Model evaluation
The visual assessment of goodness of fit plot models showed that both models adequately described the observed tobramycin concentrations (Suppl. figure S1). Calculations of bias and inaccuracy indicated that the predictions of model 1 and 2 are slightly lower than the observed concentrations (Model 1: MDPE, 2.6 %; MDAPE, 12.2 % and model 2: MDPE, 0.9 %; MDAPE, 9.5 %). Bias and inaccuracy were reasonable and met the predetermined criteria.
Bias and inaccuracy were plotted as histograms in Fig.1A and 1B to compare model 1 and 2, respectively. In Fig.1A, bias was between -6.0% when only covariates remain (e) and 5.4% when only concentration at peak remain (c). Inaccuracies were between 12.2% and 24.8%. In Fig.1B, bias were between -40.6% when observed concentration are missing (e) and 0.9% in the complete database (a). Model 1 met the predefined criteria for each database whereas model 2 did not.
Concerning simulations-based diagnostics of model 1, the histogram of NPDEs frequency revealed that they were not normally distributed (Fig.2 and Suppl. figure S2). Visual inspection of pcVPC showed evidence of misspecification (Fig.3), particularly for early time after dose (Suppl. figure S3).