Mississippi Mud in the 1990sRisks and Outcomes of Vancomycin-Associated Toxicity in GeneralOncology Practice
Linda S. Elting, Dr.P.H.1
Edward B. Rubenstein, M.D.1
Danna Kurtin, M.P.H.1
Kenneth V. I. Rolston, M.D.1
John Fangtang, Pharm.D., Dr.P.H.1
Charles G. Martin, Ph.D.1
Issam I. Raad, M.D.1
Estella E. Whimbey, M.D.1
Ellen Manzullo, M.D.1
Gerald P. Bodey, M.D.1
1 Department of Internal Medicine Specialties, TheUniversity of Texas M. D. Anderson Cancer Center,Houston, Texas.
Presented in part at the Annual Meeting of TheAmerican Society of Clinical Oncology, Los Ange-les, California, May 20–23, 1995.
Supported in part by a research grant from thePhysician’s Referral Service, M. D. Anderson Can-cer Center, Houston, Texas.
Address for reprints: Linda S. Elting, Dr.P.H., De-partment of Internal Medicine Specialties, The Uni-versity of Texas M. D. Anderson Cancer Center,1515 Holcombe Blvd., Box 40, Houston, TX 77030.
Received October 10, 1997; revisions receivedFebruary 27, 1998, and April 28, 1998; acceptedApril 28, 1998.
BACKGROUND. Discrepancies between the severity of toxicities reported in early
clinical trials and recent clinical experience with vancomycin have led to confusion
regarding the need for routine serum vancomycin level monitoring and discontin-
uation of vancomycin when toxicities occur. Therefore, the authors examined the
incidence, outcomes, and predictive factors of vancomycin-associated toxicities in
general oncology practice with the goal of developing clinically relevant prediction
rules and guidelines.
METHODS. All 742 consecutive cancer patients who received vancomycin at a
comprehensive cancer center during a 3-month period were followed prospec-
tively for the development and outcome of phlebitis, rash, ototoxicity, and neph-
rotoxicity. Logistic regression was used to derive a multiple variable model of the
risk of nephrotoxicity. A clinical prediction rule, the Nephrotoxicity Risk Score, was
developed from the risk model and validated prospectively.
RESULTS. Phlebitis occurred in 3% of patients (95% confidence interval [95% CI],
2– 4%), predominantly those with recently inserted central venous catheters.
Rashes occurred in 11% of patients (95% CI, 9 –13%); however, all but 4 patients
also were receiving b-lactam antibiotics. Clinical evidence of ototoxicity developed
in 6% of patients (95% CI, 4 –9%) who were receiving vancomycin plus other
ototoxic agents and only 3% of patients (95% CI, 2–5%) not receiving other ototoxic
agents (P 5 0.08). Nephrotoxicity occurred in 17% of patients (95% CI, 15–20%).
Logistic regression revealed that factors associated with an increased risk of neph-
rotoxicity included administration of other mild to moderate (P 5 0.01) or severely
nephrotoxic agents (P , 0.001) or an acute physiology and chronic health evalu-
S ince its discovery in 1956 and its first introduction to general usein 1958, controversy has surrounded the potential of vancomycin
to produce toxicities, particularly nephrotoxicity. In the mid- to late1950s, mortality rates associated with staphylococcal and streptococ-cal infections had risen to nearly preantibiotic era levels as theseorganisms acquired the ability to produce penicillinase. A new agentfor the treatment of penicillin-resistant staphylococcal infection
clearly was needed. After initial in vitro studies docu-mented vancomycin’s promise in these infections (sopromising that the name was derived from the word“vanquish”), it was evaluated rapidly in human trialsand made available for general use. Even as the firsttrials were being conducted, vancomycin, which wasnicknamed “Mississippi Mud” because of its appear-ance, was being purified.1 Early preparations con-tained pyrogens and impurities that produced notonly its brownish, muddy appearance, but also highfevers, hypotension, severe, dose-limiting phlebitis,and, perhaps, the nephrotoxicity that was observedanecdotally in its first recipients. However, by the timesystematic reports of larger samples of patients whohad received the purified preparations were pub-lished, methicillin, a reportedly less toxic drug, hadbeen introduced. Vancomycin was relegated to sec-ond-line use in infections serious enough to justify theuse of a toxic agent. This enhanced its reputation as anototoxic and nephrotoxic agent because of its frequentcombination with aminoglycosides in seriously ill pa-tients. This practice persisted until methicillin-resis-tant organisms began to emerge, at which time therewas a resurgence in vancomycin use. Since that time,the use of vancomycin has increased steadily, partic-ularly in cancer patients, fueled by the increase ininfections caused by methicillin-resistant Staphylo-cocci and by the shifting spectrum of infections fromgram-negative to gram-positive.2
Although vancomycin use has declined somewhatsince the emergence of vancomycin-resistant gram-positive organisms, it remains among the most com-monly used antibiotics in many oncology centers.Some clinicians consider it a moderate to highly toxicagent and carefully monitor serum vancomycin andcreatinine levels among recipients.2–5 However, an in-creasing body of literature suggests that modern prep-arations of vancomycin, when administered in com-bination with b-lactam antibiotics, are only minimallytoxic and do not justify the costs associated with thevigilant monitoring practices of the past.6 –11 Resolu-tion of this controversy is hindered by a lack of dataregarding vancomycin’s toxicity when administered incombination with antineoplastic agents. Therefore,we conducted a prospective, observational study ofthe incidence and outcomes of toxicities (rash, phle-bitis, ototoxicity, and nephrotoxicity) in cancer pa-tients receiving vancomycin in combination withcommonly used agents. By examining the incidence ofvancomycin-associated toxicities in everyday oncol-ogy practice rather than in the controlled, artificialsetting of a clinical trial, we hoped to produce data toinform the development of clinically relevant guide-lines.
PATIENTS AND METHODSBecause we focused on vancomycin use in standardoncology practice, an observational rather than anexperimental design was used. Patients were managedby their oncologists while receiving vancomycin; theinvestigators did not intervene when patients devel-oped either elevated serum vancomycin levels or tox-icity to obtain unbiased estimates of incidence in ev-eryday clinical practice.
All 765 patients who received vancomycin therapyor prophylaxis at The University of Texas M. D. Ander-son Cancer Center between September 15 and De-cember 15, 1992 were identified at the initiation ofvancomycin and followed prospectively for the dura-tion of therapy plus 14 days for development of tox-icity. Patients who received only one dose or whoreceived vancomycin orally were excluded. For the 32patients who received . 1 course of vancomycin dur-ing the study period, only the first course of therapywas included. After these exclusions, 742 patients re-mained for analysis.
Assessment of ToxicityThe patients’ hospital records were reviewed dailyduring therapy for evidence of rash, tinnitus, conver-sational hearing loss, vertigo, and phlebitis. Patientsalso were interviewed on Day 1 or 2 of therapy, twiceweekly thereafter, and at the end of therapy for evi-dence of the same toxicities. Ototoxicity was evaluatedclinically, as in everyday clinical practice, by question-ing each patient about the presence of dizziness, ver-tigo, ringing or roaring in the ears, or difficulty hearingconversations. Those who reported the onset or wors-ening of any of these symptoms since the initiation ofvancomycin were considered to have ototoxicity. Toestimate the frequency of clinically unrecognizedhearing loss, a small sample of patients without clin-ical evidence of ototoxicity were subjected to audiom-etry. Fifteen patients who received vancomycin with-out other ototoxic drugs and 15 who receivedconcomitant ototoxic drugs were selected randomlyfor audiometry on Day 1 or 2 and at the end of ther-apy. Audiometry was performed at the bedside underthe supervision of a certified technician using a por-table Belltone 9D audiometer (Belltone ElectronicsCorp., Chicago, IL). Hearing loss was defined as aunilateral or bilateral decrease in acuity of at least 30decibels at any frequency.
Estimates of creatinine clearance were calculatedusing a standard age, gender, and weight specific for-mula.12 The estimated creatinine clearance on Day 1of therapy was compared with the highest estimateduring or 14 days after therapy. In 16 patients who
2598 CANCER December 15, 1998 / Volume 83 / Number 12
received vancomycin for 1 or 2 days, the availableinformation did not permit examination of changesduring therapy; nephrotoxicity was evaluated in theremaining 726 patients. Patients whose estimated cre-atinine clearance decreased at least 20% from baselineto , 78 mL per minute (65% of the average normalvalue) but remained . 20 mL per minute during ther-apy were considered to have mild nephrotoxicity.Those whose estimated creatinine clearance fell .20% to , 20 mL per minute were considered to haveserious nephrotoxicity.
All skin rashes observed by study personnel orrecorded in the hospital records were considered. Pa-tients whose hospital records contained reports offlushing of the face, neck, or trunk with or withouthypotension and pruritis, but related temporally to theinfusion of vancomycin were considered to have ex-perienced “red person syndrome.”13 Phlebitis was de-fined as the development of erythema and pain pluseither induration or swelling along the vein duringvancomycin therapy. Neutropenia was not assessedbecause the majority of patients already were neutro-penic at the onset of vancomycin therapy. Superinfec-tion and colonization with vancomycin-resistantgram-positive organisms were not assessed.
Confounding FactorsConcurrent aminoglycosides, cisplatin, loop-inhibit-ing diuretics, aspirin, and nonsteroidal antiinflamma-tory agents were considered ototoxic agents. Nephro-toxic agents were classified as either moderate(carboplatin, cyclophosphamide, or nonsteroidal an-tiinflammatory agents in the last 30 days) or severe(concurrent amphotericin B, aminoglycosides, poly-mixin, pentamidine, cyclosporine, foscarnet, or cispla-tin in the last 30 days).
A modified acute physiology and chronic healthevaluation (APACHE) III score, including laboratoryvalues and vital signs, was used to account for con-founding by the patient’s general condition.14 Thisscore provides an objective measure of organ functionin a single composite score and has been shown topredict outcomes in hospitalized patients.15
Peak serum vancomycin levels were obtained 2hours after a dose of vancomycin. Those considered“trough” levels were obtained immediately prior to adose of vancomycin. Serum vancomycin levels thatwere obtained after three doses of vancomycin hadbeen administered and before evidence of nephrotox-icity developed were analyzed for their value as pre-dictors of toxicity.
Statistical Analysis“Red person syndrome,” tinnitus, conversationalhearing loss, vertigo, and phlebitis occurred rarely.Rash occurred more frequently, but was associatedwith only one predictive factor. Therefore, univariateanalyses of factors associated with these toxicitieswere performed. Differences between categoric vari-ables were tested for significance using two-tailed,Yate’s corrected, Pearson’s chi square tests. Those be-tween continuous variables were tested using theMann-Whitney U test.
Nephrotoxicity occurred more frequently thanother toxicities and, by univariate analysis, was asso-ciated with several risk factors. Therefore, logistic re-gression was used to construct a multiple variablemodel of vancomycin-associated nephrotoxicity. Forthis analysis, the cohort was divided into two samplesin random fashion. A 365-patient “derivation set” wasused to derive the model and develop a clinical pre-diction rule; the rule then was validated prospectivelyin the remaining 359-patient “validation set”.
In the regression model, the outcome variable wasdevelopment of nephrotoxicity; predictor variables in-cluded age, gender, neoplasm, bone marrow trans-plantation, shock, bacteremia, baseline estimated cre-atinine clearance, administration of other moderate orseverely nephrotoxic drugs, duration and total dailydose of vancomycin, and baseline APACHE score.Stepwise logistic regression, using the maximum like-lihood estimates, was used to derive the model.
A clinical prediction rule, the Nephrotoxicity RiskScore (NRS), was developed from the regressionmodel. Each regression coefficient was divided by thesmallest coefficient and rounded to the nearest inte-ger. Using this technique, a value of 1 was assigned toeach factor; these were summed for a total score basedon the presence of each factor at initiation of vanco-mycin. The possible values of the NRS range from aminimum of 0 (0 factors present) to 3 (all factorspresent). The Nephrotoxicity Risk Index was devel-oped by comparing the observed rates of nephrotox-icity in the derivation set for each of the NRSs. Patientswith NRSs of 0 or 1 were considered low risk (noroutine serum level monitoring) and those with NRSsof 2 or 3 were considered high risk (routine serumlevel monitoring after steady state). This process wasfor classification only; no levels were ordered.
The NRS was computed and each patient in thevalidation set was classified as either high or low riskas described earlier. Using this classification, the sen-sitivity, specificity, and positive and negative predic-tive values of the risk index were computed. Thesevalues were compared with the sensitivity (the per-
Vancomycin Toxicity in Cancer Patients/Elting et al. 2599
centage of patients actually monitored by their oncol-ogist among those who ultimately developed nephro-toxicity) and specificity (the percentage of patientswho were not monitored by their oncologist amongthose who did not develop nephrotoxicity) of “stan-dard practice. The performance of the model also wastested by comparing the receiver operating character-istic (ROC) curves for the derivation and validationsets. Statistical tests were computed using BMDP/Dy-namic software, Release 7.0 (BMDP Statistical Soft-ware, Inc., Los Angeles, CA).
RESULTSOf the 742 evaluable patients, approximately 50% weremales and the average age was 51 years (Table 1).Approximately 47% had hematologic malignanciesand, among the remaining patients with solid tumors,genitourinary (12%) and breast (10%) primary tumors
were most common. Ninety-seven patients (13%) hadundergone bone marrow transplantation. A majorityof patients (83%) had indwelling central venous cath-eters that had been inserted an average of 76 daysprior to vancomycin therapy. Approximately 50% ofthe patients were neutropenic, and 40% were pro-foundly so (, 100 neutrophils per mL3). The averageAPACHE score at the onset of vancomycin therapy was26 (range, 0 –79), and 18% of patients had an APACHEscore . 40.
Vancomycin was prescribed as therapy for sus-pected or proven infection in the majority of patients(84%). Among these, 29% had gram-positive bactere-mia. The average duration of therapy was 8 days(range, 1– 41 days) (Table 2). In 120 patients (16%),vancomycin was administered prophylactically, in-cluding 97 bone marrow transplantation recipients.The average duration of prophylaxis was 16 days(range, 1– 81 days).
A majority of patients (72%) received 2 g of van-comycin per day, divided into 2 doses (Table 2). Themost common regimen for prophylaxis was 1 g perday in a single dose, which was used in 106 patientsundergoing either bone marrow transplantation orconfinement to a laminar airflow room for remission-induction chemotherapy (vancomycin prophylaxishas been abandoned since the emergence of resistantEnterococci). A b-lactam antibiotic was administeredwith vancomycin in 647 patients (87%), 87 patients
Concurrent agentsAminoglycosides 87 12b-lactam 647 87Amphotericin B 113 15
Duration of therapy (days) (mean) 8 —Range 1–41 —
Duration of prophylaxis (days) (mean) 16 —Range 1–81 —
BMT: bone marrow transplantation.
2600 CANCER December 15, 1998 / Volume 83 / Number 12
(12%) received vancomycin in combination with anaminoglycoside (generally amikacin), and 113 patients(15%) received amphotericin B. In no case was vanco-mycin administered as a single agent.
Of the 742 patients, 455 (61%) had at least 1 peakvancomycin serum level obtained. Among these, themajority of patients (85%) had multiple levels ob-tained for an average of 5 levels per patient at anaverage cost (payer’s perspective) of $270 per patient.
ToxicitiesPhlebitis developed in 21 of the 742 patients (3%),often (48%) during episodes of gram-positive bactere-mia (Table 3). Phlebitis was significantly more com-mon among patients who received vancomycin for .7 days (P , 0.01) and among those with peripherallonglines (7% vs. 1%; P , 0.01). All but 3 cases ofphlebitis occurred in patients whose central venouscatheter had been inserted in the 28 days prior tovancomycin initiation. Only 1 of the 129 patients (,1%) who received vancomycin through short cathetersor butterfly needles developed phlebitis. All patientswith phlebitis responded promptly to local therapy;none required discontinuation of the catheter or van-comycin. “Red person syndrome” developed in 10 pa-tients (1%), all of whom received subsequent doses ofvancomycin without incident after the infusion time
was increased from 1 hour to 2 or 3 hours. Skin rasheswere far more common, occurring in 82 patients(11%). However, 78 of these patients also were receiv-ing a b-lactam antibiotic, making the causative role ofvancomycin difficult to evaluate. Clearly, patients whoreceived vancomycin plus a b-lactam antibiotic werefar more likely to develop rashes than those who re-ceived vancomycin plus an aminoglycoside (18% vs.1%; P , 0.0001). Although no patient received anantibiotic to which he or she was known to be allergic,14% of patients who received a b-lactam antibiotichad experienced allergic reactions to a different b-lac-tam antibiotic previously. Despite this apparent asso-ciation, a rash developed in 12% of patients with doc-umented allergy to b-lactam antibiotics and 12% ofthose without a previous history of allergy. None of therashes required discontinuation of vancomycin, al-though the b-lactam antibiotic was discontinued in 47cases.
OtotoxicityClinical evidence of ototoxicity developed in 18 of 319patients (6%) who received other ototoxic agents and12 of 423 patients (3%) in whom vancomycin was theonly ototoxic agent (P 5 0.08) (Table 4). All clinicalototoxicity resolved within 3 weeks after vancomycintherapy. Patients receiving cisplatin reported tinnitusmost commonly, whereas conversational hearing losswas reported in 2–5% of patients receiving other oto-toxic agents. Seven of the 30 patients (23%) with oto-toxicity met our definition of nephrotoxicity; serumvancomycin levels were obtained in 19 patients. Peaklevels reached 40 – 60 mg/mL in 3 of the 19 patients(16%) and 84 mg/mL in 1 patient; in the remainingpatients, peak levels were , 40 mg/mL.
High frequency hearing loss was detected in 3 ofthe 30 asymptomatic patients who were selected foraudiometry. None of these 3 patients had renal insuf-ficiency and serum vancomycin levels were not ob-tained. One patient received vancomycin withoutother ototoxic agents whereas the other two patientsreceived either cisplatin or furosemide and tri-methoprim/sulfamethoxazole.
NephrotoxicityNephrotoxicity developed in 127 of the 726 patients(17%). In 11 patients (2%), the estimated creatinineclearance fell below 20 mL per minute; 5 of thesepatients required dialysis. By univariate analysis,nephrotoxicity was more common in older patients,those with hematologic malignancies, and thosewhose APACHE scores were . 40 at vancomycin ini-tiation (Table 5). Patients who received vancomycinfor . 14 days were at significantly higher risk of neph-
TABLE 3Relation between the Prevalence of Phlebitis and PotentialRisk Factors
Risk factors Total patientsPatients withphlebitis (%)
Gram-positive bacteremia 183 10 (5)a
No bacteremia 559 11 (2)a
Duration of vancomycin (days)# 7 402 5 (1)b
. 7 340 16 (5)b
Mean duration 10 13Type of CVC
Peripheral longline 203 15 (7)b
Subclavian 377 5 (1)b
Jugular/femoral 33 0 (0)No CVC 129 1 (1)
Duration of CVC (days), 7 235 11 (5)c
7–14 53 2 (4)14–28 50 2 (4). 28 275 5 (2)c
Mean duration of CVC(days) 76a 45a
CVC: central venous catheter.a P , 0.05.b P , 0.01.c P 5 0.11.
Vancomycin Toxicity in Cancer Patients/Elting et al. 2601
rotoxicity than those who received , 7 days’ therapy(32% vs. 12%; P , 0.0001). Paradoxically, those whoreceived only 1 g of vancomycin per day were atgreater risk than those who received a standard doseof 2 g/day (26% vs. 16%; P 5 0.02). However, thisfinding was confounded by the practice (now aban-doned) of using the 1 g/day regimen for long term (.14 days) prophylaxis in bone marrow transplant recip-ients.
The development of nephrotoxicity was not asso-ciated with elevated peak serum vancomycin levels,occurring in 19% of both those whose highest peaklevel was , 50 mg/mL and those whose levels were .50 mg/mL (Table 5). However, patients whose troughserum vancomycin levels were . 10 mg/mL were sig-nificantly more likely to develop nephrotoxicity thanthose whose trough levels were maintained at , 10mg/mL throughout therapy (28% vs. 15%; P 5 0.008).
When vancomycin was not administered withother nephrotoxic agents, only 8% of patients devel-oped nephrotoxicity (Table 5) In all these cases, thenephrotoxicity was asymptomatic and self-limited. Incontrast, 36% of patients who received highly nephro-toxic agents concurrently (amphotericin B, aminogly-cosides, cyclosporine, and cisplatin) developed neph-rotoxicity (P , 0.0001). All five patients who requireddialysis received other highly nephrotoxic agents. Nopatient died as a result of renal dysfunction and nonerequired chronic dialysis.
Multiple Variable Model of Risk of NephrotoxicityLogistic regression was used to derive a multiple vari-able model of risk of nephrotoxicity in the 365-patientderivation set. The concurrent administration of ahighly nephrotoxic drug was associated with a 3-fold
increase in the risk of nephrotoxicity (odd ratio [OR] 53.10, 95% confidence level [CI], 1.48 – 6.44; P , 0.001),whereas the concurrent administration of moderatelynephrotoxic drugs increased that risk only when . 1 ofthese agents were administered (OR 5 2.94, 95% CI,1.43– 6.03; P 5 0.01). Not surprisingly, patients whowere in poorer clinical condition (APACHE score . 40)also were at a significantly increased risk of developingnephrotoxicity (OR 5 3.04, 95% CI, 1.49 – 6.21; P 50.002).
A clinical prediction rule, the NRS, was developedfrom the multiple variable model. A “point” value wasassigned to each factor as described previously and ascore was obtained by summing the total points basedon the presence or absence of each factor (Table 6). Bycomputing the point total for each patient in the der-ivation set, an index with two risk categories was de-fined (Table 7). The Nephrotoxicity Risk Index reliablyseparates patients into clinically relevant high risk(risk score of 2 or 3) and low risk (risk score of 0 or 1)groups. All patients who developed serious nephrotox-icity (estimated creatinine clearance , 20 mL/minute)fell into the high risk group. High risk patients devel-oped asymptomatic nephrotoxicity at a rate 4 timesthat of low risk patients (38% vs. 8%; P , 0.0001).
The derived model was applied prospectively tothe randomly chosen 359-patient validation set. Thecharacteristics of patients in the derivation and vali-dation sets virtually were identical. Approximately53% in each set were males and 36% in the derivationset and 38% in the validation set were age . 50 years.Approximately 49% of patients in the derivation sethad hematologic malignancies compared with 46% inthe validation set; 14% and 12% of patients, respec-tively, had undergone bone marrow transplantation.
TABLE 4Relation between Clinical Ototoxicity as Reported by Patients and Concurrent Administration of Ototoxic Agents in Combinationwith Vancomycin
95% CI: 95% confidence interval.a The presence of “conversational hearing loss” was determined by patient interview rather than by audiometry.b 95% confidence interval around the estimated percent with any clinical ototoxicity.c P 5 0.08.
2602 CANCER December 15, 1998 / Volume 83 / Number 12
Approximately 18% of patients in the derivation sethad an APACHE score . 40 compared with 19% in thevalidation set; 41% and 39% of patients, respectively,were profoundly neutropenic (, 100 neutrophils permm3). Comparison of the ROC curves demonstratedvery similar performance; the area under the curve inthe derivation set was 0.77 compared with 0.72 in thevalidation set (Fig. 1).
The NRS was computed for each patient in thevalidation set and the sensitivity and specificity of therisk index were compared with those of standard prac-tice for these patients. The sensitivity of the risk indexwas superior to that of standard practice for any neph-rotoxicity (71% vs. 53%) and for serious nephrotoxicity(100% vs. 63%) (Table 8). Likewise, the specificity ofthe Nephrotoxicity Risk Index exceeded that of stan-
dard practice for any nephrotoxicity (90% vs. 74%) andfor serious nephrotoxicity (81% vs. 69%). The clinicaland financial implications of these findings are sub-stantial. For example, during the 3-month study, 165patients who did not develop nephrotoxicity and whowould have been considered low risk by the Nephro-toxicity Risk Index had serum levels routinely moni-tored during vancomycin therapy. The cost of thismonitoring exceeded $37,000 (payer’s perspective).More important, of the 68 patients in the validation setwho developed nephrotoxicity, the risk index wouldhave classified 48 (71%) as high risk compared withonly 36 (53%) in whom serum levels actually wereobtained (P 5 0.05). All 8 patients who developedsevere nephrotoxicity would have been selected forearly monitoring by the risk index (100%), comparedwith only 5 (63%) who actually were monitored duringvancomycin therapy. Among the remaining three pa-tients with serious nephrotoxicity, in whom routineserum vancomycin levels were not monitored, twopatients subsequently required dialysis.
DISCUSSIONA number of previous studies have addressed the in-cidence of vancomycin-associated toxicity; the results
TABLE 5Impact of Risk Factors on the Prevalence of Nephrotoxicity
Risk factorNo. ofPatients No. with NT % with NT P valuea
NT: nephrotoxicity; APACHE: acute physiology and chronic health evaluation; NS: not significant.a With the exception of the values indicated, all P values reflect comparisons to the lowest risk category.b P value reflects the comparison between the indicated dosage and a 2-g dose.
TABLE 6Computation of the Nephrotoxicity Risk Score
Risk factor “Points”
Cisplatin administration at any time in the past 30 days orconcurrent administration of any of the followingdrugs: amphotericin B, aminoglycosides, foscarnet,cyclosporine, polymixin, pentamidine
1
Two or more of the following drugs at any time in thepast 30 days: carboplatin, cyclophosphamide, NSAIDS
1
APACHE Score . 40 1Total possible score (summing all points) 3
NSAIDS: nonsteroidal antiinflammatory drugs; APACHE: acute physiology and chronic health evalua-
Scores of 0 or 1 5 low risk; scores of 2 or 3 5 high risk.
Vancomycin Toxicity in Cancer Patients/Elting et al. 2603
of several of these are detailed in Table 9 for compar-ison.16 –23 In some, the sample size was too small toderive precise estimates of rates of rarely occurringevents, as reflected in the wide confidence intervals.Our study expands on previous studies by providingprecise estimates of vancomycin-associated toxicitiesin what to our knowledge is the largest sample ofcancer patients published to date. However, a poten-tial limitation of our study is the lack of patients whoreceived only vancomycin. Although interpretation ofour results is confounded by concurrent administra-tion of other agents, our sample is representative of atypical oncology population in which toxic antineo-plastic agents and two-drug antimicrobial regimensare commonplace. Given the potential for increasedrates and severity of toxicities in the presence of othertoxic agents, we consider our results to be illustrativeof the clinical realities encountered when treating pa-tients with cancer.
Another potential limitation of our study is the useof estimated creatinine clearance to determine thepresence of nephrotoxicity rather than actual creati-nine cleared during a 24-hour period. In particular, inolder patients with decreased muscle mass, this maylead to overestimates of creatinine clearance andtherefore underestimates in our rates of nephrotoxic-ity.
Previously reported rates of ototoxicity ranged be-tween 1–9%, with lower (and more precise) estimatesobserved in studies with larger sample sizes. The rate
in our patients (4%) parallels that observed in thesestudies.24 –27 Ototoxicity generally has been reportedin association with peak serum vancomycin levels .80 mg/mL. In our patients, elevated levels were ob-served in 4 of 19 patients (21%) with clinically docu-mented ototoxicity; however, only 1 of these was . 80mg/mL. This lack of correlation between serum van-comycin levels and the development of ototoxicitymay suggest that the observed toxicity was due toother concurrently administered agents or to theireffects in combination with vancomycin. It also ispossible that in elderly or obese patients, steady statehad not been reached after the third dose of vanco-mycin; thus, interpretation of serum levels in thesepatients would be difficult. Nevertheless, the inci-dence of patient-reported ototoxicity was low and thesymptoms disappeared within a month after discon-tinuation of vancomycin.
We observed a significantly lower rate of phlebitis(3%) than that reported previously (13–37%) (Table 9).Some early authors observed rates approaching 45%25
and it occasionally was necessary to rotate infusionsites daily to avoid discontinuing vancomycin. Thisdifference may be related to impurities in early prep-arations of the drug. It is possible that the use ofcentral venous catheters reduced the rate of phlebitisin our patients compared with patients treated beforethe widespread use of subclavian and longline cathe-ters. However, the rate of phlebitis in the 129 patientswho received vancomycin through “butterfly” needlesor short catheters also was far lower (1%) than ratesreported prior to purification of the drug. Removal ofthe catheter or rotation of infusion sites was not re-quired to control phlebitis in any case. We concludethat vancomycin-associated phlebitis is uncommonand mild with modern preparations of the drug. Localtherapy without removal of the catheter usually issufficient to control phlebitis.
Conversely, rates of rash in early studies of van-comycin generally were low (range, 0 – 6%) comparedwith that observed in our patients (11%) (Table 9).However, in a majority of these previous studies, theonly other antibiotic administered was an aminogly-coside, whereas in our study 647 of all patients (87%)and 95% of patients with rashes also received a b-lac-tam antibiotic. The rate of rash in our study moreclosely resembles that observed in studies of b-lactamantibiotics and may reflect rashes due to these antibi-otics rather than to vancomycin. This hypothesis issupported by the frequency with which rashes re-sponded to discontinuation of the b-lactam antibiot-ics while vancomycin was continued. Based on theseobservations, discontinuation of vancomycin to con-trol rashes most likely should be reserved for patients
FIGURE 1. Receiver operating characteristic curves for the derivation and
validation sets. The area under the curve was 0.77 for the derivation set and
0.72 for the validation set, suggesting very similar performance of the model.
The diagonal line illustrates a model with no ability to discriminate between
high risk and low risk patients.
2604 CANCER December 15, 1998 / Volume 83 / Number 12
who fail to respond to specific therapy and discontin-uation of concurrent b-lactam antibiotics.
Over the last 30 years, the nephrotoxic potential ofvancomycin has been examined in . 20 publica-tions.
1–11, 16 –33
These studies were conducted in differentpopulations, with different doses of vancomycin anddefinitions of nephrotoxicity. Major changes have oc-curred in the other agents used and in the formulationof vancomycin. As would be expected from a diversecollection of studies, the rates of nephrotoxicity vary.However, there are commonalities among them thatmay be used to inform practice.
First, previous authors17 have suggested that se-verity of illness explained individual episodes orhigher rates of nephrotoxicity, although this was mea-sured prospectively in only a few studies that includedsubjective assessment of clinical status. We objectivelyquantified this factor by computing a modifiedAPACHE III score at the onset of vancomycin therapy.Our findings suggest that these previous authors’ clin-ical impressions were correct. Patients with higher
APACHE scores (more serious illness) developednephrotoxicity more frequently than those with lowerscores (P 5 0.01). This finding particularly is impor-tant because high APACHE scores also are associatedwith colonization and infection by vancomycin-resis-tant gram-positive organisms.34
Second, with the exception of only a few stud-ies,17,19,22 the rate of nephrotoxicity in patients whoreceive aminoglycosides in combination with vanco-mycin exceeds that observed in patients who receivevancomycin alone or in combination with nonneph-rotoxic agents, often by a factor of 3 or 4.16,18,20,21,23
Our results confirm these findings. Furthermore, ani-mal studies support this observation, having demon-strated that the addition of vancomycin to aminogly-cosides not only adds to aminoglycoside-relatednephrotoxicity, but also potentiates it.35,36 Our obser-vations suggest that this finding may apply to otherhighly nephrotoxic agents such as amphotericin B,cyclosporine, and cisplatin.
Considerable controversy exists concerning the
TABLE 8Sensitivity, Specificity, and Predictive Value of the Nephrotoxicity Risk Index Compared with Standard Practice-Validation Set
TABLE 9Comparison of Rates of Toxicities from Previous Publications
ReferenceFarber andMoellening16
Melloret al.17
Sorrel andCollignon18
Ciminoet al.19
Downset al.20
Rybacket al.21
Goetz andSayers22
Salama andRotstein23
CurrentStudy
Year of publication 1983 1985 1985 1987 1989 1990 1993 1993 1998Design Retro Pro Pro Retro Pro Pro Pro Pro ProVancomycin courses (no.) 94 33 54 81 66 231 69 91 747Population type Medical Medical Medical Cancer Medical Medical Medical Medical CancerNephrotoxicity
Vancomycin Toxicity in Cancer Patients/Elting et al. 2605
relation between serum vancomycin levels and thedevelopment of nephrotoxicity. Recent authors havesuggested that peak levels, once closely monitored toidentify nephrotoxicity, are unreliable predictors ofrenal toxicity.7–9 A number of authors have reported acorrelation between elevated trough levels (. 10 mg/mL) and the subsequent development of nephrotox-icity,16,19,21 whereas others have found no clear asso-ciation.20 Univariate analyses of our observationssupport the former findings; elevated peak levels werenot associated with the subsequent development ofnephrotoxicity whereas elevated trough levels wereassociated with a two-fold increase in risk (P 5 0.008).However, despite the association noted by univariateanalysis, this factor was not a significant predictor inthe multiple variable model. We attribute this phe-nomenon to the positive correlation between a highAPACHE score at baseline (a significant predictor inthe multiple variable model) and a subsequently ele-vated trough level. Of the 115 patients whose initialsteady state trough levels were . 10 mg/mL, 44 (38%)had an APACHE score . 40 at baseline; of 193 patientswhose initial steady state trough level was , 10 mg/mL, only 11% had an elevated APACHE score at base-line.
Vancomycin-associated nephrotoxicity in oncol-ogy practice appears to be limited to a subset of pa-tients at high risk of toxicity. However, high risk pa-tients may develop severe nephrotoxicity requiringdialysis or discontinuation of drugs that are essentialto their treatment and for which there are few, if any,nontoxic alternatives. Thus, the cost of frequent mon-itoring of serum vancomycin or creatinine levels(which may exceed the average wholesale price of a10-day course of vancomycin, approximately $320) inevery patient may not be justified, although failure tomonitor high risk patients is clinically imprudent. Inthis situation, a clinical prediction rule that estimatesindividual patients’ risks of nephrotoxicity can ensurecareful monitoring of high risk patients and avoid thecosts of unnecessary monitoring in low risk patients.
The results of the current study have demon-strated the potential benefits of this approach by de-veloping the Nephrotoxicity Risk Index, which reliablyclassifies patients into low risk and high risk groups.Using the risk index, virtually all patients at risk ofserious nephrotoxicity and 71% of those at risk ofasymptomatic nephrotoxicity should be identifiedprior to vancomycin initiation, in time to amend dosesand monitor high risk patients only. Prospective test-ing of the Nephrotoxicity Risk Index is ongoing.
In this large, prospective study of vancomycin-associated toxicity, we have demonstrated the safetyof modern preparations of vancomycin; rash, phlebi-
tis, ototoxicity, and nephrotoxicity occur in cancerpatients receiving vancomycin but generally are mildand self-limiting. Although subgroups of these pa-tients are at high risk of nephrotoxicity, we have de-veloped a method (the Nephrotoxicity Risk Index) forreliably identifying these patients prior to initiation ofvancomycin. Early identification of high risk patientsshould permit intervention prior to the most seriousmanifestations of toxicity or selection of another an-tibiotic. The latter strategy particularly may be impor-tant in institutions in which vancomycin-resistantgram-positive infections occur frequently.
At the time this study was conducted, infectionscaused by vancomycin-resistant Enterococci were ex-ceedingly rare in our institution and elsewhere. How-ever, recent increases in these infections have led todiscontinuation of the use of vancomycin prophylaxisand to restrictions of the use of oral vancomycin todecrease colonization of the gastrointestinal tract. Inmany cases these policies have stabilized the risingincidence of these infections. For example, currentlyin our institution , 1% of infections are caused bythese organisms. Thus, although concern regardingvancomycin-resistant Enterococci is justifiable, aban-doning use of this safe and effective antibiotic is notindicated. The Infectious Disease Society of Americacontinues to recommend empiric vancomycin therapyin certain clearly defined situations.37 Thus, despitethe recent muddying of the waters by the emergenceof vancomycin-resistant gram-positive organisms,“Mississippi Mud” remains a safe option for therapyfor patients with cancer.
REFERENCES1. Griffith RS. Introduction to vancomycin. Rev Infect Dis 1981;
3(Suppl):S200 – 4.2. Ena J, Dick RW, Jones RN, Wenzel RP. The epidemiology of
intravenous vancomycin usage in a university hospital: a 10year study. JAMA 1993;269(5):598 – 602.
3. Moellering RC, Krogstad DJ, Greenblatt DJ. Vancomycintherapy in patients with impaired renal function: a nomo-gram for dosage. Ann Intern Med 1981;94:343– 6.
4. Appel GB, Neu HC. The nephrotoxicity of antimicrobialagents. N Engl J Med 1977;296:722– 8.
5. Rodvold KA, Zokufa H, Rotschafer JC. Routine monitoring ofserum vancomycin concentrations: can waiting be justified?Clin Pharm 1987;6:655– 8.
6. Moellering RC Jr. Monitoring serum vancomycin levels:climbing the mountain because it is there? [editorial]. ClinInfect Dis 1994;18:544 – 6.
7. Cantu TG, Yamanaka-Yuen NA, Lietman PS. Serum vanco-mycin concentrations: reappraisal of their clinical value.Clin Infect Dis 1994;18:533– 43.
8. Freeman CD, Quintiliani R, Nightingale CH. Vancomycintherapeutic drug monitoring: is it necessary? Ann Pharma-cother 1993;27:594 – 8.
2606 CANCER December 15, 1998 / Volume 83 / Number 12
9. Edwards DJ, Pancorbo S. Routine monitoring of serum van-comycin concentrations: waiting for proof of its value. ClinPharm 1987;6:652– 4.
13. Rolston K, Hoy J. Man/woman achieves whole red person-hood through vancomycin. JAMA 1986;255:2445.
14. Knaus W, Wagner D, Draper E, Zimmerman J, Bergner M,Bastos P, et al. The APACHE III prognostic system: riskprediction of hospital mortality for critically ill hospitalizedadults. Chest 1991;100:1619 –36.
15. McMahon LF, Hayward RA, Bernard AM, Rosevear JS,Weissfeld LA. APACHE-L: a new severity of illness adjusterfor inpatient medical care. Med Care 1992;30(5):445–52.
16. Farber BF, Moellering RC Jr. Retrospective study of thetoxicity of preparations of vancomycin from 1974 to 1981.Antimicrob Agents Chemother 1983;23(1):138 – 41.
17. Mellor JA, Kingdom J, Cafferkey M, Keane CT. Vancomycintoxicity: a prospective study. J Antimicrob Chemother 1985;15:773– 80.
18. Sorrell TC, Collignon PJ. A prospective study of adversereactions associated with vancomycin therapy. J AntimicrobChemother 1985;16:235– 41.
19. Cimino MA, Rotstein C, Slaughter RL, Emrich LJ. Relation-ship of serum antibiotic concentrations to nephrotoxicity incancer patients receiving concurrent aminoglycoside andvancomycin therapy. Am J Med 1987;83:1091–7.
20. Downs NJ, Neilhart RE, Dolezal JM, Hodges GR. Mild neph-rotoxicity associated with vancomycin use. Arch Intern Med1989;149:1777– 81.
21. Rybak MJ, Albrecht LM, Bolke SC, Chandrasekar PH. Neph-rotoxicity of vancomycin, alone and with an aminoglyco-side. J Antimicrob Chemother 1990;25:679 – 87.
22. Goetz MB, Sayers J. Nephrotoxicity of vancomycin and ami-noglycoside therapy separately and in combination. J Anti-microb Chemother 1993;32:325–34.
23. Salama SE, Rotstein C. Prospective assessment of nephro-toxicity with concomitant aminoglycoside and vancomycintherapy. Can J Hosp Pharm 1993;46(2):53–9.
24. Geraci JE, Heilman FR, Nichols DR, et al. Some laboratoryand clinical experiences with a new antibiotic vancomycin.Staff Meet Mayo Clin 1956;31(21):564 – 82.
25. Dangerfield HG, Hewitt WL, Monzon OT, Kudinoff Z, Black-man B, Finegold SM. Clinical use of vancomycin. Antimi-crob Agents Annu 1960;61:428 –37.
26. Kirby WMM, Perry DM, Bauer AW. Treatment of staphylo-coccal septicemia with vancomycin: report of thirty-threecases. N Engl J Med 1960;262(22):49 –55.
27. Waisbren BA, Kleinerman L, Skemp J, Bratcher G. Compar-ative clinical effectiveness and toxicity of vancomycin, ris-tocetin, and kanamycin. Arch Intern Med 1960;106:179 –93.
28. Geraci JE. Vancomycin. Mayo Clin Proc 1977;52:631– 4.29. Cook FV, Farrar WE Jr. Vancomycin revisited. Ann Intern
Med 1978;88:813– 8.30. McHenry MC, Gavan TL. Vancomycin. Pediatr Clin North
Am 1983;30(1):31– 47.31. Geraci JE, Hermans PE. Vancomycin. Mayo Clin Proc 1983;
58:88 –91.32. Cheung RPF, DePiro JT. Vancomycin: an update. Pharma-
cotherapy 1986;6(4):153– 69.33. Bailie G, Neal D. Vancomycin ototoxicity and nephrotoxic-
ity: a review. Med Toxicol 1988;3:376 – 86.34. Shay DK, Maloney SA, Montecalvo M, Banerjee S, Wormser
GP, Arduino MJ, et al. Epidemiology and mortality risk ofvancomycin-resistant enterococcal bloodstream infections.J Infect Dis 1995;172(4):993–1000.
35. Wold JS, Turnipseed SA. Toxicology of vancomycin in labo-ratory animals. Rev Infect Dis 1981;3(Suppl):S224 –9.
37. Hughes WT, Armstrong D, Bodey GP, Brown AE, Edwards JE,Feld R, et al. 1997 Guidelines for the use of antimicrobialagents in neutropenic patients with unexplained fever. ClinInfect Dis 1997;25:551–73.
Vancomycin Toxicity in Cancer Patients/Elting et al. 2607
