Population Pharmacokinetics Modeling
The population pharmacokinetic (PPK) profile of nemonoxacin was evaluated using plasma concentration data obtained from the patients with severe renal impairment and healthy controls. A nonlinear mixed-effects model software (NONMEM 7.4, ICON Development Solutions, USA), Modeling and Simulation Studio (Mas Studio 1.2.6 stable, BioVoice & BioGuider Ltd., Shanghai, China) and Perl-speak-NONMEM (PsN, version 5.0.0, Uppsala University, Sweden) were used for PPK analysis and model validation [18]. R (version 3.6.1) and RStudio (1.2.5001) software were used for statistical tests and plotting. The first-order conditional estimation with interaction approach was adopted for model development. The modeling strategy included establishment of the base model and full model development, assessment of final model adequacy, and model predictive performance and validation.
The structural base model was initially fitted using a compartment disposition model based on the PK data. The final base model was selected by the statistical significance between models using goodness-of-fit plots, the objective function value (OFV), twice the negative log-likelihood (-2LL), and Akaike’s information criterion. The inter-individual variabilities for PK parameters were assumed to follow the multiplicative exponential random effects of the formθi = θ × eηi , whereθi is the value of the parameter as predicted for the individual and θ is the population typical value of the parameter. The variability of inter-individual random effect η is a normal distribution with N (0, ω2). The residual error was tested using the constant coefficient of variation model and expressed as Cobs = Cpred× (1 + ε ), where Cobs is the observed value of an individual, Cpred is the predicted value, and ε is the intra-individual deviation with N (0, σ2).
The fixed effects were evaluated for statistical significance in a stepwise manner using a stepwise covariate model building procedure. A decrease of 3.84 in the OFV was considered a significant improvement for the forward inclusion step based on Chi-square test (α < 0.05). Meanwhile, the full model was subjected to a backward elimination step with a significance level of α = 0.01. The potential covariates of PPK parameters were screened. Age, sex, body weight, BMI, total body water (TBW), eGFR, creatinine clearance (CrCl), and albumin were treated as candidate variables. TBW was obtained using the classic Watson formula (for males, TBW = 2.447 - 0.09156 × age + 0.1074 × height + 0.3362 × weight; for females, TBW = -2.097 + 0.1069 × height + 0.2466 × weight), where age is in years, height in centimeters, weight in kilograms, and water in liters [19,20].
The final PPK model was validated by diagnostic plots and visual predictive check (VPC) techniques comprising 1000 simulations. The median, upper and lower bounds of the 95% prediction interval for PK profiles were compared against the observed plasma concentrations. The nominal 95% confidence intervals (CIs) around the point estimates were generated from 1000 bootstrap samples.