Predictive performance of an antibiotic precision dosing software program in critically ill adults with infection

BACKGROUND: Critically ill patients exhibit altered pharmacokinetics, rendering antibiotic dosing challenging. Achieving therapeutic antibiotic exposures may be improved with the use of precision dosing software programs. OBJECTIVES: To quantify and compare both a priori and a posteriori predictive performance of an antibiotic precision dosing software program in a heterogenous cohort of critically ill adults with infection. METHODS: The precision dosing program, ID-ODSTM, was used to predict a priori and a posteriori concentrations for piperacillin, meropenem, cefepime, flucloxacillin and vancomycin using clinical and demographic data derived from a previous study in critically ill adults (GUIDE trial). Predicted concentrations were compared to observed concentrations using pre-specified acceptance criteria (median predictive error (MDPE) ≤20% and median absolute predictive error (MDAPE) ≤30%), along with F20 and F30 metrics. Furthermore, the impact of predictions on theoretical dosing recommendations to achieve pre-defined drug exposures was assessed. RESULTS: The a priori predictive performance in 81 patients administered beta-lactams did not meet any pre-specified acceptance criteria (pooled MDPE -35% and MDAPE 58%). However, all beta-lactams met accuracy acceptance criteria for the a posteriori approach (pooled MDPE -3.6%), although only cefepime demonstrated acceptable precision and F20 and F30 acceptance. In 25 patients administered vancomycin, a priori predictive performance met all acceptance criteria for accuracy and precision.When assessing a priori theoretical dosing recommendation concordance, approximately one in three predictions led to an unnecessary dosing action. Concordance was similar with an a posteriori approach, however, overprediction was observed in 20% of meropenem predictions. CONCLUSIONS: In a heterogenous adult ICU population, the a priori predictive performance of vancomycin was shown to be acceptable. Conversely, the a priori predictive performance for the beta-lactams was not acceptable. Although predictive performance improved with the a posteriori approach for beta-lactams, only cefepime met acceptance criteria.