 1,999 (note that 1,807 individuals were common to both cohorts); and AusDiab 2000–2005, n = 5,039.
The AusDiab survey protocols were approved by the ethics committee of the International Diabetes Institute and Monash University's Standing Committee on Ethics in Research involving Humans. The Mauritius survey protocols were reviewed and approved by the Alfred Healthcare Group Ethics Committee (Melbourne, Australia) as well as the Ministry of Health, Mauritius. Informed consent was obtained from all participants of both surveys.
Statistical analysis was conducted using Stata 9.0 (STATA, College Station, TX). Means (±s.d.) or proportions of various physical and demographic characteristics were calculated for those attending baseline and follow-up studies. The technique used to assess temporal order was a cross-lagged panel design, where the theoretical model is that the probable cause (obesity) of a cluster of conditions is not itself caused by these conditions. Therefore the probable cause will predict the development of the conditions, but the conditions will not predict the development of the probable cause. Separate univariate linear regression models were used to assess the association between individual baseline MetS variables and the change in each of the other MetS variables, respectively, between baseline and follow-up (each model also included the dependent variable measured at baseline), with standardized regression coefficients reported. Further multivariate analysis included all six MetS components as covariates in these models. The mean variance inflation factor for these multivariable models was calculated (in AusDiab, mean variance inflation factor = 1.46, in Mauritius, mean variance inflation factor = 1.34), indicating an absence of colinearity in the models. Analysis of change in parameters between baseline and follow-up was based on the model suggested by Vickers and Altman (17). It should be noted that for all regression analyses, the glucose, waist circumference, lipid, homeostasis model assessment of insulin sensitivity (HOMA-S), and systolic blood pressure variables were entered as continuous variables, and not as dichotomous variables as are used in clinical definitions of the MetS. To account for the possibility that significant results may simple be due to multiple comparisons, P < 0.01 was considered significant. All analyses were adjusted for age, sex, and (in Mauritius) ethnic group. It is acknowledged that some of the relationships between different MetS variables presented are unlikely, for instance, blood pressure predicting change in HDL cholesterol and triglycerides. These relationships are included to emphasize the difference between these relationships and those involving more plausible predictors such as obesity and insulin sensitivity, and for completeness. If the relationships seen for plausible predictors and change in the other MetS variables were due to chance, or because of the nature of the variables themselves, similar relationships would also be expected for implausible associations.
The baseline characteristics for those who attended both the baseline and follow-up AusDiab (2000 and 2005) and Mauritius (1987 and 1992; 1987 and 1992 and 1998) surveys are presented in Table 1. At baseline, subjects in Australia compared to subjects in Mauritius were on average older, having higher BMI, waist circumference, and HOMA-S, and having lower diastolic blood pressure and 2-h postload plasma glucose.
Table 1. Baseline and follow-up characteristics of participants in 2000 and 2005 (AusDiab), 1987 and 1992 (Mauritius), and 1987 and 1998 (Mauritius)
Impact of individual baseline MetS components on change in other MetS components over 5 and 11 years
The univariate associations between each of the individual MetS components at baseline with change in other MetS components between baseline and follow-up in Mauritius (1987–1992 and 1987–1998) and AusDiab (2000–2005), adjusted for age, sex, and ethnic group (in Mauritius only), but not other MetS components, are shown in Table 2. To account for the possibility that significant results may simple be due to multiple comparisons, P < 0.01 was considered significant. Baseline waist circumference was associated with change in each of the other six MetS components tested in each of the three cohorts (all P < 0.01 and in the expected direction, with the exception of triglycerides where P = 0.015 for the Mauritius 1987–1998 cohort). In Mauritius, only baseline 2-h postload plasma glucose (P = 0.002) in the 11-year cohort was significantly associated with change in waist circumference; however, the association was with a decrease rather than an increase in waist circumference. In AusDiab, baseline HDL cholesterol (P = 0.001) was associated with an increase in waist circumference, whereas HOMA-S (P < 0.0001) was associated with a decrease in waist circumference. Baseline triglycerides were also associated with an increase in waist circumference (P = 0.016); however, this did not quite reach significance at P < 0.01.
Table 2. Univariate relationships between MetS components at baseline and change in each other component between baseline and follow-up
Further multivariate regression models including all six baseline MetS components as covariates (plus age and sex (and ethnic group in Mauritius)) are presented in Table 3. A similar pattern to the univariate analysis is seen. Waist circumference at baseline was strongly associated with deterioration in each of the other six MetS components in each of the three survey periods, with the only exceptions being change in HDL cholesterol and triglycerides in Mauritius 1987–1998 (P = 0.055 and 0.085 respectively), change in HDL cholesterol and triglycerides in AusDiab (P = 0.170 and P = 0.021 respectively), and change in systolic blood pressure in Mauritius 1987–1992 (P = 0.142). Only baseline 2-h postload glucose in the 1987–1998 Mauritius cohort (P = 0.001) and HOMA-S (P < 0.0001) in AusDiab were associated with change in waist circumference; however, for 2-h postload glucose in Mauritius, this association was with a decrease in waist circumference.
Table 3. Multiple regression analysis of baseline MetS components and change in each other component between baseline and follow-up
Figure 1 represents a summary of all significant relationships from Table 3, with the central component representing an individual baseline predictor, and the six satellite components surrounding each representing the six follow-up variables. Where arrows are present, this represents significant associations (P < 0.01) between the baseline variable and change in the satellite component for each of the AusDiab and 5- and 11-year Mauritius follow-up studies. Baseline waist circumference was significantly associated with deterioration in four of six other MetS variables in AusDiab, with deterioration in five of six MetS variables in Mauritius 1987–1992, and with deterioration in four of six MetS variables in Mauritius 1987–1998. In contrast, an increase in waist circumference between baseline and follow-up was only predicted by insulin sensitivity (HOMA-S) at baseline, and only in one of the three cohorts. While baseline triglyceride levels predicted deterioration in other MetS variables on 10 occasions (compared to 13 for waist circumference), a rise in triglyceride levels was also predicted by other baseline parameters on five occasions. Similar, although slightly attenuated, findings were also found when the Mauritius cohort was divided into ethnic Asian Indian and Creole cohorts.
The association between waist circumference at baseline and the development of the MetS as a whole (defined according to the National Cholesterol Education Program-Adult Treatment Panel III definition) was also assessed. As a continuous variable, adjusted for the same variables as included in Tables 2 and 3, waist circumference was the strongest predictor of incident MetS (P < 0.0001 in AusDiab and both Mauritius cohorts). Similarly adjusted, obesity as a dichotomous variable (as defined in the National Cholesterol Education Program-Adult Treatment Panel III MetS definition) was strongly associated with incident MetS in AusDiab (odds ratio 3.9 (95% confidence interval 3.0–5.0)) and Mauritius (1987–1992 (odds ratio 2.9 (2.3–3.6)) and 1987–1998 (odds ratio 3.1 (2.4–4.1))).
Over 5 years in the AusDiab study and in cohorts spanning both 5 and 11 years in Mauritius, we have shown here a strong association between waist circumference and deterioration in other MetS parameters in both univariate and multivariate analyses. In contrast, the corresponding relationship between those other MetS parameters and increasing waist circumference is not seen. These results are suggestive of a temporal relationship whereby central obesity is the first component of the MetS to develop, preceding deterioration in each of the other components.
Whether waist circumference is in the middle of a causal pathway leading from insulin insensitivity, to obesity and then the development of the MetS is an interesting hypothesis that is not ruled out from the results presented in Table 3. If this were the case, however, repeating the analysis without including waist circumference would be expected to reveal a strong relationship between insulin sensitivity at baseline and the variables of the MetS and not vice versa. This is not seen, however (data not shown), suggesting that the absence of a strong temporal relationship between insulin sensitivity and variables of the MetS is not due to the presence of waist circumference in the model. In addition, whether insulin sensitivity precedes hypertension has been a controversial question and our results are inconclusive on this topic (18).
Support for the role of central obesity in the pathogenesis of the MetS has come from both epidemiology and physiology. The physiological link between adipose tissue and metabolic deterioration is based on evidence including the role played by factors such as free fatty acids and tumor necrosis factor-α in impairing insulin action in skeletal muscle (19,20). Obesity-induced inflammation is also being increasingly implicated in insulin resistance (21), with the adipokine adiponectin shown to have antidiabetic, antiatherosclerotic, and anti-inflammatory functions (22), and high levels of this hormone being negatively associated with obesity, insulin resistance, and type 2 diabetes (23).
Elevated concentrations of leptin, another adipokine, have consistently been independently associated with an increased diabetes, cardiovascular disease risk, and the MetS (12,24,25,26); in addition, visfatin, as well as peptides such as plasminogen activator inhibitor-1, has also been shown to play a key role in the process of metabolic deterioration (27,28).
More directly, assessment of intra-abdominal fat mass through computed tomography has been shown to be strongly and independently associated with each of the five components of the Adult Treatment Panel III definition of the MetS (29). The complexity of the physiological pathways connecting central obesity and other MetS risk factors mean that the precise process by which central obesity and/or visceral fat interacts with insulin resistance and the MetS is far from clear at present. Our results do not rule out the possibility that an unmeasured variable affects each of the components of the MetS individually, but that obesity simply develops earlier; however, the physiological evidence outlined above directly linking obesity and each of the MetS components suggests that this is unlikely to be the case.
Epidemiologic evidence directly supporting the positioning of obesity at the start of the MetS pathway is not substantial. Several longitudinal studies have examined the role of obesity in the development of the MetS but none has explicitly addressed the question of whether obesity precedes the development of the other components. A 4.5-year longitudinal factor analysis placed BMI at the center of the three MetS factors identified (2). Three other studies have all concurred that obesity appears to be the central feature of the MetS, with a French study assessing change in weight and parallel change in other MetS parameters; the IRAS study being used to determine predictors of incident MetS as a whole; and the observation from the ARIC study that obesity was the best predictor of the development of one or more MetS components. None of these studies, however, has provided compelling data on the temporality of this association (3,4,6). Evidence for the association between obesity and deterioration in individual MetS components has been presented for hypertension and hyperinsulinemia (7,30), although in Pima Indians, an inverse relationship between obesity and hyperinsulinemia was observed, perhaps reflecting an early expression of a thrifty genotype before obesity in this population (31).
The most recent MetS definition proposed by the International Diabetes Federation acknowledged the importance of central obesity in the MetS by making this a required component for clinical diagnosis (32). Our results suggest that although central obesity is clearly important as a component of clinical MetS definitions, it should be emphasized that it appears to precede development of the other component abnormalities and when present in isolation is therefore likely to be an important warning sign for future MetS. To prevent the rising global tide of diabetes and the MetS, our results would suggest that priority be given to reducing rates of overweight and obesity.
The similar result observed in two contrasting national, population-based samples including multiple ethnic groups (Asian Indians, Creoles, and Europids in particular), in a developed and a developing nation and with different prevalence of the MetS, incidence of diabetes, and follow-up period (13,15,33), strengthens the validity of the findings and the generalizability of them to the pathogenic process that is the MetS. To our knowledge this is the first attempt to explicitly investigate a temporal relationship between all of the components of the MetS. Cross-lagged panel (or unbalanced reciprocity) designs have been shown to have limitations in determining temporal order if measurements are unreliable and if the possibility that concurrent change is responsible for an observed association is not accounted for (34). The objective nature of the variables of interest, the significant amount of time between surveys in the Mauritius and AusDiab studies, and the strength of the observed associations all reduce the likelihood that these issues are of concern in this analysis. Future research utilizing multiple time points and using discrete-time survival modeling approaches may assist in further exploring the precise temporal relationships that have been suggested here, and in more precisely quantifying the magnitude of the relationships observed.
In AusDiab, fasting insulin measurement at baseline utilized a human insulin-specific radioimmuoassay kit, while at follow-up, insulin was measured using a Chemiluminescence method, with the two methods having been shown to result in different mean values in a comparison of insulin assay types (35). For this reason, the differences in the values of HOMA-S between baseline and follow-up in AusDiab (Table 1) are not easily interpreted. The important aspects of the statistical analyses presented here, however, are the relative changes between individuals, not the actual difference between measurements at baseline and follow-up. Consequently, the strength of association observed should not be affected by differences in measurement techniques. The exclusion of those on blood pressure-lowering medication and with diabetes could be criticized due to the likelihood that these individuals are the most likely to have undergone significant decline in MetS parameters. In analyses where those on medication for hypertension and those with diabetes at baseline were included (results not shown), similar patterns were observed that if anything were stronger in their support of waist circumference as the precursor to metabolic deterioration.
If the null hypothesis was true (that all parameters were equally related to outcomes) but the reason for our findings was simply that waist circumference was more accurately measured than the other parameters, we should also expect that deterioration in waist circumference would be predicted by several baseline parameters. Because this was not seen, however, this does not seem to be a plausible explanation of the findings.
This work adds support to the hypothesis that central obesity is at the core of the MetS, and is the antecedent to deterioration in its other components. From a clinical and public health perspective, the clear implication is that preventing or reducing obesity must be the primary focus of efforts to prevent the now well-documented global epidemic of diabetes and its related complications. This work clearly suggests that elevation in waist circumference occurs before deterioration has occurred in the other components of the MetS, and that central obesity is therefore the flag bearer for the MetS. Compelling evidence from lifestyle intervention trials (36,37,38,39) suggests that it is indeed possible to reduce the risk of developing diabetes and features of the MetS, and we have added further evidence to suggest that reduction in adiposity, and in particular central obesity, should be the goal for those wishing to reduce their risk of the MetS and its associated outcomes.
We are most grateful to the many people involved in organizing and conducting the AusDiab and Mauritius studies, and especially the participants for volunteering their valuable time. A.C. is supported by Postgraduate Research Scholarship PP04M1794 from the National Heart Foundation of Australia. AusDiab was supported by a grant from the National Health and Medical Research Council of Australia (#233200). The AusDiab study, co-coordinated by the International Diabetes Institute, gratefully acknowledges the generous support given by: the Australian Government Department of Health and Ageing, Abbott Australasia Pty Ltd, Alphapharm Pty Ltd, AstraZeneca, Aventis Pharma, Bristol-Myers Squibb, City Health Centre Diabetes Service—Canberra, Department of Health and Community Services—Northern Territory, Department of Health and Human Services—Tasmania, Department of Health—New South Wales, Department of Health—Western Australia, Department of Health—South Australia, Department of Human Services—Victoria, Diabetes Australia, Diabetes Australia Northern Territory, Eli Lilly Australia, Estate of the Late Edward Wilson, GlaxoSmithKline, Jack Brockhoff Foundation, Janssen-Cilag, Kidney Health Australia, Marian & FH Flack Trust, Menzies Research Institute, Merck Sharp & Dohme, Novartis Pharmaceuticals, Novo Nordisk Pharmaceuticals, Pfizer Pty Ltd, Pratt Foundation, Queensland Health, Roche Diagnostics Australia, Royal Prince Alfred Hospital—Sydney, and Sanofi Synthelabo. The Mauritius study was undertaken with the support and collaboration of the Ministry of Health (Mauritius), the World Health Organization (Geneva, Switzerland), International Diabetes Institute (Melbourne, Australia), the University of Newcastle upon Tyne (UK), and the National Public Health Institute (Helsinki, Finland). This study was partially funded by the National Institutes of Health grant DK-25446.