Relationship Between Bilirubin, Apolipoprotein B, and Coronary Artery Disease*

Transcription

1 ANNALS O F CLINICAL AND LABORATORY SCIE N C E, Vol. 27, No. 3 Copyright 1997, Institute for Clinical Science, Inc. Relationship Between Bilirubin, Apolipoprotein B, and Coronary Artery Disease* STANLEY S. LEVINSON, Ph.D. Department o f Pathology, University o f Louisville, Louisville, KY and Laboratory Service, Department of Veterans Affairs Medical Center, Louisville, KY ABSTRACT Lipoprotein lipids and apo B from 254 male patients were compared with bilirubin as a risk factor for coronary artery disease (CAD). Patients were classified as: 1, Normal, all vessels <20% stenosis, n = 83; or 2, CAD, at least one vessel >70%, n = 171. Diagnostic accuracy was assessed by receiver operator characteristic curves (ROC). Corrections were made for possible confounding variables in the multivariate analysis. Upon ROC analysis, bilirubin showed an inverse relationship with risk of CAD, with areas under the ROC curve comparable to lipoproteins; however, bilirubin showed no discrimination below false positive frequencies of approximately 0.3. Logistic regression indicated that bilirubin was a weaker global marker than the lipoproteins and interacted with apo B. A highly significant correlation was found between bilirubin and apo B (p = ), but not with cholesterol, triglycerides, or high density lipoprotein cholesterol. Compared to lipoprotein markers, bilirubin provides little practical power of discrimination for CAD. Further studies of the affect of bilirubin on CAD must take its interaction with apo B into consideration. Introduction Bilirubin is a catabolic product of hemoglobin which is formed in the cells of the reticuloendothelial system. Recent studies have shown that there is an inverse relationship between bilirubin concentrations and risk of coronary artery disease (CAD).1,2,3 The process responsible for this relationship is unclear, * Send reprint requests to Stanley S. Levinson, Ph.D., Laboratory Service, Department of Veterans Affairs Medical Center, 800 Zom Avenue, Louisville, KY but endogenous antioxidant properties of bilirubin have been proposed as a possible mechanism. It has been suggested that bilirubin might be an additional marker for CAD with a diagnostic power equivalent to that of high density lipoprotein cholesterol (HDLC) and that bilirubin analogues might have a role in antioxidant therapy.3 In the present study, the diagnostic performance of bilirubin as a marker for CAD was assessed by construction of receiver operating characteristic (ROC) curves. Bilirubin was also compared with lipoprotein lipids and apolipoproteins as a marker for CAD. The data sug /97/ $01.20 Institute for Clinical Science, Inc.

2 186 LEVINSON gests that there is a significant interaction betw een apo B and bilirubin. Since both appear to affect CAD, this interaction must be taken into consideration in studies attempting to delineate better the relationship between bilirubin and CAD. Methods S u b je c t s, B l o o d C o l l e c t io n, a n d A n g io g r a p h y The procedure was approved by the Veteran Administration Medical Center review board. Treatment of the patients, samples, and angiography are sim ilar to th a t previously described.5,6 Briefly, patients were males, 40 to 70 years old, entering the hospital for clinically indicated angiographic studies, such as chest pain, dyspnea, syncope, abnormal electrocardiogram, and/or abnormal stress test. Blood was drawn from consecutively examined patients. The following were exclusion criteria: Patients on known lipid altering medications (gemfibrozil, hydroxymethyl-glutaryl coenzyme A reductase inhibitors, bile acid resins, niacin, and heparin), diabetics, persons with chronic kidney disease, and persons experiencing a myocardial infarction within 3 months of the procedure were excluded. Samples with elevated triglycerides (>2,000 mg/l) were tested by lipoprotein electrophoresis, and those shown to have chylomicrons w ere assumed to be non fasting and excluded. Patients who refused to participate were also excluded. Angiography was perform ed by standard methods, using the femoral or brachial artery approach. Patients were categorized into two groups7,8 1, Normal, patients with normal arteries, <20% stenosis, number = 83. In the great majority of cases, these patients showed no detectable stenosis. 2, Severe CAD, those having 5:70% stenosis in at least one major vessel, number = 171. An additional 33 patients showed interm ediate degrees of stenosis (>20% but <70%). These w ere excluded because of the small number. In order to correct for the effect of possible confounders and risk factors other than lipoproteins, a standardized questionnaire was designed to assess smoking, alcohol intake, family history of myocardial infarction (mother or father with myocardial infarction at age ^ 5 5 years), age, body mass index (BMI), which was calculated from the formula BMI = weight (kg)/[height (m)],2 and medications (3-adrenergic agonists, calcium channel blockers, angiotensin converting enzyme (ACE) inhibitors, and diuretics. Nominal variables were classified as yes or no, except for alcohol intake which was classified as none, light to moderate, and heavy, based on 0, 1 to 6, or >6 drinks per day. Blood for lipoprotein studies was drawn, without preservatives, from the femoral artery prior to the infusion of the contrast medium. The patients were fasting overnight, and, in order to avoid effects of diurnal variation,2 specimens were collected between 0900 and 1500 hours. Sera were obtained by centrifugation at room temperature. Bilirubin, cholesterol, and triglyceride assays were stored at 4 C and determ ined within 48 h. Aliquots were frozen at -70 C for apo B measurements. A ssays Bilirubin, cholesterol, HDLC, and triglycerides were assayed by routine methods with a discrete analyze.* The HDLC was separated by precipitation with dextran sulfate in the presence of magnesium ion9,10 and magnetic removal of low density lipoprotein cholesterol (LDL) and very low density lipoprotein cholesterol^ prior to measurement.11 Apo B was measured using kits by automated rate immunonephelometry,$ according to the manufacturer s instructions. It has previously been shown that this assay is linear over a wide range of apo B concentrations, shows good parallelism when compared to isolated * Johnson and Johnson Clinical Diagnostics, Rochester, NY t Polymedco, Courtland, NY t Array, Beckman Instruments, Brea, CA.

3 (pure) LDL, and that serum samples give the same results as plasma samples collected with EDTA, except for a small difference owing to the known concentrating affect that causes proteins in serum to be about 4% higher than in plasma.12 Sta tistics Statistics w ere perform ed w ith JM P,* except for the Kolmogorov Smirnov test which was performed with Statview 4.01,f and ROC curves were calculated using the program CLABROC4 Areas and standard errors were derived from the nonparametric Mann Whitney U statistic (Trapezoidal area).13 The formula: z = (trapezoidal area - 0.5)/standard error was used to test the hypothesis that an area under the ROC curve is significantly different from chance (area = 0.5), where p is determined from a table for the normal distribution (z).14 Univariate analysis was performed with the Mann Whitney U test for continuous variables, and, using the likelihood ratio, x2 was calculated from contingency tables for discontinuous variables. Multivariate logistic regression was performed on all variables with a p value of less than 0.2 by univariate testing. A p value of 0.2 was arbitrarily chosen rather than 0.05 or 0.1 because univariate derived p values may become more significant upon multivariate testing; thus, choosing 0.2 is a better safety measure to insure that a significant interaction would not be overlooked. In order to define appropriately the interactions between variables, several different logistic models were tested. Variability of the relative odds ratios were assessed using 95% confidence intervals. Univariate regression was performed with the Spearmen correlation coefficient-rho. Two tail tests were used for significance (p = 0.05). * SAS Institute Inc., Cary, NC t Abacus Concepts Inc., Berkeley, CA t Available free of charge from Dr. Charles E. Metz, Department of Radiology, University of Chicago Medical Center, 5841 South Maryland Avenue, Chicago, IL BILIRUBIN AND LIPOPROTEINS 187 Total cholesterol and apo B concentrations were not significandy different from a Gaussian distribution at the p S 0.05 level, while triglycerides, HDLC, and bilirubin were statistically different from gaussian prior to logarithmic conversion but were not significantly different from Gaussian after conversion. Thus, natural log (Ln) was used with logistic regression for these three latter variables. Results In table 1 are shown the demographics of the subjects. The univariate difference in age between the groups was highly significant. p-adrenergic agonists showed a p value <0.2. No other possible confounders came near being significantly different between groups. The effect of age and (3-adrenergic agonists were adjusted for by fitting them into the logistic regression equations below. The ROC curves compare diagnostic accuracy of methods at various cutoff values within the test distributions by normalizing each distribution into a similar scale defined in terms of diagnostic sensitivity and specificity. The ROC curves can be used to identify a best cutoff or decision point. The areas under ROC curves are global indexes of diagnostic accuracy, and, as such, can be used to compare the diagnostic accuracy of different methods. In table II are shown the areas under the curve for all variables. The relationship between CAD and each variable is indicated. It also shows the means and standard deviations for each variable, and the range of values for bilirubin. It is indicated in table II that the area under the curve for bilirubin is the same as that for cholesterol. The ROC curve for bilirubin is shown in figure 1. For comparison, the ROC curve for apo B and cholesterol are also shown. All three exhibit statistically significant discrimination as compared to an area of 0.5 (for bilirubin and cholesterol, z = 2.16, p < 0.04, for apo B, z = 3.05, p < 0.01). Notice that although apo B and cholesterol show equal global power of discrimination according to the area, bilirubin shows no discrim ina-

4 188 LEVINSON TABLE I Subject Characteristics CAD N Significance Level (p) Continous Variables Mean (SD) Mean (SD) Mann Whitnev 60.4 (7.8) 54.1 (9.1) BMI (4.7) 27.3(5.7) 0.93 Nominal Variables % Positive * % Positive* Xe ACE inhibitors Alcohol: None Light - moderate Heavy P-blockers Calcium channel Diuretics Family history** Hypertension Race White Black Smokers Numbers of patients: Normal (N) - 83; coronary artery disease (CAD) ACE * angiotensin converting enzyme. Percent positive, except for alcohol consumption and race. **Two patients were adopted. TABLE II Areas Under ROC Curves for Bilirubin and Lipoprotein Markers Assay N Mean (SD) (mg/l) CAD Mann Whitney U Statistic Relationship Betweem Variable and CAD Area Triglycerides 1340 (682) 1755 (1083) 8633 Increases with CAD 0.61 HDLC 412 (156) 372 ( 144) 8701 Decreases with CAD 0.61 apo B 1078 (327) 1198 ( 347) 7856 Increases with CAD 0.61 Total C 1780 (399) 1891 (400) 8206 Increases with CAD 0.58 Bilirubin* 6.7 (3.2) 5.7 ( 2.1) 8246 Decreases with CAD 0.58 Number of patients: coronary artery disease (CAD) - 171; Norman (N) HDLC - high density lipoprotein cholesterol. *The range of values for bilirubin was between 2.0 and 18 mg/l for the normal group, and between 2 and 17 mg/l for the CAD group.

5 BILIRUBIN AND LIPOPROTEINS 189 BILIRUBIN (mg/l) F igure 1. R e c e iv e r o p e ra to r c h a ra c te ris tic curves for bilirubin, cholesterol, and apo B. Bilirubin area ± SE = 0.58 ± Apo B area ± SE = 0.61 ± Cholesterol area ± SE = 0.58 ± Solid diagonal line is line of no discrimination, area = 0.5. Difference between 0.5 and 0.58 for bilirubin and for cholesterol, Z = 2.16, p < 0.04; difference for apo B, Z = 3.05, p < Concentrations of bilirubin corresponding to sensitivity/specificity pairs (decision points) on the ROC curve are indicated on the axis above. FALSE POSITIVE FREQUENCY (1 - SPECIFICITY) tion below a FPF of 0.3, which corresponds to a decision point of bilirubin greater than 4.7 mg/l. Since the ROC curves test on a univariate basis, bilirubin was compared to the lipoprotein markers by multivariate logistic regression. Logistic regression provides a means to evaluate the affect of bilirubin and lipoprotein markers on CAD and on one another simultaneously. At the same time, it provides a means to adjust statistically for differences by other prognostic factors. In the present case, adjustm ent for the possible confounding factors (table I) of age and (3-adrenergic agonists ((3-blockers) was made by adding them to the equation. In table III are shown a series of logistic models from which alterations in predictive power can be determined. In model 1, bilirubin is examined by itself. The affect on bilirubin of adding the possible confounders, age and (3-blockers is shown in model 2. All the following models (models 3 to 6) show the additional affect on bilirubin of adding individual lipoprotein variables. The equations

6 190 LEVINSON TABLE III Logistic Regression Comparing Bilirubin with Lipoprotein Markers Model Variables Entered in Model P Value Relative Odds (Confidence Interval) Model 1 LnBilirubin ( ) Model LnBilirubin ( ) Model 3 LnHDLC ( ) LnBilirubin ( ) Model 4 LnTriglycerides ( ) LnBilirubin ( ) Model 5 Cholesterol ( ) LnBilirubin ( ) Model 6 apo B ( ) LnBilirubin ( ) Number: Normal (N) - 83 and coronary artery disease (CAD) = 171. HDLC = high density lipoprotein cholesterol. indicate that age and (3-blockers cause a minimal change in the odds ratio from 0.20 (model 1) to 0.24 (model 2). As in other studies, lipoprotein lipids also caused minimal changes with the relative odds ratio varying between 0.28 and 0.29 (model 3 to 5), but when apo B was incorporated into the logistic equation (model 6), the relative odds ratio increase to 0.4, with the confidence interval showing a great deal of overlap beyond 1.0 (0.07 to 2.09). When the confidence interval for the odds ratio overlaps 1.0 substantially, it implies a large degree of uncertainty.15 These changes suggest that there was a strong interaction between apo B and bilirubin which caused the affect of bilirubin on CAD to be reduced. The interactions between variables were investigated further by individually examining the nonparam etric correlation coefficients,

7 which are shown in table IV. In agreement with the results of logistic regression, bilirubin and lipoprotein lipids showed non-significant correlation coefficients, but a highly significant inverse correlation was found with apo B (p = ). Discussion Results from prior studies indicated that there is a significant inverse relationship between bilirubin and CAD.1,2 3 The results shown here generally agree with these studies (table II). However, there are two important factors noted here that were not evident previously. The first of these factors is apparent from examination of the ROC curves, which were not considered in the other studies. Expressed in this way, the data show that although bilirubin is comparable to cholesterol as a global marker (table II), with an area of 0.58 for both, bilirubin shows little discrimination below a false positive frequency of about 0.3 (bilirubin of 4.7 mg/l). Consequently, it shows lower likelihood ratios for a positive test than cholesterol, providing little practical power of discrimination, since at the decision point of 4.7 mg/l (true positive frequency = 33%), th e false positive frequency is already 30%. The second factor is evident from examination of the regression analysis. As in other Bilirubin TABLE IV Correlations Between Bilirubin and Lipoprotein Markers Lipoprotein Marker Spearmna s rho BILIRUBIN AND LIPOPROTEINS 191 P Value Bilirubin HDLC Bilirubin Cholesterol Bilirubin Triglycerides Bilirubin apo B Number HDLC = high density lipoprotein cholesterol. studies, the inverse relationship between bilirubin and CAD rem ained with m inim al change after adjustment for lipoprotein lipids (table III, models 3 to 5). However, apo B, which was not measured in other studies, showed a strong interaction with bilirubin in logistic regression, with the relative odds ratio changing from 0.24 to 0.4 (table III) and the p-value doubling (model 6). These changes suggest that the relationship between bilirubin and CAD was substantially modified by apo B. Furthermore, since apo B showed a lower p value than bilirubin ( vs. 0.28), and since the confidence interval for the relative odds ratio for apo B did not overlap 1.0, while the confidence interval for bilirubin did overlap 1.0, it is apparent that apo B exerts a stronger affect on CAD than does bilirubin. Upon univariate analysis, bilirubin showed a highly significant correlation with apo B, p = (table IV), confirming that there was a significant interaction between bilirubin and apo B. Thus, further studies of the affect of bilirubin on CAD must take this interaction into consideration. Two other points of interest follow from this study. First, prior studies differed in the extent to which the inverse relationship between bilirubin and CAD varied. In one study, there was a striking progressive decrease in risk,3 with bilirubin showing a lower relative risk ratio for CAD than HDLC. In another study, the relative risk ratio was U-shaped with little or no reduced risk at the highest and lowest bilirubin concentrations.2 Here, the logistic regression (table III) indicates that when adjustment was made for possible confounders, all of the lipoprotein variables were better global markers than bilirubin, with the lipoprotein markers showing lower p-values. Second, one of the prior studies showed a significant inverse relationship between bilirubin and HDLC,3 which was not seen here (table IV). In that report, more than 50% of the CAD patients were taking cholesterol lowering medications. The authors indicated that this might be the reason they saw no significant difference in LDLC concentrations between the CAD and normal groups, but they did not com m ent on other possible

8 192 LEVINSON effects. Many lipid lowering drugs affect HDLC and triglycerides as well as LDLC. Since none of the patients in the present study were taking lipid altering drugs, this difference may explain the discrepancy. In summary, it is concluded that although bilirubin is inversely related to CAD, a relationship that appears to be independent of lipoprotein lipids, the lipoprotein lipids and apo B appear to be clinically more important markers than bilirubin. Furtherm ore, there appears to be an interaction between bilirubin and apo B which modifies the relationship between bilirubin and CAD. Adjustment for this interaction must be made in future studies examining the relationship between bilirubin and CAD. Acknowledgments This study was supported by funding from the Department of Veteran Affairs, a generous gift of reagent from Beckman Instruments, and a grant from the American H eart Association (Kentucky Affiliate). Thanks are extended to Dale Pike, R.N. for valuable technical assistance in drawing the bloods. R eferences 1. Schwertner HA, Jackson WG, Tolan G. Association of low serum concentration of bilirubin with increased risk of coronary artery disease. Clin Chem 1994;40: Breimer LH, Wannamethee G, Ebrahim S, Shaper AG. Serum bilirubin and risk of ischemic heart disease in middle-age British men. Clin Chem 1995;41: Hopkins PN, Wu LL, Hunt SC, James BC, Vincent GM, Williams RR. higher bilirubin is associated with decreased risk for early familial coronary artery disease. Arterioscler Thromb Vase Biol 1996;16: Sniderman AD, Cianflone K. [editorial] Measurement of apolipoproteins: time to improve the diagnosis and treatment of atherogenic dyslipoproteinemias. Clin Chem 1996;42: Levinson SS, Wagner S. Assay of apolipoprotein B containing lipoproteins for routine use. Arch Path Lab Med 1992;116: Levinson SS, Wagner S. Immunonephelometric/ turbidimetric apolipoprotein B assays for routine clinical laboratory use in cardiovascular disease. Clin Chim Acta 1993;223:31^t2. 7. Maciejko JJ, Homes DR, Kottke BA, Zinsmeister AB, Dinh DM, Mao SJT: Apolipoprotein AI as a marker of angiographically assessed coronary artery disease. N Engl J Med 1983;309: Kottke BA, Zinsmeister AR, Holmes Jr DR, et al. Apolipoproteins and coronary artery disease. Mayo Clin Proc 1986;61: Finley PR, Schifman RB, Williams RJ, Lichti DA: Cholesterol in high density lipoprotein: use of Mg2"/ dextran sulfate in its enzymatic measurement. Clin Chem 1978;24: Wamick GR, Nguyen T, Albers JJ. Comparison of improved precipitation methods for quantitation of high-density lipoprotein cholesterol. Clin Chem 1985; 31: Harris N, Galpchian V, Rifai N. Three routine methods for measuring high-density lipoprotein cholesterol compared with the reference method. Clin Chem 1996;42: Maciejko JJ, Levinson SS, Markyvech L, Smith M. Evaluation of a new method for assaying APO A-I and APO B lipoproteins by rate nephelometry. Clin Chem 1987;11: Hanley JA, McNeil BJ. The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology 1982;143: Beck JR, Schultz K. The use of relative operating characteristic (ROC) curves in test performance evaluation. Arch Pathol Lab Med 1986;110: Henderson AR. Chemistry with confidence: should clinical chemistry require confidence intervals for analytical and other data. Clin Chem 1993;39:929^34.