Sensitivity analysis in quantitative modeling of toxicology

Abstract

Sensitivity analysis is not only a vital part of any risk assessment, but as a major source of uncertainty identification has a highest impact on quantitative and qualitative systematic review. Gaps in our knowledge are filled with proposed model assumptions, probability distributions, factor affecting variability, best guesses, expert opinion, etc. Sensitivity analysis seeks to discover which of the outputs of a model are more sensitive when the inputted variables change and how this sensitivity may affect final decisions. Therefore, in the risk assessment the overall output confidence of a response model will be increased where an effective sensitivity analysis is applied to the risk assessment model. Accordingly, local sensitivity analysis deals with the sensitivity and analytical analysis of variables in proposed models, while the global analysis examines the sensitivity of all parameters. A local sensitivity analysis focuses on the behavior of an input while other parts remain constant. A global sensitivity analysis examines the variance output models and determines how they could be affected by input parameters. The limitation of sensitivity analysis is local analysis because of its limited range and influence of the one input parameter which is not calculated for settings other than the base level. Although because of this reason, global sensitivity analysis might often be preferred, however, the global method may not always be the chosen method due to its higher computational complexity and time-consuming nature. In fact, a local sensitivity analysis is most likely preferred.