Statistical criteria for estimation of the cerebral autoregulation index (ARI) at rest

Abstract

The autoregulation index (ARI) can reflect the effectiveness of cerebral blood flow (CBF) control in response to dynamic changes in arterial blood pressure (BP), but objective criteria for its validation have not been proposed. Monte Carlo simulations were performed by generating 5 min long random input/output signals that mimic the properties of mean beat-to-beat BP and CBF velocity (CBFV) as usually obtained by non-invasive measurements in the finger (Finometer) and middle cerebral artery (transcranial Doppler ultrasound), respectively. Transfer function analysis (TFA) was used to estimate values of ARI by optimal fitting of template curves to the output (or CBFV) response to a step change in input (or BP). Two-step criteria were adopted to accept estimates of ARI as valid. The 95% confidence limit of the mean coherence function (0.15–0.25 Hz) () was estimated from 15 000 runs, resulting in = 0.190 when using five segments of data, each with 102.4 s (512 samples) duration (Welch’s method). This threshold for acceptance was dependent on the TFA settings and increased when using segments with shorter duration (51.2 s). For signals with mean coherence above the critical value, the 5% confidence limit of the normalised mean square error (NMSEcrit) for fitting the step response to Tieck’s model, was found to be approximately 0.30 and independent of the TFA settings. Application of these criteria to physiological and clinical sets of data showed their ability to identify conditions where ARI estimates should be rejected, for example due to CBFV step responses lacking physiological plausibility. A larger number of recordings were rejected from acute ischaemic stroke patients than for healthy volunteers. More work is needed to validate this procedure with different physiological conditions and/or patient groups. The influence of non-stationarity in BP and CBFV signals should also be investigated.