Rapid Kidney Function Decline and Mortality Risk in Older Adults

Background  Impaired kidney function is associated with increased mortality risk in older adults. It remains unknown, however, whether longitudinal declines in kidney function are independently associated with increased cardiovascular and all-cause mortality in older adults.

Methods  The Cardiovascular Health Study evaluated a cohort of community-dwelling older adults enrolled from 1989 to 1993 in 4 US communities with follow-up through 2005. Among 4380 participants, the slope of annual decline in estimated glomerular filtration rate (eGFR) was estimated using both serum creatinine (eGFR_creat) and cystatin C (eGFR_cys) rates, which were measured at baseline, year 3, and year 7 of follow-up. Rapid decline in eGFR was defined as a loss greater than 3 mL/min/1.73 m² per year, and cardiovascular and all-cause mortality were assessed over a mean of 9.9 years of follow-up.

Results  Mean (SD) levels of creatinine and cystatin C were 0.93 (0.30) mg/dL and 1.03 (0.25) mg/L, respectively; mean (SD) eGFR_creat and eGFR_cys were 79 (23) mL/min/1.73 m² and 79 (19) mL/min/1.73 m², respectively. Individuals with rapid decline measured by eGFR_creat (n = 714; 16%) had increased risk of cardiovascular (adjusted hazard ratio [AHR], 1.70; 95% confidence interval [CI], 1.40-2.06) and all-cause (AHR, 1.73; 95% CI, 1.54-1.94) mortality. Individuals with rapid decline measured by eGFR_cys (n = 1083; 25%) also had increased risk of cardiovascular (AHR, 1.53; 95% CI, 1.29-1.80) and all-cause (AHR, 1.53; 95% CI, 1.38-1.69) mortality. The association of rapid decline in eGFR with elevated mortality risk did not differ across subgroups based on baseline kidney function, age, sex, race, or prevalent coronary heart disease.

Conclusion  Rapid decline in eGFR is associated with an increased risk of cardiovascular and all-cause mortality in older adults, independent of baseline eGFR and other demographic variables.

Chronic kidney disease (CKD) is a major and growing public health problem, with an estimated 17% of the adult US population affected.¹ Multiple studies, including those in older adults, have shown that prevalent CKD defined as an estimated glomerular filtration rate (eGFR) lower than 60 mL/min/1.73 m² is a strong independent risk factor for cardiovascular disease and all-cause mortality.^2-4 However, few studies have evaluated the association of long-term changes in kidney function with future cardiovascular events. In a recent analysis from the Cardiovascular Health Study (CHS),⁵ participants with mild kidney dysfunction (cystatin C level >1.0 mg/L and eGFR >60 mL/min/1.73 m²) whose disease progressed to CKD (eGFR <60 mL/min/1.73 m²) had a higher adjusted risk for cardiovascular events and mortality than individuals whose disease did not progress. In contrast to these studies, change in serum creatinine concentration was not independently associated with cardiovascular outcomes in the Heart and Estrogen/Progestin Replacement Study⁶ after adjustment for baseline creatinine level.

The challenge of accurate determination of eGFR in older adults has complicated attempts to study longitudinal changes in kidney function. Serum creatinine levels are affected in opposite directions by kidney function and muscle mass, and both are known to decline with aging.⁷ While creatinine-based eGFR (eGFR_creat) prediction equations may partially account for this limitation, their accuracy in older adults has been challenged.⁸

The use of serum measurement of cystatin C as a marker of kidney function provides greater sensitivity for detecting decreased kidney function than do creatinine-based estimates in an elderly population.^9,10 Previous analyses from the CHS have also shown that baseline cystatin C level is more strongly associated with cardiovascular events and all-cause mortality than eGFR_creat.¹¹

Based on these observations, we hypothesized that elderly participants in the CHS with the greatest declines in eGFR over the 7 years of follow-up would be at elevated risk for cardiovascular and all-cause mortality, and that cystatin C–based estimates of GFR (eGFR_cys) decline would be a more powerful predictor of mortality risk than eGFR_creat.

The CHS is a longitudinal study of community dwelling older adults designed to evaluate risk factors for cardiovascular disease. The design of this study has been described previously.¹² In brief, participants 65 years or older were recruited from Medicare eligibility lists in 4 US communities (Forsyth County, North Carolina; Sacramento County, California; Washington County, Maryland; and Pittsburgh, Pennsylvania). An initial 5201 participants were recruited between 1989 and 1990. Blood samples were drawn from these individuals in the 1989-1990 period (baseline), the 1992-1993 period (year 3), and 1996-1997 (year 7). An additional 687 black participants were added to the study in 1992-1993; these individuals had blood samples drawn in 1992-1993 (year 3) and 1996-1997 (year 7). In total, 4380 of the 5888 participants in CHS had at least 2 blood samples from these visits available for measurement of cystatin C and serum creatinine levels and thus met criteria for inclusion in this analysis.

All participants provided written informed consent, and the institutional review boards of the University of California, San Francisco, University of Washington, Seattle, and University of Pittsburgh, Pittsburgh, Pennsylvania approved the study.

Frozen serum samples stored at −70°C from the visits at baseline, year 3, and year 7 were available for measurement of cystatin C levels. Cystatin C levels were measured using a particle-enhanced immunonephelometric assay (N Latex Cystatin C; Dade Behring, now Siemens Healthcare Diagnostics Inc, Deerfield, Illinois) with a nephelometer (BNII, Siemens Healthcare Diagnostics Inc). Previous work has shown this assay to be stable through several freeze-thaw cycles.¹³ For measurement of cystatin C levels, intra-assay coefficients of variation (CVs) ranged from 2.0% to 2.8%, and interassay CVs ranged from 2.3% to 3.1%. The1992-1993 blood samples for cystatin C were assayed first in 2003. Subsequently, the 1989-1990 and 1996-1997 levels were assayed in 2006. Creatinine levels were measured immediately after the samples were drawn using a colorimetric method (Ektachem 700; Eastman Kodak, Rochester, New York). The mean CV for monthly controls was 1.94% (range, 1.16%-3.60%).

To compare changes in kidney function as measured by eGFR_creat and eGFR_cys, we used estimating equations to transform each measurement into a GFR estimate. To determine eGFR_cys, we used an equation derived from a pooling of cohorts that used iothalamate clearance as the criterion standard (eGFR_cys = 76.7 × cystatin C^−1.18).¹⁴ While several different equations have been proposed for eGFR_cys, this equation is based on the largest cohort of subjects derived from multiple data sources that all used nephelometric methods for cystatin C evaluation. We indirectly calibrated the measured serum creatinine level in the CHS cohort to the Cleveland Clinic Laboratory using NHANES III data, as described previously.^15,16 To determine eGFR_creat, we used the 4-variable equation of the Modification of Diet in Renal Disease Study Group¹⁷ (eGFR_creat = 186.3 × serum creatinine^−1.154 × age^−0.203 × 1.212 [if black] × 0.742 [if female]).

Rates of change were calculated using the 2 or 3 available cystatin C and creatinine measurements. Annualized change in eGFR was calculated using a least-squares regression slope. We examined the association of change in eGFR with mortality risk on a continuous scale by creating cubic spline plots of the middle 95% of eGFR_cys change vs mortality. The 25% of the cohort with the largest decline in eGFR_cys corresponded approximately to an annual loss of 3 mL/min/1.73 m²; we used this as a cutoff value for “rapid decline,” representing a magnitude of change that is 3 times the rate previously described in studies of normal aging, and that was beyond the range of noise in measurement.¹⁸

Events that occurred after at least 2 completed measures of cystatin C and creatinine levels were included. Events were ascertained by annual examinations and interim 6-month telephone interviews through June 30, 2005, with a median follow-up time of 9.9 years (maximum follow-up of 11.1 years). The methods of ascertaining and adjudicating events have been described previously.¹⁹ Cardiovascular mortality was defined as death from coronary heart disease, heart failure, peripheral vascular disease, or cerebrovascular disease. Medical records, death certificates, obituary review, household contacts, and the Centers for Medicare and Medicaid Services health care utilization database were used to determine vital status. Through these methods, outcome measures were obtained in 100% of participants.

We chose covariates as potential confounding factors based on their biological plausibility or based on prior studies. The following covariates were examined: (1) baseline cystatin C and creatinine levels; (2) demographic variables (age, sex, and race); (3) cardiovascular risk factors (body mass index and weight, hypertension defined by history and use of antihypertensive agents, or an average of 3 blood pressure measurements >140/90 mm Hg; diabetes mellitus defined by the use of insulin or an oral hypoglycemic agent or a fasting serum glucose level >126 mg/dL [to convert glucose to millimoles per liter, multiply by 0.0555]; levels of total cholesterol, low-density lipoprotein cholesterol, high-density lipoprotein cholesterol and triglycerides; (4) novel cardiovascular risk factors (C-reactive protein, fibrinogen, and hemoglobin concentrations); (5) subclinical vascular disease indicators (atrial fibrillation and left ventricular hypertrophy by electrocardiography, ankle-arm index, and carotid intima-media thickness); and (6) occurrence of cardiovascular disease (coronary artery disease, stroke, transient ischemic attack, and congestive heart failure) prior to the last cystatin C or creatinine level measurement. For the multivariate models, blood pressure was included using both a categorical variable and systolic and diastolic blood pressure readings at the time of the first measurement of cystatin C and creatinine.

All covariates were compared between participants with and without rapid decline in kidney function defined by eGFR_cys and separately by eGFR_creat; χ² tests were used for discrete variables, and t tests for continuous variables.

Cox proportional hazards models were used to evaluate the association of each measure of kidney function decline with the 2 mortality end points. Covariate selection was conducted as in previous reports from CHS using cystatin C.¹¹ Briefly, candidate variables were evaluated as determinants of each outcome in univariate analyses. Those that were significant were entered into multivariate proportional hazards models. Variables were selected for inclusion in the final adjusted models based on whether they changed the primary predictor's coefficient (change in eGFR_cys) by 5% or more. Variables that met this definition of confounding were retained.

We constructed separate models for eGFR_cys and eGFR_creat, but with identical covariates. We presented hazard ratios (HRs) for those with rapid decline defined by change in eGFR_cys and eGFR_creat. In addition, we compared mortality risk among participants who had rapid decline by eGFR_creat only, eGFR_cys only, neither, or both using unadjusted and multivariate-adjusted proportional hazards models.

Interactions between baseline level of kidney function and change in kidney function were evaluated, and stratified analyses were performed by baseline kidney function using low, medium, and high cystatin C categories (<1.0 mg/L, 1.0-1.28 mg/L, ≥1.29 mg/L),¹¹ and by the presence of CKD (eGFR_creat <60 mL/min/1.73 m²). In addition, because of concern about regression to the mean, we considered 2 separate models: one that did not adjust for baseline eGFR, and a second model that adjusted for mean eGFR.

We examined the association of rapid decline based on the association of eGFR_cys and eGFR_creat within subgroups based on age, sex, race, and coronary heart disease. Finally, we evaluated these associations within the subgroup of participants who were at low risk, defined as the absence of diabetes, hypertension, or cardiovascular disease at baseline, and compared the association with those participants with 1 or more of these risk conditions. The purpose of this subgroup analysis was to evaluate whether decline in kidney function is associated with mortality risk even in a relatively low-risk subgroup.²⁰

Analyses were performed using S-Plus (release 6.1; Insightful Inc, Seattle, Washington) and SPSS statistical software (release 14.0.2; SPSS Inc, Chicago, Illinois).

The 4380 individuals with 2 or 3 measurements of cystatin C and creatinine levels were younger, more likely to be female, and had substantially fewer cardiovascular risk factors, better baseline kidney function, and less cardiovascular disease at baseline than those with 0 or 1 measurement of cystatin C level. Mortality prior to the third clinical visit accounted for 51% of those who had 0 or 1 measurement of cystatin C, and loss to follow-up accounted for 4%.

Among the 4380 participants with 2 or 3 measurements of cystatin C, mean (SD) age was 72 (5) years; 9% of the cohort had diabetes mellitus at baseline, and 21% had prevalent cardiovascular disease. Mean (SD) baseline creatinine and cystatin C levels were 0.93 (0.30) mg/dL and 1.03 (0.25) mg/L, respectively (to convert creatinine to micromoles per liter, multiply by 88.4). Mean (SD) eGFR_creat and eGFR_cys were 79 (23) mL/min/1.73 m² and 79 (19) mL/min/1.73 m², respectively. Mean (SD) annualized change in creatinine and cystatin C levels and eGFR_creat and eGFR_cys were 0.01 (0.07) mg/dL and 0.03 (0.06) mg/L and −0.4 (3.6) mL/min/1.73 m² and −1.8 (2.6) mL/min/1.73 m², respectively.

Those with decline of eGFR_cys of greater than 3 mL/min/1.73 m² per year were older and more likely to have hypertension, diabetes mellitus, and cardiovascular disease at baseline than those with smaller interval changes (Table 1). First occurrences of cardiovascular disease (coronary heart disease, stroke, transient ischemic attack, or congestive heart failure) between the 1989-1990 (baseline) and 1996-1997 (year 7) testings were more common in the group with rapid decline in eGFR_cys. Similar relationships were seen between those with and without rapid decline in eGFR_creat (Table 1).

Of the 4380 individuals in the study group, 2219 died during the follow-up period (51%), 834 from cardiovascular causes (19%). The relationship between change in eGFR and mortality during follow-up is depicted in Figure 1, with the vertical line marking the cut point for rapid decline (3 mL/min/1.73 m² per year). When eGFR_cys was used as the measure of rapid decline, 1083 individuals had rapid decline (25%). The percentages of participants with all-cause and cardiovascular mortality were 63% and 26%, respectively, among persons with rapid decline, and 47% and 17%, respectively, among participants without rapid decline. After multivariate adjustment, rapid decline by eGFR_cys remained associated with a greater than 50% increase in all-cause mortality and cardiovascular mortality (Table 2). Using eGFR_creat, we found that 714 individuals had rapid decline (16%); after multivariate adjustment, rapid decline by eGFR_creat was associated with a greater than 70% increase in all-cause mortality and cardiovascular mortality. Models that did not adjust for baseline eGFR yielded similar HRs in adjusted analysis, with a 50% increase in all-cause or cardiovascular mortality using eGFR_cys and a 60% increase in both outcomes using eGFR_creat. The HRs were, in fact, slightly higher after adjustment for average eGFR across all available measurements (80% increased hazard of all-cause or cardiovascular mortality using eGFR_cys, and >90% increased hazard using eGFR_creat).

We examined the association of rapid kidney function decline with each mortality outcome within subgroups defined by baseline kidney function (as defined by an eGFR_creat <60 mL/min/1.73 m² and by initial level of cystatin C), age, sex, race, and baseline coronary heart disease (Figure 2). The point estimates were similar and statistically significant within each subgroup in Figure 2. When results were stratified by baseline CKD, rapid decline was associated with similar increases in risk for all-cause and cardiovascular mortality. There were no significant interactions between rapid kidney function decline by eGFR_cys or eGFR_creat and age (dichotomized at the median), race, sex, or baseline kidney function with either mortality end point. Furthermore, among participants who were free of diabetes mellitus, hypertension, and cardiovascular disease at baseline (n = 1918), similar associations were seen between rapid decline in eGFR and risk of all-cause mortality: for rapid decline by eGFR_cys, the adjusted HR was 1.56 (95% confidence interval, 1.32-1.84); for rapid decline by eGFR_creat, the adjusted HR was 1.83 (95% confidence interval, 1.51-2.21). These HRs were not different compared with the subgroup with at least 1 of these comorbidities at baseline (Figure 2).

The group of participants with rapid decline by both eGFR_creat and eGFR_cys had all-cause and cardiovascular mortality rates nearly twice those of the group that did not meet either criterion (Table 3). Overall mortality was similar in each group that met only 1 criterion of rapid decline but approximately 50% higher than the rate for the group that met neither criterion.

In this prospective cohort of community dwelling older adults, we observed an association between rapidly declining kidney function (loss of eGFR >3 mL/min/1.73 m² per year) and elevated risk for all-cause and cardiovascular mortality. These associations with elevated risk were observed regardless of the initial eGFR. Rapid decline by either eGFR_creat or eGFR_cys was associated with elevations in mortality risk, and these associations were similar across subgroups of age, sex, race, cardiovascular disease, and baseline risk status.

While CKD is a well-established risk factor for all-cause and cardiovascular mortality, loss of kidney function within a population with relatively well-preserved kidney function has not previously been associated with adverse outcomes. One prior study²¹ examined progression to CKD over 13 years of follow-up among a cohort of 281 individuals with hypertension and baseline eGFR higher than 90 mL/min/1.73 m²; the authors noted an association, albeit without controlling for level of baseline eGFR, between progression to CKD and future cardiovascular disease events. It is noteworthy that this was a select population with a known risk factor for progression, whereas the CHS population was substantially larger, more heterogeneous, and more generalizable to the general population of older adults.

There are several reasons why loss of kidney function may be associated with increased mortality risk. Ongoing decline in kidney function may play a causal role by exacerbating cardiovascular disease risk factors such as hypertension and dyslipidemia or by causing retention of inflammatory solutes leading to oxidative stress and vascular damage. In addition, worsening kidney function may lead to decreased appetite, loss of lean body mass, decreased physical function, and overall frailty.^22,23 Alternatively, loss of kidney function may be a marker of global atherosclerotic disease progression, with vascular disease leading to multi-organ dysfunction—including that in the kidneys—and predisposing to higher mortality risks.

Rapid decline measured by eGFR_creat and eGFR_cys were both associated with adverse outcomes. Overall, eGFR_cys detected more individuals with decline in kidney function in this cohort than did eGFR_creat, possibly reflecting that cystatin C is more sensitive to small changes in eGFR. However, participants with only eGFR_creat rapid decline also had elevated mortality risk, which implies that change in eGFR_cys has imperfect sensitivity for detecting high-risk individuals. Individuals with rapid decline in kidney function detected by both measures had the highest risk (nearly 10% annual mortality rate), suggesting that the 2 measures may provide complementary risk assessment in older adults.

Although the CHS comprises a community-based, representative sample of older adults in the US, the individuals included in this analysis were selected for having survived long enough to have at least 2 measures of kidney function. These participants were healthier and had better kidney function on average than the full CHS cohort, so their magnitude of decline in kidney function was likely less than would have been observed in the overall cohort. Nevertheless, even in this healthier subgroup, there was substantial variability in kidney function loss over time, and rapid declines of kidney function were strongly associated with mortality risk.

Our study has several limitations. We do not have direct measurements of GFR and therefore cannot assess whether the changes we observed in eGFR by either cystatin C or creatinine are truly reflective of changes in kidney function. Because 85% of the black subjects in this study were added to the cohort after the first measurement of cystatin C and creatinine, most had only 2 measurements of creatinine level; thus, we cannot determine change in eGFR as accurately in this population. Albuminuria was not measured at baseline in CHS, so we cannot comment on the role of this marker of kidney disease in mediating or modifying the relationships we observed. While regression to the mean may have led to misclassification bias, we considered several different statistical adjustments for baseline kidney function to account for this and noted that our results were robust regardless of the method used. Although cystatin C appears to be a very sensitive marker of kidney function, it may be affected by thyroid dysfunction, corticosteroid use, and body composition, and cystatin C measurements may not reflect actual GFR as accurately at the extremes of body mass.^24,25 In addition, we cannot determine whether the association of rapid kidney decline with mortality risk is causal or rather a marker of other physiologic processes that impact the kidney and other organs. Despite adjustment in multivariate models for both hypertension and diabetes mellitus, we cannot exclude the possibility of residual confounding from lifetime exposure to these conditions. However, we observed similar associations with increased mortality for the subgroup without hypertension, diabetes mellitus, or cardiovascular disease, suggesting that residual confounding was unlikely to be the primary explanation for our findings.

There are several potential implications of the results. First, decline in kidney function exceeding 3 mL/min/1.73 m² per year is independently associated with adverse outcomes. Second, changes in serum creatinine and cystatin C levels appear to provide complementary information in older adults, with the 2 measures capturing overlapping subgroups at increased mortality risk. Third, these findings were robust across multiple subgroups, including level of baseline kidney function. These associations should be confirmed in other cohorts of older adults as well as in other populations at risk for kidney disease progression. Recognition of early loss of eGFR may be important as a prognostic tool in older adults regardless of their baseline comorbid conditions.

Correspondence: Dena E. Rifkin, MD, Tufts Medical Center, 750 Washington St, Box 391, Boston, MA 02111 (drifkin@tuftsmedicalcenter.org).

Accepted for Publication: April 8, 2008.

Author Contributions: Dr Katz had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Rifkin, Shlipak, Katz, Fried, Siscovick, Chonchol, Newman, and Sarnak. Acquisition of data: Katz, Siscovick, and Newman. Analysis and interpretation of data: Rifkin, Shlipak, Katz, Fried, Siscovick, Chonchol, and Sarnak. Drafting of the manuscript: Rifkin, Shlipak, Chonchol, and Sarnak. Critical revision of the manuscript for important intellectual content: Rifkin, Shlipak, Katz, Fried, Siscovick, Chonchol, Newman, Sarnak. Statistical analysis: Katz. Obtained funding: Shlipak, Siscovick, Newman, and Sarnak. Administrative, technical, and material support: Newman. Study supervision: Shlipak, Chonchol, Newman, and Sarnak.

Financial Disclosure: None reported.

Funding/Support: The research reported in this article was supported by contracts R01-AG-027002, K24DK078204, N01-HC-35129, N01-HC-45133, N01-HC-75150, N01-HC-85079 through N01-HC-85086, N01-HC-15103, N01-HC-55222, and U01 HL080295 from the National Heart, Lung, and Blood Institute (NHLBI), Bethesda, Maryland, with additional contribution from the National Institute of Neurological Disorders and Stroke, Bethesda.

Role of the Sponsor: The NHLBI was substantially involved in data collection, analysis, and interpretation and manuscript preparation.

Previous Presentation: This work was presented in part at the American Society of Nephrology meeting; November 3, 2007; San Francisco, California.

Additional Information: A full list of participating CHS investigators and institutions can be found at http://www.chs-nhlbi.org.