Purpose. Seven methods for estimating vancomycin pharmacokinetic parameters were studied to determine which method best predicted measured concentrations for patients at a community teaching hospital.
Methods. Data from adult patients who were given vancomycin and had at least one steady-state trough concentration measured were retrospectively reviewed. Data analyzed included laboratory test values, concomitant medications, weight, height, sex, age, laboratory cultures, medical procedures performed, vancomycin dose and interval, measured vancomycin concentrations, and time of measurement. Relevant data were used in seven predictor methods that estimate volume of distribution, vancomycin clearance, and elimination rate constant to determine which yielded the best predictions of actual measured concentrations in the patient population.
Results. Data from 189 patients were included in the analyses. The coefficients of determination for the methods ranged from 0.114 to 0.234. Bias ranged from −5.90 to 0.69 mg/L, and precision ranged from 6.05 to 8.08. The Matzke method had the best combination of the least bias and best precision. Predictions were within 2.5 and 5 mg/L of measured concentrations 18.0–43.9% and 43.4–66.1% of the time, respectively. The percentage of predictions within 25% and 50% of measured concentrations ranged from 7.9% to 31.2% and from 18.0% to 48.1%, respectively. Ten (5.3%) patients had trough concentrations exceeding 20 mg/L, and 11 (5.8%) had trough concentrations of ≤3 mg/L.
Conclusion. The seven methods studied for estimating vancomycin pharmacokinetic parameters varied widely in predicting vancomycin trough concentrations compared with measured serum concentrations and were not sufficiently reliable to replace therapeutic monitoring of vancomycin serum concentrations.
- Blood levels
- Drugs, body distribution
- Rate constants
Vancomycin is a tricyclic glycopeptide antibiotic that is largely excreted unchanged in the urine.1 Because of this, patients with reduced renal function have decreased clearances of vancomycin and need reduced doses or increased dosing intervals. Common adverse reactions to vancomycin include hypotension, flushing, erythematous rash, and chills.2 More worrisome but much less common reactions include ototoxicity and nephrotoxicity, which are weakly associated with peak vancomycin concentrations above 45 mg/L, though patients developing ototoxicity or nephrotoxicity with vancomycin were often also taking concomitant medications that can cause these adverse effects.3–5 Unfortunately, the studies did not account for underlying diseases or conditions that could cause hearing loss or renal damage.
Patients’ medical problems and interpatient variation in pharmacokinetic parameters due to other causes necessitate the use of doseprediction methods that lead to desirable vancomycin serum concentrations. 6 Over the years of vancomycin use, dosing and monitoring approaches have been developed and evaluated, with some methods working better than others at predicting desired concentrations.
Some controversy surrounds the relationship between vancomycin concentration and therapeutic response. 4 There is also considerable controversy over the use of drug concentration measurements to guide vancomycin dosing decisions.5–7 In the past, it was fairly routine to measure both peak and trough concentrations; now, many clinicians monitor only the trough concentration or do not monitor drug concentrations at all.
Using vancomycin concentrations to monitor patients’ therapeutic response is not typically suggested if the duration of therapy is expected to be less than 72 hours or for patients receiving oral vancomycin. Drug monitoring is more commonly recommended for patients receiving other nephrotoxic drugs, burned patients, patients with central nervous system infections or endocarditis, i.v. drug abusers, patients with sepsis, and patients with rapidly changing renal function.8
There are a variety of methods to facilitate the dosing of vancomycin. These methods were designed to help clinicians treat their patients effectively while avoiding toxicities. For example, some methods have the potential for reducing drug therapy costs when longer intervals or smaller doses can be used. Four initial dosing approaches for vancomycin include (1) information found in the package insert, (2) the Nielsen method,9 (3) the Rotschafer method,10 and (4) the Lake method.11 These methods can be used to quickly start a patient on vancomycin. Initial doses and dosing intervals can then be individualized for patients by using measured concentrations when deemed necessary.
A number of predictor methods to determine dosing regimens based on estimations of a patient’s pharmacokinetic parameters for vancomycin have been developed. Some of the more commonly cited predictors are the Birt,8 Matzke,12 Burton,13 and Rodvold14 methods. Depending on the specific patient population, one method may be better than another for estimating that population’s vancomycin concentrations. Some of the calculation challenges associated with these predictors include selecting the best dosing weight and estimating vancomycin clearance from creatinine clearance (CLcr) (also estimated), especially in the presence of diminished renal function. Determining the appropriate weight to estimate CLcr and volume of distribution is increasingly important in the United States, as the population becomes increasingly obese.
Knowing how dosing methods perform for a given patient population can be helpful. One study evaluating the utility of three methods found that the Lake–Peterson method typically provided the best vancomycin dosing estimates for individuals with a CLcr above 15 mL/min, while the Matzke method was best for CLcr of ≤15 mL/min.15
Because vancomycin pharmacokinetic parameters can vary widely among individuals, it may be necessary to develop institution-specific, population-based dosing methods and monitoring approaches. Many clinicians now measure a single trough concentration to monitor therapeutic response, which does not allow the pharmacokinetic parameters (i.e., vancomycin clearance [CLvanco], volume of distribution [V], and elimination rate constant [k]) for individuals to be determined. Thus, use of previously developed nomograms and dosing approaches could be helpful if they consistently produced concentrations in desired ranges.
The purpose of this study was to determine whether any of seven methods for estimating the vancomycin pharmacokinetic parameters (V, CLvanco, and k) accurately predicted measured concentrations for patients at a 421-bed community teaching hospital in South Carolina.
This study retrospectively reviewed data from hospital patients who were given vancomycin and had at least one trough vancomycin concentration measured approximately 30 minutes before the administration of a steady-state dose (usually the third or fourth dose). This study was reviewed and approved by the human subjects committee of the University of Arizona where the data were analyzed. A site authorization letter was obtained from the hospital in South Carolina for use of the unidentifiable patient pharmacokinetic and dosing data. A convenience sample of patients treated with vancomycin between June 1, 2001, and October 8, 2004, and monitored by the pharmacokinetics service at the hospital was reviewed. Patients could have comorbid conditions and be taking other medications. Patients under age 18 years were excluded from this study because the predictor methods chosen were not intended for use in children or neonates. When patients had more than one reliably measured concentration, the first reliable trough concentration was used in the evaluation. Vancomycin concentrations were measured by competitive immunoassay (Advia Centaur, Bayer Health Care), which measures concentrations up to 90 mg/L and has a minimum detectable concentration of 0.67 mg/L.
All identifying information (patient names, history numbers, and physician names) was removed by hospital pharmacists before evaluation by the researchers. The data available included laboratory test values, concomitant medications, patient weight, height, sex, age, laboratory cultures, vancomycin dose and interval, time of administration, serum vancomycin concentrations, and time of vancomycin concentration measurement. The relevant information was used in seven predictor methods (appendix) to determine which best predicted patients’ actual measured vancomycin concentrations.
The patient information and dosing data were entered into an Excel spreadsheet (Microsoft, Redmond, WA) and used to estimate CLcr and dosing weight for each patient. When a model was clear on how CLcr was to be predicted (method and weight to be used), that CLcr was used for the primary prediction. In addition, CLcr was predicted using actual body weight (ABW), ideal body weight (IBW), and adjusted body weight (BWadj) for each patient. To predict CLcr using BWadj, ABW was used until ABW divided by IBW equaled 1.2. If ABW divided by IBW exceeded 1.2, BWadj was used. To determine if estimates of the concentrations could be improved by the use of different weights in the CLcr estimates, all three measures of CLcr were used to calculate CLvanco for each patient.
The steady-state trough concentrations were predicted using patient dosing information, the estimated pharmacokinetic parameters determined from each method, and the following equation:
where t′ = the duration of infusions (set at the hospital’s usual standard of 1 hour for doses of 500–999 mg, 1.5 hours for doses of 1000–1499 mg, and 2 hours for doses of 1500 mg or greater), τ = the dosage interval used, and t = the time when the concentration was measured after the end of the infusion.
The concentrations predicted by each method were compared with the measured concentrations using a variety of approaches. The coefficient of determination (r2) between predicted and actual concentrations was determined for each CLcr prediction for the group as a whole, for patients whose ABW divided by IBW was 1.2 or less, and for those whose ABW divided by IBW exceeded 1.2 to determine whether improvement occurred. The methods of Sheiner and Beal22 were used to see which method was most accurate (i.e., least biased and most precise) for this population. Precision (root mean-squared error [RMSE]) is an indication of how close predicted concentrations are to measured concentrations, and bias (mean error [ME]) is an indicator of whether predictions are, on average, higher or lower than the measured concentrations. Smaller numbers for precision and bias indicate greater precision and less bias when one method is compared with another. Relative precision (ΔRMSE) and relative bias (ΔME) were also calculated between the method judged to have the best combination of precision and the least bias overall and the other methods.
The predicted concentrations and percentages that were within 2.5 and 5 mg/L and within 25% and 50% of the measured concentrations were also determined to provide another measure of fit.
The records of 243 patients were examined. A total of 189 patients, 94 men and 95 women, were included in the final analyses. The remaining 54 were excluded because at least one piece of necessary information was missing. Patient demographics and dosing information are shown in Table 1⇓.
Of the 189 patients, 116 (61%) were more than 20% above their ideal weight, 58 (31%) were more than 50% above their ideal weight, and 30 (16%) were more than 75% above their ideal weight. The most popular dosing intervals were 12, 24, and 48 hours (n = 68, 68, and 33, respectively). The most common doses were 1000 mg (n = 117), 1500 mg (n = 23), and 1250 mg (n = 23). Other doses included 750 mg (n = 15), 500 mg (n = 8), 450 mg (n = 1), 1275 mg (n = 1), and 1750 mg (n = 1). Seventy-four patients (40%) had trough concentrations above 10 mg/L. Of these, 20 (11%) had trough concentrations above 15 mg/L, 10 (5%) of whom had trough concentrations above 20 mg/L. Eleven (6%) patients had trough concentrations of ≤3 mg/L.
The coefficients of determination for each of the seven methods and the various subgroups are shown in Table 2⇓. Use of the different weights (i.e., ABW, IBW, and BWadj) for CLcr estimation did not improve final prediction to a great degree (r2 ranged from 0.114 to 0.234). The r2 values improved in all prediction methods by removing the patients whose ABW divided by IBW was 1.2 or less. Though there was not a substantial difference between most of the methods, the Bauer method using IBW had the highest r2 for all patients, and the Ambrose method using ABW had the highest r2 when patients whose ABW divided by IBW was 1.2 or less were removed.
Table 3⇓ provides the precision, bias, relative precision, and relative bias for the predicted versus measured serum concentrations of vancomycin for the primary approach suggested by each method. Bias (ME) ranged from a 5.87-mg/L underprediction with the Birt method to a slight overprediction of 0.69 mg/L with the Burton primary method. Bias was fairly similar between the primary methods of Matzke and Burton (−0.84 and 0.69, respectively), with both fairly close to zero. Precision (RMSE) ranged from 6.05 with the Matzke method to 8.08 with the Burton method. Because of the similar bias and better precision than the Burton method, the Matzke method was judged to have the best overall combination of higher precision and the least bias. Mean absolute error (MAE), another measure of predictability, where the difference between measured and predicted concentrations are reported as the absolute value of the difference, ranged from 4.4 mg/L with the Matzke method to 6.4 mg/L with the Birt method.
Using the primary methods for determining V, CLcr, and CLvanco, the number and percentage of predictions for each method that were within 2.5 and 5 mg/L and 25% and 50% of the measured concentrations are compared in Table 4⇓. The Matzke method performed best for all measures. The Birt–Chandler method was within these limits least often.
To show the predictability at potentially toxic and subtherapeutic vancomycin concentrations, the seven methods were compared at high and low measured trough concentrations. No method consistently predicted concentrations of ≥20 and ≤3 mg/L.
All the methods examined were developed in select and often small groups of patients, and considerable variability existed among these groups. Achieving similar outcomes using these methods in different patients assumes that the new patients are similar to the original study group and that there is limited variation in pharmacokinetic parameters. No method was a particularly good predictor of measured vancomycin trough concentrations for this patient population. All of the methods demonstrated relatively poor precision and correlation, and the Rodvold, Birt, and Ambrose methods had large underprediction biases. It was not expected that the coefficient of determination would improve for all the methods when patients whose ABW divided by IBW was 1.2 or less were removed from the group. Since the original methods were developed with lower percentages of overweight patients, we assumed that the predictions would better correlate in the lower-weight patients. Our findings indicate that the methods can be used in overweight patients, though considerable variability should be expected.
In a study conducted to determine a consensus on therapeutic drug monitoring of vancomycin, 93% of respondents thought trough concentrations should be less than 10 or 5–10 mg/L, and 74% thought trough concentrations of 10 mg/L or higher were potentially toxic.23 The American Thoracic Society and the Infectious Diseases Society of America recently published guidelines for the treatment of pneumonia, which suggest target vancomycin trough concentrations of 15–20 mg/L.24 Concentrations that might be considered excessively high (≥20 mg/L) or low (≤3 mg/L, since vancomycin’s minimum inhibitory concentration is ≤4 mg/L) were observed in 21 (11%) patients receiving vancomycin by the initial dosing approaches used in our hospital. For those who consider potentially toxic trough concentrations to be lower than 20 mg/L, the potential risk of toxicity would be larger. For eight of the concentrations that were 20 mg/L or larger, no method would have identified these patients as having a concentration above 20 mg/L. At the low end of the scale (concentrations of ≤3 mg/L), similar variability existed. These results indicate that the use of measured concentration to guide therapy when starting with any predictor method will be of benefit.
Zokufa et al.25 examined five methods (Matzke, Moellering, Nielsen, Lake–Peterson, and the manufacturer’s) for dosing vancomycin and achieving desired peak and trough serum concentrations. In their study of 37 patients, the simulated concentrations from the methods were accurate only within 10% of the actual measured concentration 3–16% of the time. In our study, the methods were within 25% of the measured concentration 7.9–31.2% of the time.
Lee et al.26 compared two proctor methods for determining vancomycin concentrations, one of which was an adaptation of the Ambrose–Winter method used in this study.27 They found similar levels of predictability to the methods evaluated in this study, with RMSE ranging from 7.83 to 8.07 mg/L and ME ranging from −3.36 to 3.13 mg/L.
It is possible that other methods of determining the initial dosing of vancomycin may yield better predictions or better concentration outcomes. They may also be easier to use; however, this study did not investigate the predictability of initial dosing methods.
A number of issues should be considered when evaluating the results of this study. The exact duration of infusion was not documented on many patient data sheets so we had to assume that the hospital’s protocol was followed. We also assumed that the doses were given and the measured concentrations were drawn at the stated times and that the assays were accurate. A one-compartment model dosing approach was used to predict concentrations; it is possible that other pharmacokinetic modeling approaches may have yielded better results.
The seven methods studied for estimating vancomycin pharmacokinetic parameters varied widely in predicting vancomycin trough concentrations compared with measured serum concentrations and were not sufficiently reliable to replace therapeutic monitoring of vancomycin serum concentrations.
Appendix—Pharmacokinetic parameter prediction methodsa
Creatinine clearance (CLcr) units are stated for each method. The elimination rate constant (k) is determined from k = CLvanco/V, where CLvanco = vancomycin clearance and V = volume of distribution.
↵a V estimated using the patient’s actual body weight (ABW).
↵d CLcr estimated using the Cockcroft–Gault method,18 but the weight to be used is not stated. Elsewhere in the same textbook, it is recommended that an adjusted body weight (BWadj) be used, so this approach is assumed and used. The method for estimating BWadj is IBW + 0.4(ABW –IBW), where IBW = ideal body weight.
↵e V estimated using BWadj if ABW > IBW; ABW used if ≤ IBW. IBW = 0.73 × height (cm) − 59.42.
↵g Uses the Salazar–Corcoran19 approach to estimate CLcr in obese patients (defined as ABW/IBW ≥ 1.3). Uses Cockcroft–Gault and ABW to predict CLcr for nonobese patients. V estimated using ABW up to ABW/IBW = 1.3. After that, IBW is used.
IBW is calculated using the formula of Devine20 for all methods except Burton’s.
IBW (males) = 50 kg + 2.3(height [in] − 60) kg
IBW (females) = 45.5 kg + 2.3(height [in] − 60) kg
It should be noted that some references have provided simplified versions of several of these methods (e.g., giving a fixed V of 0.9 L/kg for all CLcr levels with the Matzke method).21 The versions used for this study were those quoted in the original articles.
- Copyright © 2006. American Society of Health-System Pharmacists, Inc. All rights reserved. 1079-2082/04/0602-1242$06.00