A simulation model can be categorized into three submodels: input-output (IO), covariate distribution, and execution models. These models are generally developed from previous data sets that may include preclinical data. However, the values of the model parameters (both structural and statistical elements) and the structure used in the simulation of a proposed trial may be different from those that were originally derived from the analysis of the previous data. This may be due to perceived differences between the design of previous trials and how the intended trial may be performed, e.g., patients versus healthy volunteers. In all cases, justification must be provided for the choice of any particular model and parameter estimates.
Input-output (IO) models include pharamacokinetic/pharmacodynamic (PKPD) models and disease progress models. The structural and statistical models and their parameter estimates to be used in simulation should be specified. Building PKPD models has been discussed extensively in the literature [see Gabrielsson and Weiner (17) for a recent discussion]. In contrast, disease progress models are less abundant, but may be derived from the placebo group data; in some (rare) circumstances databases may be available that provide the natural course data of disease progress (18). The statistical models should include those levels of random effects that are pertinent to the drug in question; these will usually include between-subject variability and residual unexplained variability. See Chapter 2 for more detail.
Covariate distribution models describe the characteristics of the subjects that might affect the drug behavior in the body. The model may include demographic, physiologic, and pathophysiologic aspects that are both representative of the patient type that is likely to be enrolled in the actual trial and pertinent to the simulation model. The distributions may be assumed to be similar to those of previous trials or based on clinical experience. The relevant covariates such as body weight, height, frequency of concomitant drug use, and baseline measurements are identified in the model development process. In addition, correlation between covariates should be considered, where appropriate, in order to avoid generating unrealistic representations of virtual patients. The distribution function of covariates may be altered in the what-if scenarios of simulation. See Chapter 3 for more detail.
Deviations from the protocol will occur inevitably during execution of a real clinical trial. Similarly, such deviations should be introduced into the virtual trial in order to make the virtual trial as representative of reality as possible. To simulate a clinical trial that accounts for trial protocol deviations, an execution model that describes these deviations must be developed. Typically, execution model(s) will include patient-specific effects such as dropouts and adherence of the patient to the dosing regimen, and investigator effects such as measurement error. Ideally, models need to be developed from prior data that describe these uncontrollable factors so that the probability of such occurrences can be mimicked for the virtual trial.
In some circumstances, execution models may not be able to be developed from prior data or extrapolated from similar drugs. In these circumstances, it is possible to either assign a probability of adherence and dropout rate, etc. based on prior experience or to design a trial that is relatively insensitive to likely protocol deviations. See Chapter 4 for more detail.
Was this article helpful?