Modelli meccanicistici: modelli compartimentali nell'analisi individuale e di popolazione

Mechanistic models use several key mathematical components: Xi represents known values, Ci represents observed concentrations, εi represents measurement errors, ϕj represents model parameters, and ƒi represents the related function. These components work together in equations to predict drug behavior, though data collection errors prevent perfect prediction of observed data.

Ordinary least squares (OLS) is biased toward larger observations, limiting prediction accuracy. Weighted least squares (WLS) and maximum likelihood or extended least squares (ML/ELS) methods improve OLS by incorporating a weighting factor that adjusts for observation size differences. This weighting enhances the accuracy of predicted versus observed value comparisons.

Population analysis models account for interindividual variability—differences between individuals—by using the same structural model to fit all subjects' data for a specific drug. Random variability, represented by ηj in mathematical equations, describes the relationship between mean and individual pharmacokinetic parameters, enabling predictions for multiple subjects simultaneously.

Population compartmental analysis includes three main approaches: naïve-average data, naïve pooled data, and the two-stage approach. The two-stage approach further includes standard two-stage (STS) and global two-stage (GTS) methods, which obtain population parameter estimates through iterative processes to improve accuracy.

Least-squares metrics quantify differences between predicted and observed values in mechanistic models. These metrics measure how well model predictions match actual concentration observations, helping researchers evaluate model performance. Different least-squares approaches—ordinary, weighted, and maximum likelihood—provide varying levels of accuracy depending on data characteristics.

Using the same structural model for all individuals in population analysis allows researchers to identify common pharmacokinetic patterns while accounting for interindividual variability through random effects. This approach enables efficient analysis of multiple subjects' data for a specific drug, providing both individual and population-level predictions.

Data collection errors prevent mechanistic models from achieving perfect prediction of observed data. Measurement errors (εi) introduce uncertainty into observed concentrations, requiring statistical methods like weighted least squares to account for these imperfections and improve model reliability in individual and population pharmacokinetic analyses.