Challenges in practical computation of global sensitivities with application to a baroreceptor reflex model

Background: Sensitivity analysis is the assessment of the effect on a model output from changing model parameters. Quantitatively, local sensitivity analysis can be performed by computing the derivative of the model output with respect to the model parameters. Measures obtained are called local sensitivities as the derivatives are evaluated at some nominal parameter values. Hence, if the uncertainty in the parameter values is high or the model is highly nonlinear, local sensitivity analysis may be inadequate. Results: This study shows how global sensitivities can be computed using simple integration of local sensitivities over the allowed parameter space utilizing quasi-Monte Carlo integration with low-discrepancy sequences. In addition we show that analysis of parameter sets displaying specific behaviors can lead to a better understanding of the role of certain parameters and be used to set bounds for the parameter space. We illustrate the approach for global sensitivity computation using a baroreceptor reflex model, which describes heart rate regulation during head-up tilt. Conclusions: Global sensitivities were calculated for an improved baroreceptor reflex model, by averaging local sensitivities. Parameter sets causing rapid fluctuations in local sensitivities, as well as parameter sets causing infeasible model behavior or solver errors, were recorded. Analysis revealed nontransparent model dynamics and