Sensitivity analysis for biomedical models.

Area of Science:

• Biomedical modeling
• Computational biology
• Systems biology

Background:

• Sensitivity analysis (SA) is crucial for evaluating variable importance in diverse scientific models.
• SA quantifies contributions of model factors to output distributions using variance decomposition.
• Its application in biomedical modeling aids in assessing model quality and understanding mechanisms.

Purpose of the Study:

• To present variance-based SA derivations and properties within the biomedical context.
• To introduce computationally efficient SA methods for large-scale biomedical models.
• To provide an objective criterion for parameter reduction in model analysis.

Main Methods:

• Statistical derivations of variance-based SA.
• Development of two numerical approximate methods using unscented transformation (UT) for SA.
• Proposal of a statistical criterion for parameter nominality assessment.

Main Results:

• Demonstration of SA's utility in biomedical model quality assessment, reduction, and factor prioritization.
• Validation of UT-based methods for achieving a balance between computational cost and precision in SA.
• Introduction of an objective criterion for determining parameter discardability in model reduction.

Conclusions:

• SA is a valuable tool for enhancing understanding and quality of biomedical models.
• The proposed UT-based methods offer efficient alternatives to computationally intensive Monte Carlo approaches for SA.
• An objective statistical criterion facilitates informed model reduction without significant information loss.