Vancomycin practical antimicrobial TDM in Japanese critical care: correcting overestimated concent

Objectives

This study aimed to improve the accuracy of initial dosing design for vancomycin when using a free web application for practical antimicrobial therapeutic drug monitoring (PAT). Specific objectives were to: identify patient subgroups in emergency and critical care settings whose predicted population mean serum vancomycin concentrations (PRED) were overestimated and to calculate appropriate PRED scaling factors to reduce prediction error.

Materials and methods

Vancomycin serum concentrations were simulated for patients treated at the Emergency and Critical Care Centers of the University of Miyazaki Hospital (Japan) using a population pharmacokinetic model. The simulations produced predicted population mean serum vancomycin concentrations (PRED) that were compared with observed concentrations (OBS) to evaluate prediction accuracy.

The study used the PAT web application for vancomycin dosing support; further specifics of the web application workflow and the underlying population pharmacokinetic model parameters are not reported in the abstract. Likewise, exact dosing regimens, sampling times for vancomycin concentrations, and additional model covariates are not detailed in the provided source text.

Patient classification and subgroup definitions

Patients were classified by the ratio of observed to predicted serum vancomycin concentrations (OBS/PRED). Two groups were defined:

• Low-trough group: OBS/PRED < 0.67
• Control group: OBS/PRED ≥ 0.67

This threshold was used to identify patients for whom the PAT-derived PRED values substantially overestimated actual measured concentrations.

Analytical approach and error metrics

Changes in patient data from admission to immediately before vancomycin administration were analyzed to identify factors associated with PRED overestimation. For subgroups that were more common in the low-trough group, scaling factors were calculated to minimize the mean squared error (MSE) between PRED and OBS. The performance of these scaling adjustments was evaluated by comparing MSE and root mean squared error (RMSE) values before and after scaling.

Results: patient distribution and subgroup characteristics

A total of 47 patients were included in the analysis. Using the OBS/PRED threshold, 23 patients (49%) were assigned to the low-trough group and 24 patients (51%) to the control group.

Compared with the control group, the low-trough group showed a significantly higher prevalence of the following characteristics:

• Creatinine clearance (CCR) < 15 mL/min
• Albumin change (ΔALB) < –0.5 g/dL
• Trauma as a clinical context

These factors were therefore considered candidates for subgroup-specific PRED adjustment because they were associated with systematic overestimation of predicted concentrations by the PAT-derived model.

Results: derived PRED scaling factors and RMSE changes

For the subgroups more frequent in the low-trough group, the investigators derived PRED scaling factors that reduced prediction error. The reported range of scaling factors was 0.53 to 0.64 across the identified subgroups.

The optimal scaling factor identified in the abstract was 0.53 for the subgroup defined by ΔCCR > 15 mL/min. Applying this scaling factor to PRED values reduced the RMSE between predicted and observed vancomycin concentrations from 5.1 μg/mL to 4.0 μg/mL, indicating improved concordance after adjustment.

Conclusions and clinical implications

The authors conclude that applying PRED scaling factors could potentially improve the accuracy of initial vancomycin dosing when using the PAT web application. Key predictors associated with overestimated PRED values were changes in renal function (ΔCCR > 15 mL/min), decreases in albumin (ΔALB < –0.5 g/dL), and trauma. For these situations, applying subgroup-specific scaling factors (reported range 0.53–0.64) may reduce prediction error and improve initial dosing decisions.

Clinicians using PAT for vancomycin in emergency and critical care settings may consider closer evaluation of recent changes in renal function, albumin, and presence of trauma when interpreting PRED outputs. Where relevant, applying an empiric PRED scaling factor (for example, 0.53 for marked increases in CCR) may decrease the expected discrepancy between predicted and observed concentrations.

Reported limitations and missing details

The abstract does not provide several methodological details necessary for full appraisal and implementation, including:

• Complete specifications of the population pharmacokinetic model used for simulation (structural model, covariates, parameter estimates).
• Details of the PAT web application workflow and how PRED values are generated in practice.
• Exact vancomycin dosing regimens, timing of concentration sampling relative to dosing, and whether samples represented trough or other timepoints.
• Full statistical reporting (confidence intervals, p-values) for subgroup comparisons beyond the summary statements in the abstract.

Because these elements are not reported in the abstract, users should consult the full article for implementation details and validation before adopting specific scaling factors in clinical practice.