Chronic Kidney Disease, Mortality, and Treatment Strategies among Patients with Clinically Significant Coronary Artery Disease

Description of Data Set

The Duke Databank for Cardiovascular Diseases, described in detail previously (14–20⇓⇓⇓⇓⇓⇓), contains information on patient-specific demographics, clinical history, physical examination, selected laboratory measures, and cardiac catheterization and coronary artery revascularization procedures gathered prospectively on patients undergoing cardiac catheterization at Duke University Medical Center. The Duke Databank for Cardiovascular Diseases has systematically collected information on the clinical experience of all patients undergoing cardiac catheterization since 1969 incorporating information for after-hospital follow-up. Patients are contacted 6 and 12 mo after their enrollment cardiac catheterization and annually thereafter to obtain information on follow-up, including survival status, medications, and rehospitalizations. The follow up in our study population is 97% complete.

All patients who underwent cardiac catheterization for the assessment of CAD at Duke University Medical Center between 1995 and 2000 with documented stenosis ≥75% of at least one major coronary artery and with serum creatinine values measured within 30 d before angiography were included in this analysis. Assignment to either PCI or CABG was made on the basis of initial revascularization procedure during the 30-d period after cardiac catheterization. For the purposes of this analysis, PCI refers to all patients who underwent percutaneous intervention, including balloon angioplasty with or without coronary stenting. The majority of the patients who underwent PCI in this analysis (91.5%) underwent coronary stent placement. Medical therapy refers to patients whose initial treatment strategy (within 30 d after catheterization) did not include either PCI or CABG. Patients who had a prior PCI or CABG, or patients with valvular heart disease, congenital heart disease, obstructive or restrictive cardiomyopathy, and hemodynamic instability or poorly visualized coronary anatomy on catheterization were excluded. Patients treated medically who died early (within 5 d after index cardiac catheterization) were also excluded to eliminate patients who did not have the opportunity to undergo PCI or CABG, reducing the bias that might be attributed to medical therapy as a result of early death. Patients were also excluded if it was apparent that cardiac catheterization took place primarily for pretransplantation assessment.

Descriptive Analysis

Demographic and clinical factors were compared among cohorts of patients stratified on renal function and treatment strategy (PCI, CABG, or medical management). Renal function was evaluated using creatinine clearance (CrCl) as a continuous measurement. CrCl was calculated using the formula derived by Cockcroft and Gault (21). Patients who had undergone renal transplantation were excluded. For descriptive purposes, normal renal function was defined as CrCl ≥90 ml/min; mild renal insufficiency, 60 to 89 ml/min; moderate renal insufficiency, 30 to 59 ml/min, and severe renal insufficiency, 15 to 29 ml/min (22). Patients with CrCl <15 ml/min and patients who were receiving dialysis were considered to have ESRD. Patients were stratified into five levels of renal function: normal, mild, moderate, and severe renal dysfunction; and ESRD. Patients undergoing dialysis were not included in the survival analysis because serum creatinine and calculated CrCl are not valid measures of renal function in this population.

Demographic and clinical factors, as well as CAD treatment strategy were compared among cohorts of patients described by categories of renal dysfunction. Categorical variables were compared between groups using the Pearson χ² tests, and continuous variables were compared using the nonparametric Kruskal-Wallis test.

Statistical Analyses

Patient survival time was calculated from the time of initial cardiac catheterization to the date of death or last contact. The unadjusted distribution of patient survival stratified by management strategy was estimated in all renal function cohorts using the Kaplan-Meier method. Groups with moderate and severe CKD were combined for graphical display because of the small sample size among the group with severe CKD.

Cox proportional-hazard regression was used to estimate the relationships between survival and selected clinical and demographic variables. Candidate variables for this analysis included CrCl; CAD treatment strategy; CAD severity; CAD symptom severity; duration of CAD; family history of CAD; comorbidities, including hypertension, diabetes mellitus, and congestive heart failure (CHF); severity of CHF symptoms; initial clinical examination findings, including carotid bruit and ventricular gallop; left ventricular ejection fraction; and previous myocardial infarction. The models incorporated a modified Charlson index to adjust for comorbid diseases (23), such as rheumatic diseases, hepatic impairment, cerebrovascular disease, peripheral vascular disease, tumor or cancer, chronic obstructive pulmonary disease, and AIDS. History of renal disease and diabetes mellitus with end-organ damage were not included in the comorbidity index so that their associations with survival could be modeled independently. Continuous baseline clinical variables were tested for linearity in relation to the log hazard and were transformed where appropriate to meet this linearity assumption. Those variables with univariable model log likelihood χ² of 2.5 or greater, as well as those thought to be clinically relevant, were included in a stepwise multivariable selection process.

The presence of effect modification of CrCl on both treatment strategy and diabetes mellitus were examined in separate multivariable models to determine if the relationship between CrCl and survival differed by the presence of diabetes mellitus or if the association between treatment strategy and survival differed by level of renal dysfunction. Once the final model was developed, the proportional-hazard assumption was checked for violations. Additional assessment of goodness of fit of the model was evaluated graphically by plotting log(−log(Survival)) versus log(time) and time and determining whether the relationships were linear. The c index, or probability of concordance between predicted probability and response, was used to quantify the predictive ability of the Cox model. A value of c = 0.5 indicates purely random predictions, whereas c = 1.0 designates perfect concordance (24). The final model, including significant interactions, was validated using a bootstrapping method, and the corrected concordance index was compared with the original model’s c index. (25,26⇓).

In addition to standard covariate adjustments, factors favoring treatment selection were identified using propensity score analysis. Propensity scores were developed from binary logistic regression models and were represented in Cox regression models by three variables consisting of the linear score or logit from each of the three logistic models (i.e., CABG versus PCI, CABG versus medical therapy, PCI versus medical therapy). This step in the analysis adjusts for likelihood that patients will receive a given therapy in an observational database when treatment is not randomly assigned. (27–29⇓⇓). Finally, because of recent concerns about possible misclassification of renal function among patients with normal GFR using prediction equations (30), the robustness of the final model was validated by repeating the final model but substituting serum creatinine for CrCl as a predictor variable.

Hazard ratio (HR) plots of CAD treatment strategy associated with mortality risk adjusted for all of the variables in the final survival model were categorized by levels of renal function. To determine the CrCl value above which no further improvement in mortality was seen (“threshold”), spline functions (31) were tested in Cox regression models. Analyses were performed using SAS version 8.2 (SAS Institute, Cary, NC). All P values reported are 2-sided, and all confidence intervals reported are 95% intervals.

Population Characteristics

A total of 4584 patients were included in the analysis; 923 patients were excluded because they did not have a serum creatinine measurement available. Compared with those included in the analysis, patients without serum creatinine measurements were similar in age, gender, race, CAD severity, left ventricular ejection fraction, CAD treatment strategy, and prevalence of comorbidities, including hypertension, CHF, peripheral vascular disease, and previous myocardial infarction. They were likely to have diabetes mellitus (32.6% versus 28.5%, P = 0.02).

Compared with patients with normal renal function, patients with CKD were in general older; more often black; more often female; and more likely to have comorbidities such as hypertension, diabetes mellitus, cerebrovascular disease, peripheral vascular disease, and CHF (Table 1). Left ventricular ejection fraction was lower among patients with worsening CKD. The median ejection fraction among patients with normal renal function was 55.9%, compared with 50.0% among patients with severe renal dysfunction. A similar pattern was noted when cardiac function was assessed by means of the conventional clinical sign of an audible ventricular gallop as a surrogate. Patients with worsening CKD also had increasing clinical severity of CHF by New York Heart Association criteria, greater frequency of left main disease, and greater frequency of three-vessel or severe CAD than those with normal renal function.

Among patients with normal renal function, 47.6% underwent PCI as initial intervention after angiography, 27.4% were treated initially with CABG, and 25.0% were managed medically. With increasing severity of CKD, medical therapy was chosen with greater frequency, from 29.2% among patients with mild CKD to 51.4% among patients with severe CKD. Additionally, as renal function worsened, patients were less likely to undergo PCI, decreasing from 35.7% of patients with mild CKD to 15.9% of patients with severe CKD. In spite of more severe CAD, patients with severe CKD underwent CABG at a similar frequency (32.7%) compared with those with normal renal function. Among patients with ESRD, PCI and CABG were performed at a similar frequency to patients with severe CKD (14.5% and 28%, respectively). Adjusted logistic regression analyses evaluating factors that predicted treatment choice suggested that symptom severity, severity of CAD, older age, and ejection fraction were the most important factors that influenced treatment selection (data not shown).

Survival Analysis

Patient survival longer than 5 yr was greatest for patients with normal renal function and gradually worsened for patients with mild, moderate, and severe renal insufficiency (Figure 1). In the adjusted main-effects survival model, declining CrCl was a strong predictor of mortality (Table 2). The increased mortality risk associated with decreased renal function was apparent up to a threshold CrCl of 85 ml/min, above which the association was attenuated. Consequently, for these analyses, CrCl was arbitrarily truncated at 85 ml/min. Among patients with a calculated CrCl ≤85 ml/min, a decline of 10 ml/min resulted in a 14% increase in mortality risk (HR, 1.14; 95% CI, 1.09 to 1.20; P < 0.0001). When these analyses were repeated with the evaluation of serum creatinine as a predictor variable instead of CrCl, increasing serum creatinine was found to be an important mortality predictor, and the remainder of the survival model was essentially unchanged.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1. Unadjusted Kaplan-Meier survival curves for cohorts of patients defined by degree of chronic kidney disease (CKD) by renal function group (severe, moderate, mild insufficiency, and normal function) based on creatinine clearance.

Patients without serum creatinine measurements available were noted to have greater mortality than those with serum creatinine and missing serum creatinine greater mortality (HR, 2.37; 95% CI, 1.81 to 2.75; P < 0.0001). Other predictors of increasing mortality include medical management compared with CABG or PCI, left ventricular impairment characterized by ejection fraction, increasing age, increasing severity of CAD, the presence of valvular disease, the history and severity of CHF, a history of myocardial infarction, a history of diabetes mellitus, and an S3 ventricular gallop as revealed at physical examination. When propensity scores were added to quantify the likelihood a treatment intervention was chosen for a particular patient, the model did not achieve a significantly better fit to the data (data not shown).

The regression diagnostics that were undertaken for the Cox model revealed that the proportional-hazard assumption was not grossly violated and the fit of the hypothesized survival distribution was met because log − log(survival) versus log(time) was linear. The bootstrapping process revealed that the Cox survival model has high predictive accuracy. There was minimal change in the corrected c index (c = 0.7765) compared with the original (c = 0.7815). Thus, the constructed model was validated by using nearly unbiased estimates of model performance.

The presence or absence of a clinical history of diabetes mellitus modified the effect of CrCl on survival (P = 0.004 for this interaction). The association between CrCl and mortality was more pronounced among patients without diabetes mellitus. These patients experienced a 22% increase in risk with each 10-ml/min decrease in CrCl (HR, 1.22; 95% CI, 1.14 to 1.30; P < 0.001). Although a similar trend was present, among patients with diabetes mellitus, a decline in CrCl was not associated with an increased risk of mortality that reached conventional levels of statistical significance (HR, 1.08, 95% CI, 0.98 to 1.20; P = 0.11).

Association between Treatment Strategy and Mortality

To illustrate the relationship between treatment strategy and CKD, CrCl was grouped into severity levels as in the descriptive analysis, and HRs (with 95% CI) characterizing relative treatment differences were reported. Treatment with PCI was independently associated with a beneficial effect on mortality when compared with medical management of CAD (HR, 0.65; 95% CI, 0.52 to 0.81; P = 0.0001). The effect of renal function on mortality interacted significantly with treatment (PCI versus medical management) (P = 0.04). PCI was associated with a survival benefit among patients with mild renal insufficiency (Figure 2A). As renal function declined further, PCI failed to offer a significant survival benefit over medical management among patients with severe renal insufficiency. Treatment with CABG was associated with a survival benefit when compared with medical management (HR, 0.42; 95% CI, 0.34 to 0.52; P = 0.0001). The association between CABG and mortality was similar among patients across all levels of renal function (Figure 2B) (P = 0.64 for interaction between CrCl and CABG versus medical management).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2. (A) Hazard ratios (HRs) for mortality for medical management versus percutaneous coronary artery intervention (PCI) for cohorts of patients defined by severity of chronic kidney disease (CKD) (adjusted). (B) HRs for mortality for medical management versus coronary artery bypass grafting (CABG) for cohorts of patients defined by severity of CKD (adjusted). (C) HRs for mortality for CABG versus PCI for cohorts of patients defined by severity of CKD (adjusted).

Treatment with CABG compared with PCI was associated with a HR of mortality of 0.65 (95% CI, 0.5 to 0.9; P = 0.002) among the entire cohort. When patients were stratified on renal function, CABG (compared with PCI) had a beneficial association with survival among patients with moderate and severe renal insufficiency not demonstrated among patients with normal renal function and mild renal insufficiency (Figure 2C).