Efficient sensitivity analysis for biomechanical models with correlated inputs.

Area of Science:

• Biomechanical modeling
• Computational biology
• Sensitivity analysis

Background:

• Variance-based sensitivity analysis (SA) often assumes input independence in biomechanical models.
• Input correlations can alter SA interpretations, impacting model development and personalization.
• High computational costs and unknown correlation structures limit the use of correlated SA.

Purpose of the Study:

• To propose an efficient surrogate model-based approach for correlated global sensitivity analysis.
• To demonstrate the interpretation and application of correlated SA in model development.
• To guide modelers in handling input correlations, even when the structure is not fully known.

Main Methods:

• Developed a surrogate model-based approach for efficient correlated global sensitivity analysis.
• Applied the methodology to a pulse wave propagation model.
• Demonstrated interpretation and guidance for modelers using correlated SA.

Main Results:

• Achieved accurate SA results at a theoretically 27,000x lower computational cost compared to non-surrogate methods.
• Input correlations were shown to significantly affect SA outcomes.
• The method effectively guides input prioritization, fixing, reduction, and dependency assessment.

Conclusions:

• The proposed surrogate-based SA approach enables efficient correlated SA for complex biomechanical models.
• Investigating input correlations is essential for accurate SA and reliable model development.
• The method supports critical aspects of model refinement and understanding parameter relationships.