Search was conducted to identify articles up to April 10, 2009.
Open data markers indicate reference values for each plot; error bars, 95% confidence intervals. ARIC indicates Atherosclerosis Risk in Communities; EPIC, European Prospective Investigation Into Cancer and Nutrition.
Fourteen data points are included for the 13 studies because results for men and women are shown separately in the Hoorn study. Size of squares corresponds to the weight of each study in the meta-analysis. CI indicates confidence interval.
Customize your JAMA Network experience by selecting one or more topics from the list below.
Li S, Shin HJ, Ding EL, van Dam RM. Adiponectin Levels and Risk of Type 2 Diabetes: A Systematic Review and Meta-analysis. JAMA. 2009;302(2):179–188. doi:10.1001/jama.2009.976
Context The association of obesity with development of type 2 diabetes may be partly mediated by altered secretion of adipokines by adipose tissue. Greater adiposity down-regulates secretion of adiponectin, an adipokine with anti-inflammatory and insulin-sensitizing properties. The strength and consistency of the relation between plasma adiponectin and risk of type 2 diabetes is unclear.
Objective To systematically review prospective studies of the association of plasma adiponectin levels and risk of type 2 diabetes.
Data Sources A systematic search of the MEDLINE, EMBASE, and Science Citation Index Expanded databases using adiponectin and diabetes and various synonyms and reference lists of retrieved articles up to April 10, 2009.
Study Selection We included prospective studies with plasma adiponectin levels as the exposure and incidence of type 2 diabetes as the outcome variable.
Data Extraction Two reviewers independently extracted data and assessed study quality. Generalized least-squares trend estimation was used to assess dose-response relationships. Pooled relative risks and 95% confidence intervals were calculated using random-effects models to incorporate between-study variation.
Results Thirteen prospective studies with a total of 14 598 participants and 2623 incident cases of type 2 diabetes were included in the meta-analysis. Higher adiponectin levels were monotonically associated with a lower risk of type 2 diabetes. The relative risk of type 2 diabetes was 0.72 (95% confidence interval, 0.67-0.78) per 1–log μg/mL increment in adiponectin levels. This inverse association was consistently observed in whites, East Asians, Asian Indians, African Americans, and Native Americans and did not differ by adiponectin assay, method of diabetes ascertainment, duration of follow-up, or proportion of women. The estimated absolute risk difference (cases per 1000 person-years) per 1–log μg/mL increment in adiponectin levels was 3.9 for elderly Americans and 30.8 for Americans with impaired glucose tolerance.
Conclusion Higher adiponectin levels are associated with a lower risk of type 2 diabetes across diverse populations, consistent with a dose-response relationship.
Quiz Ref IDAdiponectin is a 244–amino acid collagen-like protein that is solely secreted by adipocytes and acts as a hormone with anti-inflammatory and insulin-sensitizing properties.1 Findings from animal studies and metabolic studies in humans suggest several mechanisms through which adiponectin may decrease the risk of type 2 diabetes, including suppression of hepatic gluconeogenesis, stimulation of fatty acid oxidation in the liver, stimulation of fatty acid oxidation and glucose uptake in skeletal muscle, and stimulation of insulin secretion.1,2 These effects may be partly mediated by stimulatory effects of adiponectin on signaling pathways for 5′ adenosine monophosphate–activated protein kinase and peroxisome proliferator–activated receptor α.1,2 Adiponectin secretion, in contrast to secretion of other adipokines, is paradoxically decreased in obesity.3 This may be attributable to inhibition of adiponectin gene transcription by inflammatory and angiogenic factors secreted by hypertrophic adipocytes.3,4
Currently, the prevalence of type 2 diabetes in the United States and many other countries in the world has reached epidemic proportions.5 Various signaling molecules secreted by adipocytes have been implicated in the development of insulin resistance and type 2 diabetes, based on results from animal models and metabolic studies in humans.2 Epidemiologic studies can provide insight into the potential importance of these signaling molecules as determinants of the incidence of type 2 diabetes in human populations. In addition, these studies can identify biological markers that may be useful for the prediction of type 2 diabetes and the identification of high-risk groups. To date, no systematic review has been conducted that evaluates the available evidence for an association between adiponectin levels and risk of type 2 diabetes across different populations. The objective of our systematic review was to assess the consistency of the association of adiponectin levels and risk of type 2 diabetes in prospective cohort studies and to summarize results in a meta-analysis.
Three investigators (S.L., H.J.S., R.M.v.D.) identified articles through a systematic search of the MEDLINE (Pubmed), EMBASE, and Science Citation Index Expanded (ISI Web of Science) databases and through reference lists of selected articles up to April 10, 2009. The following terms were used for the MEDLINE search: (adiponectin[MeSH Terms] OR adiponectin[All Fields] OR ADIPOQ[All Fields] OR ACDC [All Fields] OR ACRP30[All Fields] OR APM1[All Fields] OR GBP28[All Fields] OR (C1q[All Fields] AND (collagen[MeSH Terms] OR collagen[All Fields]) AND domain-containing[All Fields] AND (proteins [MeSH Terms] OR proteins [All Fields] OR protein[All Fields])) OR (30[All Fields] AND kDa[All Fields] AND (adipocytes [MeSH Terms] OR adipocytes[All Fields] OR adipocyte[All Fields]) AND complement-related[All Fields] AND (proteins [MeSH Terms] OR proteins[All Fields] OR protein[All Fields])) OR adipose most abundant gene transcript 1[All Fields] AND (diabetes mellitus[MeSH Terms] OR diabetes [All Fields]). Similar search terms were used for EMBASE and the Science Citation Index. No language restriction was applied for searching and study inclusion. Our systematic review was conducted according to the Meta-analysis of Observational Studies in Epidemiology (MOOSE) guidelines.6
We only included prospective studies of plasma adiponectin concentrations and type 2 diabetes (ie, studies with adiponectin levels measured in blood collected before the onset of diabetes) in humans with a minimum follow-up duration of 1 year. We excluded literature reviews, cross-sectional studies, studies on animals or cell lines, studies of determinants of adiponectin levels, studies of genetic variation in adiponectin-related genes, and studies of type 1 diabetes or gestational diabetes. We also excluded studies on populations with specific diseases or using specific medications. Four studies were excluded because data were not reported separately for diabetes and impaired glucose tolerance (IGT).7-10 We included conference abstracts in our search, but the identified abstracts on adiponectin levels and risk of type 2 diabetes included data that were also reported in a full article or reported insufficient information to extract effect estimates.
Data extraction was conducted independently by 2 authors (S.L., H.J.S.) using a standardized data extraction form. To resolve discrepancies, a third investigator (R.M.v.D.) was consulted. We did not contact authors of the original studies in the case of missing data. For each included article, we extracted information on the title, authors, publication year, name of the study, sample size, number of diabetes cases, study design, mean (standard deviation) for the adiponectin level, duration of follow-up, mean age, country, race/ethnicity, proportion of women, year the study started, mean body mass index (BMI, calculated as weight in kilograms divided by height in meters squared), assay for measuring adiponectin levels, covariates controlled for by matching or multivariable analysis, and statistical methods used for the analysis.
The original articles used tertiles, quartiles, or quintiles as categories for adiponectin levels. For each category, we extracted median values, numbers of cases/noncases, relative risks (RRs), and 95% confidence intervals (CIs). For studies that reported several multivariable-adjusted RRs, we extracted the effect estimate that was most fully adjusted for potential confounders except for other metabolic biomarkers. If available, we also extracted the effect estimate that was additionally adjusted for other metabolic biomarkers.
We used the multivariable-adjusted odds ratios or hazard ratios reported in the original articles. For the Indian Diabetes Prevention Program study,11 we calculated crude cumulative incidence ratios of type 2 diabetes for tertiles of adiponectin level. Because incidence density sampling was used12 or because the incidence of type 2 diabetes was sufficiently low for the rare disease assumption to apply, odds ratios could be assumed to be accurate estimates of risk ratios. We therefore refer to all these estimates as RRs.
We estimated dose-response associations based on data for categories of adiponectin levels on median dose, number of cases and participants, and effect estimates with corresponding standard errors using generalized least-squares trend estimation (GLST) analysis based on the methods developed by Greenland and Longnecker.13-15 We used the 2-stage GLST method that estimates study-specific slope lines first and then derives an overall average slope.15 We chose the 2-stage method because this allowed us to combine the GLST-estimated study-specific slopes with the results from studies not eligible for GLST but that reported effect estimates for continuous associations. If medians for categories of adiponectin levels were not reported, approximate medians were estimated using the midpoint of the lower and upper bounds (or using the mean when the midpoint could not be estimated). For the open-ended categories, we estimated the median value assuming a normal distribution density function. Ten studies reported RRs for categories of adiponectin levels and were eligible for GLST dose-response analysis.11,16-24
In addition, 3 studies reported continuous results for log-transformed adiponectin levels12,25,26 and could be included in our meta-analysis of log-transformed adiponectin levels and risk of type 2 diabetes. Two studies reported only continuous results for adiponectin levels without log transformation and could only be included in our secondary analysis of adiponectin levels per 5 μg/mL and risk of type 2 diabetes.27,28 Records for the Hoorn Study22 were entered separately for men and women, because only sex-specific RRs were presented for that study. Therefore, the primary analysis of log-transformed adiponectin levels and diabetes risk was based on 14 data points and the secondary analysis of nontransformed adiponectin levels and diabetes risk on 13 data points.
Random-effects estimates models as described by DerSimonian and Laird were used to incorporate between-study heterogeneity in addition to sampling variation for the calculation of summary RR estimates and corresponding 95% CIs.29 The Cochran Q test and the I2 statistic were used to evaluate heterogeneity between studies. I2 was calculated based on the formula I2 = 100% × (Q − df)/Q.30
Stratified analyses were performed by meta-regression, examining the statistical significance of the difference in RRs according to race/ethnicity (Pima Indian and Aboriginal Canadians, Asian Indians, whites, Japanese Korean, mixture of whites and African Americans); sex (continuous and proportion of women), mean age (continuous and 3 categories) and BMI of the study population (continuous and 2 categories); duration of follow-up (continuous); publication year (continuous); assay used to measure adiponectin levels (enzyme-linked immunosorbent assay, radioimmunoassay, or other); ascertainment of diabetes (self-reports, plasma glucose measurements, or a combination); duration of follow-up (continuous and 2 categories); number of diabetes cases (continuous and 2 categories), and measure of association (odds ratios or risk ratios). Race/ethnicity was considered in the analysis to evaluate whether adiponectin levels are predictive of type 2 diabetes development in diverse populations.
A sensitivity analysis was performed to assess the influence of each individual study by omitting 1 study at a time and calculating a pooled estimate for the remainder of the studies. We also conducted sensitivity analyses using a fixed-effects model, modeling adiponectin levels without log transformation and excluding the study that did not adjust for adiposity and that included only participants with IGT at baseline.11 The Egger regression test, Begg adjusted rank correlation test, and visual inspection of a funnel plot were performed to assess publication bias.31,32 All tests were 2-sided; P < .05 was considered statistically significant. STATA version 9 (StataCorp, College Station, Texas) was used for all statistical analyses.
We identified 15 prospective studies that met our inclusion criteria, of which 13 could be used in our meta-analysis of adiponectin levels expressed per 1 log μg/mL11,12,16-26 and 12 in our meta-analysis of levels expressed per 5 μg/mL11,16-24,27,28 (Figure 1). Table 1 shows the characteristics of the 15 identified prospective studies of adiponectin levels and risk of type 2 diabetes. The studies included 10 cohort studies, 1 case-cohort study, and 4 nested case-control studies. Study participants were from the general population, except for those in the Diabetes Prevention Project27 and the Indian Diabetes Prevention Program,11 which only included participants with IGT at baseline (Table 1). Six studies included largely whites,16,20,22-24,28 2 included a mixture of whites and African Americans,19,25 2 included Pima Indians or Aboriginal Canadians,12,26 1 included Asian Indians,11 3 included Koreans or Japanese,17,18,21 and 1 included African Americans, Hispanics, American Indians, Asian Americans, and whites.27 Type 2 diabetes was ascertained using an oral glucose tolerance test in 7 studies,11,12,17,18,21,22,27 self-reported information in 4 studies,16,20,23,24 and a combination of plasma glucose concentrations and self-reported information in 4 studies.19,25,26,28Table 2 shows the results of the individual studies for adiponectin levels and risk of type 2 diabetes. Figure 2 shows the RRs according to levels of adiponectin for the studies that used at least 4 categories of adiponectin levels and thus provided more information on the shape of the association. These graphs were consistent with a monotonically declining risk of type 2 diabetes for increasing adiponectin levels, suggesting that analyses modeling adiponectin level as a continuous variable are reasonable.
For the meta-analysis, we used RRs that were most completely adjusted for covariates, including adiposity, if these covariates were available without adjustment for other metabolic biomarkers that could be biological intermediates. Our meta-analysis of log-transformed adiponectin levels and risk of type 2 diabetes included a total of 13 studies (14 data points because results for men and women were reported separately for the Hoorn Study22) including 14 598 participants and 2623 cases of type 2 diabetes. Figure 3 shows the study-specific RRs and the pooled estimate. The pooled RR of type 2 diabetes was 0.72 (95% CI, 0.67-0.78; P < .001) per 1–log μg/mL increment in adiponectin levels. This increment was approximately equivalent to the difference between the medians of the highest and the lowest tertiles of the included studies. The absolute risk difference corresponding to the pooled RR will be greater for high-risk than for low-risk populations. Based on the pooled RR and the incidence rates observed in different studies, the absolute risk difference (cases per 1000 person-years) per 1–log μg/mL increment in adiponectin levels was 3.9 for a population of elderly white and black Americans,25 7.5 for Japanese Americans,21 and 30.8 for Americans of various races/ethnicities with IGT.27 The P value for heterogeneity from the Cochran Q test (Q13 = 22.9) was 0.04, suggesting that the variation in study results was not solely attributable to sampling variation. The I2 was 43% (95% CI, 0%-70%), suggesting moderate between-study heterogeneity. The funnel plot was symmetrical, and neither the Begg test (P = .11) nor the Egger test (P = .77) suggested publication bias.
We evaluated potential sources of heterogeneity in stratified analyses. Quiz Ref IDThe inverse association between adiponectin levels and diabetes risk was consistently observed in whites, East Asians, Asian Indians, Pima Indians/Aboriginal Canadians, and populations with whites as well as African Americans (Table 3). Associations tended to be stronger for Pima Indians/Aboriginal Canadians as compared with other races and for younger as compared with older study populations, but these differences were not statistically significant (Table 3). Associations did not differ substantially by mean BMI of the study population, assay for adiponectin measurement, method of diabetes ascertainment, measure of association, duration of follow-up, or number of diabetes cases (Table 3). In addition, publication year (P = .74) and proportion of women (P = .28) were not significantly associated with the strength of the association.
We conducted a sensitivity analysis omitting 1 study at a time and calculating the pooled RRs for the remainder of the studies. This analysis showed that none of the individual studies dramatically influenced the pooled RRs, which ranged from 0.71 (95% CI, 0.66-0.77) to 0.74 (95% CI, 0.69-0.80) per 1–log μg/mL increment in adiponectin levels. In line with this finding, excluding the study that only included persons with IGT at baseline and that was the only study without adjustment for adiposity11 did not appreciably change the pooled RR (0.72; 95% CI, 0.67-0.79). Only 5 studies adjusted for several lifestyle factors (Table 2), but restricting analyses to these studies did not appreciably change the results (pooled RR, 0.71; 95% CI, 0.65-0.78). Furthermore, using a fixed-effects instead of random-effects model yielded essentially the same results (pooled RR, 0.70; 95% CI, 0.67-0.74). We also conducted an analysis modeling adiponectin levels without log transformation. This analysis included 12 studies11,16-24,27,28 with a total of 12 802 participants and 2494 cases of type 2 diabetes. The pooled RR of type 2 diabetes per 5-μg/mL increment in adiponectin levels was 0.74 (95% CI, 0.68-0.80), with a larger between-study heterogeneity than for the primary analysis (P = .004 for heterogeneity; I2 = 59%; 95% CI, 24%-78%).
Several studies presented RRs for adiponectin levels and risk of type 2 diabetes with and without adjustment for metabolic variables, including markers of glycemia and insulin sensitivity, plasma lipid levels, and inflammatory markers.19,20,22-24,26-28 We did not pool these results, because each study adjusted for different metabolic markers. As shown in Table 2, substantial associations between adiponectin levels and diabetes risk remained after adjustment for these biomarkers, with few exceptions.20,22
This meta-analysis focused on total adiponectin levels, because only the Nurses' Health Study24 and the Hawaii-LA-Hiroshima study21 reported associations for levels of high-molecular-weight adiponectin and risk of type 2 diabetes. In both studies, associations with diabetes risk were slightly stronger for high-molecular-weight adiponectin as compared with total adiponectin. In the Nurses' Health Study, the multivariable-adjusted RR for the highest as compared with the lowest quintile of high-molecular-weight adiponectin was 0.10 (95% CI, 0.06-0.15; P < .001 for trend). In the Hawaii-LA-Hiroshima study, the multivariable-adjusted RR for the highest as compared with the lowest tertile was 0.40 (95% CI, 0.22-0.75; P = .005 for trend).
In our meta-analysis of prospective studies, we observed a substantial inverse association between plasma adiponectin level and incidence of type 2 diabetes. Quiz Ref IDRisk of type 2 diabetes appeared to decrease monotonically with increasing adiponectin levels. The association was consistent for whites, East Asians, Asian Indians, African Americans, and Native Americans. The results did not differ substantially by adiponectin assay, method of diabetes ascertainment, study size, follow-up duration, BMI, or proportions of men and women.
Our study has several strengths. The included original studies were all prospective, which greatly reduces the likelihood of selection bias and reverse causation (possible effects of diabetes on adiponectin levels). In addition, dose-response relationships were evaluated through GLST analysis, allowing the combination of comparable estimates. The consistency of inverse associations between adiponectin levels and risk of type 2 diabetes across multiple strata and our sensitivity analyses indicate that our conclusions were not dependent on arbitrary decisions in our meta-analysis.
A potential limitation of our study is residual confounding. As shown in Table 2, many of the included studies adjusted for a wide range of potential confounders, including demographic and lifestyle factors. The strength of the adjusted RRs for adiponectin levels and diabetes risk and the consistency of associations across diverse populations reduce the likelihood that residual confounding by these variables can explain the findings. Another issue is whether adiponectin has a causal effect on diabetes or is only a surrogate marker for other biological risk factors. Different studies adjusted for different factors, including levels of plasma lipids and inflammatory markers. Although these variables could not fully explain the association between adiponectin levels and diabetes risk, we cannot exclude the possibility that other metabolic factors are responsible for such an association in epidemiologic studies. In addition, misclassification of type 2 diabetes in the original studies may have affected the results. However, results were similar for studies using self-reported diabetes information and those using oral glucose tolerance tests. Publication bias can affect the results of every meta-analysis. However, tests for publication bias did not indicate that this bias substantially affected our results, and results were essentially the same after restricting analyses to larger studies less likely to be affected by publication bias.
We observed significantly more variation in study results than would be expected as a result of chance, which is not surprising given the substantial differences in study populations and methods. Data from previous studies suggested that associations between adiponectin levels and risk of type 2 diabetes were stronger in women than in men22 and stronger in obese than in leaner persons.23 However, other studies did not report significant differences by sex16,19,27 or obesity,19,24 and we did not observe significant associations between the proportion of women in the study or mean BMI and the strength of the association between adiponectin levels and risk of type 2 diabetes. In our analyses, associations tended to be stronger in Pima Indians/Aboriginal Canadians as compared with other racial/ethnic groups and in younger as compared with older populations, but these results were not statistically significant. Differences in the proportion of high-molecular-weight adiponectin, the hexamer most strongly associated with diabetes risk, may also have contributed to variation in study results.21,24 Further evaluation of effect modification is needed in larger studies or individual-participant pooled analyses with more power to detect effect modification than our analyses based on study-level characteristics.
Adiponectin may exert its effects on glucose metabolism through adiponectin receptor 1 and adiponectin receptor 2. Prior study demonstrates that expression of these receptors is decreased in mouse models of insulin resistance.1 Animal studies show that adiponectin receptor 1 knockout results in the abrogation of adiponectin-induced activation of 5′ adenosine monophosphate–activated protein kinase and increased glucose production and insulin resistance.1,2 Targeted disruption of adiponectin receptor 2 leads to decreased activity of peroxisome proliferator–activated receptor α signaling pathways and insulin resistance.1 Animal studies suggest increased susceptibility to diet-induced insulin resistance among adiponectin knockout mice, and injection of recombinant adiponectin dramatically improves hepatic insulin sensitivity.1,2Quiz Ref IDIn rhesus monkeys, changes in adiponectin levels are closely associated with changes in insulin sensitivity33; in humans, higher adiponectin levels are associated with higher insulin sensitivity.34,35Some studies have associated variation in the adiponectin gene with insulin resistance and risk of type 2 diabetes.36,37 Using a mendelian randomization approach, larger studies of variants in the adiponectin gene that are related to adiponectin levels can evaluate whether associations with risk of type 2 diabetes are causal.38
Our findings show that higher adiponectin levels are consistently associated with a lower risk of type 2 diabetes in prospective studies of diverse populations. Currently, adiponectin is among the strongest and most consistent biochemical predictors of type 2 diabetes.39 Although these epidemiologic studies cannot establish causality, the consistency of the association across diverse populations, the dose-response relationship, and the supportive findings in mechanistic studies indicate that adiponectin is a promising target for the reduction of risk of type 2 diabetes.
Quiz Ref IDRecent studies have shown that adiponectin levels can be increased through pharmaceutical and lifestyle interventions.40,41 In addition, adiponectin levels may be useful for identifying persons likely to benefit most from interventions to treat “dysfunctional adipose tissue” and its metabolic complications.3 Future studies should also evaluate whether adiponectin is useful for prediction of type 2 diabetes in addition to established risk factors using statistical techniques appropriate for prognostic analyses.42
Corresponding Author: Rob M. van Dam, PhD, 655 Huntington Ave, Boston, MA 02115 (firstname.lastname@example.org).
Author Contributions: Dr van Dam had full access to the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Financial Disclosures: None reported.
Funding/Support: Dr Ding is supported in part by a postdoctoral fellowship grant from the American Diabetes Association and by a grant from the Paul and Daisy Soros Fellowships for New Americans. Dr van Dam is supported in part by a grant from the Boston Obesity Nutrition Research Center (P30 DK4600).
Role of the Sponsors: The funding organizations had no role in the design and conduct of the study; the collection, management, analysis, and interpretation of the data; or the preparation, review, or approval of the manuscript.
Create a personal account or sign in to: