Evaluation of a two-compartment Bayesian forecasting program for predicting vancomycin concentrations

The application of a two-compartment Bayesian forecasting program for vancomycin was tested retrospectively in 45 adult patients with stable renal function. Serial blood samples from 25 of these patients were used to determine population-based parameter estimates. The predictive performance of the Bayesian program was assessed by using both non-steady-state and steady-state vancomycin concentrations as feedback information. Overall, the program tended to underpredict peak and trough steady-state vancomycin serum concentrations. A larger mean prediction error (ME) was seen when non-steady-state feedback serum concentrations were used compared with using population-based parameter estimates (no feedback). In contrast, a marked improvement in ME (peaks: -1.03 versus -2.61; troughs: -1.60 versus -2.07) was seen when steady-state feedback serum concentrations were used compared with no feedback data. Precision improved when either feedback serum concentrations were used to predict steady-state peak and trough vancomycin concentrations. The results from this clinical evaluation demonstrate that the initial pharmacokinetic parameter estimates for a two-compartment Bayesian model provided accurate prediction of steady-state vancomycin concentrations. Prediction bias and precision were improved when steady-state vancomycin concentrations were used to determine individualized pharmacokinetic parameters.     

Evaluation of a Bayesian regression program for predicting warfarin response

The ability of a Bayesian regression program (Warfcalc) to predict warfarin response was evaluated retrospectively in 48 inpatients and prospectively in 10 inpatients. The prothrombin ratio (PR) on the last day of inpatient therapy was predicted using zero (naive) to five sequential, daily PR feedbacks. Bias and precision were measured using mean error (ME) and mean absolute error (MAE), respectively. Root mean squared error (RMSE) was used as a combined measure of bias and precision. In the retrospective group, the use of five PR feedbacks yielded the lowest ME, MAE, and RMSE (0.22, 0.30, and 0.45, respectively). The use of two and three daily PR feedbacks resulted in larger prediction errors compared with the use of naive parameters. Further evaluation of the retrospective patient data indicated that deletion of PR feedbacks associated with an activated partial thromboplastin time greater than 100 s and exclusion of metabolic inhibitors in the estimation of warfarin clearance resulted in more reliable predictions (ME = 0.07, MAE = 0.20, RMSE = 0.28). Similarly, deletion of such PR feedbacks and metabolic inhibitors from the prospective data and use of PRs for the first 5 days of therapy yielded ME, MAE, and RMSE values of 0.07, 0.21, and 0.27, respectively. The variance for prothrombin complex activity (PCA) as a function of the variance in the prothrombin time (PT) was investigated using Monte Carlo simulation assuming four different random error models for the PT measurements. These error models yielded functions that exhibit a maximum coefficient of variation at PCA values of 40-70%.(ABSTRACT TRUNCATED AT 250 WORDS)     

Evaluation of a Bayesian regression-analysis computer program for predicting phenytoin concentration

A microcomputer program using Bayesian regression analysis to predict serum phenytoin concentrations was evaluated. Phenytoin concentration-time data from nine healthy male volunteers and one male patient were obtained from published studies. For two different dosage regimens that each subject received, the last available predose concentration on the sixth day of the regimen was predicted using observed predose concentrations on both the morning of the third day and on each of the first three days of phenytoin administration. In nine subjects who received at least 10 days of phenytoin therapy, observed concentrations after more than 10 days of therapy were predicted using both one and three observed serum concentrations. Also, in six subjects, the observed predose concentrations for the first three days of an initial phenytoin regimen were used to predict the last predose concentration observed during each subject's second regimen. Predictive performance of the program was evaluated using mean error (m.e.) as a measure of bias, mean absolute error (m.a.e.) as a measure of precision, and root mean square error (r.m.s.e.) as a composite measure of bias and precision. The majority of the predicted serum concentrations were accurate. Predictions of serum concentrations after six days and after more than 10 days of phenytoin therapy were somewhat more accurate when three serum concentrations were used than when only one concentration was used. In the six subjects for whom concentrations from an initial regimen were used to predict those in a second regimen, the largest prediction error was 5 mg/L (m.e. 0.88, m.a.e. 1.9, and r.m.s.e. 2.4).(ABSTRACT TRUNCATED AT 250 WORDS)     

Comparison of three assay procedures for gentamicin determination

We examined 50 serum samples, submitted to our laboratory for gentamicin levels, by the enzyme multiplied immuno technique (EMIT), by a radioimmunoassay procedure and by a latex agglutination method. Significant differences in concentrations, rates of throughput, and costs between the latex agglutination and the other two procedures were observed. The precision of the assay ranged from 5.21 to 15.70%, whereas recovery values ranged from 83.8 to 110.0%. The latex agglutination procedure provided simplicity and speed at a relatively low cost. However, the significant analytical difference between this and the other two procedures indicates that the latex agglutination procedure is not adequate for measuring accurate gentamicin levels.     

Bayesian forecasting of serum gentamicin concentrations in intensive care patients

This study retrospectively evaluated the predictive performance of a 1-compartment Bayesian forecasting program in adult intensive care unit (ICU) patients with stable renal function. A comparison was made of the reliability of 3 sets of population-based parameter estimates and 2 serum concentration monitoring strategies. A larger mean error for prediction of peak gentamicin concentrations was seen with literature-derived parameters than when ICU population-based parameter estimates were used. Bias and precision improved when non-steady-state peak and trough concentrations were used to predict those at steady-state; the addition of steady-state values did not provide additional information for predictions once non-steady-state feedback concentrations were incorporated. The addition of 4 serial gentamicin concentrations obtained at both non-steady-state and steady-state did not noticeably improve the predictive performance. The results demonstrate that initial ICU pharmacokinetic parameter estimates for a 1-compartment Bayesian model provide accurate prediction of steady-state gentamicin concentrations. Prediction bias and precision showed the greatest improvement when non-steady-state gentamicin concentrations were used to determine individualised pharmacokinetic parameters.     

Evaluation of a microcomputer program (OPT) for parameter optimisation in clinical pharmacokinetics: gentamicin and tobramycin.

1. Several nomograms and computer programs are available to aid in aminoglycoside dosing. 2. Due to the variability in the relationship between dosage and serum drug levels, monitoring through the acquisition of serum drug levels is mandatory. 3. All clinical data, including serum drug levels, are subject to errors. 4. The program we have evaluated, OPT, calculates the most likely set of pharmacokinetic parameter estimates for individual patients by applying Bayes' theorem and the principle of Maximum Likelihood Estimation. Through a feedback process all available data are used, taking possible errors into account. 5. Our study shows that OPT is able to predict serum aminoglycoside levels accurately in the routine clinical setting. It may thus contribute to the quality of aminoglycoside therapy.     

Accuracy of three methods for predicting concentrations of free phenytoin

Predictions of free (unbound) serum phenytoin concentration by three methods were compared with results obtained by the Abbott TDx Free Phenytoin ultrafiltration and fluorescence-polarization immunoassay technique. Data were obtained for hospitalized adults who had been receiving phenytoin for at least five days and were free of renal or hepatic disease. Total phenytoin concentration was determined, and free phenytoin concentration was measured in ultrafiltrate at 25 degrees C. For each patient, measured concentrations of total phenytoin and albumin were used to predict free phenytoin concentrations by the Gugler method, the Sheiner-Tozer nomogram, and the Sheiner-Tozer equation. Mean measured percentages of free phenytoin were 17.79%, 12.13%, and 8.73%, respectively, for patients with albumin concentrations of less than 2 g/dL (n = 5), 2-3 g/dL (n = 18), and greater than 3 g/dL (n = 26). There was a strong correlation between actual and predicted free phenytoin concentrations for each of the methods, but all methods were found to lack precision. All methods also exhibited bias, as demonstrated by overprediction of the free concentration; however, none of the methods exhibited bias when the difference between the in vitro temperature of 25 degrees C and the in vivo temperature of 37 degrees C was considered. Because of their poor precision, the three methods evaluated in this study are not recommended for predicting free phenytoin concentration.     

Evaluation of a Bayesian regression-analysis computer program using non-steady-state phenytoin concentrations

The predictive performance of a Bayesian regression-analysis computer program that uses non-steady-state phenytoin data was evaluated. Forty patients receiving phenytoin or phenytoin sodium who had two or more non-steady-state serum concentrations were selected for study. Additional serum concentrations and dosing data were collected as they became available, but no effort was made to control the number or timing of serum concentration determinations. Patients were categorized into four groups for evaluation of the effect of potential bioavailability problems and length of dosing history (time over which serum concentration-time data were collected) on the ability to predict subsequent phenytoin concentrations. Population parameters for phenytoin maximum rate of elimination (Vmax), apparent Michaelis-Menten constant (Km), volume of distribution (V), and bioavailability (F) were obtained from the literature. Predictions based on serum phenytoin concentrations and dosing histories (information intervals) of 5 or 10 days were compared with predictions based on naive (population-based) estimates using prediction-error analysis. In each patient group, the use of either 5-day or 10-day information intervals resulted in a significant increase in precision and a significant reduction in bias compared with naive estimates. For the group of patients who initially had two or more serum concentrations within the first five days of monitoring, predictions showed a marked increase in bias and a decrease in precision as the time interval from the last measured concentration to the time of prediction increased.(ABSTRACT TRUNCATED AT 250 WORDS)     

An evaluation of Bayesian microcomputer predictions of theophylline concentrations in newborn infants

Determination of appropriate theophylline maintenance doses in preterm infants is confounded by interpatient variability. This study evaluated the performance of an IBM PC computer program applying Bayesian regression before and during steady state in 37 preterm infants. Prior population estimates of clearance and distribution volume in preterm infants and Bayesian estimates of clearance and distribution volume based on one to three theophylline plasma concentrations were used to predict subsequent concentrations (drawn 1-17 days later). We assessed the accuracy and precision of the predictive performance of the Bayesian program with the mean prediction error and the mean absolute prediction error. The absolute prediction error (mean absolute error +/- SEM) significantly decreased with increasing feedback concentrations from 3.54 +/- 0.45 micrograms/ml (population estimates) to 2.74 +/- 0.42 (one feedback) and 2.02 +/- 0.35 micrograms/ml (two feedback concentrations). Mean prediction errors (+/- SEM) based on one to three feedbacks (-1.5 +/- 0.40 micrograms/ml) were significant improvements over population predictions (-2.63 +/- 0.72 micrograms/ml, p less than 0.05), although a small but significant average overprediction remained. Absolute prediction error was correlated with postconceptional and postnatal age when zero or one but not two feedback concentrations were available. Computer program predictions based on one measured feedback concentration were more accurate and precise than population-based predictions. Refinement of population parameters or two feedback concentrations further improved performance.     

Comparison of the predictions of one- and two-compartment microcomputer programs for long-term tobramycin therapy

A retrospective study of patients receiving tobramycin compared the accuracy of predictions of actual trough serum concentrations using two commercially available microcomputer software programs. Twelve patients met the study criteria of intravenous tobramycin treatment for more than 10 days with serum concentration monitoring within the first 5 days and after 10 days of therapy. No patients received dialysis. Twenty-five serum concentrations were compared. Predictions within 0.2 microgram/ml were considered clinically "exact." No significant differences were found by chi-square analysis for any of the four possible choices (p less than 0.3). One of the programs, distributed by Dista Pharmaceuticals, offers a one-compartment model, a two-compartment model, and a two-compartment prenephrotoxic option. SIMKIN, a program marketed by Medical Engineering, Inc., uses a two-compartment model. Overall, the predictions errors were small, but occasionally were clinically significant. Further evaluation of microcomputer programs for therapeutic drug monitoring is necessary to document their impact on predicting drug efficacy and toxicity.     

Bayesian pharmacokinetics of gentamicin in a haemodialysis population

BACKGROUND: Aminoglycosides are commonly used in the haemodialysis population. Standard pharmacokinetic approaches require multiple sampling to describe the parameters of drug distribution and elimination in the intra- and interdialytic periods. OBJECTIVE: To characterise the pharmacokinetics of gentamicin in a haemodialysis population by using Bayesian pharmacokinetic methods and only two plasma concentrations. DESIGN AND PARTICIPANTS: Prospective case series of 13 adult (aged 36-70 years) haemodialysis patients (Fresenius F80 dialysers were used) receiving gentamicin. METHODS: Patients with suspected or confirmed Gram-negative infections were given gentamicin. At 48 hours after receiving the dose (at the next haemodialysis session), patients provided two blood samples, one immediately before the dialysis session and another 1 hour after haemodialysis. Data on dosage, timing and plasma concentrations for all subjects were analysed with PASTRX version 10.6 and Bayesian pharmacokinetic analysis. Volume of distribution (Vd), interdialytic elimination rate constant (k(inter)), interdialytic elimination half-life (t1/2beta, inter)) and interdialytic clearance (CL(inter)) were determined from a single predialysis plasma concentration. Elimination rate constant (k(dial)), elimination half-life (t1/2beta, dial)) and clearance (CL(dial)) during 3.5-4 hours of dialysis were also determined from the pre- and post-plasma concentrations. RESULTS: Pharmacokinetic parameters (mean +/- SD) were: Vd 0.288 +/- 0.002 L/kg, k(inter) 0.015 +/- 0.004h(-1), t1/2beta, inter) 48 +/- 11h, CL(inter) 5.9 +/- 2.4 mL/min, k(dial) 0.25 +/- 0.05 h(-1), t1/2beta, dial) 3.0 +/- 1.0h and CL(dial) 91 +/- 24 mL/min. CONCLUSIONS: The rate of elimination of gentamicin was 17-fold greater (95% CI 13.7-20.7) on haemodialysis with a Fresenius F80 than off haemodialysis. All of the pharmacokinetic parameters of interest were determined using Bayesian pharmacokinetic procedures and only two plasma gentamicin concentrations.     

Bayesian forecasting of serum lithium concentrations. Comparison with traditional methods

Twelve pharmacokinetic methods of estimating lithium maintenance dosage requirements were compared in 21 patients with bipolar illness. Methods which were compared included the single- and multiple-point methods of Perry, 4 non-linear regression and 6 Bayesian methods. The REVOL algorithm was employed for converging on to estimates of clearance and apparent volume of distribution for the non-linear regression and Bayesian methods. Data analysis was based on an evaluation of prediction error as a measure of bias, and absolute prediction error as a measure of precision. In a direct comparison, there were no statistically significant differences in bias or precision between any of the methods.     

Accuracy of Bayesian and Sawchuk-Zaske dosing methods for gentamicin

The derived pharmacokinetic variable estimates from a Bayesian aminoglycoside dosing program were compared with those from the Sawchuk-Zaske method to determine which variable estimates were the most accurate in fitting the test dose and in predicting subsequent peak and trough serum concentrations. Data on 17 patients with moderately impaired but stable renal function were analyzed. All patients received gentamicin sulfate for treatment of their infections. To determine the individualized variables using the Bayesian program, demographic data, dosing history, and one (midpoint), two (peak and trough), or four serum drug concentrations were entered into the program. The Sawchuk-Zaske method used three serum concentrations determined following a first dose or four concentrations before and after a subsequent dose to derive individualized pharmacokinetic variables. The estimates of pharmacokinetic variables determined using the Bayesian method with one, two, or four serum concentrations did not differ significantly from those obtained using all the available serum concentrations with the Sawchuk-Zaske method. Although the actual numeric differences of prediction, absolute, and squared errors for fitting the test dose were minimal, significant differences were seen. All methods were similar in predicting serum concentrations from continued dosing. For the prediction error from continued dosing, a slight but significant difference was observed with the Bayesian method using one serum concentration when compared with the other methods. The Bayesian method using one, two, or four serum gentamicin concentrations individualized pharmacokinetic variables as well as the Sawchuk-Zaske method.     

Comparison of a fluorescent immunoassay with an enzyme immunoassay and a radioimmunoassay for gentamicin

A fluorescent immunoassay (FIA) for gentamicin was evaluated and compared with an enzyme multiplied immunoassay (EMIT) and a radioimmunoassay (RIA) for gentamicin. Pooled human serum that contained low, medium, and high concentrations of gentamicin sulfate (approximately 2.5, 5, and 10 micrograms/ml of gentamicin) were analyzed for actual gentamicin concentration by FIA. Samples were assayed 10 times during the same day to evaluate within-run precision and on 10 different days to evaluate between-run precision. Serum samples obtained from patients receiving gentamicin therapy were analyzed for gentamicin concentration using FIA and EMIT, and separate serum samples were analyzed using FIA and RIA. Results were compared by regression analysis. Within-run coefficients of variation were 8.47%, 6.84%, and 2.62%, respectively, for the low, medium, and high concentrations of gentamicin, and the respective between-run coefficients of variation were 10.81%, 6.31%, and 1.64%. The correlation coefficient for the comparison of FIA with EMIT was 0.943, and the correlation coefficient for the comparison of FIA with RIA was 0.970. The fluorescent immunoassay is a reliable method for determining the concentration of gentamicin in serum. Although the results obtained by FIA correlate well with those obtained by EMIT and RIA, variability exists between concentrations determined by each of these methods.