Cardiovascular Mortality Risk in Chronic Kidney Disease:  Comparison of Traditional and Novel Risk Factors FREE

The National Kidney Foundation, American Heart Association, and the Seventh Joint National Committee on Prevention, Detection, Evaluation, and Treatment of High Blood Pressure have classified the presence of chronic kidney disease as a cardiovascular risk factor.^1- 3 Chronic kidney disease is associated with substantially increased risk for cardiovascular disease morbidity and mortality, independent of traditional cardiovascular risk factors such as diabetes, hypertension, lipoprotein levels, and tobacco use.^1,4- 6 In addition, certain novel cardiovascular risk factors are more prevalent in persons with chronic kidney disease, including elevated inflammatory and prothrombotic factors (C-reactive protein [CRP], fibrinogen, interleukin 6 [IL-6], and factor VIII), and lipoprotein(a) (Lp[a]), and decreased hemoglobin levels. These novel risk factors have been discussed as potential mechanisms for the elevated cardiovascular risk of chronickidney disease,^7- 9 but few studies have evaluated their association with cardiovascular events in persons with chronic kidney disease or compared the strength of association of traditional and novel cardiovascular risk factors. Though the National Institutes of Health and the National Kidney Foundation have prioritized the reduction of cardiovascular disease burden in persons with chronic kidney disease, prevention efforts will first require an in-depth understanding of the determinants of cardiovascular risk in persons with chronic kidney disease.¹⁰

In the Cardiovascular Health Study (CHS), a well-characterized cohort of elderly persons with a high prevalence of chronic kidney disease, we compared the association of traditional and novel risk factors with cardiovascular mortality among subgroups of participants with and without chronic kidney disease at baseline. In addition, we estimated the absolute risk associated with each candidate risk factor and constructed receiver operating characteristic (ROC) curves to estimate the aggregate predictive utility of traditional and novel risk factors.¹¹

The CHS is a prospective cohort study of risk factors for cardiovascular disease in elderly men and women. The study recruited eligible persons who resided in the households of individuals identified from an age-stratified random sample from Medicare eligibility lists in Forsyth County, North Carolina; Sacramento County, California; Washington County, Maryland; and Pittsburgh, Pa. Household members and spouses of the person being recruited were also invited to participate in the CHS if they met the following inclusion criteria: (1) at least 65 years, (2) not institutionalized, (3) expected to remain in the current community for 3 years or longer, (4) not under active treatment for cancer, and (5) gave written informed consent without requiring a proxy respondent at entry. Among those who met the eligibility requirements and were invited to participate, 57% were enrolled. The initial 5201 participants (original cohort) were enrolled from 1989 to June 1990; an additional 687 black participants (African-American cohort) were recruited and enrolled in 1992-1993. Race was self-reported in 5 categories: white, black, American Indian/American Native, Asian/Pacific Islander, other. A separate question addressed Hispanic heritage. Race was subsequently collapsed into white, black, and other because the nonwhite or nonblack proportion was very small. The baseline examination for each cohort included a medical history, physical examination, laboratory testing, and assessments of cardiovascular disease status. The study design, quality-control procedures, laboratory methods, and blood pressure measurement procedures have been published previously.¹²

This study was approved by the internal review boards of the University of California, San Francisco, University of Washington, and University of Pittsburgh.

This analysis included all participants with creatinine measures at baseline (5808). Serum creatinine was measured using the Kodak Ektachem 700 Analyzer (Eastman Kodak, Rochester, NY), a colorimetric method. We estimated glomerular filtration rate (GFR) using the modified formula of Levey et al,¹⁰ and categorized chronic kidney disease as a GFR less than 60 mL/min per 1.73 m², based on recommendations of the National Kidney Foundation.³ Serum creatinine in the Modification of Diet in Renal Disease (MDRD) study and subsequently in a subset of individuals included in the Third National Health and Nutrition Examination Survey (NHANES III) were both analyzed at the Cleveland Clinic laboratory. Because creatinine values vary across clinical laboratories, we calibrated the creatinine concentration in our current study to the Cleveland Clinic laboratory indirectly using a method previously reported by Manjunath and colleagues,¹³ and others.^14- 16 This calibration technique is based on the population-based recruitment of CHS and assumes that the mean serum creatinine comparable for a given age, race, and sex strata in CHS should be comparable with NHANES III. A linear regression of data combining each study individually with NHANES III showed that serum creatinine values were 0.11 mg/dL (9.72 μmol/L) higher in the original cohort of CHS and 0.04 mg/dL (3.53 μmol/L) higher in the CHS African American cohort. These values were then subtracted from measured creatinine levels before use in the current study.

Traditional risk factors were measured and categorized on all CHS participants. Each was dichotomized to facilitate comparisons of their association with the outcome with other risk factors. These variables were defined as follows: systolic hypertension (systolic blood pressure ≥140 mm Hg); diabetes (history of diabetes, use of hypoglycemic agent or insulin, or fasting glucose ≥126 mg/dL [>7.0 mmol/L]); current smoking; high-density lipoprotein ([HDL] ≤40 mg/dL [<1.04 mmol/L]); low-density lipoprotein ([LDL] ≥130 mg/dL (>3.37 mmol/L)]; triglycerides ≥200 mg/dL (2.26 mmol/L); regular alcohol use (≥2 drinks/wk); obesity (body mass index [BMI] ≥30 [calculated as weight in kilograms divided by the square of height in meters]); physical inactivity (lowest quartile of CHS reported energy expenditure)¹⁷; and left ventricular hypertrophy (LVH) by electrocardiogram.¹⁸ Subsequent analyses evaluated systolic blood pressure, HDL, LDL, and triglyceride levels, alcohol use (drinks/wk), BMI, and physical activity as continuous variables. All serum measures were conducted from baseline specimens.

Novel risk factors were measured on stored serum or plasma among either theentire cohort of CHS participants (CRP, fibrinogen, IL-6, and hemoglobin) or among the original cohort only (Lp[a] and factor VIII coagulant activity) from the baseline visit. The specific assays used for each measure were as follows: fibrinogen, BBL Fibrometer (Becton-Dickson, Cockeysville, Md)^19- 20; factor VIIIc, Coag-A-Mate (Organon Teknika, Dublin, Ireland)^19- 21; and IL-6, ultrasensitive enzyme-linked immunosorbent assay (R&D Systems, Minneapolis, Minn). C-reactive protein was measured using an enzyme-linked immunosorbent assay developed by the CHS central blood laboratory (CHS Blood Laboratory, Colchester, Vt). Lipoprotein(a) was measured with a monoclonal antibody–based enzyme-linked immunosorbent assay.²² The reagents for Lp(a) were obtained from Genentech (South San Francisco, Calif). Results are expressed in terms of the Lp(a) lipoprotein protein concentration (excluding lipid), with reference to a purified standard calibrated by quantitative amino acid analysis. The overall coefficient of variation for Lp(a) lipoprotein measurements in this study was 7.5%. Anemia was defined as hemoglobin of 12 mg/dL or less for both men and women. Abnormal levels for each of the other novel risk factors were defined by the highest quartile. Additional models analyzed hemoglobin, CRP, fibrinogen, IL-6, Lp(a), and factor VIIIc as continuous variables per standard deviation.

All fatal events were reviewed and classified by a mortality review committee using information from death certificates, autopsy and coroners' forms, hospital records, and interviews with attending physicians, next of kin, and witnesses. Deaths were classified as cardiovascular or noncardiovascular; cardiovascular death was defined as mortality caused by coronary heart disease, heart failure, peripheral vascular disease, and cerebrovascular disease.¹³ The outcome of this analysis was time to cardiovascular death. Participants were censored when they left the study or died from a noncardiovascular cause.

The prevalence or the level of each traditional and novel risk factor was initially compared among persons with and without chronic kidney disease, using χ² tests or t tests as appropriate. To compare the association of each dichotomized risk factor with cardiovascular mortality among persons with and without chronic kidney disease, proportional hazards models were constructed, stratified by chronic kidney disease presence. In model 1, these models were adjusted for age, sex, race, education, medication use (candidates were β-blockers, calcium channel blockers, angiotensin-converting enzyme inhibitors, and diuretics) and prevalent cardiovascular disease (defined by prior myocardial infarction, coronary revascularization, or cerebrovascular disease), and the individual risk factor of interest. In model 2, the stratified models were also adjusted for all the other traditional risk factors defined above. Candidate novel risk factors were not combined in multivariate analyses.

To evaluate for the presence of multiplicative interactions, we created a product term for each risk factor with chronic kidney disease and evaluated its significance in the multivariate model. These same methods were repeated for the risk factors when analyzed as continuous variables. C-reactive protein and IL-6 levels were log-transformed for these analyses because of their skewed distribution. Interactions of race and sex were tested with each risk factor for predicting the outcome of cardiovascular mortality.

We next estimated the absolute risk for each dichotomized traditional and novel risk factor, stratified by the presence of chronic kidney disease. The absolute risks were estimated by multiplying the incidence rate of the unexposed group by the adjusted relative risk increase (adjusted hazard ratio − 1) of the predictor. For example, the absolute risk associated with diabetes in the no–chronic kidney disease stratum was calculated by multiplying the incidence rate for cardiovascular death in participants with neither diabetes nor chronic kidney disease by the adjusted relative risk increase for diabetes within the no–chronic kidney disease strata. Similarly, the absolute risk associated with diabetes within the chronic kidney disease stratum was determined by the product of the cardiovascular mortality incidence in nondiabetics with chronic kidney disease and the adjusted relative risk increase for diabetes within the chronic kidney disease subgroup.

To compare traditional and novel risk factors in aggregate, we constructed ROC curves and determined their area under the curve (AUC). This method evaluated the overall ability of the risk factor collection to discriminate participants who would or would not subsequently die from a cardiovascular cause. Stratified by the presence of chronic kidney disease, we first determined the AUC for traditional risk factors alone; then, we calculated the AUC for the combined traditional plus novel risk factors. We compared the AUC of the 2 ROC curves within each chronic kidney disease strata using the method described by DeLong et al.²³

Intercooled Stata 8 (StataCorp LP, College Station, Tex) and SPSS 12 (SSPS, Chicago, Ill) were used for the statistical analyses. P values <.05 were considered statistically significant.

At study entry, 1249 (22%) participants had chronic kidney disease, defined by an estimated GFR of less than 60 mL/min per 1.73 m². Participants with chronic kidney disease were on average 3 years older; were more likely to be men and white; were less likely to have more than a high school education; and had a greater prevalence of cardiovascular disease than those without chronic kidney disease (Table 1). Those with chronic kidney disease also had higher levels of triglycerides and lower HDL cholesterol levels. Participants with chronic kidney disease used less alcohol, were less physically active, and had a greater prevalence of LVH. Among the novel risk factors,mean levels of CRP, fibrinogen, IL-6, factor VIIIc, and Lp(a) were higher and hemoglobin was lower among those with than among those without chronic kidney disease.

During an average follow-up time of 8.6 years, the annual risk of cardiovascular mortality was 32 per 1000 person-years among participants with baseline chronic kidney disease (342 events) and 16 per 1000 person-years among participants without chronic kidney disease (750 events). Diabetes, systolic hypertension, current smoking, low physical activity, and LVH were strongly associated with elevated risk among participants with and without chronic kidney disease, and alcohol use was associated with decreased risk even after adjustment for all other traditional risk factors (Table 2). Low HDL, elevated LDL, and triglyceride levels and obesity were not associated with the outcome in persons with or without chronic kidney disease. Although the point estimates for diabetes and LVH were modestly smaller and for low physical activity modestly larger among participants with chronic kidney disease, we observed no significant interactions among the 10 traditional risk factors with chronic kidney disease for predicting cardiovascular death.

In contrast to our findings for traditional risk factors, certain novel risk factors appeared to have weaker associations with cardiovascular mortality among participants with chronic kidney disease compared with the associations among those without chronic kidney disease (Table 2). Among the inflammatory and procoagulant factors evaluated, elevated levels of CRP, fibrinogen, and factor VIIIc had strong associations with cardiovascular mortality in participants without chronic kidney disease, but none was a significant predictor of the outcome in the chronic kidney disease group after multivariate analysis. We observed a significant interaction (P<.05) between presence of chronic kidney disease and both CRP and factor VIIIc for predicting cardiovascular mortality, whereas the interaction of chronic kidney disease with fibrinogen was of borderline significance (P = .09). Elevated levels of IL-6 had a somewhat stronger association with the outcome in the non–chronic kidney disease subgroup, though the test for interaction was not significant (P = .23). The association of elevated levels of Lp(a) with cardiovascular mortality was similar in both groups, but only significant in persons without chronic kidney disease. Anemia was not significantly associated with the outcome in either subgroup.

The comparisons of traditional and novel risk factors were repeated with most risk factors analyzed as continuous variables per standard deviation (Table 3). Systolic blood pressure had similar and significant associations with increased risk and physical activity with decreased risk of cardiovascular mortality in the subgroups with and without chronic kidney disease. As with the dichotomized predictors, LDL and HDL levels, triglycerides, and BMI were not associated with the outcome in either subgroup, and no significant interactions were observed.

Among the novel risk factors, log CRP, fibrinogen, and factor VIIIc levels again appeared to have larger associations with cardiovascular mortality in the subgroup without chronic kidney disease although the tests for interaction were only significant for fibrinogen and factor VIIIc (Table 3). The association of log IL-6 with the outcome was strong and equivalent in the participants with or without chronic kidney disease. Lipoprotein(a) and hemoglobin levels were not significantly associated with the outcome in either subgroup when modeled as a continuous variable.

We tested for interactions of sex and race with each risk factor as predictors of the outcome. None of these was significant at a level of P<.20.

For the 6 traditional risk factors that were significantly associated with cardiovascular mortality and the 6 novel risk factors, we estimated the absolute risk of each for cardiovascular mortality. This metric can be interpreted as the average annual increase in cardiovascular mortality risk on an absolute scale that is attributable to the risk factor within each subgroup of participants (with and without chronic kidney disease). Among those with chronic kidney disease (average risk of 32 cardiovascular deaths/1000 person-years), traditional risk factors were associated with the largest absolute elevations in cardiovascular mortality risk: LVH, an increased risk of 25 per 1000 person-years, and current smoking, an increased risk of 20 per 1000 person-years, were associated with the greatest increments of risk, followed by physical inactivity, an increased risk of 15 per 1000 person-years; diabetes an increased risk of 14 per 1000 person-years; and systolic hypertension, an increased risk of 14 cardiovascular deaths per 1000 person-years. Alcohol use was associated with a decreased risk of 11 deaths per 1000 person-years (Figure 1). The novel risk factors were associated with smaller increases in risk that were not statistically significant: lower hemoglobin had an increased risk of 9 deaths per 1000 person-years; elevated Lp(a), an increased risk of 6 deaths per 1000 person-years; IL-6, an increased risk of 5 deaths per 1000 person-years; and CRP, an increased risk of 5 deaths per 1000 person-years (Figure 1).

Among participants without chronic kidney disease (average risk of 16 cardiovascular deaths/1000 person-years), LVH (increased risk of 17/1000 person-years), diabetes (increased risk of 15/1000 person-years), and current smoking (increased risk of 13/1000 person-years) had the largest associations with mortality risk (Figure 1). Although increased CRP, fibrinogen, IL-6, factor VIIIc, and Lp(a) were all significantly associated with cardiovascular death, the absolute increase in risk ranged only from 4 deaths per 1000 for those with elevated Lp(a) to 8 deaths per 1000 person-years for those with elevated CRP levels (Figure 1).

Among participants with chronic kidney disease, the strong associations of traditional risk factors with cardiovascular mortality were reflected by the AUC of 0.73 (95% confidence interval [CI], 0.70-0.77; Figure 2). Adding novel risk factors had only a slight effect, increasing the AUC to 0.74 (95% CI, 0.71-0.78; P for difference = .15). Similarly, ROC curves among participants without chronic kidney disease (Figure 2) did not differ significantly when developed from traditional risk factors (AUC, 0.73; 95% CI, 0.69-0.76) or traditional novel risk factors (AUC, 0.72; 95% CI, 0.68-0.75; P for difference = .16).

In a community-based sample of elderly persons with chronic kidney disease, traditional risk factors were better predictors of cardiovascular mortality than novel risk factors. In descending order by their absolute risk, LVH, current smoking, physical inactivity, diabetes, elevated systolic blood pressure, and nonuse of alcohol were associated with substantial elevations in cardiovascular mortality risk among participants with chronic kidney disease. Among the novel risk factors evaluated, none was associated with elevated cardiovascular risk in this subgroup as a dichotomized predictor, although log IL-6 and log CRP predicted risk as linear variables. In ROC analysis, the 10 traditional risk factors accrued an AUC of 0.73, similar to that observed in the Framingham Heart Study; whereas the 6 novel risk factors had no significant additional utility for predicting the outcome.²⁴ These findings may suggest that the most promising future interventions to reduce cardiovascular mortality risk in elderly persons with chronic kidney disease would be intense modification of established risk factors rather than the pursuit of interventions to reduce levels of novel risk factors.

Prior literature evaluating the cardiovascular risk of persons with chronic kidney disease has largely been composed of longitudinal studies demonstrating elevated cardiovascular risk in persons with chronic kidney disease compared with persons with normal renal function and cross-sectional studies demonstrating that a variety of traditional and novel cardiovascular risk factors are elevated in the setting of chronic kidney disease. The relative importance of elevations in novel risk factors for determining cardiovascular risk in persons with chronic kidney disease has been relatively unexplored until recently.⁸ Muntner and colleagues²⁵ published a report from the Atherosclerosis Risk in Communities (ARIC) cohort of middle-aged adults that evaluated predictors of chronic heart disease events (myocardial infarction, cardiovascular death, or coronary revascularization) in participants with chronic kidney disease. These investigators also found strong associations of systolic blood pressure, diabetes, and current smoking with chronic heart disease events in persons with chronic kidney disease although their findings differed from ours because anemia, increased fibrinogen, and lower albumin levels did predict chronic heart disease events in analyses adjusted for age, race, sex, current smoking, diabetes, hypertension, total cholesterol, and clinical site. Muntner et al, however, did not present results with more extensive multivariate adjustment. Our cohort differs from ARIC in exclusively recruiting elderly persons, having a higher prevalence of chronic kidney disease (22% vs 5%), and having a different array of traditional and novel predictors. A recent case-control analysis by Knight et al²⁶ from the Nurses’ Health Study found an opposite interaction from ours because inflammatory factors (CRP, IL-6, and soluble tumor necrosis factor receptors I and II) were associated with greater coronary artery disease risk among participants with estimated GFR lower than 75 than in those with a GFR higher than 75. Other than the inclusion of men, the primary differences between our study and that of Knight et al are that the CHS had a higher prevalence of participants with chronic kidney disease, included exclusively elderly persons, and had a different cardiovascular outcome.

The potential implication of our article is that traditional risk factors may be the optimal targets for cardiovascular risk reduction in elderly patients with chronic kidney disease. Not only were these risk factors the strongest predictors of cardiovascular mortality, but several of them have proved and established interventions to modify their risk. In combination, the additive associations of these traditional risk factors with cardiovascular mortality in elderly persons with chronic kidney disease suggest great potential for beneficial interventions. On the other hand, we cannot assume that treating each of these traditional risk factors would lower cardiovascular mortality risk by the amount suggested in these analyses. Among these 6 traditional risk factors that predicted the outcome, only systolic blood pressure control has been shown in clinical trials to be a modifiable risk factor for cardiovascular events in the elderly. Although many health benefits are known to result from tobacco cessation, exercise, and glycemic control, we cannot assume that such interventions would necessarily reduce cardiovascular mortality in elderly persons with chronic kidney disease. Nevertheless, our findings raise the hypothesis that cardiovascular prevention efforts would be most productive if focused on aggressive lifestyle interventions and blood pressure control in this population.

Our finding that levels of CRP, fibrinogen, and factor VIIIc had larger associations with cardiovascular mortality in persons without chronic kidney disease is intriguing. In contrast, increased levels of IL-6 had a nearly equivalent association with the outcome in participants with or without chronic kidney disease. One possibility is that the levels of CRP, fibrinogen, and factor VIIIc are increased in the setting of chronic kidney disease, not because of increased production but rather due to their decreased renal clearance or degradation. Thus, the statistical association of their plasma levels with cardiovascular risk might be diminished in the setting of chronic kidney disease. An alternative explanation for these findings of interaction between the 3 biomarkers and chronic kidney disease for predicting cardiovascular mortality is that they could be chance findings.

The absence of an association between hemoglobin and cardiovascular mortality in our study differs from the associations of anemia with cardiovascular events observed in previous studies.^15,27- 28 One possibility is that the prevalence of anemia (defined as hemoglobin <12 g/dL in men and women) was too low at 4.2% to allow a precise estimate; however, we also found no association with hemoglobin evaluated as a linear variable. In addition, an association of anemia with the outcome could have been obscured by our adjustment for prevalent cardiovascular disease.

This study has several important limitations to consider. In our primary analyses, we dichotomized all of the risk factors to define absolute risks and to compare them across traditional and novel risk factors. Although this method may have reduced statistical power for detecting an association, the findings were similar when risk factors were evaluated as continuous variables. Another important issue is that the population consisted of community-dwelling elderly persons with chronic kidney disease; these findings may not generalize to younger populations with primary renal diseases. The observation that chronic kidney disease was more common in white than black participants raises the concern that our white and black participants may not be equally representative of the general community. We also defined chronic kidney disease based on estimated GFR rather than on actual measures of renal function; however, direct measures of GFR are expensive and cumbersome and are rarely used in large epidemiologic studies.²⁹ Finally, this study involved multiple comparisons and subgroup analyses with a single outcome, and thus had the potential for chance positive and negative findings.

In conclusion, we report that traditional cardiovascular risk factors had larger associations with cardiovascular mortality than novel risk factors in a community-based cohort of elderly persons with chronic kidney disease. Our findings suggest that interventions that aggressively target control of systolic blood pressure and glucose levels, tobacco cessation, and increased physical activity may have the greatest potential to reduce cardiovascular risk in this high-risk population. Although certain inflammatory and procoagulant factors were less strongly associated with cardiovascular risk in participants with chronic kidney disease, this finding requires confirmation in future studies. Future research should investigate whether aggressive lifestyle intervention in patients with chronic kidney disease can reduce their substantial cardiovascular risk.