Correlates of Vascular Access Occlusion in Hemodialysis

Correlates of Vascular Access Occlusion in Hemodialysis Philip Goldwasser, MD, M.M. Avram, MD, James T. Collier, MD, Marie-Alex Michel, MD, Sara-Ann Gusik, MS, and Neal Mittman, MD • Vascular access occlusion results in significant morbidity in hemodialysis patients. Age, diabetes, and synthetic grafts (polytetrafluoroethylene [PTFE]) have been associated with vascular access occlusion in univariate analysis. However, the independent risk associated with each of these factors has not been assessed adjusting for confounding among the factors or by other variables, such as blood pressure (BP) or hematocrit. The influence of serum lipoprotein(a) [Lp(a)) and fibronectin on vascular access occlusion has not been widely studied despite their theoretical or demonstrated importance in vascular bypass occlusion. In a cohort study of 124 hemodialysis patients monitored for up to 14 months, we reported that Lp(a) values in the upper tertile (;", 57 mg/dL) were associated with vascular access occlusion risk in white and Hispanic patients, but not in black patients. We now report an expanded analysis of this data set to determine the independent correlates of vascular access occlusion. Variables tested included age, race, gender, diabetes, access type (PTFE v endogenous), treatment time, systolic BP, hematocrit, heparin and erythropoietin dosage, and serum levels of Lp(a) and fibronectin. In univariate analysis, access occlusion was associated with age, diabetes, PTFE, Lp(a) ;", 57 mg/dL, serum fibronectin, and reduced BP. The independent correlates of first access occlusion were determined with the Cox proportional hazards model. Since the overall model included a significant race x Lp(a) interaction term, we stratified by race. In black patients, risk correlated directly with PTFE (P < 0.01) and inversely with systolic BP (P < 0.001), whereas for white and Hispanic patients, age (P = 0.04) and Lp(a) ;",57 mg/dL (P = 0.05) were associated with increased risk. In summary, vascular access occlusion was found to be associated with a number of factors. Important independent correlates were PTFE and lower BP in black patients, and age and serum Lp(a) ;",57 mg/dL in white and Hispanic patients. Diabetes mellitus and increased serum fibronectin may contribute additional risk. © 1994 by the National Kidney Foundation, Inc. INDEX WORDS: Lipoprotein(a); fibronectin; hemodialysis; vascular access; thrombosis.

R

ELIABLE vascular access has been described as the Achilles heel of modem hemodialysis. 1 In addition to being an important cause of morbidity for hemodialysis patients, access occlusion is responsible for a sizeable part of the cost of the end-stage renal disease program. 2 One fourth of all admissions and hospital days of hemodialysis patients are attributable to access placement and complications. 1 The pathology and risk factors for vascular access occlusion have been studied,l but several important questions remain to be explored. The first concerns the difficulty of identifying independent risk factors in the clinical setting. For example, age, diabetes, and use of synthetic grafts, such as expanded polytetrafluoroethylene (PTFE), have been identified as predisposing to access occlusion. 2-6 However, PTFE is more likely to be used in older patients and patients with diabetes, ie, patients with veins inadequate to support an endogenous fistula. s-? Confounding among age, diabetes, and PTFE makes it difficult to estimate the independent risk associated with each factor. Other potentially relevant factors, such as hematocrit8 and blood pressure (BP), have not been systematically examined or controlled for in previous studies. Second, no anatomic cause can be found in up

to one third of access thromboses. 1In these cases, hypercoagulability, sluggish flow due to hypotension, and extrinsic compression may be contributory.1,9.1O Deficiencies of anticoagulant proteins have not been confirmed in patients whose accesses clot frequently. II In most cases with anatomic lesions, the pathology has been found to be runoff vein stenosis due to vascular smooth muscle ("neointimal") hyperplasia leading to thrombosis. 6.10.1 2 The pathogenesis of this neointimal hyperplasia is poorly understood. Endothelial injury due to excessive intraluminal pressure, turbulent flow, or calcification may be contributory.6.13 It has been hypothesized that upstream release of growth factors (such as platelet-derived growth factor) may contribute to the prolifFrom Division of Nephrology, The Long Island College Hospital, Brooklyn, NY. Received March 3, 1994; accepted in revised form July 8, 1994. Supported in part by grants from The National Kidney Foundation of New YorklNew Jersey, the Nephrology Foundation of Brooklyn, and the Guild (Gift Shop Division) of The Long Island College Hospital. Address reprint requests to M.M. Avram, MD, The Long Island College Hospital, Atlantic Ave and Hicks St, Brooklyn, NY 11201. © 1994 by the National Kidney Foundation, Inc. 0272-6386/9412405-0006$3.00/0

American Journal of Kidney Diseases, Vol 24, No 5 (November), 1994: pp 785-794

785

786

"

.

eration of venous smooth muscle in the runoff vein.6.12.14 It is also logical to speculate that circulating levels of matrix elements, such as fibronectin, could contribute to venous neointimal hyperplasia. Soluble fibronectin is a circulating glycoprotein that can deposit in tissues (although it is not identical to "cellular" fibronectin).15-17 Fibronectin promotes the adhesion, proliferation, spreading, and migration of various cell types, including vascular smooth muscle and mesangial cells,IS-21 yet it inhibits the migration of endothelial cells. 20 Aside from a potential role in neointimal hyperplasia, fibronectin may also contribute directly to thrombosis. Fibronectin is an integral component of the fibrin gel/clot that initiates tissue repair. 22.23 Platelets, fibrin, and plasminogen can bind to fibronectin. 15 ,IS,24 Fibronectin coating of PTFE increases its thrombogenicity.25 Serum fibronectin concentration has not been examined in relation to vascular access occlusion. Serum lipoprotein(a) [Lp(a)] is a low-density lipoprotein-like particle with the potential to inhibit fibrinolysis. It has been found to be an independent predictor of coronary risk and coronary bypass stenosis in white and Asian populations. 26-3o Serum levels in the top third of the population are associated with increased risk. 26,27 Lp(a) has been found in association with areas of intimal proliferation and atherosclerosis in coronary bypass venous grafts. 31 Serum Lp(a) levels were recently found to correlate directly with the accelerated coronary disease of heart transplants, a disorder typified by extreme hyperplasia of vascular smooth muscle?2 Lp(a) may promote vascular smooth hyperplasia by inhibiting the generation of transforming growth factor-f3?3 A possible interaction of Lp(a) with fibronectin is suggested by the observation that the apoprotein(a) portion of Lp(a) can bind to the heparin-binding domain of fibronectin. 24 Endstage renal disease is associated with markedly increased levels of serum Lp(a)?4,35 In a cohort study of 124 hemodialysis outpatients followed prospectively for up to 14 months, we reported that Lp(a) levels in the upper tertile (2::.57 mg/dL) were associated with increased risk of vascular access occlusion in white and Hispanic patients, but not black patients, and confirmed that serum Lp(a) is significantly higher in black patients. 36 We now expand this analysis to include variables such as diabetes, age, BP, and fibronectin, and

GOLDWASSER ET AL

possible confounding among them, using the Cox proportional hazards model to identify the independent factors that may predispose to access occlusion. MATERIALS AND METHODS

Patients A group of 124 hemodialysis outpatients at The Long Island College Hospital was enrolled in the study on February 4 and 5, 1991, and monitored for up to 14 months. Clinical and laboratory data were collected at the time of enrollment. Follow-up to first occlusion event was censored on death. transplantation, or transfer to continuous ambulatory peritoneal dialysis or to another dialysis center. The mean length of follow-up was 10 months and the median length of followup was 14 months. The study received approval by the Institutional Review Board at The Long Island College Hospital, and written informed consent was obtained from each patient after a discussion of the protocol.

Clinical and Demographic Data Clinical and demographic data included age, race, gender. diabetic status, etiology of end-stage renal disease and total months on dialysis, estimated dry weight, treatment time, access type (endogenous fistula v PTFE graft), heparin dose, dialyzer membrane, and predialysis systolic and diastolic BPs (mean of three consecutive treatments). The mean age of the population was 60 ::t: 15 (SD) years, and the mean duration of end-stage renal disease was 56 ::t: 52 months. Forty percent of the patients were diabetic and 52% were female. The causes of end-stage renal disease were hypertensive nephrosclerosis (34.9%). diabetes mellitus (34.9%). chronic glomerulonephritis (9.5%), polycystic kidney disease (8.7%), obstruction (4.8%). miscellaneous (4.8%), and unknown (2.4%). The access type was PTFE in 89 cases (71.8%) and endogenous fistula in 35 cases (28.2%). Access creation at our center was performed by a small group of experienced vascular surgeons. Some of the patients studied were transfers who originated their dialysis and vascular access at other centers.

Description of Dialysis Prescription All patients were on thrice-weekly hemodialysis with bicarbonate bath using Cobe Centrysystem 3 dialysis machines (Cobe, Lakewood, CO). Dialyzer membranes were cellulose acetate (55%), saponified cellulose ester (32%), polysulfone (6%), hemophane (3%), and cuprarnrnonium rayon (1%). Blood flows ranged from 300 to 500 mUmin. Dialysis prescription was guided in part by a goal of achieving a value of 2:2.5 for the ratio of pretreatment blood urea nitrogen (BUN) to 5-minute posttreatment BUN. 37 ,38 The posttreatment BUN was drawn 5 minutes after treatment to minimize the problem of spuriously low post-BUN values that results from graft recirculation and from disequilibrium, The mean BUN ratio was 2.73 ::t: 0.54 (range, 1.37 to 4.60), The mean treatment time was 3.01 ::t: 0.36 hours (range, 2 to 4 hours). At enrollment, 105 patients were receiving heparin with each dialysis treatment and 70 patients were receiving erythropoietin therapy at the conclusion of each dialysis session. For

787

VASCULAR ACCESS OCCLUSION

heparin-free treatments, 100- to 2oo-mL saline flushes were administered hourly and as needed to monitor circuit patency. No patients were receiving coumadin or aspirin. Elevated venous pressure on dialysis, edema of the upper extremity containing the vascular access, or abnormally low or high monthly BUN ratio values were used by clinicians as indications for investigation of the vascular access with recirculation studies. 39 A recirculation ratio of 20% or above on two determinations (or 15% or above for blood flow rates of :5350 mUmin) was used by clinicians as an indication for angiography. For the purpose of this study, stenosis was defined as narrowing of the graft anastomosis or runoff vein that was considered to be significant by the radiologist. Although the recirculation ratio and venous line pressure were used clinically during the study, baseline values of these parameters were not obtained for testing as risk factors because the focus of the study was on potential causes of occlusion.

Biochemical Data On enrollment, nonfasting baseline bloods were drawn immediately before the first treatment of the week from the arteriovenous access. A multiphasic biochemistry screen was obtained for each patient using the SMAC autoanalyzer (Technicon, Tarrytown, NY). Total cholesterol was measured by the SMAC autoanalyzer (reference range, 120 to 200 mgt dL). High-density lipoprotein cholesterol was measured on the DuPont ACA (Wilmington, DE) using an assay from EM Diagnostic Systems (Gibbstown, NI) (reference range, 32 to 96 mg/dL). Hematocrit was measured using the Technicon HI analyzer. Prealbumin (reference range, 17 to 42 mg/dL in a normal healthy population; linear range, 7 to 50 mg!dL), apoprotein A-I (reference range, 96 to 176 mg!dL in normal males and 109 to 203 mg!dL in normal females; linear range, 25 to 225 mg/dL), and apoprotein B (reference range, 43.0 to 128 mgt dL in normal males and 42.0 to 112 mg!dL in normal females; linear range, 30 to 200 mg/dL) were determined by rate nephelometry on a Beckman Array Protein System (Beckman Instruments Inc, Brea, CAl. Appropriate blanks were used to correct for turbidity in samples. Serum Lp(a) was determined by Dr Daniel Levine of The Rogosin Institute. The New York Hospital-Cornell Medical Center. Serum Lp(a) (reference range, 0 to 24 mg!dL; linear range, 5 to 105 mg!dL) was measured by immunoturbidometric analysis (SPQ II. kits; INCST AR Corp, Stillwater, MN) on the Roche COBAS FARA n clinical chemistry analyzer (Roche Diagnostic Systems, Branchburg, NJ). The Lp(a) assay has been standardized to the double monoclonal enzyme-linked immunoassay used at the Northwest Lipid Research Laboratory, University of Washington (Seattle, WA).40 The antibodies used to measure Lp(a) do not cross-react with plasminogen up to 230 mg/dL and apoprotein B up to 436 mg!dL. There is no interference in the Lp(a) assay from triglycerides up to 4 gIL. Renal patients often have elevated triglyceride levels; therefore, patient samples with levels above 4 gIL were diluted prior to analysis. Fibronectin (linear range, 55.6 to 500 mgIL) was determined by radial immunodiffusion (The Binding Site. San Diego, CAl.

Study Design and Data Analysis In the univariate analysis, the association of baseline characteristics with access occlusion was examined with a nested case-control design. Continuous variables, reported as the mean :t SD, were compared using ANOVA or, where indicated in the text, the Mann-Whitney U-test. Proportions were compared by the chi-squared test or, where indicated in the text, by the Mantel-Haenzel test for linear association. Correlations among baseline characteristics are reported as the Pearson correlation coefficient (r) . In the multivariate analysis, the independent correlates of first occlusion (thrombosis or stenosis) after study entry were determined with the Cox proportional hazards mode!.41"2 Variables considered for the model included baseline age, race, gender, diabetes, systolic BP, access type, treatment time, dosages of heparin and erythropoietin, hematocrit, and serum levels of Lp(a) and fibronectin. The variables chosen reflected our findings in the univariate analysis as well as reports in the literature. Subsets of variables were selected for the models based on our prior experience and to balance the need to control for confounding while at the same time not overfitting the model by selecting too many variables. No more than four variables were included in the models (one per 10 outcome events)." The reduced models containing only significant terms included no more than two variables. The proportional hazard assumption was tested by plotting baseline hazards across strata. If the hazards were parallel among the strata, the variable was modeled as a covariate; otherwise, the analysis was stratified. The fit of the models were assessed by comparing predicted survival with that of the observed probabilities of survival for selected subgroups of patients. Observed survival was computed by the Kaplan-Meier method.44 Data presented include the relative risk ratios and their 95% confidence intervals, and the actual probability values for the risk ratios. The risk ratios reflect the level of risk associated with a variable, assuming other factors are held constant. For analysis of time to first vascular access occlusion, outcome was censored at the time of death, transplantation, or transfer to peritoneal dialysis or to another institution. Calculations were performed using SPSS for Windows 5.0.2 (SPSS Inc, Chicago. IL) and Egret (SERC, Seattle, WA) statistical software.

RESULTS

Baseline Characteristics and First Access Occlusion Thirty-eight patients had at least one thrombosis (n = 33) or stenosis (n = 5). A nested case-control comparison of baseline characteristics of patients with and without occlusive events is presented in Tables I and 2. The group that had vascular occlusion events was older (P = 0.03), included more diabetic (P < 0.02) and female (P oS 0.13) patients, had a higher mean serum level of fibronectin (P < 0.04), lower mean systolic and diastolic BP (P < 0.02), longer treatment times (P

788

GOLDWASSER ET AL

Table 1. Demographic and Clinical Comparison of Patients With and Without Access Occlusion Occlusion Events

Age (yr) Race (%) White Black Hispanic Male(%) Diabetes (%) Systolic BP (mm Hg) Diastolic BP (mm Hg) PTFE (%) Erythropoietin (U/wk) Hematocrit (%) Heparin (U/session) Time (hr) Crude mortality (%)

None (n = 86)

.. 1 Event (n = 38)

Probability Value

58::':: 16

65::':: 13

0.03 0.46

35% 50% 15% 54% 33% 139 ::':: 22 78::':: 11 63% 4,105::':: 3,798 27.4::':: 5.3% 1,516::'::942 2.97::':: 0.37 20%

24% 58% 18% 37% 58% 128 ::':: 25 73::':: 10 92% 3,026 ::':: 3,018 27.7::':: 5.1% 1,420::':: 893 3.10::':: 0.32 16%

0.13 < 0.02 < 0.02 <0.02 <0.002 0.11' 0.80 0.66' < 0.03' 0.80

, By Mann-Whitney U-test.

< 0.03), and were more likely to have PTFE grafts < 0.002) and serum Lp(a) values in the upper tertile (P < 0.05). There were no differences in (P

other nutritional and/or lipoprotein markers, such as serum albumin, creatinine, cholesterol, and apoprotein B; nor were differences present for hematocrit or dosages of erythropoietin or heparin. The relationship of several of the variables to the crude risk of having at least one access occlusion is shown in Figs 1 to 5. The relationship of access type to occlusion

risk was potentially confounded by other risk factors (Table 3). Po1ytetraftuoroethylene cases were older than patients with endogenous fistulae (P < 0.001), received longer treatments (P < 0.04), and included more patients with diabetes (P < 0.01) and more female patients (P < 0.03). The PTFE group also had a smaller proportion of black patients (P < 0.04); the associations of access type with the other risk factors were similar when black patients and white and Hispanic patients were examined separately.

Table 2. Biochemical Comparison of Patients With and Without Access Occlusion Occlusion Events None (n

Lp(a) (mg/dL) Lp(a) 2: 57 mg/dL (%) Cholesterol (mg/dL) Apoprotein B (mg/dL) High-density lipoprotein cholesterol (mg/dL) Apoprotein A-I (mg/dL) Albumin (g/dL) Prealbumin (mg/dL) Creatinine (mg/dL) Fibronectin (mg/L) BUN ratio (pre/post)

= 86)

43.8::':: 40.8 27% 171 ::':: 50 64.8::':: 23.9 38::':: 14 97.5::':: 23.1 3.8::':: 0.3 26.5::':: 6.7 12.1 ::':: 3.2 182 ::':: 54 2.70::':: 0.40

.. 1 Event (n

= 38)

Probability Value

54.3::':: 42.6 45% 174::':: 40 67.4::':: 19.9 39::':: 12 99.0 ::':: 15.4 3.8::':: 0.4 26.5::':: 6.0 11.8 ::':: 4.4 204::':: 52 2.79::':: 0.77

• By Mann-Whitney U-test; respective median Lp(a) values for the two groups are 33.9 and 44.3 mg/dL. t By Mantel-Haenszel test for linear association.

< 0.13' < 0.05t > 0.70 0.55 >0.70 >0.70 >0.90 >0.90 > 0.70 < 0.04 0.40

789

VASCULAR ACCESS OCCLUSION % With Occlusion

% With Occlusion ~ r---------------------------------,

~ -------------------,

P< O.oe

P
40

40

30 ..

N-24 ....... .

30

20 ..... .

20

•

10

10

oL------- <40.0

Fig 1.

40.0-54.9

55.0-64.9

65.0-74.9

;<75.0

o

:s163.0

163.1·213.9

;<214.0

Fibronectin mg/l

Age (years)

Crude rate of first occlusion by age. P

< 0.05 by Mantel-Haenszel test for linear association.

Similarly, there were significant associations among the other risk factors identified in univariate analysis. Diabetic patients were older than nondiabetic patients (66 ± 12 years v 56 ± 16 years; P < 0.001), included more females (62% v 45%; P = 0.09), and had longer treatment times (3.11 ± 0.38 hours v 2.94 ± 0.33 hours; P < 0.02) and lower diastolic BP (74 ± 9 mm Hg v 79 ± 12 mm Hg; P < 0.01). Female patients were older than male patients (64 ± 12 years v 56 ± 17 years; P < 0.01), and had higher Lp(a) (52.7 ± 39.9 mg/dL v 40.9 ± 42.6 mg/dL; P < 0.03 by Mann-Whitney U-test) and lower diastolic BP (74 ± 11 mm Hg v 80 ± 10 mm Hg; P < 0.002). Age correlated inversely with diastolic BP (r = -0.28, P = 0.002), but did not correlate with systolic BP (P > 0.90). Systolic and diastolic BP were highly correlated (r

Fig 3. Crude rate of first occlusion by serum fibronectin. P < 0.06 by Mantel-Haenszel test for linear association.

0.65, P < 10-4). The mean serum level of fibronectin was significantly higher in nondiabetic (197 ± 54 mg/L v 177 ± 53 mg/L; P < 0.05) and black (200 ± 56 mg/L v 177 ± 49 mg/L; P < 0.02) patients, but was not significantly correlated with gender, age, access type, Lp(a), or BP. By ANOV A, a higher concentration of fibronectin was independently associated with access occlusion (P < 0.05), black race (P < 0.02), and nondiabetes (P < 0.02). Serum fibronectin was correlated with the concentrations of other serum nutritional markers, such as prealbumin (r = 0.37, P < 10-3), albumin (r = 0.17, P < 0.07), creatinine (r = 0.17, P < 0.07), and apoprotein B (r = 0.22, P < 0.02). =

% With Occlusion 70 . - - - - - - - - - -- - -- - - - - - ,

% With Occlusion ~

P
60 ~

40

30

30

20 10

10

o

o :s120.0

121·149

;<150.0

<21 .5

21.5-56.9

;<57.0

Upoprotein(a) mg/dl

Systolic Blood Pressure mm Hg

Fig 2. Crude rate of first occlusion by systolic BP. P < 0.06 by Mantel-Haenszel test for linear association.

Fig 4. Crude rate of first occlusion by Lp(a) in white and Hispanic patients. P < 0.01 by Mantel-Haenszel test for linear association.

GOLDWASSER ET AL

790

interaction, the analysis was stratified by race (Tables 4 and 5). For black patients, PTFE grafts were associated with a sevenfold to eightfold increased risk (P < 0.01), whereas risk decreased by approximately 3% for each 1 mm Hg increase in baseline systolic BP (P < 0.(01). Serum Lp(a) did not influence risk in this group. In contrast, for white and Hispanic patients, risk increased nearly threefold for Lp(a) :257 mg/dL (P :5 0.05) and by 4% for each year of age [P = 0.04, adjusting for Lp(a)]. Diabetes and fibronectin were associated with increased risk in both the white/Hispanic and black groups, although neither variable was a statistically significant independent predictor, adjusting for age, systolic BP, access type, and Lp(a).

% With Occlusion ~ r-----------------------------------. P < O.75

30

20

10

o

21 .5·56.9

",57.0

lipoprotein(a) mg/dl

Fig 5. Crude rate of first occlusion by Lp(a) in black patients. P < 0.75 by Mantel-Haenszel test for linear association.

Correlates of First Access Occlusion: Multivariate Analysis

DISCUSSION

The independent correlates of first access thrombosis or stenosis were derived using the Cox proportional hazards model. Independent variables tested were age, diabetes, gender, race, systolic BP, access type, Lp(a) :257 mg/dL, fibronectin, treatment time, hematocrit, and epogen and heparin dosage. When examining the entire group, we detected an interaction term between race and Lp(a) (P = 0.05). Because of this

The morbidity and economic burden related to maintaining vascular access in hemodialysis patients are increasingly being recognized. Age, diabetes, and PTFE have been associated with access occlusion in case-control and cohort studies. However, few studies investigated the independent risk associated with each of these factors controlling or adjusting for confounding among them. Certain variables, such as BP and hemato-

Table 3. Baseline Characteristics by Access Type Access Type

Age (yr) Race (%) White and Hispanic Black Male(%) Diabetes (%) Systolic BP (mm Hg) Diastolic BP (mm Hg) Erythropoietin (U/wk) Hematocrit (%) Heparin (U/session) Time (hr) Lp(a) (mg/dL) Lp(a) ~ 57 mgldL (%) Fibronectin (mglL) , By Mann-Whitney U-test.

Endogenous Fistula (n = 35)

PTFE (n = 89)

Probability Value

53:<::: 15

63:<::: 15

<0.001 <0.04

31% 69% 66% 20% 135:<::: 28 78:<::: 14 4,257 :<::: 3,845 29.0:<::: 5.6% 1,695 :<::: 1,072 2.91 :<::: 0.32 49.4:<::: 43.4 29% 188 :<::: 61

54% 46% 42% 48% 135 :<::: 21 76:<::: 10 3,584 :<::: 3,503 27.0:<::: 5.0% 1,406 :<::: 854 3.05:<::: 0.37 46.0:<::: 40.9

34% 190:<::: 52

<0.01 <0.01 >0.90 0.60 0.29' <0.04 0.18' <0.04' 0.80' 0.58 0.90

VASCULAR ACCESS OCCLUSION

791

Table 4. Predictors of Vascular Access Occlusion Over 14 Months in 65 Black Patients* Predictor

Relative Risk

95% Confidence Interval

Probability Value

Age (per yr) Systolic BP (per mm Hg) Lp(a) ;;;,: 57 mg/dL (v < 57) PTFE (v endogenous)

1.013 0.968 0.949 7.810

0.976-1.051 0.949-0.986 0.394-2.286 1.749-34.92

0.498 <0.001 0.910 0.007

• In the reduced model, containing only significant variables, the predictors and estimated relative risks are PTFE (relative risk, 8.38; P = 0.005) and systolic BP (relative risk, 0.97; P = 0.0006).

crit, have not been systematically studied. The influence of biologically relevant circulating factors, such as Lp(a) and fibronectin, on vascular access occlusion has not been widely studied despite their theoretical or demonstrated importance for vascular bypass occlusion in coronary and other settings. To address these issues, we prospectively monitored vascular access occlusion in a cross-sectional sample of 124 hemodialysis patients for up to 14 months. Multivariate methods were used to determine the independent influence of a broad range of clinical and biochemical factors on the risk of vascular access occlusion. Since the occlusive outcome studied was thrombosis or clinically suspected stenosis, the findings of this study only apply to patients with clinically detectable events. In univariate analysis, numerous characteristics were associated with first vascular access occlusion. These included older age, diabetes, PTFE, increased serum Lp(a) and, previously unreported, lower BP, increased serum fibronectin, and longer treatment time. However, significant relationships were also demonstrated between many of these baseline variables and each other. Consequently, their independent relationship to occlusion was examined with multivariate methods. In the Cox proportional hazards analysis, we examined black patients separately from white and

Hispanic patients. Age and Lp(a) ?57 mg/dL were statistically significant only for white and Hispanic patients, while PTFE and systolic BP were statistically significant only in the black group (Tables 4 and 5). Although different variables were significant in the two groups, the relationships to risk of age, PTFE, and BP were consistent for both groups. It is not surprising that age, PTFE, and systolic BP did not achieve statistical significance for both black patients and white and Hispanic patients given the small size of the subgroups as well as inevitable differences between the groups in absolute range of values of these variables and in their degree of intercorrelation. The findings confirm and extend earlier univariate studies that found increased risk with PTFE and age. 3-5 Diabetes was significantly associated with access occlusion risk in univariate analysis but not in multivariate analysis. Diabetic patients were older, had lower BP, and were more likely to have PTFE accesses. Therefore, the lack of an independent predictive value of diabetes in this data set is due to confounding. Windus et al found diabetes to be significantly associated with risk of thrombosis of new synthetic grafts. 3 These investigators compared access survival in diabetic and nondiabetic cohorts matched for age, gender, race, access location, and erythropoietin

Table 5. Predictors of Vascular Access Occlusion Over 14 Months in 59 White and Hispanic Patients* Predictor

Relative Risk

95% Confidence Interval

Probability Value

Age (per yr) Systolic BP (per mm Hg) Lp(a) ;;;,: 57 mg/dL (v < 57) PTFE (v endogenous)

1.040 0.987 2.925 1.861

0.999-1.084 0.962-1.012 1.024-8.354 0.226-15.30

0.057 0.320 0.045 0.564

• In the reduced model, containing only significant variables, the predictors and estimated relative risks are age (relative risk, 1.04; P = 0.04) and Lp(a) ;;;,: 57 mg/dL (relative risk, 2.74; P = 0.05).

792

usage. The mean BP and hematocrit values of the cohorts were not reported. Feldman et al recently reported an analysis of the risk of access-related hospitalizations in a very large incident group of hemodialysis patients. 2 Using the Cox proportional hazards model, these investigators assessed the independent prognostic value of demographic characteristics and cause of renal failure with respect to the risk of access-related hospitalization. Diabetes, age, female gender, and black race were associated with small, albeit statistically significant increases in risk. However, the analysis did not differentiate between hospitalizations for occlusion and those for other reasons (eg, infections, aneurysm repair). More important, their data set did not allow differentiation of synthetic grafts from endogenous fistulae or venous catheters. Thus, the relative risks reported in their study for diabetes and other predictors may be confounded by access type. In the multivariate analysis, we used systolic BP as an independent variable instead of diastolic BP because the latter was confounded by age and diabetes in this data set. Although reduced systolic BP seems like an obvious cause ofhemodialysis fistula occlusion (eg, by leading to sluggish flow), it has not been previously reported. The finding that Lp(a) ~57 mgldL is an independent risk correlate in the white and Hispanic group but not in the black group extends our previous analysis?6 The lack of independent effect of Lp(a) in black patients despite higher serum concentration is compatible with the speculation that Lp(a) is less thrombogenic/atherogenic in black patients?8,4S-47 Of therapeutic importance, nicotinic acid derivatives and apheresis have been reported to lower serum Lp(a) in hemodialysis patients. 48,49 Fibronectin was a significant correlate of access occlusion in univariate analysis (Table 2 and Fig 3), but not in the multivariate analysis. Given the potential of fibronectin to contribute to thrombosis and hyperplasia, fibronectin may have a role in vascular access occlusion. In the cross-sectional analysis, fibronectin concentration correlated with other markers of visceral or somatic protein status, such as prealbumin, apoprotein B, creatinine, and albumin. These findings confirm the correlations first reported by Hou et al in 24 hemodialysis patients. so We also found that serum fibronectin was lower in dia-

GOLDWASSER ET AL

betic patients, which may be related to their overall reduced protein status. 36 Similarly, LabatRobert et al reported lower serum fibronectin levels, but increased capillary wall fibronectin, in diabetic patients. sl Since Lp(a) and fibronectin are "visceral" protein markers, it could be argued that their association with occlusion is not specific, but reflects a general tendency of more robust, wellnourished patients to clot or to undergo venous smooth muscle hyperplasia, a process somewhat analogous to atherosclerosis.14.s2-s4 Against this possibility, however, is the finding that other markers of visceral protein status and/or atherosclerotic risk, such as albumin, prealbumin, total cholesterol, and apoprotein B, did not correlate with the risk of access occlusion (Table 2). The relationship of serum Lp(a) and possibly fibronectin with access occlusion merits confirmation in other hemodialysis populations. The baseline values of hematocrit, erythropoietin dose, and heparin dose were not associated with occlusion risk despite their potential for influencing thrombosis8,ss,s6 and hyperplasia,14,s7-61 the latter, for example, by affecting the level of mitogens such as endothelin. It cannot be excluded that subsequent changes in these variables may influence risk. However, the possibility seems of diminished importance gi ven the marked similarity in baseline values of these variables between the two groups (Table 1). This hypothesis can be tested in future studies using the Cox model with time-dependent covariates. Although this study used multivariate methods to minimize the problems related to confounding and multiple comparisons, certain limitations must be noted. First is the possibility of confounding by current access site and geometry or prior treatment history (eg, access duration, previous access failures, infections, and revisions). It is unlikely that this would have impact on variables such as age, BP, and Lp(a). Second, because a cross-sectional sample was used, the patients and vascular accesses present on study entry are subject to length-selection bias, eg, clotters being switched to peritoneal dialysis. If this occurred, it would tend to minimize the associations reported. Third, there is a potential leadtime bias because a few of the occlusions (13%) were stenoses, diagnosed somewhat before a thrombosis might have occurred. However, if one

VASCULAR ACCESS OCCLUSION

recomputes the Cox models after adjusting the event-free survival of the cases with stenotic outcome by adding, for example, 90 days to each of such cases, the results are virtually identical to the original models in Tables 4 and 5. Fourth, there is a potential bias from overfitting multivariate models when many variables are examined relative to the number of occlusions, although we minimized this by testing a priori hypotheses and by limiting the number of variables examined to "the rule of 10."43 In summary, in a prospective study of 124 hemodialysis patients, access occlusion was associated with a number of risk factors. In an analysis using Cox's regression stratified by race, the independent correlates of vascular access occlusion were PTFE and reduced systolic BP for black patients, and age and increased Lp(a) for white and Hispanic patients. Diabetes and serum fibronectin may contribute additional risk. If confirmed, these findings may help clinicians to understand why certain patients are at increased clotting risk and allow investigators to better stratify baseline risk for future observational and interventional studies. ACKNOWLEDGMENT The authors thank Dr Joseph Feldman for his assistance in the statistical analysis; Dr Daniel Levine of The Rogosin Institute, New York Hospital-Cornell Medical Center, to whom we are much indebted for the Lp(a) determinations; INCSTAR Corp for their donation of Lp(a) assay kits; Antoinette Antignani for her expert technical assistance; and Marilyn Collins for her assistance in preparing this manuscript.

REFERENCES 1. Fan P-Y, Schwab SJ: Vascular access: Concepts for the 1990s. J Am Soc Nephrol 3:1-11, 1992 2. Feldman HI, Held PJ, Hutchinson JT, Stoiber E, Hartigan MF, Berlin JA: Hemodialysis vascular access morbidity in the United States. Kidney Int 43:1091-1096, 1993 3. Windus DW, Jendrisak MD, Delmez JA: Prosthetic fistula survival and complications in hemodialysis patients: Effects of diabetes and age. Am J Kidney Dis 19:448-452, 1992 4. Windus DW, Audrain J, Vanderson R, Jendrisak MD, Picus D, Delmez JA: Optimization of high-efficiency hemodialysis by detection and correction of fistula dysfunction. Kidney Int 38:337-341, 1990 5. Winsett OER, Wolma FJ: Complications of vascular access for hemodialysis. South Med J 78:513-517, 1985 6. Swedberg SH, Brown BG, Sigley R, Wight TN, Gordon D, Nicholls SC: Intimal fibromuscular hyperplasia at the venous anastamosis of PTFE grafts in hemodialysis patients: Clinical, immunocytochemical, light and electron microscopic assessment. Circulation 80: 1726-1736, 1989

793 7. Tordoir JHM, Herman JMMPH, Kwan TS, Diderich PM: Long-term follow-up of the polytetrafiuoroethylene prosthesis as an arteriovenous fistula for hemodialysis. Eur J Vasc Surg 2:3-7, 1987 8. Livio M, Gotti E, Marchesi D, Mecca G, Remuzzi G, deGaetano G: Uremic bleeding: Role of anaemia and beneficial effect of red-cell transfusions. Lancet 2:1013-1015, 1982 9. Galbraith S, Fan P, Collins D, Schwab S: Hemodialysis fistula thrombosis: A prospective evaluation of anatomic vs nonanatomic causes. J Am Soc Nephrol 3:369, 1992 (abstr) 10. Mohaideen AH, Tanchajjah S, A vram MM, Mainzer RA: Arteriovenous access for hemodialysis: Using polytetrafiuoroethylene grafts. N Y State J Med 80:190-195, 1980 11. Domoto DT, Bauman JE, Joist JH: Combined aspirin and sulfinpyrazone in the prevention of recurrent hemodialysis vascular access thrombosis. Thromb Res 62:737-743, 1991 12. Diskin CJ, Stokes TJ, Jr, Pennell AT: Pharmacologic interventions to prevent hemodialysis vascular access thrombosis. Nephron 64:1-26, 1993 13. Bush HL, Jr, Jakubowski JA, Curl GR, Deykin D, Nabseth DC: The natural history of endothelial structure and function in arterialized vein grafts. J Vasc Surg 3:204-215, 1986 14. Majack RA: Toward a molecular understanding of vascular smooth muscle cell growth control. Cardiovasc Rev Rep 7:731-737, 1986 15. Khan MY: Structure and properties of fibronectin. Indian J Biochem Biophys 27:63-68, 1990 16. Hayman EG, Ruoslahti E: Distribution of fetal bovine fibronectin and endogenous rat cell fibronectin in extracellular matrix. J Cell Bioi 83:255-259, 1979 17. Oh E, Pierschbacher M, Ruoslahti E: Deposition of plasma fibronectin in tissues. Proc Nat! Acad Sci USA 78:3218-3221, 1981 18. Von der Mark K, Von der Mark H, Goodman S: Cellular responses to extracellular matrix. Kidney Int 41 :632-640, 1992 19. Madri JA, Marx M: Matrix composition, organization and soluble factors: Modulators of microvascular cell differentiation in vitro. Kidney Int 41:560-565, 1992 20. Madri JA, Kocher 0, Merwin JR, Bell L, Tucker A, Basson CT: Interactions of vascular cells with transforming growth factor {3. Ann N Y Acad Sci 593:243-258, 1990 21. Boudreau N, Clausell N, Boyle J, Rabinovitch M: Transforming growth factor-{3 regulates increased ductus arteriosus endothelial glycosaminoglycan synthesis, and a posttranscriptional mechanism controls increased smooth muscle fibronectin, features associated with intimal proliferation. Lab Invest 67:350-359, 1992 22. Dvorak HF: Tumors: Wounds that do not heal: Similarities between tumor stroma generation and wound healing. N Engl J Med 315:1650-1659, 1986 23. Hansen MS: Platelets and plasma proteins in the early phase of thrombus formation. Dan Med Bull 33:96-105, 1986 24. Enholm C, Jauhianen M, Metso J: Interaction of lipoprotein(a) with fibronectin and its potential role in atherogenesis. Eur Heart J 11:190-195, 1990 (suppl E) 25. Kempczinski MD, Ramalanaona GR, Douville C, Silberstein EB: Thrombogenicity of a fibronectin-coated, experimental polytetrafiuoroethylene graft. Surgery 101 :439-444, 1987 26. Murai A, Miyahara T, Fujimoto N, Matsusada M, Ka-

794 meyama M: Lp(a) lipoprotein as a risk factor for coronary heart disease and cerebral infarction. Atherosclerosis 59: 199204, 1986 27. Dahlen GH, Guyton JR, Attar M, Farmer JA, Kautz JA, Gotto M: Association of levels of lipoprotein Lp(a), plasma lipids and other lipoproteins with coronary disease documented by angiography. Circulation 74:758-765, 1986 28. Scanu A: Lipoprotein(a): A genetic risk factor for premature coronary disease. JAMA 267:3326-3329, 1992 29. Rosengren A, Wilhemsen L, Eriksson E, Risberg B, Wedel H: Lipoprotein(a) and coronary heart disease: A prospective case control study in a general population sample of middle aged men. BMJ 301:1248-1251, 1990 30. Hoff HF, Beck OJ, Skibinski CI, Jurgens G, O'Neil J, Kramer J, Lytle B: Serum Lp(a) level as a predictor of vein graft stenosis after coronary bypass surgery in patients. Circulation 77:1238-1244, 1988 31. Cushing GL, Gaubatz JW, Nava ML, Burdick BJ, Bocan TMA, Guyton JR, Weilbaecher D, DeBakey ME, Lawrie GM, Morrisett JD: Quantitation and localization of apolipoproteins[al and B in coronary artery bypass vein grafts resected at reoperation. Arteriosclerosis 9:953-603, 1989 32. Barbir M, Kushwaha S, Hunt B, Macken A, Thompson GR, Mitchell A, Robinson D, Yacoub M: Lipoprotein(a) and accelerated coronary artery disease in cardiac transplant recipients. Lancet 340: 1500-1502, 1992 33. Kojima S, Harpel PC, Rifkin DB: Lipoprotein(a) inhibits the generation of transforming growth factor beta: An endogenous inhibitor of smooth muscle migration. J Cell Bioi 113:1439-1445, 1991 34. Parra HJ, Mezdour H. Cachera C, Dracon M, Jacquet A, Fruchart JHC: Lipoprotein(a) in patients with chronic renal failure treated by hemodialysis. Clin Chern 33:721, 1987 (letter) 35. Cressman MD, Heyka RJ, Paganini EP, O'Neil J, Skibiniski CI, Hoff HF: Lipoprotein(a) is an independent risk factor for cardiovascular disease in hemodialysis patients. Circulation 86:475-482, 1992 36. Goldwasser P, Michel A-M, Collier JC, Mittman N, Fein PA, Gusik S-A, Avram MM: Prealbumin and Jipoprotein(a) in hemodialysis: Relationships with patient and vascular access survival. Am J Kidney Dis 22:215-225, 1993 37. Daugirdas JT, Dumler F, Zasuwa GA, Levin NW: Chronic hemodialysis prescription, in Daugirdas JT, Ing S (eds): Handbook of Dialysis (ed 1). Boston, MA, Little, Brown, 1988, pp 839-857 38. Kesheviah PR, Hanson GL, Berketh RO, Collins AJ: A simplified approach to monitoring in vivo therapy prescription. ASAIO Trans 34:620-622, 1988 39. Sherman RA: Dialysis access recirculation, in Nissenson AR, Fine RN (eds): Dialysis Therapy. Philadelphia, PA, Hanley and Belfus, 1993, pp 19-21 40. Levine DM, Sloan BJ, Donner JE, Lorenz JD, Heinzerling RH: Automated measurement oflipoprotein(a) by immunoturbidimetric analysis. Int J Clin Lab Res 22: 173-178, 1992 41. Cox DR: Regression models and life tables (with discussion). J R Stat Soc Series B 34:187-220, 1972 42. Kelbfleisch J, Prentice R: The proportional hazards model, in The Statistical Analysis of Failure Time Data. New York, NY, Wiley, 1980, pp 70-117 43. Concato J, Feinstein AR, Holford TR: The risk of

GOLDWASSER ET AL

determining risk with multi variable models. Ann Intern Med 118:201-210, 1993 44. Kaplan EL, Meier P: Nonparametric estimation from incomplete observations. J Am Stat Assoc 53:457-481, 1958 45. Guyton JR, Dahlen GH, Ratsch W, Kautz JA, Grotto AML: Relationship of plasma lipoprotein Lp(a) levels to race and apolipoprotein B. Arteriosclerosis 5:265-272, 1985 46. Srinivasan SR, Dahlen GH, Jarpa RA, Wevver LS, Berenson GS: Racial (black-white) differences in serum lipoprotein(a) distribution and its relation to parental myocardial infarction in children. Circulation 84:160-167, 1991 47. Sorrentino MJ, Vielhauer C, Eisenbart JD, Pless GM, Scanu AM, Feldman T: Plasma lipoprotein(a) protein concentration and coronary artery disease in black patients compared with white patients. Am J Med 93:658-662, 1992 48. Bosch T, Thiery J, Sambtleben W, Seidel D: Combined simultaneous LDL-apheresis and hemodialysis in hypercholesterolemic ESRD patients: A new mode of therapy. J Am Soc Nephrol 2:233, 1991 (abstr) 49. Nakahama H, Miyazaki M, Imai N, Yokokawa T, Okada M, Kubori S, Okamural K, Tanaka Y, Shirai D: Reduction of plasma lipoprotein(a) levels by niceritrol in patients on maintenance hemodialysis. J Am Soc Nephro12:340, 1991 (abstr) 50. Hou F-f, Zhang X, Wang A-L, Wu J-G: Fibronectin in patients with chronic renal failure undergoing dialysis. Nephron 55:45-48, 1990 51. Labat-Robert J, Kern P, Robert L: Modification of the biosynthesis of type-land type-II collagens and fibronectin during diabetes and atherosclerosis. Z GerontoI24:66-69, 1991 52. Avram MM: Similarities between glomerular sclerosis and atherosclerosis in human renal biopsy specimens: A role for lipoprotein glomerulopathy. Am J Med 87:39N-41N, 1989 53. Ross R, Glomset JA: The pathogenesis of atherosclerosis: I and II. N Engl J Med 295:369-377, 420-425, 1976 54. Avram MM, Goldwasser P, Burrell D, Antignani A, Mittman N: The uremic dyslipidemia: A cross-sectional and longitudinal study. Am J Kidney Dis 20:324-335, 1992 55. Kaupke CJ, Butler GC, Vaziri ND: Effect of recombinant human erythropoietin on platelet production in dialysis patients. J Am Soc Nephrol 3:1672-1679, 1993 56. Moia M, Manucci PM, Vizzoto L, Casati S, Cattaneo M, Ponticelli C: Improvement in the haemostatic defect of uraemia after treatment with recombinant human erythropoietin. Lancet 2:1227-1229, 1987 57. Koenig W, Ernst E: The possible role ofhemorheology in atherothrombogenesis. Atherosclerosis 94:93-107, 1992 58. Yokokawa K, Tahara H, Kohno AK, Yanagisawa M, Takeda T: Heparin regulates endothelin production through endothelium-derived nitric oxide in human endothelial cells. J Clin Invest 92:2080-2085, 1993 59. Carlini RG, Dusso AS, Obialo CI, Alvarez UM, Rothstein M: Recombinant human erythropoietin increases endothelin-1 release by endothelial cells. Kidney Int 43: 10101014, 1993 60. Carlini RG, Alvaro R, Alvarez U, Rothstein M: Human recombinant erythropoietin stimulates angiogenesis. J Am Soc Nephrol 4:425, 1993 (abstr) 61. Gogusev J, Zhu D-L, Marche 0, Drueke T: Recombinant human erythropoietin stimulates rat vascular smooth muscle growth in culture. J Am Soc Nephrol 4:512, 1993 (abstr)