Use of Sensitivity Analysis to Assess Reliability of Metabolic and Physiological Models

Because ethical considerations often preclude directly determining the human health effects of treatments or interventions by experimentation, such effects are estimated by extrapolating reactions predicted from animal experiments. Under such conditions, it must be demonstrated that the reliability of the extrapolated predictions is not excessively affected by inherent data limitations and other components of model specification. This is especially true of high-level models composed of ad hoc algebraic equations whose parameters do not correspond to specific physical properties or processes. Models based on independent experimental data restricting the numerical space of parameters that do represent actual physical properties can be represented at a more detailed level. Sensitivities of the computed trajectories to parameter variations permit more detailed attribution of uncertainties in the predictions to these low-level properties. S-systems, in which parameters are estimated empirically, and physiological models, whose parameters can be estimated accurately from independent data, are used to illustrate the applicability of trajectory sensitivity analysis to lower-level models.