Estimating confidence intervals for cerebral autoregulation: a parametric bootstrap approach.

ABSTRACT

Cerebral autoregulation (CA) refers to the ability of the brain vasculature to control blood flow in the face of changing blood pressure. One of the methods commonly used to assess cerebral autoregulation, especially in participants at rest, is the analysis of phase derived from transfer function analysis (TFA), relating arterial blood pressure (ABP) to cerebral blood flow (CBF). This and other indexes of CA can provide consistent results when comparing groups of subjects (e.g. patients and healthy controls or normocapnia and hypercapnia) but can be quite variable within and between individuals. The objective of this paper is to present a novel parametric bootstrap method, used to estimate the sampling distribution and hence confidence intervals (CIs) of the mean phase estimate in the low-frequency band, in order to optimise estimation of measures of CA function and allow more robust inferences on the status of CA from individual recordings. A set of simulations was used to verify the proposed method under controlled conditions. In 20 healthy adult volunteers (age 25.53.5 years), ABP and CBF velocity (CBFV) were measured at rest, using a Finometer device and Transcranial Doppler (applied to the middle cerebral artery), respectively. For each volunteer, five individual recordings were taken on different days, each approximately 18 min long. Phase was estimated using TFA. Analysis of recorded data showed widely changing CIs over the duration of recordings, which could be reduced when noisy data and frequencies with low coherence were excluded from the analysis (Wilcoxon signed rank testp= 0.0065). The TFA window-lengths of 50s gave smaller CIs than lengths of 100s (p< 0.001) or 20s (p< 0.001), challenging the usual recommendation of 100s. The method adds a much needed flexible statistical tool for CA analysis in individual recordings.