The therapy of critically ill patients with aminoglycoside antibiotics requires a careful dosing to achieve effective drug concentrations and to avoid toxic effects.
Objective: To evaluate if a single blood sample per dosing interval 3 or 8 hours after infusion of tobramycin is appropriate to estimate the actual serum concentration 30 minutes after infusion (Cpeak30) and at the end of the dosing interval (C22h) in critically ill patients.
Methods: A total of 32 patients of the intensive care unit (ICU) with an individualized once-daily dosing regimen involved in an intensified drug monitoring of tobramycin were analyzed retrospectively. The first day, serum concentrations 3 and 8 hours after the end of infusion (C3h and C8h) were both used for calculations of Cpeak30 and C22h performed by a one-compartmental Bayesian estimation method (ABBOTTBASE). The subsequent days, each calculation included a single blood sample (C3h or C8h) as well as the corresponding available monitoring data of the previous dosing intervals. Estimation error was analyzed including bias and precision. The influence of some factors such as severity of illness (APACHE II score) and renal function (creatinine clearance) on the estimation error was investigated by multiple linear regression analysis (MLRA).
Results: In the first monitored dosing interval, Cpeak30 was overestimated and C22h underestimated by 1.72 and -0.29 microg/ml on median, respectively. The maximum deviation from the true Cpeak30 and C22h was -9.20/+8.57 and -1.90 microg/ml, respectively. The first day, prediction of potentially toxic C22h above 1 mg/ml by an estimation value above 1 mg/ml was possible in 4 of 6 cases only. The subsequent days, mean error (ME) of peak estimations of -0.34 microg/ml and a root mean squared error (RMSE) of 1.84 microg/ml indicated a small underestimation when C3h was used and an overestimation when C8h was used for calculation (ME 1.04 and RMSE 2.83 microg/ml). The difference on ME and RMSE was statistically significant. C22h values outside of the target range (a total of 10 observations) were predicted with sensitivity of 60% in each case. Prediction error of increased C22h was partly considerable and showed the trend to greater underestimation with increasing C22h in MLRA. Increasing serum creatinine (beta = 0.429, p = 0.011) and decreasing creatinine clearance (beta = -0.324, p = 0.039) were only identified as variables affecting the trough level in MLRA.
Conclusions: In the critically ill, C3h but not C8h of tobramycin permitted the estimation of the Cpeak30 in most cases with satisfactory bias and precision starting with 2nd monitored dosing interval. However, high trough levels requiring an adjustment of dose or dosing interval could not be safely predicted; prediction error was intolerable when C22h exceeded the target range of trough level. Data indicate that at least in patients with any sign of renal dysfunction, the measuring of the interesting levels should be preferred to the tested single-point based estimations.