Cerebral Autoregulation (CA) maintains a constant cerebral blood flow despite changes in blood pressure within an individualized range that is specific to each patient [1]. The mean arterial pressure (MAP) value at which cerebral blood flow (CBF) begins to decrease is called the Lower Limit of Autoregulation (LLA). During cardiopulmonary bypass (CPB) in the context of cardiac surgery, maintaining a MAP below the LLA is associated with an increased risk of postoperative morbidity [2, 3, 4, 5, 6]. The large interindividual variations in LLA make it necessary to use an individualized technique for the correct calculation of each patient’s LLA in order to define the “best” MAP, allowing for appropriate CBF [2].

Using continuous transcranial doppler to measure the CBF alongside an invasive arterial catheter to measure the MAP, the LLA can be determined by a continuous calculation of the correlation between cerebral blood flow velocity ( mean velocity of the mean cerebral artery (MV)) and MAP, also known as the mean velocity index (Mx) [2, 5, 7]. The Mx is a moving Pearson correlation coefficient and approaches the value of 1 when there is a high correlation between MAP and MV (outside of autoregulation) and approaches 0 when the MAP is on the plateau of autoregulation. In order to determine the LLA, a range of MAP values are required to obtain a correlated and uncorrelated relationship with the MV [8]. Classically, the LLA was mainly used within the neuro intensive care unit and was calculated using a large time window (at least five minutes to get the first value of Mx, with a new value each minute) [9]. This large averaging time has pros and cons. Extreme and/or aberrant values are less impactful and can be included in the final calculation. Conversely, if these extreme values are real, they will have a smaller effect on the calculated Mx. Additionally, a minimum of 15–20 minutes for the standard LLA calculation may be too long for cardiac surgery setting. Increasing the recording frequency of paired data (MAP and MV) may allow for a faster assessment of LLA and increase the integration of such values into clinical decisions [10].

The main objectives of our study were to demonstrate the clinical feasibility of a quick determination of LLA (qLLA) throughout a 15-minute period during CPB, and to compare this qLLA with the reference LLA (rLLA), which was calculated throughout the entire CPB period. Secondary objectives were to analyse the interindividual variability of qLLA and rLLA.

117 patients were eligible while 55 patients were enrolled (Fig. 3). Patient’s characteristics are shown in the Table 1. qLLA and rLLA were calculable in 78% (n = 43) and 47% (n = 26) of the cases, respectively. Mx${}_{\text{2s}}$ was consistently $>$0.4 in 7% (n = 4) (NT rate) for qLLA determination. NT rate was 35% (n = 20) for rLLA determination with an Mx${}_{\text{10s}}$ consistently $>$0.4 for 18 cases. Non calculable (NC) LLA was found in 15% (n = 8) for qLLA and 16% (n = 9) for rLLA. Average MAP was under the rLLA during 63% of the CPB time (25th–75th: 41–80%). 22 pairs of simultaneous qLLA and rLLA were available. Mean bias between qLLA and rLLA was –3 mmHg (95% CI: –7; +0) and limits of agreement between qLLA and rLLA were –19 mmHg (95% CI: –13; –25), and +13 mmHg (95% CI: +6; +19).

Fig. 3.

Flow chart. Notes: qLLA, MAP threshold below the cerebral autoregulation plateau calculated with Mx${}_{\text{2s}}$; rLLA, MAP threshold below the cerebral autoregulation plateau calculated with Mx${}_{\text{10s}}$. MAP, mean arterial pressure; qLLA, quick determination of LLA; rLLA, reference LLA.

Experimenters agreed on identical values of LLA in 48% of the cases for qLLA and in 73% of the cases for rLLA. Coefficient kappa was 0.36 (95% CI: 0.20–0.52), and 0.45 (95% CI: 0.24–0.66), for qLLA and rLLA, respectively (fair and moderate agreement). The dispersion for qLLA and rLLA ranged from 46 to 85 mmHg and 60 to 85 mmHg, respectively (Fig. 4).

Fig. 4.

Individualized qLLA (blue) and rLLA (orange) values. Notes: qLLA, lower limit of autoregulation determined by the Mx${}_{\text{2s}}$ in a short period of 15 minutes; rLLA, lower limit of autoregulation determined by the Mx${}_{\text{10s}}$ during the CPB. MAP, mean arterial pressure; CPB, cardiopulmonary bypass; qLLA, quick determination of LLA; rLLA, reference LLA.

This feasibility study demonstrates that the calculation of cerebral autoregulation with transcranial doppler was possible in 78% of the cases using a quick assessment of LLA on 15 minutes recording compared to only 47% of the cases using the standard analysis on the total CPB recording. The Gaussian distribution of qLLA values (Fig. 4) similar to the distribution of rLLA values suggests that these values are physiological [2].

The large limits of agreement between qLLA and rLLA ($\pm$15.8 mmHg) prevent any interchangeability between the two calculations. The poor interobserver agreements and the high frequency of the inability to calculate LLA raises the concern for any bedside applicability of this cerebral autoregulation analysis in this context of observational study. On the other hand, the dispersion of qLLA values was greater, especially in the lower values, which supports the hypothesis of a loss of data by excessive averaging with the reference technique.

Of note, two studies determined the CA with the rLLA for a Mx value of 0.4, but they precised that rLLA was also chosen with the lowest Mx value when the value of 0.4 was not reached [12, 13]. Unfortunately, the authors did not precise the rate of these observations.

Compared to the current use of the longer sampling (Mx${}_{\text{10s}}$), we hypothesized that a shorter sampling (Mx${}_{\text{2s}}$) would be more suitable for periods with rapid and intense variations in MAP, such as during cardiac surgery. By analysing these extreme MAP and MV values during a shorter averaging period, we thought that the determination of LLA could be done faster and potentially more effectively. We also considered that if a determination of LLA was able to be performed within a short period of time it may prove valuable for personalizing the subsequent MAP values to avoid dropping below each patient’s LLA, potentially reducing perioperative morbidity.

There are a few potential reasons for our mitigate results. First, the software automatically calculates the LLA according to a predefined period, a predefined sampling of Mx and a predefined threshold value (0.4 for the Mx). As any potential artifacts are incorporated into the Mx calculation, an observer needs to “clean out” these specific periods to ensure appropriate values and then choose individually the 15 minutes period for calculating qLLA. This could explain the large interobserver variability, as mentioned before. Second, the observational design of this study prevented from controlling numerous parameters influencing cerebral autoregulation during cardiopulmonary bypass. We hypothesize that the true individualized LLA is not constant throughout the perioperative period, as the impact of PaCO${}_2$, PaO${}_2$, temperature, type of flow, and level of flow likely modify the LLA [14, 15]. Finally, we also excluded patients presenting a significant arterial carotid stenosis. This population could further be assessed because of a potential higher sensitivity of the MV to the MAP.

Moreover, as we did not deliberately change the MAP, ranges of MAP were sometimes narrow which could explain why no threshold was obtained in 35% of the cases for rLLA. Most of these cases presented indeed a Mx${}_{\text{10s}}$ $>$0.4 (Fig. 2). Furthermore, patients with a high PaCO${}_2$ may have a significantly impaired cerebral autoregulation threshold [16].

This study has limitations. Firstly, there is a limited sample size, due in part to the high rate of unusable MV tracings. We were sometimes transiently faced with a poor Doppler signal despite the initial modifications of the gain, the filtering, and the power of the signal. Hence, the calculation of the mean velocity may have been partially incorrect because of an impropriate signal to noise ratio. Secondly, our institution has only recently started using perioperative transcranial doppler (TCD) technology. As our experience grows, the efficiency and accuracy of the recorded signals will likely improve. During cardiac surgery, an appropriate TCD signal can be challenging to maintain due to patient mobilization and use of electrosurgical unit [17].

Of note, analysis of Mx is one way to calculate cerebral autoregulation and is not considered to be the gold standard despite strong correlation with postoperative complications [2, 4, 5]. Other calculation techniques have been proposed [7], as well as other CBF monitoring techniques, such as cerebral oximetry [18]. The clinical challenge is to find an appropriate method that can continuously analyze the relationship between MAP and CBF despite confounding factors such as hemodilution or PaCO${}_2$ [16]. This also could explain why there were 35% of NT cases during rLLA assessment (long period of calculation, with a mean CPB time of 66 minutes) and only 7% during qLLA assessment (15 minutes period of calculation). Interestingly, 63% of the aortic clamping period was spent with MAP values lower than the rLLA, which seems correlated to NT cases under the cerebral autoregulation plateau (Mx${}_{\text{10s}}$ $>$0.4) during all the CPB. In our study, we retrospectively collected five cases of postoperative delirium and one haemorrhagic stroke. However, in the current practice, we do not assess systematically postoperative cognitive function, which probably led to miss significant neurological insults. Protocols aiming at maintaining a MAP above the LLA have shown conflicting results, with one having a beneficial impact on postoperative delirium [13], and another demonstrating no impact on long term neurologic complications [12].

Determination of qLLA during CPB is feasible. However, the large limits of agreement between qLLA and rLLA prevent any interchangeability. Additionally, interobserver variation limits bedside applicability for both qLLA and rLLA in a non-controlled environment. Further studies aimed at modifying MAP to actively determine the LLA may limit the impact of confounding factors.

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

All authors contributed substantially to the study and were involved in scientifically significant ways. All authors gave their final approval to the manuscript. They are all agreed to be accountable for all aspects of the work. OD, LG, and RM designed the research study. JN, OD, EB, MD recruited patients and performed the study. JB, AJ and BA provided help and advises about the transcranial doppler technic. All authors contributed to the redaction of the article. All authors read, modified and approved the manuscript.

This study had been submitted to the ethic review board of Ramsay Sante, which accepted the study (The Ethic number is 00010835). The patients were informed their datas would be used in this study and nobody gave an opposition.

Not applicable.

This research received no external funding.

The authors declare no conflict of interest.