An extension of the indirect response models to ordered categorical pharmacodynamic data using latent variables
Hutmacher, Matthew M, Sriram Krishnaswami
Pfizer
Background/Aims: Indirect response models provide a semi-mechanistic framework to address drug-induced temporal delays in response-time profiles relative to drug exposure profiles. Currently, this methodology is developed for continuous response data only. No general theory for addressing drug-induced delay in ordered categorical response data exists. An extension of the indirect model methodology for ordered categorical data is proposed using the concept of a latent variable.
Methods: The approach is motivated by the statistical concept of a latent variable – an underlying and unmeasurable continuous process (such as inflammation or disease state), which is mapped into the measurable ordered categorical data. The four indirect response mechanisms are applied to this latent variable to derive a set of indirect, latent variable response models (ILVRM).
Results: Stochastic simulations are implemented to produce expected (mean) longitudinal response profiles, which are presented graphically for the four ILVRM as a function of exposure (or dose) and time. Ultimately, the ILVRM simulation results characterize the drug-induced delay in effect, which can be used to discriminate between potential model types (a priori) when performing data analysis.
Conclusion: ILVRM methodology provides a natural (pharmacologically interpretable) way to extend indirect response mechanisms to ordered categorical data.