Limitations
The database collected prior to the evaluation of the models is a retrospective database, derived from pharmacological therapeutic monitoring data. Hospital staff may imprecisely document the time of collection and administration. This imprecision can be up to 30 minutes. Moreover, only population PK models developed on NONMEM® with covariates readily available to us were evaluated. This reduces the number of models that were assessed. This retrospective cohort study lacks randomization, however, efforts to minimize this bias through the addition multiple variables in the analysis were done. APEs were only identified as of 2015 and only when treated with IV tobramycin. APEs prior to 2015 or not treated with tobramycin were not included in the study and may underestimate the absolute number of APE occurrences per patient. MICs were not readily available and susceptibility was classified as sensitive or resistant, as per our center’s microbiology reports. This prevents the ability to capture variability inside a susceptibility group, which is likely to underpower our ability to identify a significant trend. Also, the smaller sample size for clinical outcomes may further impede the likelihood of detecting trends.