EPIBACDIAL: A Multicenter Prospective Study of Risk

Factors for Bacteremia in Chronic Hemodialysis Patients�’

BRUNO HOEN,* AGNES PAUL-DAUPHIN,t DOMINIQUE HESTIN,� and

MICHELE KESSLER�*Se?1,ice de Maladies Infectieuses et Tropicales, #Sen,ice d ‘inforinatique M#{233}dicale, Epidemiologic et

Statistiques, and �Sen’ice de N#{233}phrologie, University of Nancy Medical Center, Vandoeut’re, France.

Abstract. Baeteremic infections are a major cause of mortality

and morbidity in chronic hemodialysis patients. New develop-

ments in managing these patients (erythropoietin therapy. nasal

mupirocin, long-term implanted catheters, and synthetic mem-

branes) may have altered the epidemiologic patterns of bacte-

remia in dialysis patients. This multicenter prospective cross-

sectional study was carried out to determine the current

incidence of and risk factors for baeteremia in chronic hemo-

dialysis patients in France. A total of 988 adults on chronic

hemodialysis for I mo or longer was followed up prospectively

for 6 mo in 19 French dialysis units. The factors associated

with the development of at least one bacteremic episode over

6 mo were determined using the multivariate Cox proportional

hazards model. Staphylococcus aureus (ii = 20) and coagu-

lase-negative staphylococci (ii = 15) were responsible for most

of the 5 1 bacteremic episodes recorded. The incidence of

bacteremia was 0.93 episode per 100 patient-months. Four risk

factors for bacteremia were identified: (I) vascular access

(catheter versus fistula: RR = 7.6: 95% CI, 3.7 to 15.6); (2)

history of bacteremia (�2 versus no previous episode: RR

7.3: 95% CI. 3.2 to 16.4); (3) immunosuppressive therapy

(current versus no: RR = 3.0; 95% CI, I .0 to 6. 1); and (4)

corpuscular hemoglobin (per 1 g/dl increment: RR 0.7: 95%

CI. 0.6 to 0.9). Catheters, especially long-term implanted cath-

eters, were found to be the leading risk factor of bacteremia in

chronic hemodialysis patients. There was a trend toward re-

currenee of bacteremia that was not associated with chronic

staphylococcal nasal carriage. Synthetic membranes were not

associated with a lower risk of bacteremia in this population of

well dialyzed patients. but anemia linked to resistance to eryth-

ropoietin appeared to be a possible risk factor for bacteremia.

(J Am Soc Nephrol 9: 869-876, 1998)

Bacterial infections are still a major cause of morbidity and

mortality in chronic hemodialysis patients. Bacteremie infec-

tions have an incidence of approximately one episode per 100

patient-months ( I ) and may be life-threatening in these pa-

tients. Several risk factors for bacterial infections in chronic

hemodialysis have been identified or suggested, including

Staphylococcus aureus nasal carriage (2,3), a history of bacte-

rial infection (4), the use of a central venous dialysis catheter

rather than an arteriovenous fistula ( 1 .5,6), iron overload

(I ,7,8), hypoalbuminemia (9). and the use of bioincompatible

membranes ( 10. 1 1 ). A previous multivariate analysis identified

three parameters as significant and independent risk factors for

bacterial infection in chronic hemodialysis (4). These were a

history of bacterial infection (at least one previous bacterial

infection), the type of vascular access (catheter versus native

fistula), and elevated serum ferritin (>500 p.g/L). Since that

study was conducted, new developments in the management of

dialysis patients have included the widespread use of recom-

Received July 9. 1997. Accepted October 8. 1997.

“See Appendix for participating centers and laboratories.Correspondence to Dr. Bruno Hoen. Service de Maladies lnfectieuses et

Tropicales. University of Besan#{231}onMedical Center. F-25030 Besancon Cedex.France.

b046-6673/0905-0869$03.0()/()

Journal of the American Society of Nephrology

Copyright iD 1998 by the American Society of Nephrology

binant erythropoietin and nasal mupirocin ointments. New

dialysis techniques also have been implemented. such as the

use of long-term dialysis catheters and synthetic-presumably

more biocompatible-membranes. This investigation was con-

dueted to assess the influence of these changes on the epide-

miobogie patterns of bacteremia in chronic hemodialysis pa-

tients.

Materials and MethodsStudy Design

A prospective survey of all patients undergoing chronic hemodial-

ysis in hospital dialysis centers or in self-care dialysis units was

conducted in 19 French dialysis units. Patients undergoing dialysis at

home were excluded. All adult patients who were on a regular

hemodialysis program for more than 30 d on September 1 , I994, wereincluded in the study. All patients were followed for 6 mo. unless

follow-up was censored by renal transplantation. the patient’s moving

to a dialysis center not participating in the study, switch to peritoneal

dialysis. or death. The patients� physicians were asked not to change

the patient’s dialyzer throughout the study period. and to do only

minimal changes in other dialysis procedures. such as number and

duration of dialysis sessions. They also were asked to take at least two

blood cultures. to swab the vascular access (either the needle puncture

site of fistulas or the exit site of catheters. over a I -cm2 skin area) and

the patient’s nostrils for Staphylococcus aureu.s in anyone suspected

of developing a bacteremic infection (fever, signs of inflammation at

the vascular access site). The coordinating investigators were notified

of all confirmed diagnoses of bacterensia. The microbiobogists from

870 Journal of the American Society of Nephrology

the laboratory affiliated with each dialysis unit were also asked to

notify the coordinators of all positive blood cultures originating from

the dialysis unit. All notification forms were examined by the coor-

dinating investigators, who classified each ease according to the

following definitions.

Case Definitions

Only bacteremie episodes were considered. If a single blood culturewas positive for coagulase-negative staphylococci, Corvnebacteria, or

Bacillus sp., the culture was considered contaminated or representa-

tive of a transient bacteremia but not of a bacteremie episode. All

other situations in patients in whom at least one blood culture was

positive were defined as bacteremie episodes. Patients who developed

such a bacteremie episode within I mo before the start of the study

were not considered for incidence calculation and risk factor analysis.

Data Collection

The following data were recorded for each patient enrolled in the

study, at the time of inclusion, in September 1994: birth date: gender;

cause of renal failure; comorbid factors: date of start of dialysis;

history of bacteremia, kidney transplantation or peritoneal dialysis;previous and current erythropoietin treatment; current immunosup-pressive therapy; any iron supplementation, desfemoxamine treat-

ment, or nasal mupirocin ointments within the past 6 mo; results of

nasal swab culture; surgical procedures within the past month; and

history of blood transfusions and dialysis procedures (number ofdialysis sessions per week, duration of each session, dialysis mem-

brane, vascular access). Dialysis membrane types were listed assynthetic (polysulfone. AN69#{174},and pobymethylmethacrylate) or eel-

lubosic (hemophane, Cuprophane, and cellulose acetate). Baseline

laboratory data included serum protein, albumin, ferritin, and C-reac-tive protein levels, corpuscular hemoglobin. and blood urea nitrogen

concentrations before and after dialysis. Samples for blood ureanitrogen were taken immediately before the start of dialysis and at the

end of the session, after reducing the blood pump rate to 50 ml/min.

The post/predialysis plasma urea nitrogen ratio (R) was used to

calculate the urea reduction rate (URR = I - R) and to estimate KtIV,

from the formula: KtJV = - 1.18 X ln(R) ( I 2).

When a bacteremie episode was diagnosed, the patient’s attending

physician filled out a case record form that included: the date of firstpositive blood culture; the number of blood cultures performed and

number of positive blood cultures; identification of the micro-organ-

isms recovered from blood cultures; vascular access and nasal swabs;

possible administrations of nasal mupiroein ointments; and changes in

dialysis procedures between the beginning of the study and the onset

of the baeteremic episode.

The data collected at the end of the study period included the

number of bacteremic episodes, the number and type of any surgical

procedures, changes in vascular access and dialysis procedures. and

any changes in supportive therapy (erythropoietin, iron supplementa-

tion, desferrioxamine, blood transfusion, nasal mupirocin) since the

beginning of the study. The same laboratory data as those recorded atinclusion were also checked. The date of patient death, renal trans-

plantation, or moving was also recorded.All of the data were entered into a computerized database (Dbase

IV#{174})by two independent operators. The two database files were then

automatically compared, and all discrepancies were checked andcorrected.

Bacteriologic Techniques

The swab samples taken from the patients’ nostrils were inoculated

onto Columbia and Chapman agar plates and incubated for 48 h.

Plates were then evaluated for the presence of Staphylococcus aureus.

The plates inoculated with the swabs taken from the patients’ vascular

access sites were evaluated for the presence of any bacteria. Blood

cultures were performed using each center’s standard techniques.

Patients

A total of 988 adult patients (597 men and 391 women; mean age,

60 yr; range, 16 to 88 yr) was enrolled. Most patients (is = 839. 85%)

attended hospital dialysis centers, and 149 (15%) were treated inself-care units. The causes of renal failure were chronic gbomerubo-

nephritis (n = 223), angioselerosis (a = 176), chronic interstitial

nephritis (n = 149), polycystie renal disease (n = 105), diabetes

mellitus (a 88), connective tissue disease and systemic vaseulitides

(a = 42), myeboma and amyboidosis (ii = 23), other causes (ii 77),

and unknown (n = 105). A comorbid condition known to increase the

risk of infection was found in 191 patients and included diabetes

melbitus (a = 120), current (under treatment and/or not cured) malig-

nancy (n = 72), and HIV infection (a = 1), with two patients having

both a malignancy and diabetes. One hundred fifty-four patients had

undergone renal transplantation at least once, and 99 patients had hadat least one bacteremie episode within the past 18 mo. The mean time

that elapsed between the start of dialysis and the time of inclusion was5.5 yr (range, I mo to 30 yr). Most patients had two or three dialysis

sessions per week. The mean dialysis duration was 12 h per week. A

synthetic dialysis membrane was used in 566 (57.3%) patients. Only

18 of the 988 patients (1 .8%) had their dialysis membrane changed

during the study period and only two of them were switched from a

cellubosic to a synthetic membrane; the remaining 16 patients had

changes within the same category (cellubosie or synthetic). No other

changes (number of sessions per week, dialysis duration) took place

during the study period. Dialyzers were not reused in any center.Vascular accesses were distributed as follows: arteriovenous fistula

(Brescia type. a = 806), arteriovenous prosthetic device (ii = 124,

including 103 PTFE grafts, 13 autologous or homologous venous

grafts and eight external Thomas’ shunts), and dialysis catheter (ii =

58, including 47 long-term implanted, dual-lumen central venous

silicone catheters, and 1 1 single-lumen polyethylene catheters). Forty-

five patients were treated with immunosuppressive agents; 41 were

given corticosteroids alone (a = 33) or combined with other immu-

nosuppressive drugs (a = 8), three were given cyclosporine or aza-

thioprine alone, and one was treated by anticancer chemotherapy. Fivehundred eight patients (5 1 .4%) were given erythropoietin. Five hun-

dred and thirty patients (53.6%) had been given either oral or intra-

venous iron supplementation. and 57 patients (5.7%) had been givennasal mupirocin ointments within the past 6 mo. Baseline nasal swab

culture was positive for Staphylococcus aureus in 282 of the 952

patients (29.6%) in whom it had been performed. A surgical proce-

dure had been performed within the month before inclusion in 45

patients; 16 of these procedures involved the vascular access.

Statistical Analyses

Calculation of the Incidence of Bacteremia. The incidence of

bacteremia (1) was calculated as shown below and expressed in

number of episodes per 100 patient-months.

I Total number of bacteremie episodes X 100- Sum of all individual follow-up times

The individual follow-up time was calculated as the difference

between the census date and the start-of-study date.

Risk Factor Analysis. By the end of the study period, 50 pa-

tients (group I ) had had at least one bacteremic episode, whereas 935

Bacteremia in Chronic Hemodialysis Patients 871

a num, numerical; eat, categorical; Rx, prescription; S. aureus, Staphylococcus aureus.

patients (group 2) had remained free of bacteremia. The remaining

three patients showed symptoms of bacteremia and had been treated

accordingly, although blood cultures remained negative (one patient.

who had staphybococeal arthritis) or had not been performed (two

patients). These three patients were subsequently excluded from risk

factor analysis.

The two groups were compared using Pearson �, Mann-Whitney,

and Fisher tests as appropriate. Risk factors associated with the

development of at beast one bacteremic episode over 6 mo were

determined using the Cox proportional hazards model. Outcome was

defined as the time to the first episode of bacteremia that occurred

during the study period. All of the parameters that were considered

potential risk factors were first analyzed by univariate Cox analysis.

These variables and their format are displayed in Table 1. Those

variables that reached the 20% significance level by univariate anal-

ysis were then considered for multivariate Cox analysis. This was

performed using enter and remove limits of 0.05 and 0. 15, respee-

tively, and the asymptotic covarianee estimate method.

ResultsDescription of Bacteremic Episodes

Fifty-one bacteremic episodes occurred in 50 patients. Pa-

tients were equally distributed with respect to the location of

dialysis centers ( 13 of 237 in university hospitals and 37 of 748

in nonuniversity hospitals; P = 0.74). The mean number of

positive blood cultures per episode was 2.0 (range, I to 5).

The 5 1 causative bacteria were Staphylococcus aureus (n =

20), coagulase-negative staphylococci (n I 5), Escherichia

coli (ii = 5), other aerobic Gram-negative bacilli (n = 8),

Streptococcus oralis (,i 1 ), Listeria rnonocvtogenes (n 1),

and Bacteroidesfragilis (n = 1). One patient had two episodes

of bacteremia, more than 3 mo apart, both due to Staph ylococ-

cus aureus. In the 20 episodes of Staphylococcus aureus bac-

teremia, Staphylococcus aureus was cultured from I 0 of the 18

vascular access swabs taken and from six of the I 2 nasal swabs

taken. Staphylococcus aureus was never cultured from vascu-

bar access in patients with bacteremia due to pathogens other

than Staphylococcus aureus. Four of the 19 patients with

Staphylococcus aureus bacteremia had received nasal mupiro-

cm since the beginning of the study, whereas 16 had not.

A relapse within 2 wk after discontinuation of antibiotics

occurred in four patients. in whom the causative bacteria were

Staphylococcus aureus (ii 2), Staphylococcus epidermidis (n

= 1 ), and Stenotrophornonas maltophilia (n 1 ). Three of

these patients had a long-term dialysis catheter.

Three bacteremic episodes were complicated with secondary

septic foci, including one case of septic arthritis due to Esch-

erichia coli, one case of endocarditis due to Staphylococcus

epiderinidis, and one case of multiple septic pulmonary infarcts

due to Staphylococcus aureus in a patient with long-term

dialysis catheter.

Six bacteremic episodes resulted in death, which was con-

sidered to be directly related to bacteremia. The pathogens

responsible for these six episodes were Staphylococcus aureus

(n 2), Pseudomonas sp. (ii 2), Escherichia coli (n 1),

and Listeria monocytogenes (ti = 1).

Table 1. Variables used for risk factor analysi?

Code Name Format Unit or Category

AGE Age num Years

SEX Sex cat Male versus female

DIAB Diabetes mellitus cat Present versus absent

MALI Malignancy eat Present versus absent

DDIA Duration of dialysis num Years

MDIA Mode of dialysis cat Self-care versus hospital center

MEMB Dialysis membrane cat Synthetic versus cellubosic

XTRA History of transplantation cat Yes versus no

TRAN History of blood transfusion num Red cell packs in past 9 mo

HIBE History of bacteremic episode cat One, two, or more episodes versus no episode

IMMS Current immunosuppressive Rx cat Yes versus no

ERYP Current erythropoietin Rx num Current weekly dosage

IRON Iron Rx within past 6 mo cat Yes versus no

MUPI Nasal mupirocin within past 6 mo cat Yes versus no

NASA Nasal swab culture positive for S. aureus cat Yes versus no

SURG Surgical procedure within past mo cat Yes versus no

ACCS Vascular access cat Catheter, A/V graft versus A/V fistula

PROT Serum protein level num g/L

ALBU Serum albumin level num g/L

CRP C-reactive protein level num mg/L

FERR Ferritin num ng/L

HB Corpuscular hemoglobin num g/db

KTV Kt/V num -

872 Journal of the American Society of Nephrobogy

., P value of g2. or Mann-Whitney U test, as appropriate. Abbreviations as in Table I.

Incidence of Bactereinic Episodes

The follow-up was terminated before 6 mo in 1 10 patients

because ofdeath (ii = 61), transplantation (ii = 35). or moving

(ii = 14). The mean follow-up time in the 988 patients was

therefore 5.6 mo. The incidence of baeteremic episodes was

calculated to be 0.93 episode per 100 patient-months.

Risk Factor Analysis

Univariate Analyses. The main characteristics of the pa-

tients with and without bacteremie episodes are shown in Table

2. Significantly more patients with a bacteremic episode had a

comorbidity (diabetes or malignancy), a history of blood trans-

fusion or of bacteremia, previous nasal mupirocin ointments,

and ongoing immunosuppressive therapy than patients without

bacteremia. Approximately 9 times more of them had a dialysis

catheter. The two groups did not differ regarding dialysis

membrane at the time of inclusion (27 patients [54%] in group

1 versus 539 patients [58%] in group 2 had a synthetic dialysis

membrane; P = 0.61 ). Table 3 shows the laboratory data for

both groups of patients. Serum albumin and corpuscular he-

moglobin were significantly lower, and C-reactive protein was

significantly higher in patients who had a bacteremic episode

than in those who did not.

Each of the variables in Table I was then tested in univariate

Cox analysis. There was no discrepancy in terms of statistical

significance between the results of Cox analysis and those of

two-group comparison. Thirteen variables reached the 20%

level of statistical significance: age (RR = 1.02, P = 0.09),

diabetes (RR = 1.63, P = 0.18), malignancy (RR = 3.0, P =

0.002), mode of dialysis (RR = 0.34, P = 0.06), number of red

cell packs transfused since January 1 , I 994 (RR = 1 . 1 1 , P <

0.0001). history of bacteremic episode (�2 episodes versus no

episode: RR = 7.35. P < 0.0001), current immunosuppressive

therapy (RR = 3.01, P = 0.008), current weekly erythropoietin

dose (RR = I . 1 1, P = 0.0006), previous nasal mupirocin

(RR = 3.24, P = 0.0013), vascular access (catheter versus

fistula: RR = I 1 .9, P < 0.0001 ; graft and shunt versus fistula:

RR = 1.84, P = 0.14), serum albumin (RR = 0.92, P =

0.006), C-reactive protein (RR = 1 .0 1 , P = 0.0 1 2), and cor-

puseubar hemoglobin (RR = 0.67, P = 0.0002).

Multivariate Cox Analyses. All of these 13 variables

were considered for multivariate analysis. At the end of the

stepwise multivariate process that considered only 865 cases

(821 cases without bacteremia and 44 cases with bacteremia,

I 20 cases being excluded because of missing data), four van-

ables were retained as significant risk factors for bacteremic

episode. These were, in order of entrance into the model,

vascular access, history of bacteremia, corpuscular hemoglo-

Table 2. Characteristics of patients with (a = 50) and without (n 935) bacteremic episode (BE)

Characteristic Patients with BE Patients without BE P Value”

Age, years (mean ± SD) 63.4 ± 15.7 59.8 ± 15.5 0.08

Gender. male: ii (%) 26 (52.0) 569 (60.9) 0.21

Comorbidity

diabetes mellitus: n (%) 9 (18.0) 1 1 1 (1 1.9) 0.20

malignancy: n (%) 9 (18.0) 63 (6.7) 0.003

Time since start of dialysis. years (median and [range]) 2.43 [0.2 to 24j 3.62 [0.1 to 291 0.18

Mode of dialysis

hospital center: ii (%) 47 (94.0) 789 (84.4) 0.064

History of transplantation: ii (%) 5 (10.0) 148 (15.8) 0.267

Number of red cell packs transfused within the past 9 mo: 2.95 ± 4.6 1 .05 ± 2.7 <0.0001

(mean ± SD)

History of BE

�l within the past 18 mo: ii (%) 17 (34.0) 82 (8.8) <0.0001

no. of previous BE (mean ± SD) 0.80 ± 1.63 0.14 ± 0.57 <0.0001

Current immunosuppressive therapy: ii (%) 6 (12.0) 39 (4.2) 0.009

Current erythropoietin treatment

no. of patients treated: ii (%) 30 (60.0) 477 (5 1 .0) 0.22

current dose, kU/wk: (mean ± SD) 4.64 ± 4.99 2.85 ± 3.57 0.02

Iron Rx within past 6 mo: ii (%) 27 (54.0) 502 (53.7) 0.96

Nasal mupirocin within past 6 mo: a (%) 8 (16) 49 (5) 0.004

Nasal swab positive for S. aureus: ii (%) 15 (31.3) 265 (29.4) 0.79Surgical procedure within I mo: ii (%) 2 (4.0) 43 (4.6) 0.84Vascular access

A/V fistula 25 (50.0) 778 (83.2)

A/V graft and shunt 7 (14.0) 1 17 (12.5) <0.0001

dialysis catheter 18 (36.0) 40 (4.3)

a Brackets around the P value indicate the variable selected at each step. Variable codes are defined in Table I.

Bacteremia in Chronic Hemodialysis Patients 873

Table 3. Laboratory data for patients with (n = 50) and without (ii = 935) BE at inclusion

Variable Patients with BE Patients without BE P Value

Serum protein (gIL) 67.5 ± 7.0 68.7 ± 6.3 0.28

Serum albumin (g/L) 37.8 ± 5.5 39.9 ± 5.1 0.03

C-reactive protein (mgIL) 23.2 ± 29.3 14.6 ± 24.8 0.006

Serum ferritin (ngIL) 346 ± 502 353 ± 434 0.44

Corpuscular hemoglobin (g/dl) 9.0 ± 1.3 9.8 ± 1.5 0.0001

Predialysis urea nitrogen (mmollL) 32.9 ± 15.1 34.0 ± 15.9 0.56

Postdialysis urea nitrogen (mmol/L) 1 1 .0 ± 6.5 1 1 . 1 ± 6.8 0.71

Urea reduction rate 0.67 ± 0. 1 2 0.68 ± 0. 1 0 0.73

Kt/V 1.40 ± 0.52 1.39 ± 0.38 0.73

a p value of �2 or Mann-Whitney U test, as appropriate.

bin, and current immunosuppressive therapy. Table 4 summa- study (I ) using similar methods gave a slightly lower mci-

rizes the multivariate stepwise process and shows the P values dence, but the recording of bacteremic episodes might not have

for each variable at each of the five steps. The relative risks been as exhaustive as in the present study.

associated with these four risk factors are displayed in Table 5. Again, staphylococci were found to be the major causative

Each of these four variables was graphically evaluated for pathogens and the vascular access the main portal of entry for

adherence to the assumption of proportional hazards. The infection. Staphylococci accounted for more than two-thirds of

survival curves exhibited constant differences between the causative bacteria, but coagulase-negative staphylococci were

strata of each variable, suggesting that the assumption of almost as often responsible for bacteremia as were Staphvlo-

proportional hazards was not violated. Multivariate Cox anal- coccus aureus.

ysis was repeated, using only these four variables for 982 cases Previously, we identified prior history of bacterial infection,

and with only three cases excluded for missing data. The same the use of a dialysis catheter. and iron overload as three

four variables were entered into the Cox model in the same significant and independent risk factors for bacterial (not only

order as before, without significant changes in the relative risk bacteremic) infections in chronic hemodialysis patients, re-

values. gardless of any patient-related factor, such as age or time

elapsed since the start of dialysis (4).

Discussion The current study again shows that the use of a dialysis

The incidence of0.93 episode ofbacteremia per 100 patient- catheter is the greatest risk factor for bacteremia in hemodial-

months is in the range of that reported by others. A previous ysis patients. However, our data confirm that policies for

Table 4. Summary of the multivariate stepwise analysis showing the P values for each variable at each step�’

p Values before StepVariable

I 2 3 4 5

ACCS, graft versus fistula 0.935 0.180 0.437 0.414 0.576

ACCS, catheter versus fistula [<0.0001] <0.0001 <0.0001 <0.0001 <0.0001

AGE 0.048 0.389 0. 155 0.078 0.073

HIBE. I versus no episode 0.250 0.997 0.482 0.745 0.707

HIBE. �2 versus no episode <0.0001 [<0.00011 <0.0001 <0.00()b <0.0001

IMMS 0.001 0.036 [0.004] 0.006 0.015

HGLB 0.0008 0.006 0.005 [0.009] 0.009

DIAB 0.207 0.315 0.270 0.433 0.284

MALI 0.004 0.018 0.1 10 0.214 0.456

MDIA 0.036 0.201 0.266 0.268 0.253

MUPI 0.001 0.526 0.787 0.561 0.708

ERYP 0.0005 0.006 0.040 0.097 0.273

IRAN 0.0001 0.090 0.032 0.366 0.208

ALBU 0.021 0.324 0.326 0.404 0.707

CRP 0.013 0.219 0.753 0.931 0.447

874 Journal of the American Society of Nephrology

Table 5. Risk factors for baeteremia as identified by Cox

multivariate analysis of data from 865 chronic

hemodialysis patients, 44 with and 821 without

bacteremic episode”

.Risk Factor

Relative Risk (95%.

confidence interval)

Vascular access

fistula I

graft and shunt 1 .29 (0.50 to 3.34)

catheter 7.64 (3.73 to 15.67)

History of bacteremia

no previous episode 1

I previous episode 1.19 (0.44 to 3.23)

�2 previous episodes 7.33 (3.27 to 16.43)

Current immunosuppressive treatment

no 1

yes 3.01 (1.20 to 7.56)

Corpuscular hemoglobin (per I g/dl 0.75 (0.60 to 0.93)

increment)

a Risk factors are ranked according to their order of entrance into

the model during the stepwise process.

dialysis catheters have changed in recent years, with increased

use of long-term, percutaneous implanted dual-lumen central

venous silicone catheters. In this study, 47 of the 58 patients

receiving dialysis via a central venous catheter had such cath-

eters, either Quinton Permcath#{174} or Twin-cath#{174} (1 3). And 16 of

the 50 (32%) bacteremic patients had a long-term implanted

catheter, compared with only 3 1 of the 935 (3.3%) nonbacte-

remic patients, a strikingly significant difference (P < l0�),

whereas there was no difference for short-term catheters (2 of

50 versus 9 of 935; P 0. 10). Additionally, three of the four

relapsing bacteremie episodes occurred in patients who had

long-term implanted catheters. These findings show that al-

though the incidence of short-term catheter-related bacteremias

seems to have decreased (perhaps because of successful infec-

tion control measures), long-term implanted catheters have

emerged as a leading risk factor of bacteremia in chronic

hemodialysis patients. Nephrologists should be aware of this

and should seriously consider removing the catheter in cases of

bacteremia ( I 4), especially if a relapse occurs.

The second most important risk factor is a history of bacte-

remia. The occurrence of bacteremia during the study period

and a history of bacteremia were strongly linked, whatever the

variable format considered (at least two versus no previous

bacteremic episode; or the total number of episodes over the

past 18 mo). In a previous study (4), we suggested that this

tendency toward recurrence of bacterial infections in chronic

hemodialysis patients, in combination with the predominant

responsibility of Staphylococcus aureus, might be due to

chronic staphylococcal nasal carriage, as evidenced by others

(3,15). There is also convincing evidence that eradication of

staphylococcal nasal carriage, especially when using mupiro-

cm nasal ointments, may help to reduce the incidence of

staphylococcal bacteremia in hemodialysis patients (16). In the

present study, which was designed to assess whether staphy-

lococcal nasal carriage is a risk factor for bacteremia, not only

was staphylococcal nasal carriage not an independent risk

factor for bacteremia, but it was far from statistically signifi-

cant in univariate analysis (see Table 2). Even the proportion of

patients with Staphylococcus aureus bacteremia who had

Staphylococcus aureus nasal carriage (38.9%) was not signif-

icantly higher than in nonbacteremic patients (29.4%, P =

0.38). Nasal swab cultures, which were done again at the

diagnosis of bacteremia, were not more often positive in pa-

tients with Staphylococcus aureus bacteremia (6 of 19) than in

patients with bacteremia due to other pathogens (4 of 3 1 , P =

0. 1 1). These negative results might be explained partly by the

low rate of staphylococcal nasal carriage in our study popula-

tion. Besides, we performed only one nasal culture, at the time

of inclusion, whereas recent studies suggest that continuous

surveillance for Staphylococcus aureus nasal carriage is more

effective than cross-sectional screenings for identifying pa-

tients at risk of developing bacteremias (15, 17). However,

another study, designed to evaluate the influence of Staphylo-

coccus aureus nasal carriage on the incidence of bacteremia,

showed that diabetes and the use of a central venous catheter

were significant risk factors for bacteremia, whereas staphylo-

coccal nasal carriage was not ( 1 8). Thus, our results, plus the

fact that resistance to mupirocin may be favored by its use

( 16, 19), argue against the widespread use of nasal mupirocin in

hemodialysis units. We recommend that mupirocin be reserved

for patients with repeatedly positive Staphylococcus aureus

nasal cultures and other well demonstrated risk factors for

bacteremia.

The intricate influences of iron overload, erythropoietin

therapy, and anemia are more complex to analyze. Before the

advent of erythropoietin therapy, we confirmed that iron over-

load increases the risk of bacterial infection in dialysis patients

( I ,4), as had been established previously (7,8). Many of the

patients in our former study were severely iron overloaded, as

indicated by serum ferritin levels > 1000 p�g/L in more than

10% of the patients. More than half of the patients in the

present study were on erythropoietin therapy at the time of

enrollment, and their mean serum ferritin levels were far lower,

with only 5% of the patients having ferritin > 1000 p�g/L.

Consequently, iron overload was no longer a risk factor for

bacteremia in this study. On the other hand, anemia appeared

to be significantly associated with bacteremia. Boelaert et a!.

showed that recombinant erythropoietin helps to reverse the

polymorphonuclear granulocyte dysfunction in iron over-

loaded dialysis patients (20). They attributed this effect to the

decrease in iron overload, which was confirmed in another

study that showed a correlation between hematocrit and gran-

ulocyte phagocytic function, but no correlation with serum

ferritin, suggesting that erythropoietin does not improve gran-

ubocyte-deficient phagocytosis simply by reducing iron load

(2 1 ). The bacteremic patients in our study were given eryth-

ropoietin more often, although not significantly so, and at

higher doses than nonbacteremic patients. Nevertheless, ane-

mia was significantly more profound in bacteremic patients.

The same was true when only the 507 patients given erythro-

Bacteremia in Chronic Heniodialysis Patients 875

poietin at the start of the study were considered. This suggests

that anemia was linked to resistance to erythropoietin. How-

ever, we failed to identify a known cause of resistance to

erythropoietin that could lead us to interpret anemia as a

confounding factor. Thus, inadequacy of dialysis was not a

factor, because the mean urea reduction rates were more than

65%, and the mean Kt/V indexes were greater than 1 .2 in all

patients, without any significant difference between the bacte-

remic and nonbacteremic patients (Table 3). Iron deficiency

can be firmly ruled out (see Tables 2 and 3). Aluminum

overload is very unlikely because the patients in this study had

not been given aluminum hydroxide for years. However, no

specific tests were carried out to firmly exclude aluminum

intoxication. Also not systematically performed were serum

parathyroid hormone assays to rule out the possible role of

hyperparathyroidism. Chronic low-grade inflammatory disease

could be involved, because the initial C-reactive protein levels

of patients who subsequently developed a bacteremia were

significantly higher than in nonbacteremic patients. However,

the C-reactive protein levels were rather low, even in the

bacteremic patients (see Table 3). Additionally. when consid-

ering only the 99 subjects who had developed at least one

bacteremic episode within the past 1 8 mo, mean C-reactive

protein levels were not significantly higher in the 17 bactere-

mic patients than in the 82 nonbacteremic patients (22.6 versus

2 1 .8 mgIL; P = 0.9). To our knowledge, this is the first study

that suggests that anemia could be a risk factor for bacteremia.

Consistent with these findings are those of Keane and Collins,

who noted that patients with a hematocrit <29% had a higher

mortality rate than those with a hematocnit �30% (22). Inter-

estingly, the higher mortality rate was linked to an increase in

infectious causes of death.

Our results do not demonstrate that cellulosic membranes

increase susceptibility to bacteremic infections in hemodiabysis

patients, a fact that had been suggested by several experiments

(23,24). However, there is little clinical evidence to confirm

these experimental data. Two retrospective studies reach op-

posite conclusions (10,1 1). A prospective comparative evalu-

ation of two dialysis membranes over 20 wk found that three of

eight patients treated with Cuprophane developed septicemia,

whereas none of seven patients treated with polysulfone did

(23). However, the small number of patients studied makes

these results difficult to interpret. Our findings are likely to be

reliable because they are based on the prospective longitudinal

study of a large number of patients who were treated with the

same membrane throughout the study period. However, they

do not permit us to generate hypotheses on the effect of

membrane biocompatibility and porosity on the risk of deveb-

oping bacteremia. In fact, in our study, unsubstituted cellulosic

membranes with high complement activation and low ultrafil-

tration coefficients were grouped with modified cellubosie

membranes that have a wide range of complement activation

and water flux rates. Likewise, synthetic membranes included

both low-flux and high-flux membranes. However. although

our grouping criteria were not primarily based on biocompat-

ibility, it is generally admitted that synthetic membranes as a

whole are more biocompatible than cellulosic membranes. In

summary, we can say from our results that if dialysis mem-

branes play a role in the development of infection, it is only

marginal and not significant compared with the other major

risk factors we disclosed. In this respect. we totally agree with

Bonomini et a!. , who showed that the basic structure of dialysis

membranes has no impact on long-term general morbidity ( 1 1).

and with Churchill, who pointed out that trials examining the

influence of dialysis membrane on the incidence of bacterial

infection must also take into account other important determi-

nants, such as the type of vascular access and serum albumin

level (25).

Although mubtivariate Cox models are only models that

should never be regarded as definitive. our study reliably

identified four significant predictors of bactereniia in chronic

hemodialysis patients. We found that vascular accesses still are

the major source of bacteremia in chronic hemodialysis pa-

tients and that long-term implanted catheters represent the

leading risk factor for bacteremia in these patients. Although

staphylococci were predominantly responsible for the bactere-

mic episodes, staphylococcal nasal carriage did not appear to

increase the risk of their occurrence. Finally, the use of syn-

thetic membranes did not appear to be a protective factor in this

population of well dialyzed patients.

AcknowledgmentsThis study was funded by a Clinical Research Program grant from

the University of Nancy Medical Center and by grants from Smith-

Kline Beecham (Nanterre, France), Boehringer Mannheim (Rueil-

Malmaison, France). and Hospal (Lyon. France). We are grateful to

Prof. Serge Brian#{231}on. who acted as scientific adviser, and to Prof.

Jean-Fran#{231}ois Vieb for his helpful contribution to the manuscript

reviewing process.

AppendixParticipating Clinical C’enters

CHU de Besan#{231}on (M. Jamali, J.M. Chabopin); H#{244}pitalde

Chalon sur Sa#{244}ne(J.F. Cabannes, P. Dubot); H#{244}pitalde Char-

leville M#{233}zi#{232}res(J.J. Dion); Clinique de Chaumont (S. Corbin,

C. Corbin); CHU de Dijon (E. Robin. G. Rifle); H#{244}pitalde

Dole (J. Guillaumie); H#{244}pitalde Macon (G. Janin); H#{244}pitalde

Metz (H. Terrasse, P. Mirgaine); H#{244}pitalde Montbdliard (P.

Hanhard, C. Bernard); CHU de Nancy (T. Cao Huu, M.

Kessler); Association Lorraine de Traitement des Insuffisants

R#{233}naux, Nancy (J. Chanliau, J. Gamberoni. P.Y. Durand);

Polyclinique de Gentilly, Nancy (J.C. Valdenaire, J.M. Ber-

theau); Polyclinique d’Essey-les-Nancy (J.M. Bertheau. A.M.

Bertheau); CHU de Reims (E. Canivet, J. Chanard); Clinique

Bethesda, Strasbourg (J.F. Marichal); CHI Toubon (C. Wolff);

H#{244}pitalde Troyes (R. Montagnac. F. Schillinger, T. Milcent);

H#{244}pitalde Verdun (B. Gilson, P. Bindi); H#{244}pitalde Vittel (E.

Prenat).

Participating Laboratories of Batteriologv

CHU de Besan#{231}on (P. Pl#{233}siat);H#{244}pitalde Chalon sur Sa#{244}ne

(C. Sire-Bidault); H#{244}pitalde Charleville M#{233}zi#{232}res(J.C. Rev-

eil); Clinique de Chaumont (D. Gaupillat); CHU de Dijon (C.

Neuwirth); H#{244}pital de Dole (C. Gauthier); H#{244}pital de Macon

876 Journal of the American Society of Nephrology

(A. Bayle); H#{244}pitalde Metz (Y. Rio); H#{244}pitalde Montb#{233}liard

(C. Febvre); CHU de Nancy (M. Weber); Laboratoire Aubert,

Nancy; Laboratoire Bnignon, Nancy; CHU de Reims (C. Chip-

paux); Clinique Bethesda, Strasbourg (J.L. Kaufmann); CHI

Toulon (E. Delbeke); H#{244}pitalde Troyes (J.C. Croix, C. Eloy);

H#{244}pital de Verdun (S. Boussard); H#{244}pital de Vittel (J.J.

Gaultier, J. Gaultier).

ReferencesI . Kessler M, Hoen B, Mayeux D, Hestin D. Fontenailbe C: Bacter-

aemia in patients on chronic haemodialysis. Nephron 64: 95-100. 1993

2. Goldblum SE, Reed WP, Which JA, Goldmann RS: Staphvlo-

coccus carriage and infections in haemodiabysis patients. Dial

Transplant 7: 1 140-1 148, 1978

3. Yu VL, Goetz A, Wagener M, Smith PB, Rihs JD, Hanchett J,

Zuravleff ii : Staphylococcus aureits nasal carriage and infections

in patients on hemodialysis. N Engi J Med 315: 91-96, 1986

4. Hoen B, Kessler M, Hestin D, Mayeux D: Risk factors for

bacterial infections in chronic haemodialysis adult patients: A

multieentre prospective survey. Nephrol Dial Transplant I 0:

377-381, 1995

5. Cheesbrough YS, Finch RG, Burden RP: A prospective study ofthe mechanisms of infection associated with hemodialysis cath-eters. J Infect Dis 154: 579-589. 1986

6. Almirall J, Gonzales J, Reblo J, Campistol JM, Montoliu J, Dela

Bellasea JP, Revert L, Gatell JM: Infection of hemodialysis

catheters: Incidence and mechanisms. Am J Nephrol 9: 454-459,

I 989

7. Boelaert JR. Daneels RF, Sehurgers ML, Matthys EG. GordtsBZ, Van Landuyt HW: Iron overload in haemodialysis patientsincreases the risk of baeteraemia: A prospective study. Nephrol

Dial Transplant 5: 130- 134, 1990

8. Tiebemans CL, Lenelud CM, Wens R, Collart FE, Dratwa M:

Critical role of iron overload in the increased susceptibility of

haemodialysis patients to bacterial infections: Beneficial effects

of desferrioxamine. Nejthrol Dial Transplant 4: 883-887, 1989

9. Churchill DN, Taylor DW, Cook Ri. LaPlante P. Barre P. CartierP, Fay WP, Goldstein MB, Jindal K, Mandin H, McKenzie JK,

Muithead N, Parfrey PS. Posen GA, Slaughter D. Ulan RA. WerbR: Canadian haemodialysis morbidity study. Am J Kidney Di.s

19: 214-234, 1992

10. Hornberger JC. Chernew M, Petersen J, Garber AM: A multi-variate analysis of mortality and hospital admissions with high

flux dialysis. J Am Soc Nephrol 3: 1227-1237, 1993

I I . Bonomini V. Cobi L, Scolari MP, Stefoni 5: Structure of dialysis

membranes and long-term clinical outcome. Am J Nephrol 15:

455-462. 1995

12. Daugirdas iT: Bedside formulas for KTIV: A kinder. gentler

approach to urea kinetic modeling. ASAIO Trans 35: 336-338,

1989

13. Canaud B, Bdraud ii, Joyeux H, Mion C: Internal jugular vein

eannulation with two silicone rubber catheters. A new and safe

temporary vascular access for haemodialysis: 30 months experi-

enee. Artif Organs 10: 397-403, 1986

14. Shaffer D: Catheter-related sepsis complicating bong-term, tun-

neled central venous dialysis catheters: Management by guide-

wire exchange. Am J Kidney Dis 25: 593-596, 1995

15. Gbowacki LS. Hodsman AB, Hammerberg 0, Meraw J, MeNeill

V. Card ML, Potters H, MeGhie K, Stitt LW: Surveillance and

prophylactic intervention of Staphylococcus aureus nasal cob-

nization in a hemodialysis unit. Am J Nephrol 14: 9-13, 1994

16. Boelaert JR. Van Landuyt HW, Godard CA, Daneels RF,

Sehurgers ML, Matthys EG, Dc Baere YA, Gheyle DW, Gordts

BZ. Herwaldt LA: Nasal mupirocin ointment decreases the mci-

denee of Staphylococcus aureus baeteraemias in haemodialysis

patients. Nephrol Dial Traiisplant 8: 235-239, 1993

17. Montagnac R, Eloy C, Sehillinger F, Croix JC, Mileent T: Etude

iterative de Ia prevalence du portage nasal de Staphylococcus

aureus ehez l’h#{233}modialys#{233}.Presse Med 24: 1075-1077. 1995

18. Roubicek C, Brunet P. Mallet MN, Dussol B, Gonzales A,

Andrieu D. Merzouk T, Jaber K. Berband Y: Portage nasal de

Staphylococcus aureus: Prevalence dans un centre

d’h#{233}modialyse et influence sur les baet#{233}ri#{233}mies.N#{233}phrologie 16:

229-232, 1995

19. Layton MC, Patterson JE: Mupirocin resistance among consec-

utive isolates of oxacillin-resistant and borderline oxacibline-

resistant Staphylococcus aureus at a university hospital. Antiini-

crob Agents Chetnother 38: 1664-1 667, 1994

20. Boebaert JR. Cantinieaux BF, Hariga CF. Fondu PG: Recombi-

nant erythropoietin reverses polymorphonuclear granulocyte

dysfunction in iron-overloaded dialysis patients. Nephrol Dial

Transplant 5: 504-507, 1990

2 1 . Veys N. Vanhobder R, Ringoire 5: Correction of deficient phago-

cytosis during erythropoietin treatment in maintenance haemo-

dialysis patients. Am J Kidney Dis 19: 358-363, 1992

22. Keane WF, Collins AJ: Influence of co-morbidity on mortality

and morbidity in patients treated with haemodialysis. Am J

Kidney Dis 24: 1010-1018, 1994

23. Vanholder R, Ringoir 5, Dhondt A, Hakim R: Phagocytosis in

uremie and haemodialysis patients: A prospective and cross

sectional study. Kidney Int 39: 320-327, 1991

24. Himmelfarb J, Zaoui P. Hakim R, Holbroock D: Modulation of

granubocyte LAM-I and MAC-I during dialysis: A prospective

randomized controlled trial. Kidney bat 41: 388-397, 1992

25. Churchill DN: Clinical impact of biocompatible dialysis mem-

branes on patient morbidity and mortalily: An appraisal of the

evidence. Nephrol Dial Transplant lO[Suppl 10]: 52-56, 1995