This single-center, experimental, randomized, controlled, double-blind, crossover trial did not demonstrate an effect of oxygen compared to air on the changes in the primary outcome cardiac output during simulated blood loss in healthy volunteers. Neither did we find an effect of oxygen on the changes in the secondary outcomes stroke volume or MCAV. However, we observed a reduction in MCAV occurring before LBNP exposure that persisted during LBNP. Interestingly, we did find that oxygen improved tolerance to simulated blood loss. In our exploratory analyses, we found an increase in ScO2 occurring before LBNP exposure that persisted during LBNP, and a small but statistically significant effect of oxygen on the changes in heart rate, yielding a less pronounced heart rate increase during LBNP.

Our findings suggest that supplemental oxygen does not affect the changes in cardiac output during simulated blood loss in healthy volunteers, which is consistent with the previous work [13]. Since we applied a greater range of LBNP than the previous study, our results also suggest that the magnitude of LBNP does not influence the effect of oxygen on cardiac output.

While few have explored how oxygen influences hemodynamics during LBNP, many studies have examined the effect of oxygen at rest. We did not find a statistically significant main effect of oxygen on cardiac output, which seemingly contrasts a meta-analysis reporting a mean reduction of 10.2% in cardiac output when inhaling 100% oxygen [5]. However, it is important to note that data used in our regression analysis included all LBNP levels, as our trial was not designed and powered to study the effect of oxygen without LBNP. Also, in contrast to most studies in this meta-analysis, the present trial had a randomized, double-blinded, crossover design.

It may seem paradoxical that we observed an effect of oxygen on heart rate but not on stroke volume and cardiac output, given that cardiac output is the product of the former two. While the effect of oxygen on stroke volume was not statistically significant, our results indicate a tendency toward an increase with oxygen.

From baseline to LBNP, we observed a reduction in MCAV with oxygen that persisted during, and was not altered by, LBNP. Despite contention, MCAV measured by transcranial Doppler ultrasound is assumed to reflect cerebral blood flow under certain conditions [25, 26]. Assuming a constant diameter of the middle cerebral artery, the present finding suggests that cerebral blood flow may be reduced by supplemental oxygen during simulated blood loss, but we observed no synergistic effect between oxygen and simulated blood loss. It has been suggested that changes in arterial oxygen content are outweighed by opposite changes in cerebral blood flow to maintain cerebral oxygen delivery at or near normal levels [27]. We did not measure arterial partial pressure of oxygen (PaO2) in the present trial. However, if we assume an increase in PaO2 from 13 to 73 kPa as shown in a similar study [28], that MCAV reflects cerebral blood flow and that SpO2 reflects SaO2, our data suggest that cerebral oxygen delivery was similar when inhaling oxygen and air (Additional file 4, Figure S2 and S3).

In contrast to the reduction in MCAV with oxygen from baseline to LBNP, there was an increase in ScO2 that persisted during LBNP. Consequently, ScO2 was elevated at higher LBNP levels when breathing oxygen compared to air. However, it is worth noting that ScO2 measured by NIRS captures tissue oxygen saturation only in a region of the anterior brain. In addition, the observed increase with oxygen could be affected by extracranial contamination due to the elevated skin oxygenation. If this indeed reflects a genuine increase in cerebral tissue oxygenation, we either underestimated the increase in CaO2 with oxygen, or the changes in MCAV were larger than true cerebral blood flow changes.

Interestingly, we did find that oxygen improved tolerance to simulated blood loss without accompanying changes in the predefined systemic hemodynamic variables. While MCAV decreased with supplemental oxygen, ScO2 increased, which could elucidate the improved tolerance to simulated blood loss with oxygen. In our exploratory analyses, we did find a small and statistically significant effect of oxygen on the changes in heart rate during LBNP. One could speculate if the reduction in heart rate could have prolonged the diastole and thereby improved cardiac filling which would fit with the statistically non-significant trend of greater stroke volumes with oxygen.

We observed no effect of oxygen on the changes in SVR and MAP during LBNP, nor did we find any effect from baseline to LBNP. The latter finding appears to diverge from prior research reporting vasoconstriction elevating SVR and MAP during inhalation of 100% oxygen at rest [5]. The suggested rise in SVR with hyperoxia has been proposed to trigger the heart rate reductions, although the underlying mechanism is not well understood [6]. However, it is crucial to note that the present study was neither specifically designed nor adequately powered to examine the isolated effects of oxygen without LBNP, which may account for the disparate findings.

Oxygen therapy in trauma patients is a debated topic and retrospective studies have reported both favorable and unfavorable associations between oxygen and outcome [29]. The TRAUMOX2 trial was initiated due to the lack of randomized trials and hypothesizes that a restrictive compared to a liberal oxygen strategy will improve outcome [30]. We did not observe significant changes in systemic hemodynamics during oxygen therapy that would indicate a disadvantageous effect of oxygen in trauma patients. On the contrary, we found an increased tolerance to simulated blood loss.

While we did not find significant influence of supplemental oxygen on systemic hemodynamics, we did find an effect on cerebral hemodynamics. Future studies should examine whether the observed effect translates to a clinically relevant effect on global or regional cerebral oxygen delivery, e.g., in patients with traumatic brain injury. We observed a reduction in cerebral blood velocity with oxygen during LBNP. Whether this reflects lower cerebral blood flow under the present conditions remains to be elucidated. Furthermore, future work should investigate whether the increase in arterial oxygen content with hyperoxia counteracts any potential decrease in cerebral blood flow, thereby sustaining adequate oxygen delivery. Based on our finding of improved tolerance to simulated blood loss with oxygen, future studies should also explore if this effect is present in bleeding trauma patients, assess its clinical relevance, and explore possible mechanism.

There are important differences between traumatic and simulated blood loss. In the LBNP model, there is no tissue damage, and the LBNP and oxygen exposure in the present trial had a maximum duration of 24 min, which is less than the typical duration of traumatic blood loss and oxygen exposure, and also not covering a reperfusion phase after volume resuscitation. Consequently, effects that may occur after the acute phase of blood loss are not examined in our study. Also, the present trial only included healthy awake subjects who were breathing spontaneously without concomitant pain or traumatic brain injury which limits the external validity of our study.

Four subjects exhibited hemodynamic decompensation during oxygen visits compared to 10 during air visits, indicating a separation between the two conditions. Since most decompensations during both oxygen and air visits were due to reductions in MAP, there was no evident effect of treatment on the cause of decompensation. Despite decompensation being defined by both subjective (symptoms) and objective (MAP and heart rate) criteria, our results suggest that the majority were driven by objective criteria. Shared frailty models account for random effects and censoring, which is important for our crossover design as LBNP tolerance is reproducible [12]. Consequently, we contend that there exists a credible and objective distinction between oxygen and air.

We employed non-invasive methods in the present trial as they involve low risk in healthy volunteers. Although non-invasive methods in general might be less reliable than invasive methods, we believe they are adequate in the present trial. Stroke volume measured by suprasternal pulsed Doppler ultrasound has been validated previously by Eriksen and Walløe [19]. In brief, blood velocity in the LVOT is maintained with a rectangular (not parabolic) velocity profile for some centimeters into the ascending aorta. Importantly, this velocity is maintained even if the diameter of the sinus of Valsalva and the proximal aorta exceeds that of the LVOT. Consequently, any change in aortic dimensions with LBNP [31] should not affect these measurements. As the ultrasound probe sits well in the suprasternal notch, a fixed angle to the ascending aorta is easy to maintain. The angle of the aorta has been demonstrated to change little with LBNP [31], and a caudal displacement of the heart with LBNP should not be of significance due to the preservation of velocity for the initial centimeters within which the measurements are performed. The diameter of the LVOT was measured only once, as this is believed to be a fixed, fibrous structure [32].

There was a large between-subject variability in cardiac output. Different body sizes, inaccuracies in LVOT measurements and different angles of suprasternal ultrasound insonation may partially explain this. However, these factors would have a minor influence on the results as we analyzed changes from baseline with mixed regression.

Due to large between-subject variability in blood pressure [33] we used a relative reduction in blood pressure, and not an absolute value as a threshold for hemodynamic decompensation. Further, we used MAP rather than systolic pressure as the former determines flow by the Poiseuille equation and thus oxygen delivery.

When calculating mean values for each LBNP level, the values were trimmed for the 5% highest and lowest values to reproducibly and objectively remove outliers caused by, e.g., motion artifacts and extrasystoles. In general, with increased trimming, the mean approaches the median. The level of trimming was largely arbitrarily chosen, but by visual inspection the calculated mean values seemed to fit the observations well. Also, changing the degree of trimming to 0%, 2.5% or 10% did not change the conclusions for the primary outcome.

While our study featured a limited number of subjects, the incorporation of a crossover design, repeated measurements, and a standardized intervention protocol increased the statistical power for each subject. We calculated the sample size to be able to detect, with a reasonable probability, what is considered a clinically relevant difference in the primary outcome cardiac output [22]. We do not believe that a type II error is likely, as the confidence interval does not encompass what we consider a clinically meaningful effect, and we therefore believe the sample size was adequate. The assumptions in the sample size calculation are best judged by the confidence intervals of the estimates [34]. Determining what constitutes a clinically significant effect is, to some extent, subjective and may also be contingent on the context. Importantly, the sample size estimation was performed on the primary outcome, and the study may have been underpowered to detect treatment effects on the secondary and exploratory outcomes.