Incidence rate ratio (IRR) (95% confidence interval) for cardiovascular death. The IRR associated with (A) decreasing kidney function (estimated glomerular filtration rate [EGFR]) and (B) increasing urine albumin-creatinine ratio (ACR). The restricted cubic spline models were adjusted for age, sex, EGFR, and ACR, and the reference (IRR = 1) was set to the median ACR or median EGFR. The distributions of EGFR and ACR in the general population are also shown (bars).
Cardiovascular mortality risk in the general population by categories of estimated glomerular filtration rate (EGFR) and urine albumin-creatinine ratio (ACR). The incidence rate ratios (IRRs) were adjusted for age and sex. The ACR is the mean of 3 samples, and optimal ACR is below sex-specific median (< 5 mg/g in men and < 7 mg/g in women), and high normal is 5 to 19 mg/g in men and 7 to 29 mg/g in women. Microalbuminuria is 20 to 199 mg/g in men and 30 to 299 mg/g in women.30 Subjects with an optimal ACR and an EGFR of 75 mL/min/1.73 m2 or higher comprised the reference group. *P < .05. †P < .01. ‡P < .001.
Hallan S, Astor B, Romundstad S, Aasarød K, Kvenild K, Coresh J. Association of Kidney Function and Albuminuria With Cardiovascular Mortality in Older vs Younger IndividualsThe HUNT II Study. Arch Intern Med. 2007;167(22):2490-2496. doi:10.1001/archinte.167.22.2490
Copyright 2007 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.2007
The cardiovascular risk implications of a combined assessment of reduced kidney function and microalbuminuria are unknown. In elderly persons, traditional cardiovascular risk factors are less predictive, and measures of end organ damage, such as kidney disease, may be needed for improved cardiovascular mortality risk stratification.
The glomerular filtration rate was estimated from calibrated serum creatinine, and the urine albumin-creatinine ratio (ACR) was measured in 3 urine samples in 9709 participants of the second Nord-Trøndelag Health Study (HUNT II), a Norwegian community-based health study, followed for 8.3 years with a 71% participation rate.
An estimated glomerular filtration rate (EGFR) at levels of less than 75 mL/min/1.73 m2 was associated with higher cardiovascular mortality risk, whereas a higher ACR was associated with higher risk with no lower limit. Low EGFR and albuminuria were synergistic cardiovascular mortality risk factors. Compared with subjects with an EGFR greater than 75 mL/min/1.73 m2 and ACR below the sex-specific median who were at the lowest risk, subjects with an EGFR of less than 45 mL/min/1.73 m2 and microalbuminuria had an adjusted incidence rate ratio of 6.7 (95% confidence interval, 3.0-15.1; P < .001). The addition of ACR and EGFR improved traditional risk models: 39% of subjects with intermediate risk were reclassified to low- or high-risk categories with corresponding observed risks that were 3-fold different than the original category. Age-stratified analyses showed that EGFR and ACR were particularly strong risk factors for persons 70 years or older.
Reduced kidney function and microalbuminuria are risk factors for cardiovascular death, independent of each other and traditional risk factors. The combined variable improved cardiovascular risk stratification at all age levels, but particularly in elderly persons where the predictive power of traditional risk factors is attenuated.
Primary prevention of cardiovascular disease (CVD) in people 70 years or older is debated.1,2 Despite the fact that most of the cardiovascular morbidity, mortality, and costs occur at older ages, there are few data on the benefits and risks of the treatment, and we lack tools for accurate prediction of cardiovascular risk. Reduced kidney function and urinary excretion of albumin are used for defining and staging chronic kidney disease,3 a common condition with a high risk of CVD in addition to the risk of progression to end-stage renal disease.4- 7 The interaction between the kidney and the heart increasingly emerges as an important factor for CVD,8 and kidney function and urinary excretion of albumin are now suggested as risk factors to be measured in assessing risk of cardiovascular death, in addition to hypertension, diabetes mellitus (DM), hypercholesterolemia, and smoking.9,10 However, these new, potentially related risk factors have seldom been evaluated together in population-based studies.
Many of the studies that found decreased kidney function increases mortality are analyses of cardiovascular intervention trials with limited measures of kidney disease.11 Early population-based studies did not find reduced kidney function to be an independent risk factor for cardiovascular death,12,13 but most recent studies7,10,14 find that a reduced estimated glomerular filtration rate (EGFR) (an EGFR <60 mL/min/1.73 m2) is a major risk factor for cardiovascular mortality and morbidity. Albumin leakage in the urine, despite its large day-to-day variation, has emerged as an important risk factor for atherosclerotic CVD.10,15 International guidelines16,17 recommend screening for microalbuminuria in subjects with DM or hypertension. However, most studies have examined either EGFR or albuminuria, but not both. So far, to our knowledge, there are only 2 reports on the combined effect of kidney function and albuminuria.18,19 These studies used a semiquantitative dipstick analysis for measuring albuminuria in a single urine sample and were able to evaluate only subjects with macroalbuminuria.
Hence, more information on the combined effect of reduced kidney function and quantitatively measured microalbuminuria is needed. Both can be considered as measures of end organ damage, and inclusion of such variables could improve risk stratification in general, and in particular among elderly persons, for whom traditional risk factors have reduced predictive power.20- 22 This could be important in screening programs and targeting preventive treatment for subjects with increased cardiovascular risk. We analyzed data from the second Nord-Trøndelag Health Study (HUNT II), a large prospective cohort study with an albumin-creatinine ratio (ACR) measured in 3 urine samples. First, we explored the association among abnormal EGFR, albuminuria, and cardiovascular mortality, with special emphasis on the near-normal levels in a general population. Second, to address the potential clinical usefulness of such measurements, we compared cardiovascular risk models with and without a combined EGFR-ACR variable in subjects younger than 70 years and those 70 years or older.
The HUNT II study is a large-scale Norwegian general health survey. From 1995 to 1997, every individual residing in the county who was at least 20 years old (n = 92 939) was invited to participate, and 70.6% of the total adult population participated. We evaluated a subpopulation that was asked to deliver urine samples in addition to the standard testing: all subjects with DM or treated hypertension (prevalence rates, 3.4% and 11.1%, respectively), plus a 5% random sample.
Nord-Trøndelag County is located in the middle of Norway and is fairly representative in terms of geography, economy, industry, age distribution, and morbidity and mortality.23 The population is ethnically homogeneous (> 97% white). A more detailed description of the objectives, contents, methods, and participation in the HUNT II study has been given elsewhere.24 The participants gave an informed consent, which included linkage to central national registries, and the study was approved by the regional committee for medical research ethics, the Norwegian Data Inspectorate, and the Ministry of Health.
The participants reported on several aspects of their current and former health, on illness in the family, socioeconomic status, and risk factors, such as physical activity and smoking. The clinical examination included measurement of height, weight, and waist and hip circumference. Three consecutive standardized blood pressure measurements were recorded in the sitting position at 1-minute intervals using an automatic oscillometric method (Dinamap 845XT; Criticon, Tampa, Florida). Fresh serum and urine samples were analyzed on a Hitachi 911 autoanalyzer (Hitachi, Mito, Japan) within 2 days. The GFR was estimated with the reexpressed 4-variable Modification of Diet in Renal Disease study formula for isotope dilution mass spectrometry traceable serum creatinine values in all subjects25:
EGFR = 175 × (Serum Creatinine in Milligrams per Deciliter)-1.154 × Age -0.203 (× 0.742 for Women) (× 1.21 for Black Persons).
Our original Jaffé-based creatinine values were recalibrated to the Roche enzymatic method to provide isotope dilution mass spectrometry traceable values, and the EGFR values have been shown to be unbiased in a general population.26 Participants were asked to deliver urine samples on 3 consecutive mornings, and those reporting urine infection during the previous week or menstruation at the time of collection were excluded. Urine albumin was measured by an immunoturbidimetric method (Dako A/S, Glostrup, Denmark), and urine ACR was used as an expression for albumin excretion.
Vital status as of January 1, 2005, was provided by the Statistics Norway database23 for all participants, and the cause of death was available in 99.7% of cases. We defined cardiovascular death as death certificates with the following International Statistical Classification of Diseases, 10th Revision (ICD-10),27 codes as underlying cause of death28: hypertensive disease (I10-I15), ischemic heart disease (I20-I25), arhythmia (I44-I49), heart failure (I50), cerebrovascular disease (I60-I69), and diseases of the arteries (I70-I77). We defined coronary heart death as caused by ischemic heart disease (I20-I25).
Statistical analyses were generated using Stata software (version 9; Stata Corp, College Station, Texas). Six subjects with an EGFR below 15 mL/min/1.73 m2 were excluded, and 7 with EGFR greater than 200 mL/min/1.73 m2, which is physiological unlikely, were given a value of 200 mL/min/1.73 m2. Associations of EGFR and ACR with mortality were examined using multivariate Poisson regression models, which express relative risk as an incidence rate ratio (IRR), and this yielded similar results to Cox proportional hazard regression analyses. Analyses addressing the general population accounted for the urinary testing sampling scheme using appropriate sample weights. We explored the continuous relationship of mortality risk associated with lower EGFR or higher ACR using restricted cubic spline models adjusted for age, sex, EGFR, and ACR. Interaction on an additive scale was used to evaluate whether EGFR and ACR are more useful for risk stratification when used together vs separately.29 A composite variable with 16 categories (combining 4 EGFR categories and 4 ACR categories) was used to evaluate the combined effect of these 2 variables. Microalbuminuria was defined as an ACR of 20 to 200 mg/g in men and 30 to 300 mg/g in women,30 but because previous studies indicate that the risk extends below this level, we also categorized ACR as “optimal” values below the sex-specific median (<5 mg/g in men and <7 mg/g in women) and as “high normal” values (5-19 mg/g in men and 7-29 mg/g in women). Age- and sex-adjusted IRRs, as well as multivariate-adjusted IRRs, were calculated (age, sex, prevalent CVD, DM, systolic blood pressure, antihypertensive medication, current smoking, cholesterol, high-density lipoprotein cholesterol, and EGFR-ACR categories). We also calculated excess risk, an absolute measure of risk increase, by categories of EGFR and ACR, using the coefficients from the multivariate-adjusted models and the observed risk in the reference group (optimal ACR and EGFR ≥75).
We assessed the risk for cardiovascular death associated with traditional risk factors and with EGFR-ACR in subjects younger than 70 years and those 70 years or older. This age stratification was chosen a priori because current risk models are based on study populations below this threshold.31,32 All-cause and coronary heart disease mortality were used as outcomes in secondary analyses. Risk models with and without EGFR-ACR were assessed with the Akaike Information Criterion,33 which is a likelihood-based measure that adds a penalty for model complexity, and with the C statistic (area under the receiver operating characteristic [ROC] curve) based on logistic regression analyses. We also compared risk estimates from Poisson regression models with the observed risk during follow-up. European guidelines recommend primary preventive treatment in subjects younger than 65 years if the 10-year absolute cardiovascular mortality risk is more than 5%,28 but higher thresholds for absolute risks may be more useful in elderly persons.34,35 We therefore classified people into low-, intermediate-, or high-risk categories when cardiovascular mortality rates were less than 5, 5 to 10, and more than 10 per 1000 person-years, respectively.
Baseline data for the study population are given in Table 1. A total of 9709 participants returned 3 urine samples, giving an overall response rate of 86.4%. The 2294 subjects without hypertension and DM selected at random had cardiovascular mortality rates similar to those not selected for urine testing (3.19 vs 3.29 per 1000 person-years; log rank test, P = .80). When adjusting for a slightly higher age, none of the baseline characteristics and cardiovascular risk factors of our study subjects were substantially different from the general Norwegian population. During a median follow-up period of 8.3 years, 1981 subjects in our study group died, and 1018 of those deaths were caused by CVD.
The continuous relationships of cardiovascular mortality risk associated with lower EGFR and higher ACR adjusted for each other and for age and sex are shown in Figure 1. The adjusted IRR started to increase as the EGFR decreased below 75 mL/min/1.73 m2. In the EGFR range of 75 to 135, the IRR was very close to 1. At EGFRs above 135, where precision is poor and estimates may reflect low muscle mass as much as higher GFR, the risk of cardiovascular mortality was higher than in the EGFR range of 75 to 135 (IRR, 1.48; 95% CI, 1.10-1.99). In contrast, the IRR increased continuously with increasing ACR. An interaction between EGFR and ACR, defined as a departure from the additivity of their absolute effects, was observed. Subjects with EGFR lower than 60 mL/min/1.73 m2 and microalbuminuria had an excess age- and sex-adjusted risk: relative excess risk owing to interaction was 1.98 (95% CI, 0.02-3.94).
The association of a combined kidney function and albuminuria variable with cardiovascular mortality using categories emphasizing the near-normal levels is illustrated in Figure 2. In the general population, age- and sex-adjusted Poisson regression analysis showed that lower EGFR categories were associated with increased relative risk within every ACR category. Likewise, increasing ACR categories were associated with increased mortality within every EGFR category. Subjects with microalbuminuria and an EGFR lower than 45 mL/min/1.73 m2 had 12 times higher cardiovascular mortality risk compared with the reference category of subjects with an EGFR higher than 75 mL/min/1.73 m2 and “optimal” ACR. However, if subjects had a low EGFR but an optimal ACR, or if they had microalbuminuria and a normal EGFR, they had only a moderately increased risk (IRR, 2.3 and 3.0).
Adjusting for age, sex, prevalent CVD, DM, systolic blood pressure, antihypertensive medication, current smoking, cholesterol, and high-density lipoprotein cholesterol attenuated the IRRs, as shown in Table 2. There was a strong trend for higher risk at lower GFR in subjects with microalbuminuria (P = .02 at age <70 years, P = .002 at age ≥70 years). At lower ACR levels, there was no significant trend for increased cardiovascular risk with decreasing EGFR (P = .91 and P = .98 at optimal ACR, and P = .31 and P = .78 at high normal ACR for both age ranges, respectively). The association of reduced kidney function and increased albumin excretion with cardiovascular mortality tended to be even stronger in participants 70 years or older compared with those younger than 70 years. Given the higher baseline risk among older participants, this translates into large differences in absolute excess risk by EGFR-ACR category. Table 2 shows that there were 4.1 more cardiovascular deaths per 1000 person-years when the EGFR was lower than 45 mL/min/1.73 m2 and microalbuminuria was present in an average person younger than 70 years, compared with those with EGFR higher than 75 mL/min/1.73 m2 and optimal ACR. The corresponding mortality risk difference for persons older than 70 years was 63.6 per 1000 person-years. The associations between EGFR-ACR and all-cause mortality, as well as coronary heart disease mortality, were assessed in secondary analyses, and the associations were similar to those for cardiovascular mortality.
The relative contribution of different risk factors to global cardiovascular risk is displayed in Table 3. Their ranking was nearly identical based on the change in the Akaike Criterion Information index or the area under the ROC curve. Adding EGFR-ACR or information on prevalent CVD to the risk model made the greatest additional improvement to the model in subjects younger than 70 years as well as in those 70 years or older. Diabetes mellitus had the next highest contribution, whereas current smoking changed from a moderately important variable in young and middle-aged persons to not being associated with cardiovascular risk in elderly individuals. Adding EGFR-ACR to a model already including all of the traditional risk factors substantially improved the model for both younger and older participants. Both ROC and Akaike Information Criterion analyses showed that the combined EGFR-ACR variable was especially important among older participants.
Reclassification of subjects, that is, the percentage of subjects initially classified as having a low (<5%), intermediate (5%-10%), or high (>10%) 10-year cardiovascular mortality risk based on a traditional model who would be reclassified to higher- or lower-risk categories by a model also including GFR-ACR, is presented in Table 4. A traditional model (age, sex, prevalent CVD, hypertension treated with drugs, systolic blood pressure, current smoking, and total and high-density lipoprotein cholesterol) and a model also including EGFR-ACR agreed that 76.6% of the general population was at low risk. However, 6.6% of the general population would be classified differently by adding EGFR-ACR to the traditional model, and the most dramatic impact was on the intermediate-risk category, which constitutes 7.7% of the general population. One-quarter of the intermediate-risk subjects were reclassified to low risk, and these individuals had a 2.9-fold lower observed risk than those classified as having an intermediate risk in both models. One-tenth of the intermediate-risk subjects were reclassified to high risk, and these individuals had a 5.67-fold higher observed risk than those classified as intermediate risk in both risk models.
In this large, population-based study, we documented that impaired kidney function and urinary albumin excretion were strongly associated with cardiovascular mortality. Both were independent risk factors with higher risk at lower EGFR below a threshold of 75 mL/min/1.73 m2 and at higher ACR with no lower threshold apparent. They were synergistic on the additive scale, suggesting better risk stratification when both EGFR and ACR are used together. The improvement of global fit of cardiovascular risk models was comparable with that obtained by adding traditional risk factors like DM, hypertension, smoking, or cholesterol, and model improvement was most dramatic in subjects who were at least 70 years old.
Until now, the combined effect of reduced kidney function and albuminuria has been uncertain. Previous data showing that patients with macroalbuminuria and EGFR levels lower than 60 mL/min/1.73 m2 have 2 to 4 times higher mortality risk than subjects without these risk factors18,19 are useful, but the impact of the much higher prevalence of lower levels of albuminuria needed to be quantified. Dipstick testing for albuminuria is a relatively insensitive method not able to detect microalbuminuria. Our study extends these previous results from Japan and from patients who have had a myocardial infarction to a white general population. Future risk in subjects with moderately to severely reduced kidney function (EGFR <60 mL/min/1.73 m2) varied dramatically by level of albuminuria. The risk remained rather low, even in elderly persons, if urinary albumin excretion was “optimal” (ACR <6 mg/g). If these data prove to be generalizable, they suggest that combined assessment with both EGFR and albuminuria will be a useful way to stratify risk among the large group of subjects with chronic kidney disease. The prevalence of chronic kidney disease is high (10%) in the United States as well as in Europe,4,5 and further risk stratification will be useful.
Current cardiovascular risk models are intended for use in subjects younger than 70 years,31,32 and all attach importance to age as a major risk factor. However, age itself is not directly causally related to CVDs but rather reflects the progressive accumulation of atherosclerosis and end organ damage. The mean risk scores for age fail to account for individual variability. Preventive treatment is increasingly offered to people older than 70 years, but applying current guidelines for primary prevention to the elderly population is problematic because it tends to indicate treatment for nearly everyone.36,37 At the same time, elderly persons are susceptible to higher risks of polypharmacy and adverse effects. Thus, more accurate risk models are needed for the elderly population, and new cardiovascular risk models should consider including information on kidney function and urinary albumin excretion.
Our study has some methodological aspects that need discussion. First, urine samples were not collected from all participants. However, ACR determinations in 3 fresh urine samples for 9709 subjects based on a stratified random sample provide a solid base for inference to the population. Second, GFR estimation based on serum creatinine level has limitations. Although we used calibrated serum creatinine values to avoid systematic bias, the GFR estimation has only moderate accuracy, especially when the GFR is greater than 60 mL/min/1.73 m2.25 This could veil a possible association to mortality and create a threshold effect. Studies of cystatin C in elderly individuals suggest that the risk associated with decreased kidney function may be strongly underestimated when one relies on creatinine.38 Third, our primary end point (cardiovascular death) was based on death certificates. Even though the Nordic cause-of-death registers have been found to be reasonably valid indicators for cardiovascular death,39,40 there might have been some misclassification. However, the unique identification number given to all Norwegian citizens at birth enabled us to determine vital status of all participants with certainty at the end of the observation period, and the effect of EGFR-ACR on all-cause mortality was similar to that observed for cardiovascular mortality. Fourth, only baseline data were available, so we could not take into account the potential effect of changing risk factors and treatments that could affect the outcomes of interest. Finally, the generalizability of our results to other races and ethnic groups may be limited.
In conclusion, decreased kidney function and increased albumin excretion, even at near-normal levels, were associated with increased cardiovascular mortality independently of each other and of established risk factors. A variable based on the combination of EGFR and ACR was especially helpful for refining risk estimates in subjects older than 70 years, and its relative contribution to global risk was comparable with that provided by individual traditional cardiovascular risk factors such as DM, hypertension, lipids, or smoking.
Correspondence: Stein Hallan, MD, PhD, Department of Cancer Research and Molecular Medicine, Faculty of Medicine, NTNU St Olav University Hospital, Olav Kyrres gt 17, Trondheim N-70006, Norway (email@example.com).
Accepted for Publication: July 1, 2007.
Author Contributions:Study concept and design: Hallan, Kvenild, and Coresh. Acquisition of data: Hallan and Kvenild. Analysis and interpretation of data: Hallan, Astor, Romundstad, Aasarød, and Coresh. Drafting of the manuscript: Hallan. Critical revision of the manuscript for important intellectual content: Astor, Romundstad, Aasarød, Kvenild, and Coresh. Statistical analysis: Hallan, Astor, and Coresh. Obtained funding: Kvenild. Administrative, technical, and material support: Romundstad, Aasarød, and Kvenild.
Financial Disclosure: None reported.
Additional Information: The HUNT Study is a collaboration between the HUNT Research Center, Faculty of Medicine, Norwegian University of Science and Technology, Verdal; the Norwegian Institute of Public Health, Oslo; Nord-Trøndelag County Council; and the Central Norway Regional Health Authority.
Additional Contributions: We thank the health service and people of Nord-Trøndelag for their endurance and participation.