Adiponectin exhibits cardioprotective properties in experimental studies, but elevated levels have been linked to increased mortality in older adults and patients with chronic heart failure (HF). The adipokine’s association with new-onset HF remains less well defined. The aim of this study was to investigate the associations of total and high–molecular weight (HMW) adiponectin with incident HF (n = 780) and, in a subset, echocardiographic parameters in a community-based cohort of adults aged ≥65 years. Total and HMW adiponectin were measured in 3,228 subjects without prevalent HF, atrial fibrillation or CVD. The relations of total and HMW adiponectin with HF were nonlinear, with significant associations observed only for concentrations greater than the median (12.4 and 6.2 mg/L, respectively). After adjustment for potential confounders, the hazard ratios per SD increment in total adiponectin were 0.93 (95% confidence interval 0.72 to 1.21) for concentrations less than the median and 1.25 (95% confidence interval 1.14 to 1.38) higher than the median. There was a suggestion of effect modification by body mass index, whereby the association appeared strongest in participants with lower body mass indexes. Consistent with the HF findings, higher adiponectin tended to be associated with left ventricular systolic dysfunction and left atrial enlargement. Results were similar for HMW adiponectin. In conclusion, total and HMW adiponectin showed comparable relations with incident HF in this older cohort, with a threshold effect of increasing risk occurring at their median concentrations. High levels of adiponectin may mark or mediate age-related processes that lead to HF in older adults.
Obesity and diabetes are foremost risk factors for heart failure (HF), which has drawn attention to the adipocyte-derived hormone adiponectin as a potential pathophysiologic mediator. Adiponectin exhibits insulin-sensitizing and antiatherogenic properties and the ability to counter ischemia-reperfusion injury, apoptosis, and hypertrophic signaling in cardiomyocytes. Clinical studies in healthy, middle-aged adults have reported inverse associations of adiponectin with left ventricular (LV) mass and diastolic function, suggesting that the adipokine could offer protection against HF. In patients with established HF, however, higher adiponectin concentrations portend in-creased mortality. This positive association is influenced by natriuretic peptides, which can directly stimulate adiponectin secretion, and by the weight loss that characterizes HF-associated cachexia, such that higher adiponectin levels in this setting may reflect underlying HF severity. This may account for the relation observed between higher adiponectin levels and worse LV systolic function in older, higher risk adults, which was attenuated by adjustment for natriuretic peptides. Still, the association between adiponectin and incident HF is not well defined. Whereas 2 modestly sized, population-based studies reported null associations for total adiponectin and new-onset HF, a larger investigation suggested a J-shaped association in men. Moreover, no prospective study to date has examined this relation for the reportedly more bioactive high–molecular weight (HMW) adiponectin. We therefore investigated the associations of total and HMW adiponectin with new-onset HF in a large older cohort, and also explored the adipokine’s relation to LV structure and function in a subset of patients with available echocardiograms.
Methods
The Cardiovascular Health Study (CHS) is a prospective survey of risk factors for cardiovascular disease (CVD) in community-living United States adults aged ≥65 years. An original cohort of 5,201 subjects was enrolled in 1989 and 1990. A second cohort of 687 African-Americans was recruited in 1992 and 1993. All subjects underwent health evaluations per standardized protocols.
Of the 5,553 subjects who participated in the 1992 and 1993 examinations (hereafter “baseline”), 4,715 had samples available for adiponectin measurement. For the present analyses, subjects with prevalent HF, atrial fibrillation, or CVD were excluded (n = 1,444). These conditions were ascertained through questionnaires, review of medical records, or adjudication of interim events. Another 43 subjects were excluded for missing laboratory measures assayed after baseline that were not part of the initial studywide imputation, leaving 3,228 eligible participants.
Anthropometry was assessed in standard fashion. Hypertension was defined as systolic blood pressure ≥140 mm Hg or diastolic blood pressure ≥90 mm Hg or by self-report and antihypertensive treatment. Diabetes was defined as fasting glucose ≥126 mg/dl or hypoglycemic therapy. Subclinical CVD was based on carotid ultrasound, ankle-brachial index, electrocardiography, and the Rose angina questionnaire. Blood samples were collected after a 12-hour overnight fast, and laboratory testing was performed as reported. Homeostasis model assessment of insulin resistance and cystatin C–based estimated glomerular filtration rate were calculated using standard methods. Total and HMW adiponectin were measured using an enzyme-linked immunosorbent assay (Millipore, Billerica, Massachusetts) on stored ethylenediaminetetraacetic acid plasma; interassay analytical coefficients of variation were 6.9% and 11.1%, respectively.
Two-dimensional and Doppler echocardiography was performed in 1994 and 1995, following standardized protocols. LV mass was indexed to gender, height and weight and anteroposterior left atrial (LA) diameter to gender and height only, on the basis of regression equations derived in a healthy subset of the cohort (body mass index [BMI] <30 kg/m 2 and absence of hypertension, diabetes, clinical and subclinical CVD, and cardiovascular medications) using previously described methods. The corresponding indexes, LV mass/(e 4.47 + 0.1[male] × height −0.64 × weight 0.72 ) and LA diameter/(e −0.16 + 0.05[male] × height 0.28 ), were then multiplied by 100 to yield percentage predicted LV mass and LA diameter. LV hypertrophy and LA enlargement were defined as percentage predicted values >95th percentile, or >144% and >126%, respectively. The LV ejection fraction was classified as normal (≥55%) or reduced (<55%). Diastolic function was assessed using the transmitral E/A ratio (<0.7, 0.7 to 1.5, or >1.5).
Semiannual contacts were used to survey for potential clinical events. All potential incident events were confirmed through review of subjects’ medical records. Diagnosis and treatment were used to determine the presence of HF. The CHS cardiovascular events committee also reviewed symptoms, signs, and chest x-ray findings of HF to classify all events.
Cross-sectional associations of adiponectin with covariates were evaluated using Pearson’s correlations or Student’s t tests. Missing data on baseline covariates (n = 273) were handled by carrying forward previous values or multiple imputation. Associations of total and HMW adiponectin with incident HF were evaluated with Cox models (with the proportional-hazards assumption verified by Schoenfeld residuals), while those with echocardiographic measures (after excluding interim HF cases) were assessed using logistic regression. The functional forms of these associations were evaluated with penalized cubic splines. On the basis of visual inspection of the cubic spline plots, and previous data, continuous associations of total and HMW adiponectin with HF were modeled using linear splines with knots at their median concentrations.
To determine the independent associations of total and HMW adiponectin with outcomes, we fit models to adjust for aggregate potential confounding. Models were initially adjusted for age, gender, and race. Thereafter, we considered an array of covariates previously linked to adiponectin levels, arriving at a parsimonious model that additionally included income, systolic and diastolic blood pressure, BMI, angiotensin-converting enzyme inhibitor use, current smoking, alcohol use, self-reported health status, and estimated glomerular filtration rate. We also adjusted for confounding by N-terminal pro–brain natriuretic peptide (NT-proBNP) in the subset with available measurements (n = 2,500). For incident HF, exploratory analyses evaluated the influence of putative mediators, consisting of subclinical CVD, diabetes, lipids, and C-reactive protein. We also examined the impact of adjusting for interim coronary heart disease as a time-varying covariate. Assessment for interaction with key covariates entailed inclusion of cross-product terms.
All analyses were performed using Stata version 11.0 (StataCorp LP, College Station, Texas) or R version 2.13.0 (R Foundation for Statistical Computing, Vienna, Austria).
Results
Eligible participants had a mean age of 74 ± 5 years, and 63.5% were women. Table 1 lists the associations of total adiponectin with baseline covariates. Adiponectin was positively correlated with age, high-density lipoprotein, and NT-proBNP but negatively correlated with BMI, waist-to-hip ratio, diastolic blood pressure, homeostasis model assessment of insulin resistance, low-density lipoprotein, triglycerides, physical activity, and C-reactive protein. In turn, total and HMW adiponectin were highly positively correlated. Furthermore, total adiponectin levels were higher in women, non-African-American subjects, alcohol drinkers, and participants with previous weight loss or LA enlargement. Concentrations were instead lower in subjects with more modest incomes; with hypertension, diabetes, or subclinical CVD; or receiving antihypertensive therapy.
| Covariate | Correlation Coefficient or Geometric Mean (95% CI) (mg/L) | p Value |
|---|---|---|
| Age | 0.20 | <0.001 |
| Gender | <0.001 | |
| Female | 14.1 (13.8–14.4) | |
| Male | 9.9 (9.6–10.2) | |
| Race/ethnicity | <0.001 | |
| Non-African-American | 13.1 (12.8–13.3) | |
| African-American | 9.4 (8.9–9.8) | |
| Income <$16,000 | 0.009 | |
| Yes | 12.0 (11.6–12.4) | |
| No | 12.6 (12.3–12.9) | |
| BMI (kg/m 2 ) | −0.33 | <0.001 |
| Waist-to-hip ratio | −0.36 | <0.001 |
| Hypertension | <0.001 | |
| Yes | 11.7 (11.4–12.1) | |
| No | 12.8 (12.5–13.1) | |
| Systolic blood pressure | 0.01 | 0.700 |
| Diastolic blood pressure | −0.06 | 0.001 |
| Diabetes mellitus | <0.001 | |
| Yes | 8.7 (8.2–9.2) | |
| No | 13.0 (12.8–13.2) | |
| Homeostasis model assessment of insulin resistance ∗ | −0.17 | <0.001 |
| Low-density lipoprotein | −0.03 | 0.049 |
| High-density lipoprotein | 0.48 | <0.001 |
| Triglycerides † | −0.32 | <0.001 |
| Current smoker | 0.081 | |
| Yes | 11.8 (11.2–12.5) | |
| No | 12.5 (12.2–12.7) | |
| ≥7 alcoholic drinks/week | <0.001 | |
| Yes | 13.4 (12.8–14.1) | |
| No | 12.2 (12.0–12.5) | |
| Physical activity b | −0.04 | 0.028 |
| Lipid-lowering medication | 0.004 | |
| Yes | 11.3 (10.5–12.0) | |
| No | 12.5 (12.2–12.7) | |
| Use of any antihypertensive medication | <0.001 | |
| Yes | 11.6 (11.2–11.9) | |
| No | 12.9 (12.6–13.2) | |
| β-blocker use | <0.001 | |
| Yes | 10.3 (9.7–10.9) | |
| No | 12.6 (12.4–12.8) | |
| Angiotensin-converting enzyme inhibitor use | <0.001 | |
| Yes | 10.8 (10.2–11.5) | |
| No | 12.5 (12.3–12.8) | |
| Weight loss >10 lb in previous 3 yrs ‡ | <0.001 | |
| Yes | 14.8 (13.8–15.9) | |
| No | 12.8 (12.5–13.0) | |
| Fair/poor health status | 0.990 | |
| Yes | 12.4 (11.8–13.0) | |
| No | 12.4 (12.2–12.6) | |
| Serum albumin | −0.03 | 0.053 |
| Estimated glomerular filtration rate | 0.01 | 0.530 |
| C-reactive protein | −0.08 | <0.001 |
| NT-proBNP | 0.17 | <0.001 |
| HMW adiponectin | 0.93 | <0.001 |
| Subclinical CVD | <0.001 | |
| Yes | 12.1 (11.8–12.4) | |
| No | 12.9 (12.6–13.3) | |
| LV hypertrophy | 0.34 | |
| Yes | 12.0 (11.2–12.8) | |
| No | 12.4 (12.1–12.7) | |
| Reduced LV systolic function | 0.73 | |
| Yes | 12.1 (11.1–13.2) | |
| No | 12.3 (12.1–12.6) | |
| LV diastolic function (transmitral E/A ratio) | 0.14 | |
| <0.7 | 12.6 (11.3–14.1) | |
| 0.7–1.5 | 12.3 (12.0–12.5) | |
| >1.5 | 11.7 (11.2–12.2) | |
| Dilated LA diameter | <0.001 | |
| Yes | 14.4 (13.3–15.7) | |
| No | 12.0 (11.8–12.3) |
∗ In subjects not receiving hypoglycemic medications (n = 3,012).
During a median follow-up period of 11.7 years, 780 incident HF events occurred. Adjusted cubic spline analyses demonstrated that the association of total adiponectin with incident HF was nonlinear (p = 0.012), with a suggestion of a similar nonlinear relation for HMW adiponectin (p = 0.12) ( Figure 1 ). The relations were such that, below median concentrations of total adiponectin (12.4 mg/L) and HMW adiponectin (6.2 mg/L), there were no significant associations with incident HF at various levels of adjustment for potential confounders, including NT-proBNP ( Table 2 , models 1 to 3). Above their median concentrations, however, total and HMW adiponectin exhibited significant direct associations with new-onset HF. After extensive adjustment for confounders (model 2), for every SD increment in total adiponectin (SD 7.9 mg/L) or HMW adiponectin (SD 5.9 mg/L), HF risk increased by about 1/4. This was attenuated slightly by additional adjustment for NT-proBNP (model 3).
| Model | Hazard Ratio per SD ∗ Increment (95% CI) | |||
|---|---|---|---|---|
| <Median | p Value | ≥Median | p Value | |
| Total adiponectin (median 12.4 mg/L) | ||||
| Model 1 † | 0.83 (0.64–1.07) | 0.14 | 1.15 (1.04–1.26) | 0.005 |
| Model 2 ‡ | 0.93 (0.72–1.21) | 0.61 | 1.25 (1.14–1.38) | <0.001 |
| Model 3 § | 0.97 (0.72–1.30) | 0.84 | 1.18 (1.05–1.32) | 0.004 |
| Model 4 || | 1.23 (0.94–1.61) | 0.14 | 1.33 (1.21–1.48) | <0.001 |
| HMW adiponectin (median 6.2 mg/L) | ||||
| Model 1 † | 0.82 (0.62–1.09) | 0.17 | 1.15 (1.05–1.25) | 0.002 |
| Model 2 ‡ | 0.97 (0.73–1.29) | 0.83 | 1.23 (1.13–1.34) | <0.001 |
| Model 3 § | 0.95 (0.69–1.30) | 0.73 | 1.16 (1.05–1.29) | 0.005 |
| Model 4 || | 1.29 (0.96–1.75) | 0.09 | 1.30 (1.19–1.42) | <0.001 |
∗ Total adiponectin, SD 7.9 mg/L; HMW adiponectin, SD 5.9 mg/L.
† Adjusted for age, race, and gender.
‡ Adjusted for age, race, gender, income, BMI, systolic blood pressure, diastolic blood pressure, angiotensin-converting enzyme inhibitor use, current smoking, alcohol use, health status, and estimated glomerular filtration rate.
§ Adjusted for covariates in model 2 plus NT-proBNP (in subset with available measures; n = 2,500).
|| Adjusted for covariates in model 2 plus diabetes, low-density lipoprotein, high-density lipoprotein, triglycerides, and C-reactive protein.
Further adjustment for putative metabolic and inflammatory mediators (model 4), however, led to higher risk estimates in the lower range of concentrations for total and HMW adiponectin ( Table 2 ). This was characterized by a tendency toward an increased risk for HF with increasing concentrations, making the association of either measure with HF more linear across its distribution. No meaningful change in risk estimates occurred at the upper range of concentrations. Findings were unchanged after additional adjustment for subclinical CVD or homeostasis model assessment of insulin resistance.
We observed no significant interaction by age, gender, race, diabetes, or health status (p >0.12). There was a suggestion of effect modification by BMI for total (p = 0.010) and HMW adiponectin (p = 0.057), however, whereby the direct association with HF observed for adiponectin levels greater than their medians appeared strongest in participants with lower BMIs. For instance, stratifying participants by World Health Organization categories of BMI, those with BMIs <25 kg/m 2 showed a significant relation for total adiponectin above its median concentration (adjusted hazard ratio per SD [model 2] 1.37, 95% confidence interval [CI] 1.21 to 1.55), but this was not apparent in participants in higher BMI strata (BMI 25 to 29.9 kg/m 2 : adjusted hazard ratio 1.13, 95% CI 0.95 to 1.34; BMI ≥30 kg/m 2 : adjusted HR 0.91, 95% CI 0.64 to 1.32). This interaction persisted after the exclusion of participants with BMIs <18.5 kg/m 2 or those with fair or poor health status.
There was no evidence of a differential effect of adiponectin on systolic versus diastolic HF in the subset of 356 participants in whom HF type could be ascertained. Furthermore, the overall findings were not meaningfully altered after adjustment for incident coronary heart disease (n = 113), nor were they materially different after the exclusion of elevated NT-proBNP or antecedent weight loss.
Unlike the associations with incident HF, there was no evidence of a nonlinear relation between adiponectin measures and echocardiographic parameters (p >0.09). There were no significant associations between total adiponectin and LV hypertrophy (adjusted [model 2] odds ratio [OR] 1.04 per SD increment, 95% CI 0.88 to 1.23) or high (>1.5) transmitral E/A ratio compared with normal (0.7 to 1.5) (adjusted OR 1.11 per SD, 95% CI 0.88 to 1.40). Total adiponectin did show a significant inverse association with low (<0.7) transmitral E/A ratio compared with normal (adjusted OR 0.88 per SD, 95% CI 0.78 to 0.99), as well as borderline direct associations with LA enlargement (adjusted OR 1.13 per SD, 95% CI 0.99 to 1.32, p = 0.097) and reduced LV systolic function (adjusted OR 1.23 per SD, 95% CI 1.00 to 1.51, p = 0.054), but these were abolished by adjustment for NT-proBNP. To mirror the approach taken with incident HF, corresponding findings stratified by the median concentration of total adiponectin are presented in Table 3 , which suggest that such associations occurred predominantly above the median. Findings were similar for HMW adiponectin. There was no interaction with BMI in these associations.
