Model dependency of multivariate autoregressive spectral analysis

A combination of simulations and experimental data analysis has been used to demonstrate that, because cardiovascular control represents a complex linking of input and output parameters, interpreting the variability of individual parameters such as heart rate and arterial pressure virtually requires the use of techniques that quantify control by relating these inputs and outputs. Transfer functions represent appropriate techniques for this purpose. Further, despite the complexities of in vivo physiological control, many of the control elements can be well characterized by only taking into account single inputs and outputs and using a bivariate AR model. However, occasionally when two control systems have a strong and simultaneous influence on a single output parameter, such as arterial pressure and respiratory activity on RR interval, an expansion of the model to the general multivariate case may be required for a complete interpretation. Finally, although not fully demonstrated here, because of the closed-loop nature of cardiovascular control it is likely that algorithms that include causality to account for this characteristic, such as the AR formulation, will most accurately identify the transfer relations.