The interaction between arterial oxygenation and carbon dioxide and hospital mortality following out of hospital cardiac arrest: a cohort study

In the UK, 30,000 out of hospital cardiac arrests (OHCA) occur each year [1]. Outcomes are poor; of those who survive to ICU admission, just 28.6% survive to hospital discharge [2]. Derangements in oxygenation and carbon dioxide (PaCO₂) following cardiac arrest (CA) may exacerbate the post-cardiac arrest syndrome [3]. Hence, there has been a recent focus on the management of arterial oxygen and PaCO₂ in an effort to improve outcomes [4‐18]. However, definitions of hypoxemia, hyperoxemia, hypocapnia and hypercapnia vary between studies making it hard to determine thresholds for benefit or harm. Furthermore, studies typically examine the impact of either oxygenation or PaCO₂ on outcomes; our knowledge of their interaction is limited [18].

Arterial hypoxemia following CA is associated with higher mortality [4‐7]. However, previously used criterion for hypoxemia has combined patients with low PaO₂ and those with abnormal PaO₂/FiO₂ ratios making it hard to determine whether absolute hypoxemia or abnormal gas transfer is implicated in the observed increase in mortality. In addition, this approach creates a heterogenous population where patients may have low PaO₂/FiO₂ ratios but consistently be exposed to normoxia [4‐7].

Conversely, hyperoxemia may exacerbate cellular injury in CA survivors [19]. The evidence surrounding the effect of hyperoxemia is conflicting. Studies have shown an association between hyperoxemia following CA and mortality [4, 5, 8]. Whilst these studies corrected for a variety of cardiac arrest features and physiological parameters, they did not use extensively validated disease-specific scoring systems or illness severity scores. Subsequent studies which used modified Acute Physiology And Chronic Health Evaluation (APACHE) scores to correct for illness severity found no association between mortality and hyperoxemia [6].

Hypocapnia is associated with worse outcomes following CA [6, 13]. The impact of hypercapnia is more uncertain. A PaCO₂ > 45 mmHg has been associated with improved neurological outcomes [13, 15]. However, higher levels of hypercapnia have been associated with poor neurological outcome or higher mortality [5, 16, 17].

An interdependence exists between ventilation parameters and oxygenation [6]. However, our understanding of how interactions between arterial oxygenation and PaCO₂ affect outcome following CA is limited and rarely investigated [6, 15, 18]. One prospective observational study found an association between high mean PaCO₂ and PaO₂ in the first 24 h following CA and good neurological outcome [15]. A recent pilot study found no difference in biomarkers of cerebral injury in patients randomized to four combinations of normocapnia, hypercapnia, normoxia or hyperoxemia. In this study, elevated regional cerebral oxygen saturation was seen with both hypercapnia and hyperoxia [18]. It is unclear whether a combination of hyperoxia and hypercapnia would overwhelm anti-oxidant systems or improve survival through increased cerebral oxygenation [18]. This is relevant as cerebral injury accounts for two thirds of deaths in patients admitted to ICU following CA [19]. Further understanding of the effects of arterial oxygenation, PaCO₂ and their interactions in CA survivors would guide future patient management and inform trial design.

A total of 74,373 patients admitted following CA were identified; 50,748 met the exclusion criteria (Fig. 1). Patient demographics for 23,625 OHCA patients are shown in Table 1. ICU mortality was 48.9% and hospital mortality 59.1% (Table 1). A total of 860 patients were excluded from the final logistic regression analysis due to missing data. The crude, unadjusted and adjusted odds ratios for hospital mortality are presented in Table 2.

In this large retrospective study of patients admitted to ICU following OHCA, we found a significant association between hypoxemia and worsening PaO₂/FiO₂ ratios and mortality. This is in keeping with the other large CA databases [4‐7, 12]. This has biological rationale as hypoxemia is a marker for pulmonary pathology and exacerbates myocardial dysfunction and cerebral injury [27]. No association between hyperoxemia and mortality was observed. Importantly, we found PaCO₂ modified the PaO₂/FiO₂–mortality and PaO₂–mortality relationships.

The post-cardiac arrest syndrome is characterized by a widespread ischaemia/reperfusion response [28]. Post-ischaemic tissue is susceptible to oxygen free radical damage, resulting in reduced left ventricular function, coronary artery vasoconstriction and myocardial ischaemia [29]. Hyperoxemia may exacerbate oxygen free radical-mediated damage in the brain and promote pulmonary inflammation [4, 30]. In contrast, a recent porcine model of CA demonstrated a reduced incidence of low brain tissue oxygenation in swine treated with a FiO₂ of 1.0 and a 20-mmHg increase in mean arterial pressure (MAP) from baseline compared to those treated with a SpO2 target of 94–98% and a MAP target of > 65 mmHg [31]. This may help explain our finding of a lower mortality associated with hyperoxia. Previous studies which demonstrated an association between hyperoxemia and mortality following CA did not use validated scores to correct for illness severity [4, 5]. Our study, derived from a high-quality database, demonstrated no association between mortality and hyperoxemia or PaO₂/FiO₂ > 300 mmHg and is in keeping with similar studies [6]. Indeed, we found patients with hyperoxemia to have the lowest mortality. A sensitivity analysis examining different thresholds of hyperoxia found no association between hyperoxia and mortality. Following CA, the risk of exposure to hyperoxia falls with time, hence the low number of patients with a lowest PaO₂ > 300 mmHg [10]. However, the risk of a type II error is high. In sensitivity analyses including those who died within the first 24 h, a PaO₂ > 300 mmHg was associated with a higher mortality. Two thirds of patients who do not survive the first 24 h following OHCA have withdrawal of life-sustaining therapy based on pre-existing co-morbidities or perceived poor neurological prognosis [32]. Thus, for the majority of deaths within the first 24 h, the risk of mortality is unrelated to exposure to oxygenation or PaCO₂. As the risk of exposure to hyperoxia falls with time during the first 24 h following OHCA [10], those who die within the first 24 h have a disproportionate risk of having a lowest PaO₂ in the hyperoxia range. In our study, these patients were excluded to avoid this important potential confounding variable.

In keeping with other studies, we found hypercapnia to be associated with lower mortality [13‐15]. Hypercapnia modified both the PaO₂/FiO₂–mortality and PaO₂–mortality relationships. The mortality benefit was seen in all PaO₂/FiO₂ categories but was confined to hyperoxic patients in the PaO₂–mortality model.

Following CA, cerebral vasoconstriction and a loss of cerebral autoregulation have been demonstrated [33, 34]. Hypercapnia may increase cerebral blood flow, improve cerebral oxygenation, exhibit direct neuroprotective effects, reduce pulmonary and systemic inflammation and reduce oxygen free radical-mediated tissue injury [13, 15, 35‐40]. In CA survivors, hypercapnia may attenuate oxygen free radical production. Alternatively, the observed benefit of hypercapnia over normocapnia may be attributable to injurious ventilation strategies used to achieve normocapnia [3].

Hypercapnia was repeatedly associated with lower mortality in the PaO₂/FiO₂ model. However, hypercapnia was not associated with lower mortality in the setting of hypoxemia or normoxia in the PaO₂–mortality model. This may represent a sick cohort of patients with respiratory failure and poor pulmonary compliance not fully corrected for in the PaO₂–mortality model [41]. Hypercapnic acidosis causes pulmonary hypertension, right ventricular strain, reduced coronary blood flow and cerebral oedema [42, 43]. These effects are accentuated by hypoxemia [44, 45].

Hypocapnia was associated with higher mortality in all categories in the PaO₂/FiO₂–mortality model and in patients with hyperoxemia. Hyperoxemia and hypocapnia both cause cerebral vasoconstriction reducing cerebral blood flow [46‐48]. Hypocapnia shifts the oxygen dissociation curve impairing oxygen delivery [47]. Together, these may exacerbate cerebral ischaemia.

The ability of PaCO₂ to modify outcomes depended on whether a PaO₂/FiO₂ or PaO₂ model was used. In addition, PaCO₂ had limited influence on outcomes in patients with hypoxemia and normoxia. It is likely that hypoxemia is the overwhelming factor determining mortality and PaCO₂ has a limited ability to modify this outcome. However, our understanding of the pathophysiology of the modifying impact of PaCO₂ is poor.

Our observational study has a number of strengths. We investigated the association between PaO₂/FiO₂ and absolute PaO₂ on mortality separately and unlike in other studies tested for interaction with PaCO₂. The cohort was larger than other studies in this area [4‐18, 42]. Thus, allowing us to treat PaCO₂ as a categorical variable and investigate whether a threshold existed beyond which hypercapnia became harmful. Our cohort was derived from a high-quality database allowing correction for confounding variables. Our findings are supported by sensitivity analyses. We used the Acute Physiology Score component of the APACHE II score to correct for illness severity having excluded oxygenation, pH and temperature as they were tested as primary exposures in our model. The Acute Physiology Score has previously been demonstrated to have a better positive predictive value in predicting mortality following cardiac arrest than the APACHE II score [49]. The use of modified APACHE scores to correct for illness severity when examining the association between oxygenation and carbon dioxide and outcomes following cardiac arrest is well established [6, 7, 13, 15] but not universally applied [4, 5, 7, 9‐11, 16, 17]. The APACHE II score has previously been shown to have a similar ability to predict mortality following OHCA as the disease-specific OHCA score [50].

Our study has a number of limitations. As a cohort study, causality cannot be inferred. It is possible that residual confounders remain. No data was available on intra-arrest characteristics. Hence, we have been unable to account for presenting rhythm, bystander CPR or defibrillation or duration of delay to ROSC, all of which significantly impact on patient outcomes [51]. For OHCA patients, APACHE III scores showed a modest ability to predict mortality, whereas delay to ROSC showed a good ability to predict mortality [52]. We acknowledge that intra-arrest characteristics, including the delay to ROSC, are better predictors of outcome following cardiac arrest than illness severity scores. Unfortunately, no such data was collected within the ICNARC-CMPD. During our study period, there may have been temporal changes in cardiac arrest management including temperature control post-cardiac arrest; to account for this, we adjusted for year of admission [53]. Our outcome measure was hospital mortality; no data was available on longer-term mortality or neurological outcomes.

Previous studies have demonstrated an association between hyperoxemia on admission to ICU and mortality following CA [4, 5]. We cannot exclude that exposure to derangements in oxygenation and PaCO₂ in the immediate post-ROSC period is more prognostically significant than derangements as recorded in the ICNARC database. However, this is unlikely, as the worst PaO₂ in the first 24 h predicts ICU mortality more accurately than PaO₂ on the first ABG in a general ICU population [22]. Additionally, cumulative exposure to hyperoxemia over the first 24 h following CA has also been associated with mortality; the ICNARC-recorded ABG provides a surrogate for cumulative oxygen exposure during the first 24 h [10].

It could be argued that the ABG data collected may not be truly representative of an individual’s exposure to derangements in oxygenation following CA. However, the ICNARC-recorded ABG uses methods similar to the APACHE methodology. The PaO₂ recorded using the APACHE methodology is more representative of the mean PaO₂ in the first 24, 48 and 72 h following CA than an ABG taken on admission to ICU [4, 7]. A further limitation is the use of the ICNARC-recorded PaCO₂; however, the PaCO₂ recorded using similar APACHE methodology correlates closely with PaCO₂ in the first 24 h in CA survivors [13].

In examining PaO₂/FiO₂ and PaO₂ separately, we have demonstrated that derangements in both were associated with higher mortality [4‐7]. Despite a large number of patients with abnormal gas transfer, patients were typically normoxic; this may reflect titration of FiO₂ and adherence to ILCOR guidelines [3, 54]. However, it may have contributed to the different behaviour of the PaO₂/FiO₂ and PaO₂ models in relation to the modifying impact of PaCO₂. In our PaO₂/FiO₂ model, we were unable to differentiate between low PaO₂/FiO₂ ratios due to ARDS, pulmonary oedema or pre-existing lung pathology. Whilst we corrected for the presence of APACHE II-defined severe respiratory co-morbidity, residual confounding due to pre-existing respiratory pathology may have remained.

Finally, in choosing PaO₂ > 100 mmHg as the threshold for hyperoxemia, we may have missed harm associated with a higher threshold [55]. To address this, we presented a sensitivity analysis. There are several reasons to justify our threshold of 100 mmHg; a PaO₂ > 100 mmHg is rarely observed in health (hence, patients where the lowest PaO₂ is > 100 mmHg have been exposed to supraphysiological levels of oxygenation over a 24-h period and are by definition hyperoxemic), a PaO₂ of 150-200 mmHg has been associated with the lowest hospital mortality in a post-cardiac arrest population, a similar threshold has previously been used when investigating hyperoxemia in a general ICU population and choosing this threshold identified a large hyperoxic patient cohort reducing the risk of a type II error [6, 32]. However, conclusions from our sensitivity analysis are limited by the small number of patients in the hyperoxemia categories, resulting in a risk of a type II error.

We found an association between hypoxemia, low PaO₂/FiO₂ ratios and hypocapnia and mortality following OHCA. As hypothesized, PaCO₂ modifies the relationship between oxygenation and mortality, though the relationships are complex. Our study emphasizes the need for future studies examining the interaction of PaO₂ and PaCO₂ in OHCA survivors.