Pharmacokinetic/Pharmacodynamic Determinants of Vancomycin Efficacy in Enterococcal Bacteremia

ABSTRACT

While pharmacokinetic-pharmacodynamic targets for vancomycin therapy are recognized for invasive methicillin-resistant Staphylococcus aureus infections, scant data are available to guide therapy for other Gram-positive infections. A retrospective single-center cohort of patients with Enterococcus bacteremia hospitalized between 1 January 2009 and 31 May 2015 were studied. The average vancomycin AUC_0–24 was computed using a Bayesian approach. The MIC was determined by gradient diffusion (Etest; bioMérieux), and the average AUC_0–24/MIC value over the initial 72 h of therapy was calculated. We assessed 30-day all-cause mortality as the primary outcome. Classification and regression tree analysis (CART) was used to identify the vancomycin AUC_0–24/MIC value associated with 30-day mortality. Fifty-seven patients with enterococcal bacteremia (32 E. faecium, 21 E. faecalis, and 4 other Enterococcus spp.) were studied. The median vancomycin MIC was 0.75 mg/liter (range, 0.38 to 3 mg/liter). All-cause 30-day mortality occurred in 10 of 57 patients (17.5%). A CART-derived vancomycin AUC/MIC_Etest value of ≥389 was associated with reduced mortality (P = 0.017); failure to achieve this independently predicted 30-day mortality (odds ratio, 6.83 [95% confidence interval = 1.51 to 30.84]; P = 0.01). We found that a vancomycin AUC/MIC_Etest value of ≥389 achieved within 72 h was associated with reduced mortality. Larger, prospective studies are warranted to verify the vancomycin pharmacodynamic targets associated with maximal clinical outcomes and acceptable safety.

INTRODUCTION

It is widely accepted that the target vancomycin trough concentration for the treatment of methicillin-resistant Staphylococcus aureus (MRSA) bacteremia is 15 to 20 μg/ml to achieve an AUC/MIC of 400 (1, 2). Vancomycin is also used in the treatment of other Gram-positive organisms, namely, Enterococcus spp., coagulase-negative Staphylococcus, and Streptococcus spp. This is especially true in the event of resistance or patient allergic reactions to beta-lactams. However, there is a paucity of evidence in the literature to guide the target vancomycin concentration in the treatment of non-MRSA Gram-positive infections.

The goal of this investigation was to determine whether vancomycin concentration targets (serum trough concentrations or ratio of the vancomycin area under the concentration-time curve to the MIC [AUC/MIC]) values within the first 72 h of therapy were associated with clinical outcomes for patients with bacteremia due to Enterococcus spp. The primary outcome measure was 30-day mortality. Secondary outcomes included days of bacteremia, relapse within 90 days, and incidence of nephrotoxicity.

RESULTS

From 1 January 2009 to 31 May 2015, there were 417 index cases of Enterococcus bacteremia. Of these, 174 had polymicrobial bacteremia, 145 received less than 3 days of vancomycin, 23 were vancomycin-resistant enterococci, 7 were without vancomycin concentrations, 5 were without vancomycin MIC data, and 6 were on renal replacement therapy. Therefore, 57 cases were included in the analysis.

The median age of this sample was 75 years (interquartile range [IQR], 62 to 84 years), and the median APACHE II score was 12 (IQR, 8.5 to 17). Three common comorbidities were diabetes 21 cases (36.8%), malignancy 18 cases (31.6%), and heart disease 14 cases (25.6%). Twenty patients (35.1%) had a baseline creatinine clearance of less than 30 ml/min. An infectious diseases specialist consult was obtained in 33 cases (57.8%), and 8 patients (14%) were admitted to the intensive care unit. Of the 57 patients, 14 (24.6%) had an unknown source of bacteremia, and the two common sources of bacteremia were intra-abdominal (18; 31.6%) and urinary (n = 13; 22.8%). Concurrent infection occurred in 37 patients.

A total of 62 vancomycin concentrations obtained within the first 96 h of vancomycin administration were reported. The median daily vancomycin dose was 1,000 mg (IQR, 500 to 1,875 mg). The median duration of total vancomycin therapy was 14 days (IQR, 9.5 to 16 days). Concomitant antimicrobial therapy was observed in 52 (91.2%) patients, and 20 (35.1%) received concurrent aminoglycosides. Thirty-five (61.4%) cases received concurrent nephrotoxins.

Of the cases, there were 32 (56.1%) Enterococcus faecium, 21 (36.8%) Enterococcus faecalis, and 4 (7.0%) consisted of other Enterococcus spp. The MIC distributions for vancomycin according to Etest and broth microdilution are displayed in Table 1. Majority (87.7%) of the MIC was <1.5 mg/liter.

TABLE 1.

Vancomycin MIC distribution according to Etest and BMD (n = 57)^a

Etest MIC (mg/liter)	No. of isolates with a BMD MIC of:			Total no. of isolates
	≤0.5 mg/liter	1 mg/liter	2 mg/liter
0.38	2	1		3
0.5	13	8		21
0.75	6	5		11
1	4	11		15
1.5		1		1
2		3	1	4
3			2	2
Total	25	29	3	57

^aBMD, broth microdilution.

The overall 30-day mortality rate in our study was 17.5% (10 of 57 cases). Baseline characteristics were similar between survivors and nonsurvivors (Table 2). There was no statistically significant difference in terms of bacteremia source, comorbidities, concurrent infections, and severity of illness. Rates of nephrotoxicity, days of bacteremia, and days to fever defervescence were similar between nonsurvivors and survivors (Table 3). There was only one case of relapse at 90 days. Survivors had statistically longer duration of vancomycin therapy, while the vancomycin trough was not significantly different between the groups (Table 3).

TABLE 2.

Baseline demographic and clinical features according to 30-day mortality and vancomycin AUC/MIC_Etest target attainment^a

Characteristic	30-day mortality			Vancomycin AUC/MICEtest
	Survivors (n = 47)	Nonsurvivors (n = 10)	P	<389.08 (n = 11)	≥389.08 (n = 46)	P
Demographics
Median age in yrs (IQR)	74.0 (61.0–82.0)	82.5 (66.3–88.5)	0.24	80.0 (60.0–88.0)	72.5 (62.5–82.3)	0.37
Male	26 (55.3)	5 (50)	1.00	6 (54.5)	25 (54.3)	0.99
Chinese ethnicity	40 (85.1)	9 (90)	1.00	10 (90.9)	39 (84.8)	1.00
Median baseline CLCR in ml/min (IQR)	42.9 (30.7–73.9)	38.8 (13.7–80.5)	0.46	22.1 (14.0–90.5)	43.0 (30.8–71.6)	0.39
Source of bacteremia
Intra-abdominal	15 (31.9)	3 (30)	1.00	1 (5.6)	17 (37.0)	0.15
Urinary	12 (25.5)	1 (10)	0.43	3 (27.3)	10 (21.7)	0.70
Unknown	10 (21.3)	4 (40)	0.24	3 (27.3)	11 (23.9)	1.00
Line related	2 (4.3)	0 (0.0)	1.00	0 (0)	2 (4.3)	1.00
Infective endocarditis	3 (6.4)	0 (0.0)	1.00	1 (9.1)	2 (4.3)	0.48
Skin and soft tissue	2 (4.3)	1 (10)	0.45	2 (18.2)	1 (2.2)	0.09
Central nervous system	2 (4.3)	0 (0.0)	1.00	1 (9.1)	1 (2.2)	0.35
Pneumonia	0 (0.0)	1 (10)	0.18	0 (0)	1 (2.2)	1.00
Comorbidities
Malignancy	17 (36.2)	1 (10)	0.15	2 (18.2)	16 (34.8)	0.47
Diabetes	17 (36.2)	4 (40)	1.00	6 (54.5)	15 (32.6)	0.30
Heart disease	11 (23.4)	3 (30)	0.69	2 (18.2)	12 (26.1)	0.71
Liver disease	4 (8.5)	2 (20)	0.28	1 (9.1)	5 (10.9)	1.00
Dementia	4 (8.5)	1 (10)	1.00	1 (9.1)	4 (8.7)	1.00
Immunosuppression	4 (8.5)	0 (0)	1.000	0 (0)	4 (8.7)	0.58
Median severity of illness markers (IQR)
APACHE II score	12.0 (8.0–16.0)	16.5 (9.0–20.3)	0.18	16.0 (13.0–19.0)	11.0 (8.0–16.0)	0.02
Pitt bacteremia score	1 (0–2)	2 (1–3.75)	0.07	2 (1–5)	1 (0–2)	0.02
Charlson's weighted index	3 (1–5)	3 (1.75–3.25)	0.95	3 (2–3)	3 (1–5)	0.71
Exposures
ICU admission	5 (71.4)	3 (28.6)	0.59	2 (18.2)	5 (10.9)	0.61
ID consult	28 (59.6)	5 (50)	0.72	5 (45.5)	28 (60.9)	0.50
Concurrent infection	31 (66)	6 (60)	0.73	9 (81.8)	28 (60.9)	0.30
Site
Urinary	14 (29.8)	2 (20)		5 (45.5)	11 (23.9)
Lung	11 (23.4)	3 (30)		2 (18.2)	12 (26.1)
Hepatobiliary	5 (10.6)	0		0	5 (10.9)
Central nervous system	1 (0.02)	1 (10)		2 (18.2)	0
Other
Concurrent antibiotics	43 (91.5)	9 (90)	1.00	10 (90.9)	42 (91.3)	1.00
Concurrent aminoglycosides	16 (34)	4 (40)	0.73	7 (63.6)	13 (28.3)	0.04
Median days of aminoglycosides (IQR)	0 (0–1)	0 (0–1)	0.96	1 (0–1)	0 (0–1)	0.06
Concurrent nephrotoxins	27 (57.4)	8 (80)	0.29	8 (72.7)	27 (58.7)	0.50
Other sites with Enterococcus spp.	14 (29.8)	1 (10)	0.26	4 (36.4)	11 (23.9)	0.46

^aAbbreviations: AG, aminoglycosides; APACHE, acute physiology and chronic health evaluation; AUC, area under the concentration-time curve; CL_CR, creatinine clearance; CWI, Charlson weighted index of comorbidity; ICU, intensive care unit; ID, infectious disease physician; IQR, interquartile range. All data are presented as the number (%) unless otherwise stated.

TABLE 3.

Outcomes and vancomycin parameters according to 30-day mortality and vancomycin AUC/MIC_Etest target attainment^a

Characteristic	30-day mortality			Vancomycin AUC/MICEtest
	Survivors (n = 47)	Nonsurvivors (n = 10)	P	<389.08 (n = 11)	≥389.08 (n = 46)	P
Outcomes
30-day mortality	0 (0)	10 (100)		5 (45.5)	5 (10.9)	0.02
90-day relapse	1 (2.1)			0 (0)	1 (2.2)	1.00
Median days of bacteremia (IQR)	4.0 (3.0–6.0)	5.0 (3.5–6.0)	0.60	5 (3–6)	4.0 (3.0–6.0)	0.61
Median days to fever defervescence (IQR)	1 (0–1)	1 (0–1.25)	0.93	1 (0–1)	1 (0–1)	0.75
Nephrotoxicity	2 (4.3)	2 (20.0)	0.14	1 (9.1)	3 (6.5)	1.00
Median vancomycin parameters (IQR)
Vancomycin duration (days)	15.0 (12.0–17.0)	10.5 (7.25–12.0)	0.03	10 (5–15)	14.5 (11.0–17.0)	0.03
Daily vancomycin dose (mg)	1,000 (500–2,000)	625 (500–1,500)	0.32	500 (500–2,000)	1,000 (500–1,562.5)	0.81
Vancomycin trough (μg/ml)	11.4 (6.9–16.0)	8.6 (6.1–13.7)	0.56	8.4 (4.5–10.3)	12.0 (7.1–16.2)	0.07
AUC0–24 (h·mg/liter)	470.1 (366.3–596.1)	377.0 (316.9–643.4)	0.45	386.7 (302.8–566.3)	474.6 (366.2–601.9)	0.17
MIC (Etest)	0.75 (0.5–1.0)	1 (0.69–1.25)	0.14	1.5 (1.0–2.0)	0.5 (0.5–1.0)	<0.001
Vancomycin AUC/MICEtest	670.0 (479.0–858.5)	418.0 (296.2–766.7)	0.07

^aAbbreviations: AUC, area under the concentration-time curve; BMD, broth microdilution; IQR, interquartile range. Data are presented as number (%) unless otherwise stated.

Using classification and regression tree (CART) analysis, the breakpoint for 30-day all-cause mortality using Etest was identified for the vancomycin AUC/MIC to be at 389. The differences in baseline characteristics between patients with an AUC/MIC_Etest value of <389 and those with an AUC/MIC_Etest value of ≥389 were not statistically different for all variables evaluated, except for APACHE II score (16 versus 11, respectively [P = 0.02]), Pitt bacteremia score (2 versus 1, respectively [P = 0.022]), and median MIC_Etest (1.5 versus 0.5, respectively [P < 0.001]). Patients with an AUC/MIC_Etest of <389 had a 3-fold increase in 30-day mortality compared to those who achieved an AUC/MIC_Etest of ≥389 (45.5 versus 10.9%, respectively [P = 0.02]) (Fig. 1). Other outcomes were not statistically different (Table 3).

FIG 1.

Distribution of AUC₂₄/MIC attained and mortality. Bars: black, survivors; gray, nonsurvivors.

The multivariable logistic regression analysis showed that an AUC/MIC_Etest value of <389 remained an independent predictor of 30-day mortality (odds ratio, 6.83; 95% confidence interval = 1.51 to 30.84; P = 0.01 [Hosmer-Lemeshow statistic for final model, P = 0.284]) (Table 4).

TABLE 4.

Final multivariate logistic regression model of factors associated with 30-day mortality^a

Characteristic	Univariate analysis			Multivariate analysis
	OR	95% CI	P	OR	95% CI	P
APACHE II score	1.09	0.98–1.20	0.13	1.06	0.93–1.20	0.40
Malignancy	0.20	0.23–1.68	0.15	0.29	0.03–2.72	0.28
Nephrotoxicity	5.63	0.69–45.90	0.14	6.59	0.66–65.42	0.11
MICEtest	2.13	0.77–5.90	0.14	0.41	0.075–2.27	0.31
AUC/MICEtest, <389.08	6.83	1.51–30.84	0.02	6.83	1.51–30.84	0.01

^aAbbreviations: APACHE, acute physiology and chronic health evaluation; AUC, area under the concentration-time curve; CI, confidence interval; OR, odds ratio. The Pitt bacteremia score was not considered for multivariable analysis (correlated with APACHE II). Hosmer-Lemeshow statistic for final model, P = 0.284.

DISCUSSION

In this study, we sought to identify the vancomycin pharmacodynamic target associated with clinical outcomes among patients with Enterococcus bacteremia. We used gradient diffusion (Etest) to determine the MICs for our study since this is readily accessible and practical for most clinical microbiology laboratories, compared to the CLSI broth microdilution (BMD) method (3). Gradient diffusion, although generally generating MICs which are higher than BMD methods, have been suggested to better predict clinical success/failure with S. aureus infections (4). In our study, we also determined vancomycin MICs via a commercial automated BMD method with the Vitek II system and found 100% categorical agreement for these vancomycin-susceptible isolates with the Etest (Table 1). Given that the lowest dilution for vancomycin on our Vitek II panel was 0.5 mg/liter, we elected to perform our AUC/MIC analysis using results from the Etest.

Using CART analysis, patients who received vancomycin treatment regimens that resulted in an AUC/MIC_Etest ratio of <389 had a 3-fold-higher 30-day mortality compared to those who had AUC/MIC_Etest ratio of ≥389 (P = 0.017). Our study is consistent with a Canadian study with respect to overall 30-day mortality rates (5).

Previous preclinical and clinical studies suggested that the AUC/MIC ratio is the pharmacodynamic parameter best associated with vancomycin effectiveness (6, 7). To our knowledge, this is the first study examining the relationship between vancomycin AUC/MIC_Etest and clinical outcomes in patients with Enterococcus bacteremia.

Besides the MIC determination method (e.g., Etest versus BMD), the AUC/MIC ratio is also dependent on the method of AUC calculation. Calculation of the actual AUC requires multiple concentration measurements, which is not feasible in routine clinical practice. Using PK models and Bayesian method, one can estimate the AUC from a few concentration measurements. For the calculation of AUC, we used a version of BestDose software made available at the Laboratory of Applied Pharmacokinetics and Bioinformatics website. The Bayesian method of AUC calculation is more accurate as it combines a priori information obtained from a population model with a posteriori individual patient's latest pharmacokinetic data (8). Furthermore, vancomycin trough concentration need not be obtained under steady-state conditions unlike in the case when using the traditional first-order equation method. This is desirable in the daily unpredictable patient care setting (9). However, pharmacokinetic knowledge is required to interpret the output of the program. To obtain appropriate pharmacokinetic results, patient-specific information such as serum creatinine levels, weight and height, drug administration details, and concentration levels need to be obtained (8).

Among the 11 cases who did not attain an AUC/MIC of 389.08, the median vancomycin MIC was significantly higher (1.5 versus 0.5 mg/liter; P < 0.001); on the other hand, these patients also received a lower daily vancomycin dose (500 mg versus 1,000 mg, P = 0.805); this could have led to the lower AUC/MIC attained. The shorter duration of vancomycin therapy in the nonsurvivors was driven by mortality. It did not have an impact on the AUC/MIC attained since the AUC/MIC was estimated within the first 72 h of vancomycin therapy, and all patients received at least 3 days of vancomycin in this study.

Vancomycin trough concentration has been used as a surrogate marker for efficacy. Guidelines recommend that for intermittent dosing, target trough concentrations be maintained at 15 to 20 μg/ml for treatment of complicated MRSA infections such as hospital-acquired pneumonia (2). In this present study, median vancomycin trough concentration was not significantly different between the survivors and nonsurvivors (11.4 μg/ml versus 8.6 μg/ml; P = 0.56). Similarly, various other studies also showed no association between vancomycin tough concentration and treatment outcomes for the treatment of S. aureus bacteremia (10, 11). To ensure optimal dosing, an AUC should be calculated instead.

Neely et al. proposed that a vancomycin AUC greater than 700 mg·h/liter increases the risk of nephrotoxicity (12). In our study, four patients (7%) developed nephrotoxicity, one of whom had an AUC of 749.27 mg·h/liter. This means that in order to safely achieve AUC/MIC_Etest ratio of >389, the MIC_Etest for Enterococcus spp. has to be ≤1.5. In our sample, 89.4% of the Enterococcus spp. isolated had an MIC_Etest of ≤1.5. Therefore, an AUC/MIC_Etest of >389 can be safely attained in the majority of cases in our institution. On the other hand, given the difficulty and risk of nephrotoxicity in attaining an AUC/MIC_Etest ratio of >389 for cases with a vancomycin MIC_Etest of ≥1.5 mg/liter, perhaps an alternative agent should be considered for treatment of enterococcus bacteremia in these patients, even when the current susceptibility breakpoint for vancomycin is 4 mg/liter. Future studies should be conducted to further evaluate this.

Other than using a computer program to improve the AUC estimation, our study measured potential confounders such as comorbidities and illness severity scores. We also included patients with various sources of bacteremia. Nonetheless, our study has several limitations. Since this is a retrospective observational study, there may be confounders and bias introduced that can be avoided with a prospective study design. Since the data were collected during routine patient care, parameters such as weight and height may not have been measured precisely, although this mimics real world practice. Similarly, blood cultures were not repeated daily in some of the patients, hence the actual days of bacteremia may have been shorter. This study was carried out in one institution, and thus our results may not be applicable to other centers. It should be noted that we estimated the AUC₂₄ as an average of the AUC in the first 72 h of vancomycin therapy instead of reporting AUC attained on specific days. The sample size of this study is small and could have limited our ability to identify further significant factors that predicts mortality. Using CART analysis, even though we were able to identify the AUC/MIC_Etest breakpoint that maximize the mortality benefit, this should be confirmed in a larger prospective study.

In conclusion, we identified that the pharmacodynamic target of vancomycin AUC/MIC is associated with mortality in Enterococcus bacteremic patients. An AUC/MIC_Etest ratio of <389 was associated with an increased mortality in our study. Since this was a retrospective study, larger prospective studies with various MIC determination methods, including BMD, are warranted to determine the vancomycin pharmacodynamic target associated with maximal clinical outcomes and acceptable safety.

MATERIALS AND METHODS

Study design.

We performed a retrospective, single-center, observational cohort study. Patients hospitalized at Tan Tock Seng Hospital (TTSH) with Enterococcus bacteremia treated with vancomycin between 1 January 2009 and 31 May 2015 were included in the analysis. Bacteremia was defined as ≥2 separate blood cultures positive for Enterococcus spp. or 1 positive blood culture with an identifiable source in a clinical scenario consistent with bacteremia (13). For patients with more than one episode of Enterococcus bacteremia, only the first episode during the study period was included. This study was approved by our hospital's institutional review board.

Study population.

The exclusion criteria were as follows: patients younger than 21 years, pregnancy, patients on renal replacement therapy, the presence of polymicrobial bacteremia, isolates resistant to vancomycin, a history of vancomycin therapy for less than 3 days, and a lack of trough vancomycin concentration and vancomycin MIC data.

Data collection.

A list of patients with Enterococcus bacteremia was generated through retrospective query of the Antimicrobial Resistance and Utilization Surveillance (ARUS) database at TTSH. Electronic inpatient medication records (eIMR) were accessed to determine whether the patients had received intravenous vancomycin within 72 h of the infection. Patients who met any exclusion criteria were removed from the study.

Data collected from patients' medical records included demographic characteristics, comorbidities, APACHE II (14), Pitt bacteremia score (15), and Charlson weighted index of comorbidity (CWI) score (16) within 24 h of the day of positive index blood culture, source of bacteremia, antimicrobial treatment data, concurrent use of potential nephrotoxins such as aminoglycosides, nonsteroidal anti-inflammatory drugs and radioactive contrast agents administered within 1 week of vancomycin, response to vancomycin as measured by duration of bacteremia, relapse of bacteremia within 90 days, or the occurrence of nephrotoxicity and mortality; it was also noted if an infectious disease specialist consult was requested.

The 30-day mortalities were counted from the date of first positive Enterococcus blood culture. Baseline renal function was estimated using the Cockcroft-Gault equation to estimate creatinine clearance (17). Nephrotoxicity was defined as an increase of 44.2 μmol/liter or 50% or more in the baseline serum creatinine level in two consecutive laboratory tests, whichever was greater (2), from the initiation of vancomycin to 7 days after the cessation of vancomycin therapy. The duration of bacteremia was determined by the number of days from first positive blood culture to the date of the first negative Enterococcus blood culture (18). A relapse was considered if cultures became negative for more than 2 days and then reappeared positive within 90 days (19).

Microbiological data.

All blood culture isolates were stored at −80°C in the department of laboratory medicine at TTSH. Only the first positive blood culture from each selected patient was retrieved for subsequent testing. All isolates were confirmed as Enterococcus spp. according to standard methods and vancomycin susceptibility was determined according to Clinical Laboratory and Standards Institute (CLSI) guidelines, via disk diffusion testing, at diagnosis (20).

For purposes of this study, further vancomycin MIC determinations were performed on retrieved archived isolates using a commercial automated broth microdilution (aBMD) method using the Vitek II system (bioMérieux Vitek, Inc., Hazelwood, MO, USA) and by gradient diffusion (Etest, bioMérieux) in accordance with the manufacturer's instructions. All MICs were read and confirmed by 2 operators who were blinded with respect to the clinical outcomes of the patients from whom the isolates had been obtained.

Vancomycin dosing and pharmacodynamic data.

Patients were initially treated with 15 to 20 mg/kg per dose of vancomycin, according to actual body weight. The dosing interval was selected and adjusted based on renal function according to our institutional guideline: patients with creatinine clearances (CL_CRs) of ≥50 ml/min, 30 to 49 ml/min, 10 to 29 ml/min, and ≤10 ml/min were dosed at intervals of 12, 24, 48, and 96 h, respectively. Dosing regimens were adjusted by clinical pharmacists to achieve a serum trough concentration of 15 to 20 μg/ml based on hospital guidelines. Our institution guideline recommends monitoring vancomycin trough concentrations (to be obtained ≤1 h before administration of a dose) and within 96 h of vancomycin initiation. Since this was an observational study, the vancomycin concentrations were determined at the discretion of the clinicians. The dates and times of vancomycin dose administration and blood draw for vancomycin concentrations were recorded and entered into the BestDose software.

The 24-h vancomycin AUC (AUC₂₄) was calculated using the Bayesian approach. Individual pharmacokinetic parameters were estimated for each patient by BestDose software version 1.118 (21). This software has been used to individualize vancomycin dosing in patients (22). It requires (i) a model consisting of equations and probability distributions for the values of the variables (“parameters”) in equations that describe the population pharmacokinetics of vancomycin in adults (the “Bayesian prior”) and (ii) an individual patient's details such as weight, height, and serum creatinine values (“covariates”), administered vancomycin regimen, and observed drug concentrations (23). BestDose uses nonparametric statistics, such that the Bayesian prior consists of discrete “support points,” each of which is a collection of values for the parameters in the model (e.g., volume of distribution) and an associated probability of that set of parameter values. Based on the patient data, the probabilities of the support points were recalculated as the “Bayesian posterior” joint probability distribution for that patient (24). In effect, the population model (Bayesian prior) was updated for the patient.

The AUC in the first 72 h (AUC₇₂) was obtained from BestDose, which used the trapezoidal approximation algorithm on concentrations calculated approximately every hour from the median of the Bayesian posterior parameter values and the patient's dosing and covariate data. The AUC₇₂ of treatment was evaluated because early attainment of the target is important, especially in cases of bacteremia (25). The average AUC₂₄ was calculated by dividing the AUC₇₂ by 3. The AUC₂₄/MIC was calculated using the gradient diffusion MIC method (MIC_Etest). Since the vancomycin trough concentration may not have been obtained at the correct time, vancomycin serum trough concentrations were obtained from the software. The average of predicted vancomycin concentrations before each administered vancomycin dose was considered the vancomycin trough concentration.

Statistical analysis.

The relationships between the various vancomycin concentration targets and clinical outcomes were explored as both a continuous and a categorical variable. CART analysis was used to identify the AUC₂₄/MIC_Etest of vancomycin associated with 30-day mortality. CART is a set of analytic methods used to identify subgroups within a particular population that share similar characteristics that influence the dependent variable (26).

Categorical variables were compared using χ² test or the Fisher exact test. Continuous variables were tested for normality using Shapiro-Wilk test. A Mann-Whitney U test or Student t test was used for continuous variables where appropriate. Potentially significant variables identified on univariate analysis (P ≤ 0.2) were included into a multivariate model to identify the risk factors associated with 30-day mortality. If variables were correlated, only one was included into the analysis. The CART-derived two-group AUC/MIC_Etest ratio cutoff was used to create a dichotomous variable for predicting 30-day mortality with both uni- and multivariable regression. Goodness of fit of the final model was evaluated with the Hosmer-Lemeshow statistic. All statistical analyses were performed using SPSS Statistics, IBM SPSS software (version 23.0; SPSS, Inc., Chicago, IL), except for CART analysis, in which JMP software version 12.2.0 (SAS Institute, Inc., Cary, NC) was used.