Interpreting population pharmacokinetic-pharmacodynamic analyses - a clinical viewpoint

Abstract

The population analysis approach is an important tool for clinical pharmacology in aiding the dose individualization of medicines. However, due to their statistical complexity the clinical utility of population analyses is often overlooked. One of the key reasons to conduct a population analysis is to investigate the potential benefits of individualization of drug dosing based on patient characteristics (termed covariate identification). The purpose of this review is to provide a tool to interpret and extract information from publications that describe population analysis. The target audience is those readers who are aware of population analyses but have not conducted the technical aspects of an analysis themselves. Initially we introduce the general framework of population analysis and work through a simple example with visual plots. We then follow-up with specific details on how to interpret population analyses for the purpose of identifying covariates and how to interpret their likely importance for dose individualization.

Figure 1

Observed concentration–time data for 30 individuals

Figure 2

(A) The observed concentration–time data overlaid with the median predicted concentration from the PK model. (B) Observed concentration–time data overlaid with the median and 2.5^th and 97.5^th percentiles of the predicted concentrations from the PK model

Figure 3

Individual estimates of systemic drug CL vs. creatinine clearance. The line is the regression line and the intercept represents non-renal clearance and the slope represents the fraction of drug cleared unchanged by the kidneys. The vertical difference of any individual from the regression line represents the difference of that individual from the population average and is given by c (Equation 3)

Figure 4

The distribution of values of CL in the population. (A) Allows the distribution to be normal. The verticle line at 0 indicates the lower bound of biologically acceptable values. Here 2.3% of the predicted values of CL are unrealistic. (B) Allows the distribution to be lognormal. Negative (or zero) values of predicted CL are not allowed

Figure 5

Predicted concentrations from the PK model that does not include the covariate CL_CR (solid line). Predicted concentrations from a PK model that includes the covariate CL_CR as a covariate on CL (dashed lines). Including the covariate CL_CR in the model reduces the unexplained variability in the model predictions and hence improves the reliability of the model predictions