Adiponectin and Risk of Stroke
Prospective Study and Meta-analysis
Background and Purpose—The favorable cardiovascular effects attributed to adiponectin may lower risk of stroke. We investigated this in a prospective study and meta-analysis.
Methods—A case–cohort study nested within the Potsdam cohort of the European Prospective Investigation into Cancer was performed, with 170 incident cases of ischemic stroke and a randomly selected subcohort of 2155 participants without major cardiovascular disease at baseline. A random-effects dose–response meta-analysis was performed on prospective studies reporting on adiponectin and incident stroke in healthy populations up to April 2013, identified through MEDLINE and EMBASE.
Results—In European Prospective Investigation into Cancer-Potsdam, after adjustment for cardiovascular risk factors, the hazard ratio of ischemic stroke per 5-µg/mL higher total-adiponectin was 1.10 (95% confidence interval, 0.89–1.37). Participants with higher total-adiponectin had higher high-density lipoprotein-cholesterol and lower high-sensitivity C-reactive protein and triglyceride levels, and had less often diabetes mellitus. Additional adjustment for these putative mediators yielded a hazard ratio of 1.31 (95% confidence interval, 1.04–1.64). Nine studies (19 259 participants, 2960 cases), including European Prospective Investigation into Cancer-Potsdam, were meta-analyzed. Pooling relative risks adjusted for cardiovascular risk factors not including putative mediators indicated moderate between-study heterogeneity (I2=52.2%). This was explained by the smallest study, and the pooled relative risk (95% confidence interval) before and after its exclusion was 1.03 (0.98–1.08) and 0.99 (0.96–1.01) per 5 µg/mL, respectively. The pooled relative risk (95% confidence interval) additionally adjusted for potential mediators was 1.08 (1.01–1.15) and 1.05 (1.00–1.11) before and after excluding the same study, respectively.
Conclusions—Adiponectin is not associated with risk of stroke. If anything, adiponectin relates directly to stroke risk after controlling for risk factors that favorably correlate with adiponectin.
Adiponectin is a hormone derived from adipose tissue. Its levels are usually higher in women than in men and are downregulated with increasing central adiposity.1 Adiponectin has been suggested to exert anti-inflammatory, antiatherogenic, and insulin-sensitizing effects,2 to promote high-density lipoprotein-cholesterol formation,3 to reduce plasma triglyceride levels,4 and it has been inversely associated with carotid intima-media thickness.5,6 Total-adiponectin circulates in the blood stream as globular-adiponectin, and as full-length fractions of low-molecular weight, medium-molecular weight, and high-molecular weight (HMW). HMW adiponectin has been proposed as the most active fraction for glucose homeostasis,2 whereas the lower weight fractions have been associated with anti-inflammatory effects.7
The favorable cardiovascular effects attributed to adiponectin may lower risk of stroke. We evaluated this in the Potsdam cohort of the European Prospective Investigation into Cancer (EPIC). Because of the antiatherogenic properties of adiponectin, we focused on ischemic stroke (IS). To assess the consistency of the association between adiponectin and risk of stroke in apparently healthy populations, we also performed a meta-analysis on prospective studies.
EPIC-Potsdam study comprises 27 548 individuals from the general population of the Potsdam area (Germany).8 The association of plasma total-adiponectin with risk of IS was analyzed using a case–cohort design. The study-set included all incident IS cases that occurred during a mean follow-up of 8.2±2.2 years9 and a random subcohort (n=2500). When both IS and myocardial infarction (MI) occurred in the same participant (n=5), we only considered the first event. IS occurred before MI in 2 cases and MI before IS in 2 more. One participant had a stroke and MI on the same day; the MI was prioritized over stroke. After exclusion of individuals because of prevalent IS or MI, missing follow-up, or missing adiponectin or covariate information, the final study population comprised a subcohort of 2155 participants and 170 IS cases (11 being fatal and 27 belonging to the subcohort). A total of 605 participants were fasting for ≥8 hours before the blood draw. Informed consent was obtained from all study participants. The study was approved by the Ethical Committee of the state of Brandenburg, Germany.
Baseline plasma levels of biomarkers were measured at Tübingen University (Germany) in samples retrieved from frozen storage (2007–2008). Adiponectin was determined with an ELISA (Linco Research, St Charles, MO; intra-assay and interassay coefficients of variation, 0.1%–6.2% and 5.0%, respectively). High-density lipoprotein-cholesterol, triglycerides, high-sensitivity C-reactive protein, and creatinine were determined with the Siemens ADVIA 1650. N-terminal probrain-type natriuretic peptide was measured in all incident IS cases and in a random subsample of 1137 subcohort members (attributable to financial restrictions) at the Institute of Clinical Chemistry, University of Magdeburg (2012) with an electrochemiluminescence immunoassay on the Immulite analyzer (Siemens AG, Munich, Germany).
Because women had higher total-adiponectin levels than men, baseline characteristics were cross-sectionally compared across sex-specific total-adiponectin tertiles based on the distribution among the subcohort, by age-adjusted ANOVA.
Association of total-adiponectin with risk of IS was calculated as hazard ratios using Cox proportional-hazard regression, modified according to Prentice,10 using a robust estimator to compute the 95% confidence interval. Age was the underlying time metric. HRs were estimated per 5-µg/mL higher total-adiponectin and according to sex-specific tertiles of total-adiponectin in the subcohort. HRs were adjusted for sex (M1); further for common cardiovascular risk factors (waist circumference, smoking, sports activity, education, alcohol consumption, prevalent hypertension, M2); and further for fasting status and for metabolic factors that have been suggested to be intermediate factors11 (diabetes mellitus, high-density lipoprotein- cholesterol, triglycerides, and high-sensitivity C-reactive protein, M3).
Effect modification by sex and by age were, respectively, evaluated by modeling the product-term sex times total-adiponectin, and age times total-adiponectin along with the main effects in the Cox-regression. Sensitivity analyses were performed after excluding fatal IS cases; participants with total-adiponectin below/above the population-wide lower/upper deciles; with prevalent chronic diseases; central obesity; taking antihypertensive, antidiabetic or lipid-lowering medication; those >60 years; smokers or alcohol consumers. Also the first 4 years of follow-up were excluded to account for latency.
MI shares the atherosclerotic pathophysiological pathway with IS. Because we focus on IS here, HRs of MI per 5-µg/mL higher adiponectin are presented in Table I in the online-only Data Supplement. Identification and verification of incident MI cases have been described elsewhere.9 Study design, exclusion criteria, and adjustment models were the same as used for IS.
Meta-analysis of Prospective Studies
Meta-analysis of Observational Studies in Epidemiology guidelines12 were applied (Table II in the online-only Data Supplement). Two investigators independently searched MEDLINE and EMBASE for prospective studies performed in healthy populations, with adiponectin as exposure (total or either of its circulating fractions) and stroke (total stroke or IS) as outcome, published until April 2013. Search terms used in MEDLINE were (Adiponectin[Mesh] or adiponectin or acrp30 or apm) and (Cerebrovascular Disorders[Mesh] or stroke or brain infarction) and (Longitudinal Studies[Mesh] or prospective[Mesh] or prospective or nested or cohort). The equivalent search was constructed for EMBASE. Reference lists of retrieved articles were hand-searched for additional studies. Data gathered included: first author, publication year, location, study design, race/ethnicity, number of participants, proportion of women, duration of follow-up, age, adiponectin assay, mean adiponectin levels and body mass index among noncases, number of cases, case ascertainment, variables controlled for and comparison used.
Effect estimates were extracted for the least and most adjusted models, with and without inclusion of presumed mediators. We contacted authors when data needed for the dose–response meta-analysis were incompletely reported.13–15 Study-specific dose–response associations were calculated per 5-µg/mL higher adiponectin concentration by using the generalized least squares for trend estimation method, to combine comparable estimates with random-effect meta-analysis. Heterogeneity between-study results was evaluated by using the I2 statistic. Publication bias was investigated by Egger test and by visual inspection of the funnel plot. An influential analysis was performed by omitting 1 study at a time. Statistical analyses were performed using SAS Enterprise Guide 4.3 (SAS Institute Inc, Cary, NC) and STATA SE version 12.1 (StataCorp, College Station, TX).
Baseline Characteristics of Participants
Baseline median (interquartile range) total-adiponectin levels in men and women were 5.6 (4.15–7.36) and 8.6 (6.32–11.40) µg/mL, respectively. Baseline characteristics of the subcohort across adiponectin tertiles are shown in Table 1 separately for men and women. Compared with those in the lower tertiles, men and women in the highest tertiles were slightly older and showed better cardiovascular profiles.
Adiponectin and Incident Stroke
Table 2 shows HRs of incident IS per 5-µg/mL higher total-adiponectin, and across sex-specific tertiles of adiponectin. No significant associations were found after adjustment for sex and common cardiovascular risk factors. After additional adjustment for potential mediators, including high-density lipoprotein-cholesterol, high-sensitivity C-reactive protein, and diabetes mellitus, adiponectin was directly associated with IS risk. Adjustment for creatinine, a crude estimate of renal function, yielded essentially similar HRs. Adjustment for N-terminal probrain-type natriuretic peptide, a discerning marker for heart damage, only slightly attenuated the HRs (data not shown). The association between adiponectin and stroke did not differ by sex (P interaction=0.4) or age (P interaction=0.1) after adjustment for variables in M2 (Table III in the online-only Data Supplement). HRs remained similar after exclusion of 11 fatal cases of IS. Also, none of the other prespecified sensitivity analyses led to different results.
Meta-analysis of Prospective Studies
Sixty-four hits from EMBASE and 82 from MEDLINE were retrieved. After exclusion of duplicates and of articles that did not meet the inclusion criteria (Figure I in the online-only Data Supplement), 9 articles11,13–20 reporting results for 8 independent study populations were eligible for inclusion. Nine studies,11,13–16,18–20 including EPIC-Potsdam, reported on total-adiponectin. The Cardiovascular Health Study11 and the Women’s Health Initiative Study17,18 reported also on HMW-adiponectin, with findings being similar to total-adiponectin. Therefore, our analysis includes risk estimates of total-adiponectin only. Thus, together with the EPIC-Potsdam study, ≤9 independent studies were meta-analyzed (Table 3) in total 19 259 participants and 2960 incident stroke cases.
Five studies were nested case–control14–18 and 4 were cohort11,13,19,20 studies. Study populations were from Europe (predominantly white),13–15,20 Japan,4 and the United States (multiple ethnic groups),11,17–19 and included men,13,15,20 women,17,18 or both.11,14,16,19 Mean age ranged from 4713 to 74.411 and follow-up time from 4.914 to 2713 years. Adiponectin was measured in serum17 or plasma.11,13–16,18–20 Incident stroke was ascertained by review of medical records or death certificates in all studies. Relative risks (RR) were provided as odds ratio14,16–18 or hazard ratios.11,13,15,19,20 Odds ratios were assumed to reasonably approximate the RR.
Studies evaluated adiponectin in quartiles,15–21 quintiles,13 and as a continuous variable, either untransformed11,15,19 or log-transformed (Table 4).16 Outcomes were total stroke,14,19,20 IS,11,13,15,17,18 or both.16 For 1 study13 that reported RR only for a subgroup of smokers, we calculated the crude odds ratio from the reported contingency table including participants. One study14 provided separate RR for men and women (personal correspondence), and these were included in the analysis as separate RR. For another study,11 we used a fixed-effects meta-analysis to combine the RRs corresponding to before and after the knot of the spline-regression. Before and after adjustment for potential mediators, adiponectin was significantly related to stroke risk only in 115 and 3 studies,15,19 including EPIC-Potsdam, respectively.
The Figure (A) shows the pooled multivariable-adjusted RR without adjustment for mediators. Initially, we observed moderate heterogeneity across study results (I2=52.2%; P=0.027). This was mainly driven by the smallest study,15 which after exclusion reduced the I2 statistic to 3.2% (P=0.41) and yielded a pooled RR of 0.99 (95% confidence interval, 0.96–1.01) per 5-µg/mL higher adiponectin. Similar to EPIC-Potsdam, combining RR additionally adjusted for potential mediators available for 8 independent studies (Figure [B]) yielded a direct association between adiponectin and stroke (RR, 1.08; 95% confidence interval, 1.01–1.15) per 5-µg/mL higher adiponectin. However, also here we observed moderate between-study hetereogeneity (I2=63.1%; P=0.006). After omitting the smallest study, the I2 became 45.1% (P=0.08) and the pooled RR 1.05 (95% confidence interval, 1.00–1.11). The lack of symmetry of the funnel plot as well as Egger test (P<0.05) suggested lack of smaller studies reporting an inverse association between adiponectin and stroke.
In EPIC-Potsdam, higher levels of total-adiponectin were associated with better metabolic profiles. In contrast, total-adiponectin tended to be positively associated with IS risk. Interestingly, this association became stronger and significant after controlling for metabolic markers that have been suggested to be intermediates. Although (putative) overadjustment may have produced spurious results, unknown pathways underlying the increased risk cannot be excluded. Results of our meta-analysis based on 9 independent prospective studies extend those from 2 recent meta-analyses on the relationship between adiponectin and stroke, which included only 222 and 3 studies.23 Our meta-analysis indicated lack of association between adiponectin and stroke risk. If anything, adiponectin was directly related to stroke risk after controlling for metabolic factors that favorably correlate with adiponectin.
A trend toward increased risk of coronary heart disease11 and cardiovascular death21,24 associated with higher adiponectin despite better metabolic profiles has also previously been observed in studies in older populations. One possible explanation is that the progression of cardiovascular diseases may lead to adiponectin resistance.25 In the EPIC-Potsdam study, we excluded participants with a clinical history of major cardiovascular disease events at baseline. We also lagged the statistical analysis by 4 years to account for latency, which did not alter our results. It has also been suggested that the presence of diseases prompting to a hypercatabolic state may induce compensatory increased adiponectin levels.26 Our findings, however, did not substantially change after excluding prevalent cases of several chronic diseases or controlling for creatinine, and adjusting for N-terminal probrain-type natriuretic peptide only slightly attenuated the risk estimates. Also, the better metabolic profiles associated with higher total-adiponectin would argue against these explanations. We focused on total-adiponectin and not on HMW-adiponectin, which is thought to be the most biologically active form.27 Nevertheless, total and HMW-adiponectin have shown to be highly correlated,11 and recent findings do not support HMW-adiponectin to be more strongly related to risk of stroke11 or coronary heart disease.28 Adiponectin levels increase with age, and this could be a critical factor in assessing the associations between adiponectin and cardiovascular end points. However, age did not modify the association between adiponectin and incident stroke in our study population, which age was mostly between 35 and 65 years. Finally, it has been suggested that very high levels of adiponectin could have harmful effects by means of the activation of complement mechanisms.29
Limitations of our prospective study include that a single assessment of adiponectin may be susceptible to within-individual variation. Adiponectin concentrations, however, have shown to be quite stable over time.30 Furthermore, only a third of the study population was in fasting status. However, adiponectin values did not differ according to fasting status, and adjusting for it did not affect our risk estimates.
The meta-analysis included studies that differed in population characteristics, methods, variables controlled for, and outcomes considered (total stroke/IS). However, we did not find that these factors influenced results across studies. Although we observed a potential risk for publication bias, it may be unnecessary to evaluate the existence of preferential publication when very few studies reported significant results.31
This prospective study and meta-analysis shows that circulating total-adiponectin does not relate to risk of stroke. If anything, adiponectin relates directly to stroke risk after controlling for metabolic factors that correlate favorably with adiponectin. The number of studies on HMW-adiponectin is limited, and no studies on other adiponectin fractions are available. Therefore, it remains unclear whether specific molecular-weight fractions of adiponectin may influence risk of stroke.
We are grateful to W. Fleischhauer for case ascertainment; E. Kohlsdorf for data management; A. Bury, S. Herbert, and E. Eiden for biochemical analyses; and to Drs Söderberg (Umeå University Hospital, Sweden) and Prugger (Paris Cardiovascular Research Centre, University of Paris Descartes, France) for kindly providing additional data for the meta-analysis.
Sources of Funding
This study was supported by German Federal Ministry of Education; German Federal Ministry of Science (01 EA 9401); European Union (SOC 95201408 05F02 and SOC 98200769 05F02); and German Cancer Aid (70-2488-Ha I).
The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.113.001851/-/DC1.
- Received April 24, 2013.
- Accepted September 25, 2013.
- © 2013 American Heart Association, Inc.
- Qiao L,
- Zou C,
- van der Westhuyzen DR,
- Shao J
- Bevan S,
- Meidtner K,
- Lorenz M,
- Sitzer M,
- Grant PJ,
- Markus HS
- Gardener H,
- Sjoberg C,
- Crisby M,
- Goldberg R,
- Mendez A,
- Wright CB,
- et al
- Weikert C,
- Stefan N,
- Schulze MB,
- Pischon T,
- Berger K,
- Joost HG,
- et al
- Prentice RL
- Matsumoto M,
- Ishikawa S,
- Kajii E
- Ogorodnikova AD,
- Wassertheil-Smoller S,
- Mancuso P,
- Sowers MR,
- Rajpathak SN,
- Allison MA,
- et al
- Rajpathak SN,
- Kaplan RC,
- Wassertheil-Smoller S,
- Cushman M,
- Rohan TE,
- McGinn AP,
- et al
- Wannamethee SG,
- Shaper AG,
- Whincup PH,
- Lennon L,
- Sattar N
- Van Berendoncks AM,
- Garnier A,
- Beckers P,
- Hoymans VY,
- Possemiers N,
- Fortin D,
- et al
- Kaplan RC,
- Ho GY,
- Xue X,
- Rajpathak S,
- Cushman M,
- Rohan TE,
- et al
- Ioannidis JP,
- Trikalinos TA