ANTIMICROBIAL AGENTS AND CHEMOTHERAPY, Oct. 2011, p. 4581–4588 0066-4804/11/$12.00 doi:10.1128/AAC.00115-11 Copyright © 2011, American Society for Microbiology. All Rights Reserved.
Vol. 55, No. 10
Prediction of Failure in Vancomycin-Treated Methicillin-Resistant Staphylococcus aureus Bloodstream Infection: a Clinically Useful Risk Stratification Tool䌤† Carol L. Moore,1,2* Mei Lu,3 Faiqa Cheema,1 Paola Osaki-Kiyan,1 Mary Beth Perri,1 Susan Donabedian,1 Nadia Z. Haque1,2 and Marcus J. Zervos1,4* Division of Infectious Diseases,1 Department of Pharmacy Services,2 and Department of Public Health Sciences,3 Henry Ford Hospital, Detroit, Michigan, and Wayne State University School of Medicine, Detroit, Michigan4 Received 27 January 2011/Returned for modification 26 April 2011/Accepted 26 July 2011
Methicillin-resistant Staphylococcus aureus (MRSA) is a common cause of bloodstream infection (BSI) and is often associated with invasive infections and high rates of mortality. Vancomycin has remained the mainstay of therapy for serious Gram-positive infections, particularly MRSA BSI; however, therapeutic failures with vancomycin have been increasingly reported. We conducted a comprehensive evaluation of the factors (patient, strain, infection, and treatment) involved in the etiology and management of MRSA BSI to create a risk stratification tool for clinicians. This study included consecutive patients with MRSA BSI treated with vancomycin over 2 years in an inner-city hospital in Detroit, MI. Classification and regression tree analysis (CART) was used to develop a risk prediction model that characterized vancomycin-treated patients at high risk of clinical failure. Of all factors, the Acute Physiology and Chronic Health Evaluation II (APACHE-II) score, with a cutoff point of 14, was found to be the strongest predictor of failure and was used to split the population into two groups. Forty-seven percent of the population had an APACHE-II score < 14, a value that was associated with low rates of clinical failure (11%) and mortality (4%). Fifty-four percent of the population had an APACHE-II score > 14, which was associated with high rates of clinical failure (35%) and mortality (23%). The risk stratification model identified the interplay of three other predictors of failure, including the vancomycin MIC as determined by Vitek 2 analysis, the risk level of the source of BSI, and the USA300 strain type. This model can be a useful tool for clinicians to predict the likelihood of success or failure in vancomycin-treated patients with MRSA bloodstream infection. trough concentration or the area under the curve (AUC)/MIC ratio may theoretically be associated with better outcomes in MRSA BSI; however, this has yet to be proven in prospective clinical trials (27). Given the emerging information relating to outcomes of vancomycin treatment in cases of MRSA BSI, a comprehensive evaluation of patient, infection, strain, and treatment characteristics is necessary to determine which factors, or combination of factors, is most predictive of outcome and what, if any, treatment strategies can improve outcome. Traditionally, studies evaluating MRSA BSI lacked one or more of these factors. Studies evaluating outcome in cases of MRSA BSI have traditionally used the logistic regression model, which relies on classical statistical principles to determine the odds ratio as the measure of likelihood of risk for a specific outcome. However, the application of odds ratios to the clinical care of patients is often difficult, as it cannot account for patients with multiple risks or the complex interplay among the various factors unless they are specified in the analysis. The purpose of this study, therefore, was to conduct a comprehensive evaluation of the factors involved in the etiology and treatment of MRSA BSI in vancomycin-treated patients in order to develop a risk stratification tool for clinicians that characterizes those patients at high risk of failure. (This study was presented in part at the 19th European Congress of Clinical Microbiology and Infectious Diseases, Helsinki, Finland, May 2009, abstr. P-1861 [25a].)
Methicillin-resistant Staphylococcus aureus (MRSA) is a common cause of bloodstream infection (BSI) and has been associated with mortality rates of between 20 and 30% (16, 18, 24, 30, 33). Invasive infections caused by MRSA have been implicated in over 18,000 deaths annually (14). Vancomycin has remained the mainstay of therapy for serious Gram-positive infections, particularly MRSA BSI; however, therapeutic failures with vancomycin have been increasingly reported (3, 7, 24). There is also growing evidence that infection with an organism with an increased vancomycin MIC that is still within the susceptible range may also be associated with increased rates of vancomycin failure (9, 23, 29, 30). Evidence suggests that the increased risk associated with a vancomycin MIC of 2 g/ml is not mitigated by targeting higher vancomycin trough concentrations in the range of 15 to 20 g/ml (9). It has been postulated that strategies such as increasing the vancomycin
* Corresponding author. Mailing address for Marcus J. Zervos: Division of Infectious Diseases, Henry Ford Hospital, 2799 West Grand Blvd., CFP 314, Detroit, MI 48202. Phone: (313) 916-2573. Fax: (313) 916-2993. E-mail:
[email protected]. Mailing address for Carol L. Moore: Division of Infectious Diseases, Department of Pharmacy Services, Henry Ford Hospital, 2799 West Grand Blvd., CFP 519, Detroit, MI 48202. Phone: (313) 916-0777. Fax: (313) 916-2993. E-mail:
[email protected]. 䌤 Published ahead of print on 8 August 2011. † The authors have paid a fee to allow immediate free access to this article. 4581
4582
MOORE ET AL. MATERIALS AND METHODS
Study design and data collection. A retrospective cohort design was employed, using clinical failure as the primary outcome. Consecutive patients with either primary or secondary MRSA BSI were hospitalized between July 2005 and October 2007 at a 900-bed tertiary-care hospital in Detroit, MI. MRSA bloodstream isolates were prospectively identified over the study period, and the initial MRSA blood culture was collected and stored as part of another study. For inclusion into this study, subjects had to have received active treatment within 48 h of the onset of infection and had to have received at least 48 h or more of vancomycin therapy. This study was consistent with the principles of the Declaration of Helsinki and was approved by the Institutional Review Board at Henry Ford Hospital, which provided a waiver of informed patient consent due to the nature of the study. Clinical data collection was conducted using a standardized case report form. Data were also abstracted from the medical record at certain time points during therapy (days 0, 1, 2, 3, and 30 and the end of therapy). These included antibiotic agents administered, pharmacokinetic indices for vancomycin, serum vancomycin levels, serum creatinine levels, creatinine clearance, cumulative daily vancomycin dose, the presence or absence of positive or negative cultures on the day of evaluation, and the presence or absence of a removable source of infection. Vancomycin dosing strategies and assessment of vancomycin pharmacokinetics. All patients were initially dosed with vancomycin by the use of either the hospital-approved dosing nomogram, which has been adapted from a previously validated nomogram to achieve vancomycin trough values of 10 to 20 g/ml (13), or individualized dosing by pharmacy services, using both predicted empirical and calculated pharmacokinetics for definitive treatment. Hemodialysis patients were given doses determined empirically using a previously established dosing regimen that has been shown to achieve predialysis vancomycin levels of 5 to 20 g/ml (1). Predialysis follow-up levels were evaluated by the clinical pharmacist, and the maintenance dose was adjusted to achieve a predialysis vancomycin level of more than 15 g/ml. For the purposes of evaluating pharmacokinetic indices, patients on hemodialysis (n ⫽ 42) or those with acute renal failure (n ⫽ 54) at the onset of infection were excluded from the analysis. Vancomycin trough values were obtained from serum concentrations collected within the hour prior to the next dose at steady-state conditions (3 half-lives). True trough values were calculated for all patients by mathematical extrapolation using equations 1 through 3 as follows: for equation 1, estimated creatinine clearance ⫽ (140 ⫺ age in years)(IBW)/(72 ⫻ Scr) ⫻ 0.85 (for women); for equation 2, true trough ⫽ measured concentration ⫻ e(⫺ke ⫻ true time ⫺ actual time); and for equation 3, ke ⫽ 0.00117 ⫻ CrCl ⫹ 0.003. For these equations, IBW represents ideal body weight [for men, 50 ⫹ (2.3 kg ⫻ inches of height above 60); for women, 45 ⫹ (2.3 kg ⫻ inches of height above 60)], Scr represents serum creatinine concentration, ke represents the elimination rate constant, true time represents correct time for trough assessment, actual time represents actual time of trough assessment, and CrCl represents creatinine clearance. Definitions. The duration of bacteremia was calculated as the number of days the blood cultures were documented to be positive for MRSA. Epidemiologic classification was based on the presence or absence of health care risk factors and determination of whether the acquisition was the result of community or hospital onset, as previously described (15). The suspected source of the BSI was identified by chart review using CDC definitions for nosocomial infection (10). The sources of BSI were classified into 3 categories: low-risk sources (related mortality rate of ⬍10%), which included intravenous catheter, urinary tract, earnose-larynx, gynecologic sources, and several manipulation-related sources; intermediate-risk sources (associated mortality rate of 10 to 20%), which included osteoarticular sources, soft tissue sources, and unknown sources; and high-risk sources (mortality rate of ⬎20%), which included endovascular sources, lower respiratory tract, abdominal sources, and central nervous system (CNS) foci as previously described (18, 30). Patients with 2 or more possible sources of BSI (e.g., 1 low-risk source and 1 high-risk source) were classified as having the higher or highest risk source. Concomitant therapy was defined as treatment with either aminoglycoside or rifampin for 3 days or more in addition to vancomycin. Concomitant therapy was defined as early (ⱕ48 h) or late (⬎48 h) in relation to initiation of therapy. Thirty-day mortality was defined as mortality occurring in the 30-day period following acquisition of the index culture. Microbiologic failure was defined as growth of MRSA from a blood culture at ⱖ7 days from the index culture while the patient was receiving therapy. Recurrence of infection was defined as test results showing blood cultures positive for MRSA within 30 days of acquisition of the index culture and after completion of therapy. Clinical failure was defined as a composite of mortality, microbiologic failure, and/or recurrence of MRSA BSI within the first 30 days.
ANTIMICROB. AGENTS CHEMOTHER. Microbiologic methods. MRSA isolates were collected at the onset of infection and stored in the Infectious Diseases Research Laboratory at Henry Ford Hospital, Detroit, MI. Each MRSA bloodstream isolate underwent in vitro susceptibility testing to determine vancomycin MIC, vancomycin minimum bactericidal concentration (MBC), and heterogeneous vancomycin-intermediate S. aureus (hVISA) phenotype and was subjected to pulsed-field gel electrophoresis (PFGE) typing using methods previously described (4, 21). The following S. aureus strains were used in performance of the tests: ATCC 29213, Mu3, and Mu50. The vancomycin MIC was determined by automated microdilution with Vitek 2 (bio-Merieux, Durham, NC), Etest (bio-Merieux, Durham, NC), and manual broth microdilution (6). All manually determined MICs were from examinations by a single observer. Inducible clindamycin resistance was determined using the D-zone test as described by the CLSI (5). Vancomycin MBCs were determined using previously established methods (6, 26). The MBC was reported in micrograms per milliliter in the same manner as used for the MIC. Vancomycin tolerance was defined as an MBC/MIC ratio ⱖ 32 g/ml. Isolates were tested for hVISA by the use of the macrodilution Etest method (bioMerieux, Durham, NC) as previously described (34). Genomic DNA for PFGE was prepared, digested with SmaI (New England BioLabs, Beverly, MA), and processed using a CHEF-DR III system (Bio-Rad Laboratories, Hercules, CA) and a previously described method (21). PFGE patterns were compared using BioNumerics software (version 3.5; Applied Maths, Belgium). Multiplex PCR was performed to identify staphylococcal cassette chromosome mec (SCCmec) types I to V using a previously described method (36). The accessory gene regulator (agr) locus, in addition to the presence of the Panton-Valentine leukocidin (PVL) toxin genes lukS-PV and lukF-PV, was determined for all isolates (17, 31). Statistical methods. Descriptive characteristics are reported as percentages with means ⫾ standard deviations (SD) or medians with interquartile ranges (IQR). Categorical variables were analyzed using the chi-square test or the two-sided Fisher’s exact test as appropriate. Continuous variables were analyzed using the independent-sample t test or the nonparametric Mann-Whitney U test as appropriate. A P-value ⬍ 0.05 was considered statistically significant. CART analysis. We used Classification and Regression Tree (CART) analysis to develop a prediction model for clinical failure (35). The prediction model was generated using all patient, strain, and infection variables in addition to treatment variable data collected during the first 3 days of vancomycin therapy. CART analysis, or binary recursive partitioning analysis, uses a nonparametric analytical approach that identifies the variables most predictive of the outcome of interest and subsequently develops a predictive model for classification of future subjects. Unlike multivariable logistic regression analysis, it is ideally suited for generation of a clinical decision model, as it can reveal important relationships between variables that can remain hidden when using other types of analyses. CART analysis begins with the root node (all subjects) and then determines which variable has the highest predictive ability. Subsequently, it determines the optimal split (cutpoint) for this variable that partitions the population into 2 child nodes with distinctly different outcomes. Every value of the predictor variable is tested for its potential as a split value, and the program determines the most predictive splitter. Each child node can then itself become a parent node that produces its own child nodes. The process continues to classify the subjects until no further variables of predictive importance are identified (2, 8). To avoid model overfitting, we first identified a subset of variables based on their importance (from high [e.g., 100%] to low [0%]), which can be interpreted as the degree of masking of the tree structure or predictive ability for classification; this method is similar to the use of a univariate analysis approach before multivariable modeling is performed. Before the multivariable modeling, a set of battery tests were performed to select the minimum numbers of observations composing a parent or child node, the number of folds for a cross-validation, and the optimal splitting methods. We then fit the model using variables with a relative importance value of 20% or higher. The receiver operating characteristic (ROC) curve was calculated on the basis of the learning data and testing data. The ROC value was reported in the range of 0 to 1, where 0.70 or above is considered to represent a good prediction. CART analysis was conducted using CART 6.0 software (Salford Systems, San Diego, CA).
RESULTS During the study period, 288 consecutive patients with MRSA BSI were identified. Patients were excluded if they had been treated with vancomycin within ⬍48 h of presentation (n ⫽ 24), had been treated with another agent(s) (n ⫽ 30), had
VOL. 55, 2011
VANCOMYCIN FAILURE IN MRSA BLOODSTREAM INFECTION TABLE 1. Patient demographics Demographica
Valueb
Age in years (mean ⫾ SD)...........................................................57 ⫾ 17 Sex Male.............................................................................................114 (57) Races African-American.......................................................................140 (70) Caucasian .................................................................................... 55 (28) Other ........................................................................................... 5 (3) Comorbid conditions Cardiovascular disease .............................................................. Cerebral vascular disease.......................................................... Diabetes ...................................................................................... Malignancy.................................................................................. Renal failure Acute ....................................................................................... Chronic.................................................................................... ESRD on dialysis (HD or PD) ............................................ Liver disease ............................................................................... COPD .......................................................................................... HIV and/or AIDS ...................................................................... Immunosuppression................................................................... IVDA...........................................................................................
78 (39) 47 (24) 84 (42) 35 (18) 54 (27) 20 (10) 42 (21) 48 (24) 21 (11) 11 (6) 28 (14) 46 (23)
Prior hospitalizationc .....................................................................113 (57) Prior surgeryd.................................................................................. 22 (11) Nursing home resident .................................................................. 36 (18) Prior systemic antibioticse ............................................................. 82 (41) Vancomycin ................................................................................ 45 (23) Risk level of source ....................................................................... Low .............................................................................................. 53 (27) Intermediate ............................................................................... 91 (46) High ............................................................................................. 56 (28) Source of bloodstream infection CVC ............................................................................................. Infective endocarditis ................................................................ Skin or wound ............................................................................ Pneumonia .................................................................................. Intra-abdominal.......................................................................... Genitourinary ............................................................................. Graft or device ........................................................................... Multiple ....................................................................................... Undetermined ............................................................................
46 (23) 20 (10) 63 (32) 19 (10) 4 (2) 7 (4) 9 (5) 10 (5) 22 (11)
APACHE-II result at onset (mean ⫾ SD).................................15 ⫾ 8 Mean (SD) duration of bacteremia in days ............................... 2 (4) a ESRD, end-stage renal disease; HD, hemodialysis; PD, peritoneal dialysis; COPD, chronic obstructive pulmonary disease. b Values represent numbers (percent) of subjects unless otherwise indicated. c Hospitalization within the previous year. d Surgery within the previous 30 days. e Systemic antibiotics within the previous 90 days.
received no treatment (n ⫽ 6), were ⬍18 years of age (n ⫽ 2), had an incomplete medical record (n ⫽ 15), had been transferred to another facility (n ⫽ 5), or had a nonevaluable outcome (n ⫽ 6). A total of 200 consecutive subjects treated with vancomycin were evaluated. The baseline characteristics of the study population can be found in Table 1. Overall, the majority of the population was African-American (70%) and male (57%), with a mean age of 57 ⫾ 17 years. Cardiovascular disease (39%) and diabetes (42%) were common, and 57% of the population had been hospitalized within the previous year. Most infections were from an intermediate (46%)- or high
4583
(28%)-risk source, with a mean Acute Physiology and Chronic Health Evaluation II (APACHE-II) score at onset of 15 ⫾ 8. Outcome assessments conducted at 30 days from acquisition of the index blood culture found a mortality rate of 15%, a microbiologic failure rate of 7%, and a recurrence rate of 5%. Overall, 24% had clinical failure at 30 days. A bivariate (success versus failure) outcome analysis was conducted, evaluating patient, treatment, infection, and strain characteristics (see Table 2 and Table 3). Patients with treatment failure were more likely to have cardiovascular disease (34 versus 55%; P ⫽ 0.009), to have acute renal failure (22 versus 45%; P ⫽ 0.002), and to be immunosuppressed (11 versus 23%; P ⫽ 0.034). Success was more common in those with a history of intravenous drug abuse (IVDA) (27 versus 11%; P ⫽ 0.021). Source removal and initial vancomycin trough were not associated with outcome; however, a higher definitive vancomycin trough value was associated with failure (17 ⫾ 6 versus 21 ⫾ 7; P ⫽ 0.044). Early concomitant therapy with either aminoglycoside (99%) or rifampin (1%) was associated with success (33 versus 17%; P ⫽ 0.032). A higher APACHE-II score (14 ⫾ 8 versus 19 ⫾ 9; P ⫽ 0.001) and risk level of the source were also associated with failure. Infection with a USA600 PFGE strain was associated with failure (0 versus 11%; P ⫽ 0.001). CART modeling. All variables were evaluated for inclusion in the final model. These included all patient, strain, and baseline infection characteristics in addition to infection characteristics (e.g., source removal) and treatment characteristics (e.g., vancomycin serum concentrations) over the first 3 days of therapy. Eight variables were found to have a relative importance level of at least 20%: APACHE-II score; risk level of the source; USA100 PFGE type; agr locus; IVDA; vancomycin MIC as determined by Vitek 2; vancomycin MIC as determined by Etest; and USA300 PFGE type. Because the USA100 PFGE type, USA300 PFGE type, and agr locus were highly correlated (correlation coefficients ⬎ 0.80), only the USA300 PFGE type was included in the initial multivariable model. Four variables remained in the final model, which had estimated predictive-ability ROC values of 0.74 based on the learning data and 0.73 based on the testing data determined using a 50-fold cross-validation approach (see Fig. 1). The final decision tree consisted of 5 terminal nodes. Of all the factors, the APACHE-II score was found to be the variable most predictive of treatment failure; a cutpoint score of 14 split the population into two distinct groups. Forty-seven percent of the population had an APACHE-II score ⬍ 14 and a low (11%) rate of clinical failure, with an associated mortality rate of 4%. Fifty-four percent of the population had an APACHE-II score ⱖ 14 and a higher (35%) rate of clinical failure, with an associated mortality rate of 23%. Despite those with APACHE-II score ⬍ 14 having a low rate of clinical failure or mortality, the vancomycin MIC value (as determined by Vitek 2 analysis) was found to have predictive ability, splitting this population into two distinct groups. Ninety-five percent of subjects had an APACHE-II score ⬍ 14 and a vancomycin MIC ⱕ 1 g/ml, and this group also had low rates of clinical failure (8%) and mortality (5%). Subjects in this group had a mean age of 51 ⫾ 16 years, a low prevalence of cardiovascular disease (23%), diabetes (30%), and acute renal failure (18%), and a high prevalence of IVDA (42%). Community-
4584
MOORE ET AL.
ANTIMICROB. AGENTS CHEMOTHER.
TABLE 2. Comparison of outcomes by patient and treatment characteristics in MRSA BSIa Success (n ⫽ 153)
Failure (n ⫽ 47)
P
56 ⫾ 17
60 ⫾ 17
0.160
84 (55)
30 (64)
0.280
110 (72) 40 (26) 3 (2)
30 (64) 15 (32) 2 (4)
52 (34) 36 (24) 64 (42) 23 (15)
26 (55) 11 (23) 20 (43) 12 (26)
0.009 0.986 0.930 0.098
33 (22) 15 (10) 32 (21)
21 (45) 5 (11) 10 (21)
0.002 0.789 0.958
Peripheral vascular disease Liver disease COPD HIV and/or AIDS Immunosuppression
26 (17) 37 (24) 17 (11) 7 (5) 17 (11)
10 (21) 11 (23) 4 (9) 4 (9) 11 (23)
0.504 0.913 0.788 0.292 0.034
Other conditions IVDA Nursing home residence Prior vancomycin exposure
41 (27) 26 (17) 35 (23)
5 (11) 10 (21) 10 (21)
0.021 0.504 0.818
2.5 ⫾ 2.7
3.0 ⫾ 2.9
0.225
42 (28) 54 (35) 57 (37)
11 (23) 17 (36) 19 (40)
Characteristicb
Patients Age in years (mean ⫾ SD) Sex Male Race African-American Caucasian Other
0.466
Comorbid conditions Cardiovascular disease Cerebral vascular disease Diabetes Malignancy Renal failure Acute Chronic ESRD on dialysis (HD or PD)
Serum creatinine (mg/dl; mean ⫾ SD) Treatments Source removal Removable source, removed Removable source, not removed Nonremovable source Vancomycin trough (mean mg/dl ⫾ SD)c Initial Definitive
0.850
11 ⫾ 6
14 ⫾ 8
0.198
17 ⫾ 6
21 ⫾ 7
0.044
Early concomitant therapyd
51 (33)
8 (17)
0.032
a
Results reported as percentages unless otherwise indicated. ESRD, end-stage renal disease; HD, hemodialysis; PD, peritoneal dialysis; COPD, chronic obstructive pulmonary disease. Subjects with acute renal failure at the baseline of infection or those on chronic hemodialysis were excluded from the vancomycin trough evaluation. The numbers of evaluable subjects were 88 for the initial trough and 78 for the definitive trough. The initial vancomycin trough was defined as the trough attained with the first 72 h. The definitive vancomycin trough was defined as the trough attained after the first 72 h. d Early concomitant therapy was defined as concomitant treatment with either aminoglycoside or rifampin for 3 days or more within the first 48 h. b c
associated infection was common (44%), with 43% of the subjects having a skin or wound source and 17% experiencing endocarditis. Fifty-eight percent had a USA300 infection. The small (n ⫽ 5) proportion with an APACHE-II score ⱖ 14 and a vancomycin MIC of 2 g/ml had a high (60%) rate of clinical failure. All of the failures in this group were due to microbiologic failure, with a mean (⫾ SD) duration of bacteremia of 9 ⫾ 8 days. These subjects had a mean age of 55 ⫾ 17 years, and cardiovascular disease (40%), diabetes (60%), and acute renal failure (40%) were common. Most infections were associated with health care but were community onset (80%). Pneumonia (40%) and central venous catheters (CVC) (40%) were common sources of infection. All 5 of these subjects had a USA100 isolate infection.
In comparison, in those with an APACHE-II score ⱖ 14, the variable most predictive of failure was found to be the source of the BSI. Thirty-five percent of those subjects had an infection from a low-risk source, most frequently catheter related, and the subjects had lower rates of clinical failure (14%) and mortality (8%), whereas 65% had infections from an intermediate- or highrisk source and high rates of clinical failure (46%) and mortality (31%). In those with infections from a low-risk source, the strain type of the infecting strain was the variable most predictive of clinical failure. Subjects infected with a USA300 strain were found to have a 36% clinical failure rate and an associated mortality rate of 27%. These subjects had a mean age of 61 ⫾ 18 years, were commonly being treated by hemodialysis (73%), and had high prevalences of cardiovascular disease (64%) and diabe-
VOL. 55, 2011
VANCOMYCIN FAILURE IN MRSA BLOODSTREAM INFECTION
4585
TABLE 3. Comparisons of outcomes by infection and strain characteristics in MRSA BSIa Characteristic
Infections APACHE-II (mean ⫾ SD) Risk level of infection sourceb Low Intermediate High Source of bloodstream infection CVC Infective endocarditis Skin or wound Pneumonia Intra-abdominal Genitourinary Graft or device Multiple Undetermined Mean no. of days ⫾ SD of bacteremia Strains PFGE identification USA100 USA300 USA600 Other Vancomycin MIC (g/ml) Manual broth microdilution 0.25 0.50 1.00 2.00 Automated dilution (Vitek-2) ⬍1.0 2.0 Etest 1.0 1.5 ⬍2 hVISA
Treatment success (n ⫽ 153)
Treatment failure (n ⫽ 47)
14 ⫾ 8
19 ⫾ 9
47 (31) 69 (45) 37 (24)
6 (13) 22 (47) 19 (40)
39 (26) 15 (10) 51 (33) 14 (9) 2 (1) 7 (5) 3 (2) 8 (5) 14 (9)
7 (15) 5 (11) 12 (26) 5 (11) 2 (4) 0 (0) 6 (13) 2 (4) 8 (17)
1⫾2
5⫾6
⬍0.001
72 (47) 67 (44) 0 (0) 9 (6)
23 (49) 17 (36) 5 (11) 2 (4)
0.822 0.355 0.001 1.000
28 (18) 107 (70) 17 (11) 1 (1)
6 (13) 33 (70) 8 (17) 0 (0)
142 (93) 11 (7)
39 (83) 8 (17)
11 (7) 117 (77) 25 (16)
2 (4) 36 (77) 9 (19)
4 (3)
6 (13)
P
0.001 0.020
0.581
0.083 0.725
0.012
a
Results are reported as percentages unless otherwise indicated. b Low-risk sources included intravenous catheter, urinary tract, ear-nose-larynx, gynecologic sources, and several manipulation-related sources; intermediate-risk sources included osteoarticular sources, soft-tissue sources, and unknown sources; high-risk sources included endovascular sources, lower respiratory tract, abdominal sources, and central nervous system foci.
tes (46%) and low prevalences of acute renal failure (18%) and IVDA (9%). All of the infections were associated with health care, with 73% community onset. All of the infections were caused by a USA300 MRSA strain and had a CVC source, and 27% had a vancomycin MIC of 2 g/ml by Vitek 2. Subjects infected with a non-USA300 isolate had a 4% clinical failure rate and a 0% mortality rate. These subjects had a mean age of 64 ⫾ 15 years and were commonly being treated by hemodialysis (54%). Cardiovascular disease (46%) and diabetes (54%) were common, and there were low prevalences of acute renal failure (15%) and IVDA (0%). All infections were associated with health care, with 81% community onset. Most of the infections were caused by a USA100 MRSA strain (92%) and had a CVC source (81%), and 15% had a vancomycin MIC of 2 g/ml as determined by Vitek 2 analysis.
For those with an APACHE-II score ⱖ 14 and intermediate- or high-risk infection sources, there was no other characteristic that was predictive of clinical failure. These subjects had a mean age of 62 ⫾ 17 years, with high prevalences of cardiovascular disease (53%), diabetes (51%), and acute renal failure (40%) and a low prevalence of IVDA (10%). Most (79%) infections were associated with health care, with 63% community onset. Most infections were caused by either a USA100 (51%) or a USA300 (31%) strain, and common sources of infection were skin or wound (34%), pneumonia (17%), and endocarditis (7%). All the graft or device infections in the study population were found in this group (13%). As it is typically not possible to determine the drug MIC until 2 to 3 days into treatment, we conducted a post hoc
4586
MOORE ET AL.
ANTIMICROB. AGENTS CHEMOTHER.
FIG. 1. Risk stratification tool for determination of outcomes for MRSA bloodstream infections treated with vancomycin. Terminal nodes are shaded in gray. Vancomycin MICs were determined by the Vitek 2 method. The low-risk group included intravenous catheter, urinary tract, ear-nose-larynx, gynecologic, and several manipulation-related sources; the intermediate-risk group included osteoarticular sources, soft tissue sources, and unknown sources; and the high-risk group included endovascular sources, the lower respiratory tract, abdominal sources, and central nervous system foci.
evaluation to determine whether prior exposure to vancomycin at the baseline could be used to predict the likelihood of a higher vancomycin MIC by the 3 methods evaluated. The MICs and frequencies of prior vancomycin exposure determined by each of the MIC testing methods were as follows: for broth microdilution, MIC of 0.25 (18%), 0.5 (20%), 1 (40%), and 2 (100%) (P ⫽ 0.032); for automated microdilution (Vitek 2), MIC of ⱕ1 (20%) and 2 (47%) (P ⫽ 0.017); and for the Etest, MIC of 1 (15%), 1.5 (20%), and ⱖ2 (38%) (P ⫽ 0.051). DISCUSSION The present report clearly outlines the complex interplay of factors involved in the outcome of MRSA BSI and presents a clinically useful risk stratification tool that can serve to identify a distinct population of vancomycin-treated patients at both low and high risk of clinical failure. Those patients with intermediate- or high-risk sources of infection in combination with an APACHE-II score of 14 or greater, representing the patient group that comprised 35% of the total population, are particularly vulnerable to clinical failure. However, this evaluation was also able to identify a large (44% with APACHE-II score ⬍ 14 and vancomycin MIC ⱕ 1 g/ml) subset of vancomycin-treated patients who were at a markedly low risk of failure (8%). We also demonstrated that strain-related characteristics such as the vancomycin MIC and the USA300 strain
type can have important prognostic value in the prediction of clinical outcome. Importantly, the present analysis did not find any modifiable factors (e.g., treatment characteristics) that significantly influenced outcome when all factors were considered. The initial vancomycin trough value was not associated with outcome, and although higher definitive trough values were associated with failure and early concomitant therapy was associated with success in bivariate analysis, these factors did not have any predictive ability when multivariable modeling was conducted. Recent consensus guidelines recommend targeting a vancomycin trough concentration of 15 to 20 g/ml for treatment of bacteremia caused by S. aureus; however, there are no prospective and limited observational clinical data supporting any trough target in MRSA BSI (9, 27). A retrospective investigation of MRSA BSI recently found that a lower initial vancomycin trough value (⬍15 mg/dl) was an independent predictor of failure (16). However, although this evaluation included a large number of patients, it is unclear how patients in acute renal failure or those on hemodialysis were evaluated, particularly at the steady state, as initial vancomycin levels were available for most (96%) of the study patients. Additionally, targeting a vancomycin trough concentration of 15 to 20 g/ml may be associated with harm to patients, as recent data indicate a greater risk of nephrotoxicity with higher doses of vancomycin or increasing trough concentrations (9, 12, 16, 19, 20).
VOL. 55, 2011
VANCOMYCIN FAILURE IN MRSA BLOODSTREAM INFECTION
The determination of the optimal pharmacokinetic target for vancomycin dosing likely requires further prospective clinical evaluation; however, our data indicate that a higher trough concentration contributed little to the outcome of infection and other studies have demonstrated that this approach may pose a hazard to subjects (9, 12, 19, 20). Ultimately, this study has shown that, while characteristics such as vancomycin trough concentration, use of concomitant therapy, change to an alternative agent, and prompt removal of the infectious source do influence outcome, the effect is outweighed by the impact of nonmodifiable factors (e.g., the infection source) involved in the pathogenesis of MRSA BSI. We found that the vancomycin MIC can have important prognostic value for vancomycin-treated patients. Furthermore, the method for determination of the MIC was found to be of importance in this study. However, we also found significant discordance between the three methods with respect to the vancomycin MIC values determined, with the Etest commonly reporting MIC values 1.5 to 2 times higher than those reported by the gold standard, manual broth microdilution (11, 28). We evaluated all three methods and found the vancomycin MIC determined by Vitek 2 to be the value most predictive of clinical failure, a result that is also supported by a recent study by Hsu et al. (11). As recent attention has focused on poorer outcomes associated with higher vancomycin MICs that are still within the range indicating susceptibility of the organism, these findings suggest that automated MIC testing may be preferable to manual methods for the purposes of outcome prediction. In patients with higher severity of illness (APACHE-II score ⱖ 14) and low-risk sources of infection, the majority of the infections were related to catheter use, and strain type became important in the prediction of outcome. Those infected with USA300 MRSA, all of which were catheter-related infections, had particularly poor outcome rates (36% clinical failure, 27% mortality) in comparison to the rates determined for those with non-USA300 MRSA strains, who had far better outcomes (4% clinical failure rate, 0% mortality rate). Although both groups had health care-associated disease, the data clearly suggest a strain-related influence on outcome for these patients. The USA300 group also had a higher proportion of isolates with a vancomycin MIC of 2 g/ml by Vitek 2 (27%). These findings could suggest decreased activity of vancomycin or could be evidence of a changing epidemiology and adaptation of USA300 from a community-associated strain commonly causing infections of the skin and soft tissue in younger patients to that of a strain occurring in older, sicker patients with health care risk factors. Recent studies have documented the considerable presence of USA300 in the health care setting, and the emergence of resistance to other antimicrobials in USA300 MRSA was recently reported, with investigators suggesting horizontal plasmid transfer from nosocomial USA100 MRSA isolates (22, 32). Previous work by the current investigators has demonstrated higher clinical failure rates in the treatment of USA300 MRSA infections when a health care association was present compared to communityassociated disease (25). The results from this study highlight an important relationship and perhaps an emerging pattern of concern in the epidemiology of USA300 MRSA. In this study we did identify a large subset of vancomycin-
4587
treated patients who are at a markedly low risk of failure, in particular, those with an APACHE-II score ⬍ 14 and an isolate with a vancomycin MIC of 1 g/ml or less. And yet measurements of health care utilization such as length of stay and treatment duration were found to be comparable to those determined for the members of the other groups who experienced poorer outcomes. This group tended to have more community-associated disease, a skin or wound source of BSI, a higher proportion of intravenous drug users, and BSI with strains with lower vancomycin MICs. These patients may be an ideal population for the evaluation of strategies aimed at reducing unnecessary exposure to antibiotics or potential opportunities to reduce health care utilization. Examples of this might include examining shorter courses of therapy or a more expedient transition to the outpatient setting. The weaknesses of this study relate to its retrospective nature and that it was conducted at a single center, and the external validity may be limited, as the study population was younger and mostly African-American and had a high prevalence of IVDA. However, the consecutive nature of patient selection and the extensive patient, strain, infection, and treatment information collected and analyzed are strengths of the study. Previous studies have traditionally lacked complete strain information; we have found that such information is an important factor in the prediction of outcome. By developing this risk stratification model, we have uncovered the complex interplay of factors associated with outcome in MRSA BSI, particularly the outcome for patients with multiple predictive factors. The lack of such factors has been an important weakness of previously conducted studies, which traditionally used linear regression analyses, limiting their application to clinical practice. The present-day population experiencing MRSA BSI is diverse and heterogeneous, which highlights the value of a risk stratification tool that can be applied to clinical practice. In addition to identifying those at high risk of clinical failure, we have also identified a large population of vancomycin-treated patients at particularly low risk of clinical failure. Further investigation into both of these groups is warranted. This study ultimately found that the interplay of factors involved in the etiology and management of MRSA BSI is complex, which highlights the limitations of a “one size fits all” approach.
ACKNOWLEDGMENTS This work was supported through an Investigator Initiated Research grant provided by Cubist Pharmaceuticals. The design and conduct of the study, including collection, management, analysis, and interpretation of the data and preparation of the manuscript, was conducted solely by us. Cubist Pharmaceuticals reviewed and approved the final manuscript. Carol L. Moore (principal investigator [Henry Ford Hospital]) and Mei Lu (independent statistician [Henry Ford Hospital]) had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Cubist Pharmaceuticals did not have access to study data. Potential conflicts of interest are as follows. C.L.M. has received research funding from Cubist Pharmaceuticals. N.Z.H. has received research funding from Pfizer. M.J.Z. has received research funding from Cubist Pharmaceuticals, Pfizer, and Johnson and Johnson and is a member of the speaker’s bureau for Cubist Pharmaceuticals and Astellas. M.L., F.C., P.O.-K., M.B.P., and S.D. have no conflicts to disclose.
4588
MOORE ET AL. REFERENCES
1. Ariano, R. E., A. Fine, D. S. Sitar, S. Rexrode, and S. A. Zelenitsky. 2005. Adequacy of a vancomycin dosing regimen in patients receiving high-flux hemodialysis. Am. J. Kidney Dis. 46:681–687. 2. Breiman, L. 1984. Classification and regression trees. Wadsworth International Group, Belmont, CA. 3. Charles, P. G., P. B. Ward, P. D. Johnson, B. P. Howden, and M. L. Grayson. 2004. Clinical features associated with bacteremia due to heterogeneous vancomycin-intermediate Staphylococcus aureus. Clin. Infect. Dis. 38:448– 451. 4. Chua, T., et al. 2008. Molecular epidemiology of methicillin-resistant Staphylococcus aureus bloodstream isolates in urban Detroit. J. Clin. Microbiol. 46:2345–2352. 5. CLSI. 2005. Performance standards for antimicrobial susceptibility testing: 15th informational supplement, document M100-S15. CLSI, Wayne, PA. 6. CLSI. 2007. Performance standards for antimicrobial susceptibility testing: 17th informational supplement. CLSI, Wayne, PA. 7. Fowler, V. G., Jr., et al. 1999. Infective endocarditis due to Staphylococcus aureus: 59 prospectively identified cases with follow-up. Clin. Infect. Dis. 28:106–114. 8. Frankel, M. R., et al. 2000. Predicting prognosis after stroke: a placebo group analysis from the National Institute of Neurological Disorders and Stroke rt-PA Stroke Trial. Neurology 55:952–959. 9. Hidayat, L. K., D. I. Hsu, R. Quist, K. A. Shriner, and A. Wong-Beringer. 2006. High-dose vancomycin therapy for methicillin-resistant Staphylococcus aureus infections: efficacy and toxicity. Arch. Intern. Med. 166:2138–2144. 10. Horan, T., and, G. R. P. 2004. Surveillance of nosocomial infections, 3rd ed. Lippincott Williams and Wilkins, Philadelphia, PA. 11. Hsu, D. I., et al. 2008. Comparison of method-specific vancomycin minimum inhibitory concentration values and their predictability for treatment outcome of meticillin-resistant Staphylococcus aureus (MRSA) infections. Int. J. Antimicrob. Agents 32:378–385. 12. Jeffres, M. N., W. Isakow, J. A. Doherty, S. T. Micek, and M. H. Kollef. 2007. A retrospective analysis of possible renal toxicity associated with vancomycin in patients with health care-associated methicillin-resistant Staphylococcus aureus pneumonia. Clin. Ther. 29:1107–1115. 13. Karam, C. M., P. S. McKinnon, M. M. Neuhauser, and M. J. Rybak. 1999. Outcome assessment of minimizing vancomycin monitoring and dosing adjustments. Pharmacotherapy 19:257–266. 14. Klevens, R., et al. 2007. Invasive methicillin-resistant Staphylococcus aureus infections in the United States. JAMA 298:1763–1771. 15. Klevens, R. M., et al. 2006. Community-associated methicillin-resistant Staphylococcus aureus and healthcare risk factors. Emerg. Infect. Dis. 12: 1991–1993. 16. Kullar, R., S. L. Davis, D. P. Levine, and M. J. Rybak. 2011. Impact of vancomycin exposure on outcomes in patients with methicillin-resistant Staphylococcus aureus bacteremia: support for consensus guidelines suggested targets. Clin. Infect. Dis. 52:975–981. 17. Lina, G., et al. 1999. Involvement of Panton-Valentine leukocidin-producing Staphylococcus aureus in primary skin infections and pneumonia. Clin. Infect. Dis. 29:1128–1132. 18. Lodise, T., P. McKinnon, L. Swiderski, and M. Rybak. 2003. Outcomes analysis of delayed antibiotic treatment for hospital-acquired Staphylococcus aureus bacteremia. Clin. Infect. Dis. 36:1418–1423. 19. Lodise, T. P., B. Lomaestro, J. Graves, and G. L. Drusano. 2008. Larger vancomycin doses (at least four grams per day) are associated with an increased incidence of nephrotoxicity. Antimicrob. Agents Chemother. 52: 1330–1336. 20. Lodise, T. P., N. Patel, B. M. Lomaestro, K. A. Rodvold, and G. L. Drusano.
ANTIMICROB. AGENTS CHEMOTHER. 2009. Relationship between initial vancomycin concentration-time profile and nephrotoxicity among hospitalized patients. Clin. Infect. Dis. 49:507– 514. 21. McDougal, L., et al. 2003. Pulsed-field gel electrophoresis typing of oxacillinresistant Staphylococcus aureus isolates from the United States: establishing a national database. J. Clin. Microbiol. 41:5113–5120. 22. McDougal, L. K., et al. 2010. Emergence of resistance among USA300 methicillin-resistant Staphylococcus aureus isolates causing invasive disease in the United States. Antimicrob. Agents Chemother. 54:3804–3811. 23. Moise, P. A., G. Sakoulas, A. Forrest, and J. J. Schentag. 2007. Vancomycin in vitro bactericidal activity and its relationship to efficacy in clearance of methicillin-resistant Staphylococcus aureus bacteremia. Antimicrob. Agents Chemother. 51:2582–2586. 24. Moise-Broder, P., et al. 2004. Accessory gene regulator group II polymorphism in methicillin-resistant Staphylococcus aureus is predictive of failure of vancomycin therapy. Clin. Infect. Dis. 38:1700–1705. 25. Moore, C., et al. 2009. Comparative evaluation of epidemiology and outcomes of methicillin-resistant Staphylococcus aureus (MRSA) USA300 infections causing community- and healthcare-associated infections. Int. J. Antimicrob. Agents 34:148–155. 25a.Moore, C., et al. 2009. Abstr. 19th Eur. Congr. Clin. Microbiol. Infect. Dis., Helsinki, Finland, May 2009, abstr. P-1861. 26. NCCLS. 1999. Methods for determining bactericidal activity of antimicrobial agents; approved guideline. NCCLS document M26-A. NCCLS, Wayne, PA. 27. Rybak, M. J., et al. 2009. Vancomycin therapeutic guidelines: a summary of consensus recommendations from the infectious diseases Society of America, the American Society of Health-System Pharmacists, and the Society of Infectious Diseases Pharmacists. Clin. Infect. Dis. 49:325–327. 28. Sader, H. S., P. R. Rhomberg, and R. N. Jones. 2009. Nine-hospital study comparing broth microdilution and Etest method results for vancomycin and daptomycin against methicillin-resistant Staphylococcus aureus. Antimicrob. Agents Chemother. 53:3162–3165. 29. Sakoulas, G., et al. 2004. Relationship of MIC and bactericidal activity to efficacy of vancomycin for treatment of methicillin-resistant Staphylococcus aureus bacteremia. J. Clin. Microbiol. 42:2398–2402. 30. Soriano, A., et al. 2008. Influence of vancomycin minimum inhibitory concentration on the treatment of methicillin-resistant Staphylococcus aureus bacteremia. Clin. Infect. Dis. 46:193–200. 31. Strommenger, B., C. Cuny, G. Werner, and W. Witte. 2004. Obvious lack of association between dynamics of epidemic methicillin-resistant Staphylococcus aureus in central Europe and agr specificity groups. Eur. J. Clin. Microbiol. Infect. Dis. 23:15–19. 32. Tenover, F. C., and R. V. Goering. 2009. Methicillin-resistant Staphylococcus aureus strain USA300: origin and epidemiology. J. Antimicrob. Chemother. 64:441–446. 33. Wisplinghoff, H., et al. 2004. Nosocomial bloodstream infections in US hospitals: analysis of 24,179 cases from a prospective nationwide surveillance study. Clin. Infect. Dis. 39:309–317. 34. Wootton, M., A. MacGowan, T. Walsh, and R. Howe. 2007. A multicenter study evaluating the current strategies for isolating Staphylococcus aureus strains with reduced susceptibility to glycopeptides. J. Clin. Microbiol. 45: 329–332. 35. Zhang, H., and B. Singer. 1999. Recursive partitioning in the health sciences. Springer, New York, NY. 36. Zhang, K., J. McClure, S. Elsayed, T. Louie, and J. Conly. 2005. Novel multiplex PCR assay for characterization and concomitant subtyping of staphylococcal cassette chromosome mec types I to V in methicillin-resistant Staphylococcus aureus. J. Clin. Microbiol. 43:5026–5033.
