Introduction

Admission to hospital as well as 30-day survival after out of hospital cardiac arrest (OHCA) has increased in recent years and most 30-day survivors after OHCA are discharged with good neurological function [1]. Despite these advances, the proportion of patients dying after hospital admission is more than 50 % and the major causes are the primary ischemic cerebral injury sustained during the no-flow time of the OHCA and the additional secondary cerebral reperfusion injury that commences at return of spontaneous circulation (ROSC) [2, 3]. Reperfusion entails increased reactive oxygen species (ROS) production, mitochondrial dysfunction and apoptosis, and thus, exacerbates the detrimental consequences of the OHCA [3]. Targeted temperature management (TTM) has been suggested as an intervention to attenuate these effects but studies are inconclusive and current studies indicate varying use internationally [4,5,6,7,8]. Recent data suggest that elevated arterial partial pressure of carbon dioxide (PaCO2), hypercapnemia, might improve neurological outcome after OHCA. Possible underlying mechanisms include decreased cerebral vascular resistance (CVR), increased cerebral blood flow (CBF), modulation of inflammatory processes and anti-convulsive properties [9,10,11,12,13,14,15,16]. In contrast to hypercapnemia, low PaCO2, hypocapnemia, increases CVR, decreases CBF, reduces oxygen delivery (CDO2) and is associated with poor outcome [10, 16,17,18,19]. Low arterial partial pressure of oxygen (PaO2), hypoxemia, is the primary source of neuronal injury occurring during the OHCA and a determinant of neurological outcome [3, 20]. Elevated PaO2, hyperoxemia, has also been associated with poor neurological outcome, possibly due to increased lipid oxidation, production of ROS, mitochondrial damage and reduced CBF [3, 21,22,23]. The association of combinations of extreme PaO2 and PaCO2 values after OHCA with outcome have less frequently been studied, but the combination of moderate hypercapnemia and mild hyperoxemia was association with improved neurological outcome in one study [24]. Overall study results are inconsistent and other investigations trying to confirm the protective or harmful associations of exposure to extreme PaO2 and PaCO2 values with neurological outcome were unable to do so [25,26,27]. Moreover, the available studies differ in methodology, inclusion criteria and may lack sufficient power. Therefore, we conducted this study of the International Cardiac Arrest Registry (INTCAR) 2.0 database to investigate the association between exposure to extreme PaCO2 and PaO2 values and neurological outcome at hospital discharge in a large cohort of adult, unconscious patients with sustained ROSC after OHCA.

Methods

INTCAR 2.0 is an international multicenter database including cardiac arrest patients admitted to intensive care units (ICU) at 22 medical centers in the United States and Europe. The present investigation of the INTCAR 2.0 database included prospectively collected cardiac arrest and treatment data from adult (≥18 years of age), unconscious (GCS < 8), OHCA patients with sustained ROSC. All patients in this study received TTM treatment and were admitted between 2008 and 2018. Patient data collected in the database was anonymized and OHCA data was reported according to the Utstein-style protocol [28]. Ethical committees in each participating country approved the data collection and analysis. Informed consent was either waived or obtained from all participants or relatives according to national and local standards, in line with the Helsinki declaration. Reporting of our analyses was guided by the STROBE recommendations [29].

Definition of PaO2 and PaCO2 groups and data registration

In the INTCAR 2.0 protocol, extreme PaO2 or PaCO2 exposure thresholds were defined as PaO2 > 40 kPa, PaO2 < 8.0 kPa, PaCO2 > 6.7 kPa and PaCO2 < 4.0 kPa. Exposure to one or more extreme values during the first 24 h after ROSC was registered in a dichotomous manner (yes/no). The PaO2 and PaCO2 thresholds were aligned with previous studies [17, 21, 22]. Additionally, the single highest and lowest PaO2 values and the lowest PaCO2 value during the first 24 h after ROSC were documented, regardless of exposure level. In total 7 data-points (4 PaO2 and 3 PaCO2 data-points) were collected per patient. For the purpose of this study we divided patients according to their extreme PaO2 or PaCO2 value exposure into four groups defined by the extreme values in the INTCAR 2.0 protocol; hyperoxemia (PaO2 > 40 kPa), hypoxemia (PaO2 < 8.0 kPa), hypercapnemia (PaCO2 > 6.7 kPa) and hypocapnemia (PaCO2 < 4.0 kPa). Patients not exposed to extreme values were classified as PaO2 and PaCO2 no-exposure (PaO2 8.0–40 kPa and PaCO2 4.0–6.7 kPa). Patients exposed to more than one extreme value were included in all exposure groups.

Outcome

To better compare with previous analyses [21, 22, 27, 30], cerebral performance category (CPC) at discharge from hospital was chosen as primary outcome endpoint. After neurological assessment at hospital discharge by a trained health care professional OHCA patients were allocated to one of the five CPC categories, ranging from CPC1 (good cerebral performance/mild disability), CPC2 (moderate disability), CPC3 (severe disability), CPC4 (coma state) and CPC5 (brain death) [31, 32]. For this study we dichotomized outcome into good (CPC1 and 2) and poor (CPC3–5). Delayed outcomes, typically around 6 months after presentation, were also collected, by telephone interview or medical records.

In our primary analysis, we tested the association of exposure to extreme PaO2 or PaCO2 values with outcome. We conducted 8 analyses: 1.the hyperoxemia group was compared to the PaO2 no-exposure group and 2.to patients without hyperoxemia (no-hyperoxemia). The hypoxemia group was compared 3.to the PaO2 no-exposure group and 4.to patients not exposed to hypoxemia (no-hypoxemia). Patients in the hypercapnemia group were compared 5.to patients in the PaCO2 no-exposure group and 6.to patients without hypercapnemia exposure (no-hypercapnemia), while patients with hypocapnemia exposure were compared 7.to the PaCO2 no-exposure group and 8.to patients not exposed to hypocapnemia (no-hypocapnemia).

In previous studies, exposure to hyperoxemia, hypoxemia and hypocapnemia were associated with poor outcome while hypercapnemia was associated with good outcome [13, 18, 21, 22]. In our secondary analyses we therefore, a priori, defined exposure groups to investigate these findings and compared patients exposed to the combination of hyperoxemia with hypocapnemia to a PaO2 and PaCO2 no-exposure group, followed by a PaO2 no-exposure group with hypercapnemia compared to the PaO2 and PaCO2 no-exposure group. Subsequently, we designed regression models with ascending and descending PaO2 values from < 20 kPa to > 60 kPa and > 8.0 kPa to < 5.0 kPa to define a possible threshold for the onset of the association of hyperoxemia or hypoxemia and poor outcome. We also designed a regression model for the onset of the association of hypocapnemia and poor outcome with descending PaCO2 values from > 4.0 kPa to < 3.5 kPa.

Sensitivity analyses

Sensitivity analyses were performed for our primary analyses with all double exposed patients (hyperoxemia and hypoxemia or hypercapnemia and hypocapnemia) removed. Furthermore, we performed sensitivity analyses of our primary analyses, replacing outcome at discharge with long-term outcome at 6-month follow-up.

Statistical analysis

Proportions are presented as numbers and percentages and continuous variables as means with standard deviations (SD) or medians with interquartile ranges (IQR). Logistic regression analysis was used to assess the association between PaO2 and PaCO2 and neurological outcome at discharge. For the ascending analysis the odds ratio (OR) above the threshold was compared to the OR under the threshold, while for the descending analyses the OR under the threshold was compared to the OR above the threshold. All analyses were adjusted for pre-specified, and OHCA relevant co-variates: age (years), sex (male/female), previous chronic heart failure (yes/no), previous chronic obstructive pulmonary disease (COPD) (yes/no), cardiac arrest witnessed (yes/no), bystander cardiopulmonary resuscitation (yes/no), initial rhythm shockable (yes/no), time to ROSC, admission GCS-M 1 vs 2–5, circulatory shock, TTM-treatment (low, 32–34 degrees Celsius (°C) versus high, 35–37 °C) and pH on admission as fixed effects and treatment site as a random effect. None of the independent variables included in our models were highly correlated. We conducted two-sided tests and considered a P-value < 0.05 as significant. We included patients with complete data (PaO2 and/or PaCO2 and CPC at discharge registered) in the primary and secondary analyses. However, for our long term outcome sensitivity analysis, we imputed missing outcome (last observation (CPC at discharge) carried forward). Analyses were conducted using R: A language and environment for statistical Computing (version 3.3.3 R Foundation for Statistical Computing, Vienna, Austria) [33].

Results

The INTCAR 2.0 database included 2162 OHCA patients who were assessed for eligibility. Of this cohort, we excluded 27 patients who experienced OHCA but were not unconscious on admission. The remaining 2135 patients were included in our final analysis (Fig. 1). Baseline data for this group is displayed in Table 1. Baseline data for the different PaO2 and PaCO2 exposure groups are displayed in the Additional File, Tables 1 and 2. Six hundred eighteen (28.9%) patients experienced a good outcome and 1517 (71.1%) a poor outcome. Eight hundred twenty-eight (38.8%) patients were alive at discharge, while 1307 (61.2%) were dead. At 6-month follow-up the outcome of 634 (29.7%) patients was good, whereas 1501 (70.3%) patients had a poor outcome in the cohort with imputed data. The cohort without imputation showed a good outcome in 450 (24.3%) and a poor outcome in 1400 (75.7%) patients. All patients received TTM treatment during the first 24 h after ROSC, 1673 (78.4%) to target temperature 32–34 °C and 462 (21.6%) to 35–37 °C. Three hundred and fifty-seven (18.7%) patients were exposed to hyperoxemia, 343 (17.9%) patients to hypoxemia and 76 (3.9%) to both, while 670 (34.5%) patients experienced hypercapnemia, 458 (23.6%) hypocapnemia and 222 (11.4%) both. During the first 24 h after OHCA, median highest PaO2 was 25.7 (IQR 18.5–38.1) kPa, median lowest PaO2 was 10.0 (IQR 8.1–12.7) kPa and median lowest PaCO2 was 4.3 (IQR 3.7–4.9) kPa.

Fig. 1
figure 1

Patient selection pathway. OHCA = out-of-hospital cardiac arrest, n = number, PaO2 = arterial partial pressure of oxygen, PaCO2 arterial partial pressure of carbon dioxide, vs = versus

Full size image
Table 1 Baseline characteristics of patients included in the PaO2 and PaCO2 analyses, n = 2135
Full size table

In our primary analyses we found, after adjustment, neither hyperoxemia nor hypoxemia exposure in the first 24 h after ROSC to be associated with poor neurological outcome (all analyses, P = 0.13–0.44) (Table 2). Exposure to hyper- or hypocapnemia during the first 24 h after ROSC was also not associated with poor outcome (all analyses, P = 0.18–0.49) (Table 2).

Table 2 Association of exposure to extreme PaO2 and PaCO2 values with poor neurological outcome
Full size table

In our secondary analysis the outcomes for patients exposed to the combination of hyperoxemia with hypocapnemia showed no association with poor neurological outcome (P = 0.11, Table 3). The exposure combination of hypercapnemia with PaO2 no-exposure was also not associated with poor outcome (P = 0.86, Table 3). Figure 2a and b depict the adjusted OR with 95% CIs for poor neurological outcome across ascending and descending PaO2 cut off values. Figure 2c shows the adjusted OR with 95% CIs for poor neurological outcome across descending PaCO2 cut off values. We did not detect a significant threshold value for the onset of an association with poor outcome in any of these three analyses.

Table 3 Association of PaO2 and PaCO2 combinations with poor neurological outcome
Full size table
Fig. 2
figure 2

a-c. Forest plot showing the adjusted ORs (bullet points) with 95% CI (horizontal lines) for poor neurological outcome (CPC 3–5) across ascending PaO2 cut-off points (a), descending PaO2 cut-off points (b) and descending PaCO2 cut-off points (c). ORs and CIs are presented on a logarithmic scale. For (a), OR above 1.0 indicates worse outcome above the PaO2 threshold. For (b) and (c), OR above 1.0 indicates worse outcome under the PaO2 or PaCO2 threshold. OR = Odds ratio, 95% CI = 95% confidence interval, CPC = cerebral performance category, PaO2 = arterial partial pressure of oxygen, PaCO2 = arterial partial pressure of carbon dioxide. All analyses were adjusted for co-variates

Full size image

Sensitivity analyses

The results of the sensitivity analysis with all double exposed patients (hyperoxemia and hypoxemia or hypercapnemia and hypocapnemia) removed were similar to the results of our primary analyses (P = 0.07–0.29) (Additional File, Table 3). Replacing outcome at discharge with long term outcome in our primary analyses did not change our results significantly, neither in the dataset without imputed outcome measures (P = 0.14–0.89) nor in the dataset with missing outcome measures imputed (P = 0.13–0.59) (Additional File, Table 4 and 5).

Missing data

244 patients had one or more PaO2 or PaCO2 data points missing. Comparing this group with the group of patients with complete PaO2 and PaCO2 data (n = 1891) showed similar values at baseline (Additional File, Table 6).

Discussion

In this exploratory study testing the associations between exposure to extreme PaCO2 and PaO2 values and neurological outcomes at discharge of 2135 patients with OHCA, we found that exposure to extreme PaO2 and PaCO2 values was common, but not significantly associated with neurological outcome after adjusting for in the context of OHCA-relevant covariates. In our subsequent analyses, we did not show any significant associations of combinations of PaO2 and PaCO2 and poor neurological outcomes. Despite investigating PaO2 values to > 60 kPa and < 5.0 kPa and PaCO2 values to < 3.5 kPa in our ascending and descending cut-off point analyses, we did not identify a numerical threshold for the onset of the association of each variable with poor neurological outcome. These findings suggest that PaO2 and PaCO2 may not be directly associated with outcome after resuscitation from OHCA. Animal studies have shown worse neurological outcomes and increased neurological injury after exposure to hyperoxemia following resuscitation from cardiac arrest and indicate that hyperoxemia in the post cardiac arrest phase might be harmful [34]. These findings have been corroborated by retrospective observational human studies [21, 35, 36]. Moreover, a threshold for the onset of poor outcome has been proposed at 40 kPa [21]. Elmer et al. confirmed the previously suggested hyperoxemia threshold of 40 kPa for the onset of poor outcome but also showed that moderate hyperoxemia (PaO2 13.5–39.9 kPa) was associated with lower SOFA scores at 24 h, indicating a possibly beneficial effect at these levels [37]. This finding was supported by a study of Helmerhorst et al. investigating 5258 cardiac arrest patients, displaying a U-shaped relationship between PaO2 and outcome and, although not significant, the lowest probability of in-hospital death between 13.6–40 kPa [38]. However, not all investigations support these results [27, 39], and studies are frequently of retrospective design, correct for different confounders and investigate mixed IHCA and OHCA cohorts.

A recent multi-center study of 280 patients across 6 hospitals in the United States by Roberts et al., sampled blood gases at 1 and 6 h after ROSC and found that early hyperoxemia was associated with poor outcome at discharge. The investigators also substantiated the suggested threshold for the onset of poor outcome at 40 kPa. We investigated comparable PaO2 levels in our study and although our results were not significant, the point estimates of our primary analyses indicate higher probabilities for poor outcome in the hyperoxemia group but also in the hypoxemia group. We did not identify a significant threshold for the onset of poor neurological outcome in our cut-off analysis but the lowest probability of poor outcome was in the group exposed to a PaO2 of up to 20 kPa which was similar to the risk ratio analysis by Roberts et al. Nevertheless, there are noteworthy differences between the investigations; our cohort was significantly larger than Roberts et al. and we included exclusively OHCA patients in order to increase homogeneity regarding cardiac arrest etiology. Furthermore, and most importantly, Roberts et al. sampled blood gases according to a prospective protocol over the first 6 h whereas our study evaluated the most extreme blood gas values during the first 24 h after ROSC.

Exposure to hypercapnemia or hypocapnemia in the post cardiac arrest phase is common [17, 18, 25, 40] and hypocapnemia has frequently been associated with poor outcome [17, 18, 41] while hypercapnemia exposure has been associated with poor outcome [17, 30, 38, 42], good outcome [12, 13, 18, 24] or no difference in outcome [25, 41]. In an analysis of 9176 adult OHCA patients in the ROC-network, Wang et al. showed that hypercapnemia at any time-point within the first 24 h after OHCA and hypocapnemia towards the end of the first 24 h was associated with increased in-hospital mortality. Our study employed the same cut-off levels for hypercapnemia and hypocapnemia as Wang et al., but the prevalence of hypercapnemia and hypocapnemia were lower in our analysis (34.5% versus 51.0 and 23.6% versus 30.6%, respectively). The overall in-hospital mortality of Wang et al. was similar to our proportion of patients with CPC5 at discharge (67.3% versus 61.2%), but we did not achieve significant results in our analyses, and somewhat contrary to the ROC-network analysis our point estimates indicate a lower probability for poor outcome in the group exposed to hypercapnemia. However, the studies are not entirely comparable; Wang et al. included significantly more patients and analyzed the first, last or any arterial blood gas measurement during the first 24 h of hospitalization, while our study analyzed the most extreme values within 24 h of ICU admission. Moreover, Wang et al. did not correct for in-hospital care such as induced hypothermia or physiological parameters as pH.

Considering the results of the studies investigating PaO2 or PaCO2, the exposure to combinations of extreme PaO2 and PaCO2 values might also be associated with neurological outcome. Vahersaalo et al. found in a cohort of 409 OHCA patients the combination of moderate hypercapnemia and mild hyperoxemia to be associated with improved neurological outcome. We investigated hypercapnemia in combination with PaO2 8.0–40 kPa, but were not able to show an association with an improved outcome in this group. Treatment with induced hypothermia to 32–34 °C might influence CO2 solubility and represent a potential bias between analyses, but the 32–34 °C groups were of similar size in both studies (71% versus 78.4%). Nevertheless, there were significant differences, most notably, Vahersaalo et al. measured mean PaO2 and PaCO2 values in different ranges whereas we analyzed exposure to the most extreme values.

As shown above, studies investigating extreme PaO2 and PaCO2 value exposure after cardiac arrest differ in inclusion criteria and the time frame after ROSC, objectives and results. Moreover, short term variability in vascular tone and acid-balance due changes in the fraction of inspired oxygen (FiO2) or respiratory rate is commonly not accounted for [23, 43]. It seems also important to point out that the possible protective and harmful properties associated with exposure to extreme PaO2 and PaCO2 values are still of a largely hypothetical nature. Hypercapnemia increases CBF and might improve outcome by optimizing CBF and CDO2 after OHCA as suggested by Eastwood et al. [13, 15]. Hypercapnemia is also an effective anticonvulsant, suppressing neuronal activity in the central nervous system and potentially reducing neuronal metabolic demands following ROSC, but so far, hypercapnemia has failed to show an association with favorable EEG patterns after OHCA [9, 44,45,46]. The optimal dose of hypercapnemia in the post OHCA phase, if favorable, is not known. Two randomized controlled pilot-studies investigating high normal PaCO2 (5.8–6.0 kPa) and mild hypercapnemia (6.7–7.3 kPa) have used neuron specific enolase (NSE) as a surrogate marker of neuronal injury [13, 44]. NSE was significantly reduced in patients exposed to mild hypercapnemia, while high normal PaCO2 exposure was not associated with NSE levels after OHCA. Although, consistent high quality evidence is lacking, there are no indicators of harmful effects of controlled hypercapnemia exposure after OHCA [13, 25, 44]. However, the results of the present study conflict with results from previous investigations and support the need for further randomized trials [47, 48].

Neuronal metabolic failure due to hypoxemia during the no-flow period of the OHCA is the principal cause of cerebral damage, but also hyperoxemia following ROSC has been associated with neuronal injury, possibly due to increased production of ROS, lipid oxidation and decreased CBF [3, 49, 50]. In a randomized pilot trial, moderately elevated PaO2 levels (20–25 kPa) did not influence NSE levels or neurological outcome after 6 months and exposure to PaO2 levels ≥40 kPa following ROSC has not been investigated in a prospective randomized manner in humans [44]. However, randomized animal trials and observational human studies suggest harmful effects [12, 21, 34, 36, 37]. Our results do not support these findings entirely and randomized studies investigating increased levels of PaO2 in the post OHCA phase would be a possible way to further test the effect of PaO2 on outcome.

Our study has several limitations. Firstly, due to its observational design, the results are hypothesis generating and we cannot make causality statements. Secondly, we evaluated the most deviant PaO2 or PaCO2 values in the first 24 h after ROSC and were not able to analyze the exact exposure time-point, duration or to correct for acid-base parameters at the same time-point. Thirdly, in the statistical analyzes, our P-values were not significant on the 0.05 threshold level, but considering the direction of our point estimates and the width of the 95% CI’s, we cannot exclude a possible type II error and that there are associations that may have been statistically significant in a larger population [51]. We did not correct for FiO2 or PaO2/FiO2 ratios since FiO2 was not registered in the INTCAR 2.0 protocol and the PaO2/FiO2 ratio is rather an indicator for altered lung function, already accounted for by correcting for pre-existing COPD. The strengths of this study are the multicenter prospective design with 22 participating centers over two continents, a large cohort with over 2000 OHCA patients with extensive data regarding cardiac arrest characteristics and medical background and few excluded patients, as well as no missing outcome data in our primary and secondary analyses.

In summary, this study did not show an independent association of exposure to extreme PaO2 and PaCO2 values during the first 24 h after ROSC and neurological outcome at hospital discharge. The results of studies investigating exposure to extreme PaO2 and PaCO2 values vary widely and there is currently no consensus if extreme PaO2 or PaCO2 values are harmful, beneficial or innocuous to the post OHCA patient. The results of future prospective randomized studies are warranted before the existing recommendations on PaO2 and PaCO2 levels in the post OHCA phase can be revised [47, 52].

Conclusion

In a large cohort of patients resuscitated from OHCA, exposure to extreme PaO2 and PaCO2 values in the first 24 h after ROSC occurred commonly, but was not independently associated with neurological outcome at discharge.