134 A. G. Beshish et al.: J Extra Corpor Technol 2025, 57, 129 – 136 However, the delineation of risk categories in different patient populations remains unclear.
There is no generally accepted value to discriminate pathologic hyperoxia. Injurious hyperoxia may vary by patient population and clinical context [ 25 ]. Poor outcomes may occur when PaO 2 exceeds a certain threshold, where endogenous antioxidants are unable to prevent oxidative stress. When high amounts of oxygen are introduced to previously ischemic tissues, this leads to the generation of reactive oxygen species( ROS) and activation of inflammatory pathways via cytokines and other immunological signaling pathways [ 29 ]. The generation of ROS leads to lipid peroxidation and protein changes, resulting in cell injury. In patients who have experienced CA or resuscitation aftershock, an increase in ROS may deplete plasma antioxidant potential, which may lower the threshold for subsequent oxidative injury [ 14, 25 ], and this effect may be more pronounced in neonates and infants due to immature antioxidant defenses [ 4 ]. The effect of relative hyperoxia may be even more pronounced in patients with cyanotic heart disease who have significantly lower baseline PaO 2. It is not known if the antioxidant systems of the body are downregulated in patients with chronically lower baseline PaO 2. When placed on ECLS, these patients are exposed to a relative hyperoxia state. It is plausible that these patients are more vulnerable to supraphysiologic oxygen, the extent of which is yet unknown.
Because there is no generally accepted definition of hyperoxia in neonates with congenital heart disease with various physiology and complexity in different settings, we used a ROC curve analysis in this specific cohort to determine which PaO 2 values may be associated with adverse outcomes. This similar strategy was employed by Sznycer-Taub et al., and Beshish et al. in two separate reports. Sznycer-Taub and colleagues evaluated hyperoxia in pediatric cardiac patients( neonates and infants) supported on VA-ECLS and found that a PaO 2 of 193 mmHg in the first 48 h was determined to have good discriminatory ability about 30-day mortality [ 4 ]. Using a similar strategy, Beshish and colleagues showed that a PaO 2 of 313 mmHg for infants undergoing cardiac surgery utilizing CPB was independently associated with 30-day mortality [ 9 ]. Our cut-off definition of hyperoxia was very close to that identified by Sznycer-Taub et al. although the patient population that was used by Sznycer-Taub was slightly different, as the former study captured all infants who required VA-ECLS following cardiac surgery. In that study, the cutoff point remained the same( PaO 2 of 193 mmHg) in a subgroup analysis of all neonates( n = 70), and in neonates who underwent a Norwood operation( n = 35).
Limitations
Our findings are subject to all limitations inherent to singlecenter retrospective cohort studies. Although samples to measure PaO 2 were obtained at dedicated time intervals, it is not possible to discern the effect of time spent in a hyperoxia state as opposed to the effects of acutely high PaO 2 levels. Additionally, there may be some bias as to which patients are exposed to hyperoxia. The majority of our cohort had a PaO 2 level around or over 200 mmHg while on VA-ECLS, which limited our ability to study the relationship between lower oxygen tension levels and outcomes. Although we identified a cut point for PaO 2 of 233 mmHg using an AUC, the sensitivity was 36 %. The sensitivity is low, and this is clearly a limitation of our study that we think can be overcome with a larger patient population that is more homogenous. Importantly, many of these limitations can be addressed in a multicenter validation study, which our group is currently pursuing. It is acceptable that the AUC for the model is low, which to the complexity of hyperoxia, which might not be properly captured by the variables in the model. There is also the possibility of interaction between predictors, which was not accounted for. An alternative way that can be done in future analyses is to combine different sets of predictors or explore other analytic approaches that might be able to yield strong associations within certain patient populations. Also, when presenting findings in Figure 3, we show that the correlation between PaO 2 level, duration of ECLS and mortality was not significant. This could be related to many factors such as the heterogeneous patient population, patient response to ECLS exposure, severity of illness, and other Confounding factors, which can also play a crucial role, such that the important predictors might not have been accounted for in the correlation, to name a few.
Conclusions
Of the 229 VA-ECLS runs in 209 patients, hyperoxia during the first 48 h of VA-ECLS defined by a mean PaO 2 > 233 mmHg, occurred in approximately 30 % of runs. Patients in the hyperoxia group were older and had higher median initial ECLS flow in the first 2 h. The mortality rate was higher in the hyperoxia group, and although it did not reach statistical significance, patients in the hyperoxia group had higher odds of mortality( p = 0.052). Multicenter and prospective evaluation of this modifiable risk factor is imperative to improve the care of this high-risk cohort.
Funding This research did not receive any specific funding.
Conflicts of interest The authors declared no conflict of interest.
Data availability statement All available data are incorporated into the article.
Author contribution statement
A. B. and H. V. designed the study. P. R.-M., R. S., J. Q., K. K.-L., T. Z., and A. B. performed the research and analyzed the data. A. B. wrote the manuscript, and all authors contributed to the final version.
Ethics approval
Data collection was conducted as a retrospective cohort study to determine the ranges of PaO 2 exposure and the potential association between exposure to hyperoxia and poor outcomes. Our primary aim was to determine if hyperoxia while on VA-ECLS was associated with