Abstract
Abstract. Patients with end-stage renal disease treated by hemodialysis have a tremendous risk for cardiovascular complications that cannot be explained by traditional atherosclerosis risk factors. Lipoprotein(a) (Lp(a)), a risk factor for these complications in the general population, is significantly elevated in these patients. In this study, it was determined whether Lp(a) and/or the genetically determined apo(a) phenotype are risk predictors for the development of coronary artery disease in these patients. A cohort of 440 unselected hemodialysis patients were followed for a period of 5 yr independent of the cause of renal disease, duration of preceding treatment, and the preexistence of coronary artery disease at study entry. Coronary events defined as definite myocardial infarction, percutaneous transluminal coronary angioplasty, aortocoronary bypass, or a stenosis >50% in the coronary angiography were the main outcome measure. Sixty-six (15%) of the 440 patients suffered a coronary event during follow-up. In univariate analysis, patients with events were significantly older and showed a trend to lower HDL cholesterol concentrations, and higher apolipoprotein B and Lp(a) concentrations without reaching significance. Apo(a) phenotypes of low molecular weight, however, were significantly more frequent in patients with compared to those without events (43.9% versus 21.9%, P < 0.001). The other lipids, lipoproteins, and apolipoproteins were similar in both groups. Multiple Cox proportional hazards regression analysis found age and the apo(a) phenotype to be the best predictors for coronary events during the observation period, independent of whether patients with a preexisting coronary artery disease or an age >65 yr at the study entry or both were excluded from the analysis. Diabetes mellitus was a risk factor only in presence of a low molecular weight apo(a) phenotype. The genetically determined apo(a) phenotype is a strong and independent predictor for coronary events in hemodialysis patients. Apo(a) phenotyping might be helpful to identify hemodialysis patients at high risk for coronary artery disease.
Cardiovascular disease is still the most frequent cause of death in patients with end-stage renal disease (1,2,3). Besides other lipoprotein abnormalities, lipoprotein(a) (Lp(a)) has been suggested to be associated with these complications (reviewed in reference (4).
Most prospective studies in the general population have described Lp(a) as an independent risk factor for coronary artery disease (CAD) (5,6,7,8,9,10,11,12,13,14,15,16,17). Interest in this atherogenic particle was raised because of the high heritability of Lp(a) plasma concentrations. These are mainly determined by the size polymorphism of apo(a) (18), which originates from a varying number of kringle-IV (K-IV) repeats in the apo(a) gene (19,20,21). This results in a negative correlation between the number of K-IV repeats and the Lp(a) plasma concentrations. From studies considering the apo(a) size polymorphism, it was concluded that the apo(a) gene locus determines the risk for coronary heart disease through its allelic control of Lp(a) plasma concentration (12,17,22,23,24,25,26).
In recent years, numerous studies have described elevated Lp(a) plasma concentrations in patients with renal disease (4). Some of them, including one prospective investigation (7), found a correlation with atherosclerotic complications that was not confirmed by other studies (4). In cross-sectional studies, we and others observed a higher frequency of apo(a) phenotypes with low molecular weight in hemodialysis patients with atherosclerotic complications (27,28,29).
Prospective studies investigating coronary events are hard to perform in the general population because they require a high number of subjects and a long follow-up time to observe the required number of cases for reliable statistical analysis. The prospective study of the Framingham offspring cohort, for example, included 2191 men followed for 15.4 yr. During this time, 129 CAD events were observed (11). Therefore, this “gold standard” of epidemiologic study is time-consuming and costly. The situation in hemodialysis patients is completely different. Because these patients reach the end point of interest very frequently and soon, the number of patients under observation can be low compared to the general population.
In the present study, we prospectively followed a cohort of 440 unselected hemodialysis patients for a period of 5 yr. During this time, we recorded coronary events defined by stringent criteria and attempted to identify parameters associated with an increased risk for these events.
Materials and Methods
Patients
This investigation is based on a study that recruited a total of 534 adult Caucasian hemodialysis patients in seven renal units during 1991 and 1992 (30). In five of the seven renal units, 440 unselected hemodialysis patients (187 women and 253 men) with a mean age of 57 ± 15 yr and a mean duration of hemodialysis treatment of 26 ± 33 mo (range, 3 to 180) at the study entry were prospectively followed. Inclusion criteria were a duration of dialysis treatment of at least 3 mo and an age between 18 and 75 yr at the study entry. Enrollment in the study was postponed to a later time in case a patient had suffered a cardiovascular event or an infection during the preceding 2 mo. Patients were included independent of the cause of renal disease and regardless of whether they had already suffered a coronary event before the enrollment. Fifty-one (11.6%) of the 440 hemodialysis patients had already suffered a coronary event before the follow-up period started. Age and the presence of a previous coronary event, however, were considered in the final analysis (see below).
Patients were prospectively followed until a date fixed for each center, which was during the period between July 1996 and March 1997. We calculated the individual follow-up period from the time of enrollment in the study until the occurrence of a coronary event (see definition below). We censored the follow-up in case of death unrelated to a defined coronary event (n = 128), renal transplantation (n = 127), change of residence or change of renal unit (n = 14) or restoration of kidney function (n = 5) using the date these occurred; otherwise, the above-mentioned fixed date was used.
Dialysis was performed three times weekly for 3.9 h on average. Dietary recommendations for all patients included a diet with 50 to 60% carbohydrates and 30 to 35% fat of total caloric intake. A daily food intake of 35 kcal/kg body wt including 1 g protein/kg body wt was prescribed to compensate for increased protein catabolism and loss of amino acids during hemodialysis.
Patients were considered diabetic if they were being treated with insulin or oral hypoglycemic agents, and/or fasting blood glucose was higher than 140 mg/dl. Patients were classified as hypertensive if either the BP measured at the beginning of dialysis treatment was greater than 160/95 mmHg or patients were on antihypertensive treatment. Smoking history was recorded by interview. Sixteen patients were on lipid-lowering drugs. They remained in the analysis since they had similar lipid and lipoprotein levels than those without these drugs.
Definition of CAD
We used stringent criteria to define an event of CAD. These were a definite myocardial infarction with clinical symptoms and either corresponding changes in enzymatic or electrocardiogram pattern (significant Q waves) or the diagnosis of a new event during an autopsy. Discovery of signs of an old myocardial infarction on routine examinations during the follow-up were not included due to lack of accurate timely assignment. Sudden death was included only when a coronary event was confirmed as the cause of death. Additional CAD events were defined by aortocoronary bypass surgery, percutaneous transluminal coronary angioplasty (PTCA), and/or coronary angiography. Selective coronary angiography was performed in those patients who reported angina pectoris and/or other clinical hints (changes of ST segment) suggesting coronary insufficiency. Multiple projections of the left and right coronary artery were assessed, and the presence of hemodynamically relevant stenosis (stenosis >50% of the luminal diameter) was classified as CAD. One subanalysis excluded angiographically diagnosed coronary stenosis from the CAD event definition for reasons given in the Discussion. Classification of CAD events was blinded to LP(a) levels and apo(a) isoforms. In case of multiple events in a subject, only the first one counted and was taken as the end point.
A CAD event (or a cerebrovascular event) before enrollment in the study was established by medical record documentation using the same criteria used during the prospective follow-up.
Laboratory Procedures
Ethylenediaminetetra-acetic acid plasma was taken after a 12-h overnight fast before dialysis was started only when patients had not suffered a coronary event or an infection in the preceding 2 mo. After low-speed centrifugation, samples were frozen and kept at -80°C before analysis (31).
All laboratory qualifications and apo(a) phenotyping were performed during the following weeks after patients were enrolled in the study centrally in one laboratory. At this time, the laboratory staff involved in the study was unaware of patient outcome.
Lp(a) quantification was performed as described in detail (31) with a double-antibody enzyme-linked immunosorbent assay, using an affinity-purified polyclonal apo(a) antibody for coating and the horse-radish peroxidase-conjugated monoclonal 1A2 for detection. This anti-apo(a) antibody recognizes the epitope motif YYPN in kringle-IV (K-IV) type 2 (32). An Lp(a)-positive serum from Immuno (Vienna, Austria) with the same apo(a) isoforms served as standard throughout the study. Each sample was analyzed in duplicate, and intra- and interassay coefficients of variation were 2.7 and 6%, respectively.
Plasma concentrations of apoB were measured by enzyme-linked immunosorbent assay technique as described previously (33). Apo A-I was determined by commercially available electroimmunodiffusion assay (Immuno). Total cholesterol, HDL cholesterol, triglycerides, and albumin (bromcresol green method) were measured using kits from Boehringer Mannheim (Mannheim, Germany). Measurements were made on microtiter plates as described previously (31). LDL cholesterol was calculated with the Friedewald formula in patients with triglyceride concentrations below 400 mg/dl. Adjustment of lipid, lipoprotein, and apolipoprotein concentrations for hematocrit had no influence on our findings (34).
Apo(a) phenotyping was performed by sodium dodecyl sulfate-agarose gel electrophoresis (SDS agarose) under reducing conditions as outlined (35) with slight modifications, followed by immunoblotting (18) using the monoclonal antibody 1A2 for detection of apo(a) isoforms.
Statistical Analyses
Statistical analysis was performed with Statistical Package for the Social Sciences (SPSS) for Windows 7.5.2 Univariate comparisons of continuous variables between patients with and without a CAD event during the follow-up period were done by unpaired t test or the nonparametric Wilcoxon rank sum test in case of non-normally distributed variables (Lp(a), triglycerides, and time on dialysis). Dichotomized variables were compared using Pearson's χ2 test, and the relative risks for suffering a coronary event were calculated together with the 95% confidence intervals (CI).
Multivariable adjusted risk estimates were calculated for the whole patient group as well as after exclusion of patients with a preexisting CAD, age > 65 yr, or both using a multiple Cox proportional hazards regression analysis (short: Cox regression). A similar analysis in the whole patient group was performed after exclusion of angiographically diagnosed coronary stenosis from the CAD event definition. A stepping procedure was used to identify variables associated with the development of coronary events over time. P values for the inclusion and exclusion to the model were 0.10 and 0.15, respectively. Variables used in these final analyses were age, preexisting coronary event, apo(a) phenotype stratified in low molecular weight (LMW) and high molecular weight (HMW) apo(a) types, diabetes mellitus, HDL cholesterol, apoB, and logarithmically transformed Lp(a). Possible interactions between the variables were investigated by the introduction of interaction terms to the multivariate models, which were created by calculating the product of two variables. Forward and backward stepping methods as well as inclusion or exclusion of other variables (e.g., gender, total or LDL cholesterol, triglycerides, plasma albumin, creatinine, dialysis hours per week, hypertension, or smoking) to the model revealed the same results. A risk factor was only considered independent from the other risk factors in the model if its univariate and multivariate relative risks were relatively similar in magnitude and when no evidence of an interaction with other risk factors (tested by interaction terms) was observed. Multivariate adjusted time-to-event curves were generated for patients with HMW and LMW apo(a) phenotypes.
Because of the high number of detectable apo(a) phenotypes (>30), many phenotypes were represented only in low numbers. To account for this problem, we decided a priori to combine apo(a) phenotypes in steps of three K-IV repeats according to the molecular weight of the smaller apo(a) isoforms to have sufficient sample sizes in each category (36). Since subjects with 11 to 16 or >34 K-IV repeats were represented relatively rarely, we built one group by combining 11 to 19 and another by combining >31 K-IV repeats. Furthermore, we divided apo(a) phenotypes into two subgroups according to the molecular weight of the smaller apo(a) isoforms, as done in previous studies (12,17,23,27,28,29,30,33,37). The LMW group included all subjects with at least one apo(a) isoform with 11 to 22 K-IV repeats (38); the HMW group comprised all subjects having only isoforms with more than 22 K-IV repeats. If two apo(a) isoforms were detectable, we used only the smaller apo(a) isoform for categorization, which we discussed recently in detail (29).
Results
Univariate Analysis
In 66 (15%) of the 440 patients, we registered at least one event of CAD during the prospective observation period. Thirty-six of these patients suffered a definite myocardial infarction, five patients underwent a PTCA, and one patient an aortocoronary bypass. In the remaining 24 patients, the diagnosis of a coronary event was established by coronary angiography.
In a first step, we compared the laboratory parameters and the frequency of atherosclerosis risk factors between patients with and those without coronary events during the observation period (Tables 1 and 2). Those with events were significantly older (64 ± 12 versus 56 ± 15 yr, P < 0.001), had higher plasma concentrations of apoB (105 ± 34 versus 96 ± 33 mg/dl, P = 0.05), and a trend to lower HDL cholesterol (35.4 ± 13.9 versus 32.3 ± 13.5 mg/L, P = 0.09). The concentrations of other lipids, lipoproteins, and apolipoproteins (Table 1), as well as the frequencies of males, smokers, and hypertension (Table 2), were not significantly different between the two groups. A higher frequency of patients with diabetes mellitus was observed in the group with coronary events (29% versus 20%, P = 0.12) (Table 2).
Univariate comparison of continuous variables between hemodialysis patients with and without events of coronary artery disease (CAD) during the follow-up perioda
Univariate comparison of categorical variables between hemodialysis patients with and without events of CAD during the observation perioda
Lp(a) showed a trend to elevated concentrations in patients with coronary events without reaching significance (28.9 ± 34.5 versus 21.2 ± 22.9 mg/dl, P = 0.28) (Table 3). When we compared the frequency of apo(a) phenotypes, it became apparent that patients who developed a coronary event showed significantly more often apo(a) phenotypes with a low number of K-IV repeats. The results were similar whether we used only the smaller apo(a) allele as shown in Table 3 or both apo(a) isoforms for the categorization. When we categorized patients according to LMW and HMW apo(a) phenotypes, the frequency of LMW apo(a) phenotypes was twice as high in the group with compared to that without coronary events (43.9% versus 21.9%, P <0.001).
Lp(a) plasma concentrations and apo(a) size polymorphism in hemodialysis patients with and without events of CAD during the observation perioda
The total duration of observation was 857 patient-years after censoring of follow-up. We observed in both subgroups with LMW apo(a) phenotypes (11 to 19 and 20 to 22 K-IV repeats) an elevated frequency of coronary events per 100 patient-years when compared to subgroups with only HMW apo(a) isoforms (Figure 1). The results were similar independent of whether all patients or only those without a coronary event before the enrollment of the study were included in the analysis. Obviously, a clear association exists between the apo(a) phenotypes stratified in LMW and HMW and the occurrence of coronary events. Since the apo(a) phenotype group with 20 to 22 K-IV repeats represents a threshold for developing coronary events in hemodialysis patients (Figure 1), we performed further analyses simply by using the dichotomous variable of LMW (11 to 22 K-IV repeats) and HMW (>22 K-IV repeats) apo(a) phenotypes.
Number of coronary artery disease (CAD) events per 100 patient-years in hemodialysis patients in relation to the number of kringle-IV (K-IV) repeats of apo(a). Data are provided for the whole patient group as well as those free of CAD events at the start of the study. n = the number of patients in each group of K-IV repeats.
Multiple Cox Proportional Hazards Regression Analysis Including all CAD Events
We used a Cox regression-adjusted analysis of coronary event-free survival after the enrollment to the study to identify the variables with the most predictive value for the risk of developing a coronary event. In a first analysis, we used the information from all patients. The best model identified four variables associated with an event: age, LMW apo(a) phenotype, diabetes mellitus, and a preexisting CAD (Table 4). We further investigated the independence of these variables by two approaches. First, we tested several interaction terms built from the above variables that identified an interaction between diabetes mellitus and the apo(a) phenotype contributing significantly to the risk. This means that diabetes mellitus was only a risk factor for a major coronary event in the presence of an LMW apo(a) phenotype (Table 4). Introducing this interaction term to the model did not change the relative risks of age, LMW apo(a) phenotype, and preexisting CAD for the development of a major coronary event during the follow-up. Second, the univariate and multivariate Cox regression analysis resulted in similar relative risks for the variables age, LMW apo(a) phenotype, and preexisting CAD. These two approaches identified these three variables to be independent risk factors for a coronary event.
Effect of different variables on the development of a CAD event in hemodialysis patients using a multiple Cox proportional hazards regression analysisa
Age and the LMW apo(a) phenotype were also independent risk factors when we excluded those 51 patients who had already suffered a CAD event before enrollment into the study (Table 4) or those 82 patients who had suffered a CAD and/or a cerebrovascular event. Again, diabetes mellitus was only a risk factor in the presence of an LMW apo(a) phenotype. Figure 2 shows the time-to-event curves for patients with LMW and the HMW apo(a) phenotypes, separately for the total patient group (panel A) and for those without preexisting CAD (panel B). This demonstrates a higher risk for a coronary event during the entire observation period for patients with LMW compared to those with HMW apo(a) phenotypes.
Coronary event-free survival in hemodialysis patients with high (HMW) and low molecular weight (LMW) apo(a) phenotypes. Panel A shows the results for the whole patient group, and panel B includes only patients free of coronary events at the start of the study. Adjusted results are obtained from a multiple Cox proportional hazards regression analysis. Numbers near the survival curves represent the number of patients with HMW and LMW apo(a) phenotypes at risk at the times 0, 12, 24, 36, 48, and 60 mo.
The LMW apo(a) phenotype remained an equally good predictor for a coronary event when we used for the Cox regression analysis only patients with an age ≤65 yr at the start of the study regardless of whether patients with a preexisting CAD were included in the model (Table 5). The relative risk for suffering a coronary event was between 2.3 and 2.7 for patients with LMW apo(a) phenotypes and was therefore very stable in the different models we tested (Tables 4 and 5). The apo(a) phenotype became the most significant predictor even when we included only the data from 191 patients with an age ≤55 yr resulting in a relative risk of 7.7 (95% CI, 2.3 to 25.8) (P < 0.001).
The effect of different variables on the development of a CAD event in hemodialysis patients aged ≤65 yr using a multiple Cox proportional hazards regression analysisa
Patients included in this study were already on hemodialysis treatment for 26 ± 32 mo at the beginning of the observation period with a range from 3 to 180 mo. To minimize this heterogeneity, we considered in a further analysis only patients on dialysis therapy for less than 2 yr at the start of the study. In these patients, the LMW apo(a) phenotype remained a significant predictor for coronary events also.
The Lp(a) concentration usually shows a high correlation with the apo(a) phenotype (18), which has to be considered in the regression analysis. Therefore, we investigated in a similar series of analyses as those described above whether the Lp(a) concentration is predictive of a coronary event when the apo(a) phenotype is excluded from the analysis. Lp(a) concentration still failed to be a significant predictor in all models (P values between 0.15 and 0.50). Even the introduction of an interaction term of the apo(a) phenotype with the Lp(a) concentration did not show a significant correlation (P = 0.23).
We also investigated whether the lack of a significant association of Lp(a) levels with coronary events is due to patients in the group remaining free of coronary events but who suffered a stroke or peripheral atherosclerosis during the follow-up. Exclusion of these patients from the analysis did not significantly change the results (data not shown).
Multiple Cox Proportional Hazards Regression Analysis Excluding CAD Events Defined by Coronary Angiography
The inclusion of coronary events defined as angiographically diagnosed coronary stenosis is a possible cause of selection bias since not all patients under observation underwent an angiography. We therefore excluded in a final analysis angiographically diagnosed coronary stenosis from the event definition. We again observed the same four variables as predictors for CAD events as in the analysis including all 66 cases: age, preexisting CAD events, diabetes mellitus, and the LMW apo(a) phenotype (Table 6). An interaction term of diabetes mellitus with the apo(a) phenotype displaced diabetes from the model, which again means that diabetes was only a risk factor in presence of an LMW apo(a) phenotype.
The effect of different variables on the development of a CAD event excluding coronary stenosis defined by coronary angiography from the event definitiona
Inclusion of other variables (e.g., gender, total or LDL cholesterol, triglycerides, plasma albumin, creatinine, weekly dialysis duration, hypertension, or smoking) to each of the multivariate models discussed above did not significantly change the results.
Discussion
This is the first prospective study investigating the predictive value of the apo(a) phenotype for coronary events in hemodialysis patients. The results strengthen the indication we had already obtained from our recent cross-sectional investigations (27,29), i.e., that the LMW apo(a) phenotype is associated with an increased risk for atherosclerosis in hemodialysis patients. Furthermore, it supports our earlier observations (27,29) that although Lp(a) is elevated in hemodialysis patients, it is not predictive for atherosclerotic events. This is in contrast to the other prospective study of 129 hemodialysis patients by Cressman and colleagues, which found Lp(a) and the presence of a previous event as the only contributors to the risk of a clinical event attributed to atherosclerosis during the follow-up period (7). However, there are differences between these two studies, such as the classification of cardiovascular events, the ethnic populations studied, the intensity (hours per week) and duration (months) of hemodialysis treatment, and the assays used for measurement of Lp(a). However, we do not know to what extent these variables can explain the differences in the results.
A genetic parameter such as the apo(a) phenotype has a major advantage for the evaluation of an atherosclerosis risk compared with other biochemical or clinical parameters: It will not change with disease or environmental influences. Therefore, a single blood withdrawal can identify a person as one of a group with a higher risk for atherosclerosis. This is especially important in patients with renal disease, since they show during the various stages of renal disease pronounced and manifold changes in their biochemical parameters (39), including Lp(a) concentrations (4). Inflammatory processes (40) or the change of dialysis modality from hemodialysis to continuous ambulatory peritoneal dialysis (30) can have a significant influence on Lp(a) levels. This might be the reason why these parameters are not predictive of coronary events when evaluated in the dialysis phase since they will not reflect the often completely different concentrations that predominated only a few months previously during varying phases of compensated or decompensated renal retention. This makes the situation very different from non-renal disease subjects who usually do not experience these dramatic changes.
In recent investigations, we observed an apo(a) phenotype-specific elevation of Lp(a) in hemodialysis patients (30,33), which means that Lp(a) increases only in patients with HMW apo(a) phenotypes compared with isoform-matched control subjects. Patients with LMW apo(a) phenotypes show Lp(a) concentrations similar to those of matched control subjects (30,33). This observation is an unexplained phenomenon, which was recently confirmed by another study group (40). In the first study in hemodialysis patients relating atherosclerosis to Lp(a) and apo(a) phenotype, we found that the apo(a) phenotype was a better predictor for the prevalence and the degree of carotid atherosclerosis than was the Lp(a) plasma concentration (27). This is in contrast to the general population, in which the Lp(a) concentration dominates in the prediction of CAD (12,23,24). We therefore developed a model explaining the situation in hemodialysis patients based on the above-described apo(a) isoform-specific elevation of Lp(a) (30,33). In patients with HMW apo(a) phenotypes, Lp(a) concentrations increase and come closer to the concentrations seen in patients with LMW apo(a) phenotypes. Therefore, the risk for atherosclerotic complications can no longer be discriminated by means of Lp(a) concentrations. The apo(a) phenotype, however, gives approximate information about the former contribution of Lp(a) to the risk for atherosclerosis. This is probably more important since the pre-disease period with its specific atherosclerosis risk lasted longer in most of the patients than did the present situation. It is furthermore conceivable that patients with an LMW apo(a) phenotype and a more pronounced atherosclerosis preload develop a “galloping” atherosclerosis after commencement of renal insufficiency or hemodialysis treatment. This model was supported by a recent cross-sectional study, which found the apo(a) phenotype more predictive for CAD than the Lp(a) concentration in a large group of 607 hemodialysis patients (29). These patients were independent from those investigated in the present study. Therefore, end-stage renal disease treated by hemodialysis seems to be the only known condition in which the apo(a) phenotype is more predictive in multivariate analysis than the Lp(a) plasma concentration. In the present study the Lp(a) concentration did not reach significance in the Cox regression analysis even when the apo(a) phenotype was excluded from the analysis.
Similar to our recent prevalence study (29), age and diabetes mellitus were predictive of coronary events. Interestingly, diabetes mellitus was only a significant predictor when patients had an LMW apo(a) phenotype that indicates an interaction of these two conditions. ApoB and HDL cholesterol showed a tendency to be associated with CAD, which, however, did not reach statistical significance. The frequency of males was not significantly different in univariate and multivariate analysis, which might be explained by the fact that our recent study was a prevalence study. In the present study, only events occurring during the follow-up period of 5 yr were used for the analysis.
Patients were on average relatively old (57 ± 15 yr) at study entry, and hormonal protection was therefore possibly already diminished in women.
Limitations of the Study
Patients included in this study were already on hemodialysis treatment for 26 ± 32 mo at the beginning of the observation period. Therefore, the patient group was heterogeneous with respect to dialysis duration. We do not think, however, that this had a major influence on the results, because patients both with and without CAD events during the follow-up period had been on dialysis treatment for a similar time at enrollment in the study (Table 1). Even when we considered only patients on dialysis therapy for less than 2 yr at the start of the study, the main findings in the multivariate regression analysis remained unchanged.
Unfortunately, we are not able to provide measurements for dialysis adequacy like urea reduction ratio (41) or Kt/V. These data were not included in the standard examination of patients or were discussed controversially (42) when we designed this study several years ago. Albumin, creatinine, urea, and the weekly dialysis duration at the study entry, however, were similar in patients who suffered a coronary event compared to those who remained unaffected during the follow-up.
A further limitation is the definition and diagnosis of coronary events in our study. For all of the patients, a coronary angiography was not warranted. Since approximately 50% of dialysis patients with clinical angina pectoris have no significant coronary artery occlusion (43), we decided to use only stringent criteria for defining a coronary event. We thus identified only major coronary events, i.e., definitive myocardial infarction, aortocoronary bypass surgery, PTCA, and coronary artery stenosis >50% verified by angiography. However, the inclusion of events defined by angiographically verified stenosis is problematic in case not all patients undergo a coronary angiography. This could cause a selection bias in defining the event group since it is likely that some individuals in the event-free group had an asymptomatic stenosis. Because the whole patient group was not systematically angiographied, those patients might have been misclassified. To exclude such a selection bias, we also performed a Cox regression analysis excluding stenosis verified by angiography from the definition of coronary events. Although this strengthened the definition of cases, it did not change our main findings (Table 6). The decision to use homogeneous criteria for atherosclerotic complications led, on the other side, to the exclusion of cerebrovascular accidents and peripheral vascular disease from the main analysis. In contrast to our clear definition of complications, the inclusion of, for example, cerebrovascular disease would have brought several uncertainties to the analysis due to heterogeneities of the underlying pathophysiology: Ischemic and hemorrhagic strokes cannot be considered as one event group, and in some cases the classification is not entirely possible. We believe that the use of hard criteria for CAD in this study is a major strength of the investigation that might have been diluted by the inclusion of less identifiable and pathophysiologically heterogeneous atherosclerotic changes.
Finally, the study group included patients with and without coronary events at the start of follow-up as well as patients with age >65 yr. Because both factors could have an influence on the outcome, we performed the multivariate analysis separately for the whole patient group as well as for the remaining patients when those with a preexisting CAD (and in one subanalysis even those with cerebrovascular accidents) or an age >65 yr or both were excluded. Nevertheless, the apo(a) phenotype was, other than age, the best predictor for a coronary event during the observation period regardless of whether older patients or patients with a preexisting CAD or both were excluded. Since the relative risks of age and the LMW apo(a) phenotype remained stable, independent of the inclusion or exclusion of other variables and regardless of whether we calculated the relative risks in univariate or in multivariate Cox regression analysis (or after the introduction of several interaction terms), it can be concluded that age and the LMW apo(a) phenotype are independent predictors of major coronary events in hemodialysis patients. Patients with LMW apo(a) phenotypes could especially profit from regular evaluations to detect and correct the disease before it becomes fatal.
Acknowledgments
Acknowledgments
Dr. Kronenberg is supported by the Austrian Program for Advanced Research and Technology (APART) of the Austrian Academy of Science. This study was supported by grants from the Austrian Nationalbank (Project 5553) and from the D. Swarovski/Raiffeisen Foundation to Dr. Kronenberg, as well as from the Austrian Fonds zur Förderung der wissenschaftlichen Forschung to Dr. Dieplinger (P-12358). The expert technical assistance of the nursing staff of the five dialysis units is appreciated.
Footnotes
American Society of Nephrology
- © 1999 American Society of Nephrology