Intraoperative Oxygen Concentration and Neurocognition after Cardiac Surgery: A Randomized Clinical Trial

This parallel-group randomized controlled trial enrolled patients at Beth Israel Deaconess Medical Center in Boston, Massachusetts, between July 2015 and July 2017. Institutional Review Board approval was obtained on February 2, 2015, from the Committee on Clinical Investigations, and all patients provided written informed consent. Full details of the study protocol have been previously published.¹⁹  The trial was registered with ClinicalTrials.gov (NCT02591589, https://clinicaltrials.gov/ct2/show/NCT02591589, principal investigator: Shahzad Shaefi, registration date: October 29, 2015). The study protocol is available at ClinicalTrials.gov. The trial was designed to assess the effect of two intraoperative oxygen titration strategies, namely hyperoxia and normoxia, with postoperative neurocognitive function among older patients having cardiac surgery.

Patients 65 yr or older having elective or urgent coronary artery bypass graft (CABG) surgery requiring CPB were eligible for trial inclusion. Patients who were undergoing emergent CABG, procedures requiring single lung ventilation, or off-pump CABG, or patients with signs of cardiogenic shock, as dictated by preoperative inotropic, intraaortic balloon counterpulsation, or mechanical circulatory support, were excluded. Non–English-speaking patients were excluded because the neurocognitive assessments could not be administered in languages other than English.

The patients were randomized using a permuted block randomization schedule with block sizes of four, in which patients were randomly assigned in a 1:1 allocation to the normoxia or hyperoxia arm. Randomization assignments were allocated using sealed, sequentially numbered opaque envelopes. The randomization schema was generated by a unblinded statistician and concealed from study investigators. At the time of randomization, the anesthesiologist would open the next sequentially numbered envelope to obtain the randomization assignment. Study team members assessing neurocognitive function and postoperative clinical outcomes were blinded to patient arm, as were patients. Surgeons, anesthesiologists, and perfusionists involved in providing clinical care intraoperatively were not blinded due to the need for protocol adherence.

Study ventilator settings were applied after induction of general anesthesia and successful endotracheal intubation and continued throughout surgery. In the normoxia group, the fraction of inspired oxygen (Fio₂) was set at a minimum of 0.35 to maintain an arterial partial pressure of oxygen above 70 mmHg, or oxygen saturations measured by pulse oximetry greater than or equal to 92%. If required, the Fio₂ was titrated up to prevent hypoxemia (oxygen saturations less than 92%) during the pre- and postbypass periods. During CPB, a blended air/oxygen mixture was titrated to arterial blood gas analysis to maintain the Pao₂ between 100 and 150 mmHg. In the hyperoxic arm, the Fio₂ was set at 1.0 throughout the intraoperative period including CPB. Both groups received an anesthetic regimen at the discretion of the treating provider. Mechanical ventilation was based on the institutional standard of care, with a tidal volume of 6 to 8 ml/kg (employing ideal body weight) and positive end-expiratory pressure of 5 cmH₂O.

The primary outcome was postoperative cognition, as assessed by the Telephonic Montreal Cognitive Assessment score on postoperative day 2. Cognitive function was measured preoperatively as baseline and subsequently postoperatively daily until discharge, unless the participant was in the intensive care unit (ICU) and nonverbal. The telephonic assessment is an adaptation of the Montreal Cognitive Assessment, a validated screening instrument that is highly sensitive for detecting mild cognitive impairment.^20–23  The Telephonic Montreal Cognitive Assessment is evaluated on a 22-point scale, with lower scores indicating worse cognitive status. The assessment comprises an aggregate score of individual assessments of attention and concentration, executive functions, memory, language, conceptual thinking, calculations, and orientation. The items do not require writing or visual cues to complete and therefore can be easily adapted to either in-person or telephone administration. In addition to the primary outcome, neurocognitive scores at 1, 3, and 6 months were evaluated via telephone interviews as secondary outcomes. Patients were contacted by telephone for 1-, 3-, and 6-month assessments. Research staff attempted to reach patients until their call window expired (before or after 7 days for the 1-month follow-up call and before or after 14 days for the 3- and 6-month follow-up call) or for up to 10 attempts. The rationale for using the abbreviated Telephonic Montreal Cognitive Assessment was to facilitate the use of a consistent scale continuously throughout the study from the inpatient, in-person assessments to over-the-phone assessments at 1, 3, and 6 months.

Additional secondary outcomes included the incidence and severity of postoperative delirium, time to extubation, days of mechanical ventilation, length of ICU and hospital stay, and mortality at 30 days and 6 months. Postoperative delirium was assessed each postoperative day until discharge with the Confusion Assessment Method or Confusion Assessment Method–ICU for nonverbal (intubated) patients. As with the Telephonic Montreal Cognitive Assessment, delirium assessments were administered by study staff members who were blinded to group assignment. Delirium severity was scored using the long Confusion Assessment Method Severity score, which assigns points from 0 to 19, with worsening delirium characterized by higher scores.²⁴  The worst Confusion Assessment Method Severity score for their hospital stay was analyzed. Time to extubation was reported as the number of hours from when patients were initially intubated for surgery to when they were last extubated. Hospital length of stay was defined as the number of days spent in the hospital after surgery, and ICU length of stay was defined as the number of days spent in the ICU before transfer to the general inpatient cardiac surgery ward.

Because the patient population under study is by definition critically ill, we collected data on serious adverse events and unexpected nonserious adverse events that were possibly related to the study. Patients were assessed daily while in the hospital for a maximum of 3 days postoperatively. Additional adverse outcome data were collected from the Society of Thoracic Surgeons (Chicago, Illinois) database, including sternal wound infection, renal failure, myocardial infarction, reoperation, and stroke. This trial did not include any interim analyses to stop for safety, efficacy, or futility; thus, the trial was not stopped early, nor was there a dedicated data safety monitoring board. Adverse event monitoring was performed by members of the study team and reviewed by the principal investigator at regular intervals.

The minimal clinically important difference on the Mini-Mental State Examination in postoperative change after cardiac surgery has been shown to be 2 points, with postoperative day 2 reported as the in-hospital postoperative nadir time point.²⁵  Because there is no reported minimal clinically important difference for the Telephonic Montreal Cognitive Assessment, we utilized 2 points as our minimal clinically important difference based on previous validated crosswalk methods.^26,27  Using a two-sided alpha of 0.05 and 80% power, we determined that a sample size of 74 patients was needed to detect a mean difference in Telephonic Montreal Cognitive Assessment scores of two points (SD of 3) between the hyperoxia and normoxia groups. To allow for potential longer-term attrition from loss to follow-up or withdrawal, a total of 100 participants were enrolled and underwent surgery.

Descriptive statistics of the data are presented as means ± SD, medians (interquartile range), or counts and proportions, depending on variable type and distribution. Normality of continuous data was assessed with the Shapiro–Wilk test. Differences in continuous variables were assessed using independent sample t tests or Wilcoxon rank-sum tests as appropriate. Differences in categorical data were assessed with a chi-square or Fisher’s exact test. Our primary outcome, the Telephonic Montreal Cognitive Assessment score on postoperative day 2, was assessed with a nonparametric Wilcoxon rank-sum test. The Hodges–Lehmann estimation of shift is reported as the location shift and associated 95% CI. Differences in neurocognition at the follow-up time periods was also assessed between groups with the use of a Wilcoxon rank-sum test. In a post hoc analysis, the incidence of delirium was assessed among all participants, and the time to delirium and delirium severity was reported among only those who developed delirium. Full details of the statistical analysis plan were previously published in the protocol paper.¹⁹  SAS 9.4 (SAS Institute Inc., USA) was utilized for analysis. For all analyses, two-sided P values of less than 0.05 were considered statistically significant.

A total of 492 patients were screened, of whom 343 met eligibility criteria. A total of 100 patients were randomized, received the study intervention, and analyzed (fig. 1). Overall, the majority of participants were male (84%), white (95%), and had a median age of 71 (interquartile range, 68 to 75) yr and a median baseline Telephonic Montreal Cognitive Assessment score of 17 (16 to 19). All patients underwent isolated CABG only. No significant differences were found with regard to baseline demographics, medical comorbidities, or surgical characteristics between the groups (table 1). Baseline functional status was not statistically different between groups, with the majority of participants living independently and having a high school degree or greater (e-table 1 in the Supplemental Digital Content, http://links.lww.com/ALN/C521).

Protocol adherence was enforced immediately after endotracheal intubation. Reliable oxygenation separation between groups was achieved with a median intraoperative Pao₂ of 309 (285 to 352) in the hyperoxia group and 153 (133 to 168) mmHg in the normoxia group (P < 0.0001). Median intraoperative oxygen saturations were 99.2% (98.6% to 99.6%) in the hyperoxic group and 96.7% (95.7% to 97.8%) in the normoxic group (P < 0.0001; fig. 2). Other surgical characteristics are depicted in table 2. Observed intraoperative CPB-related characteristics were not statistically different between the hyperoxia and normoxia groups regarding cross-clamp (69 [54 to 80] vs. 69 [57 to 78]; P = 0.89) and total CPB minutes (83 [66 to 97] vs. 81 [68 to 91]; P = 0.67).

There was no significant difference in median Telephonic Montreal Cognitive Assessment score on postoperative day 2 between the hyperoxia and normoxia groups (18 [16 to 20] vs. 18 [14 to 20]; P = 0.42). On postoperative day 2, the between-group difference between normoxia and hyperoxia was −1 (95% CI, −2 to 1). It should be noted that the primary outcome could not be assessed in 10 (10%) of patients due to prolonged intubation, patient refusal, or withdrawal from the study. The trajectory of neurocognition between groups is portrayed between groups in figure 3. Although a slight increase was observed in both arms over time, this is likely attributable to the known and expected learning effects associated with repeated tests. When reported stratified by sex, the overall Telephonic Montreal Cognitive Assessment score on postoperative day 2 among males was 18 (14 to 20) and 18 (16 to 19) among females. There were no significant differences between groups in neurocognitive scores at 1, 3, or 6 months. These results, especially those at the longer follow-up, should be interpreted with extreme caution, because only 55% of patients could be contacted at the 6-month time period (e-table 2 in the Supplemental Digital Content, http://links.lww.com/ALN/C521).

The incidence of postoperative delirium in the hyperoxia group was 30.6% compared to 31.4% in the normoxia group (P = 0.93; e-fig. 1 in the Supplemental Digital Content, http://links.lww.com/ALN/C521). Clinical outcomes of study patients are described in table 3. No statistically significant difference was observed between those randomized to the hyperoxia or normoxia groups in regards to hospital days (8 [5 to 11] vs. 7 [5 to 10] days, respectively; P = 0.70) or ICU (2 [1 to 3] vs. 1 [1 to 3] days; P = 0.34) length of stay. No adverse events were determined to be possibly or probably related to the study intervention. The incidence of adverse events including mortality, stroke, pneumonia, acute kidney injury, reoperation, and atrial fibrillation was not statistically significantly different between groups.

In a post hoc analysis of patients who developed delirium, there was no statistically significant difference in time to delirium (1 [1 to 2] vs. 2 [1 to 3] days; P = 0.17; e-fig. 1 in the Supplemental Digital Content, http://links.lww.com/ALN/C521) or delirium severity (as assessed by Confusion Assessment Method Severity score; 11 [8 to 13] vs. 8 [7 to 11]; P = 0.23) in the hyperoxia group as compared with the normoxia group.

This randomized clinical trial assessed the effect of normoxic versus hyperoxic intraoperative oxygen conditions on postoperative cognition, measured using Telephonic Montreal Cognitive Assessment scores in an older population of patients having CABG. Although statistically and clinically significant differences in protocol-defined oxygen titration between groups were achieved, no significant difference was observed in postoperative cognition or delirium at any time point between groups. Additionally, oxygen titration strategy did not impact time to extubation, length of ICU and hospital stay, or patient mortality.

The results of this study do not provide clarity on the optimal oxygenation strategy for patients having cardiac surgery with CPB with regards to neurocognition and delirium. Potential reasons hyperoxia may be harmful in patients having cardiac surgery utilizing CPB are cardiovascular dysfunction,^28,29  enhancement of ischemia–reperfusion injury,^30,31  and direct injury from reactive oxygen species.^32–35  However, we did not find evidence of such harm in our study, at least that manifested as cognitive dysfunction. In interpreting these results, it is entirely plausible that the arterial oxygen content is a minor variable in the response to ischemia–reperfusion injury, dwarfed by that from the systemic inflammatory response to CPB. Other studies both in cardiac surgery^16,17,36  and allied specialties have not demonstrated statistically significant results regarding the detrimental effects of hyperoxia, albeit via varied outcome measurements.^12,37–39  It must be borne in mind that this was a study of moderate compared with severe hyperoxia. More contemporary studies are attempting to examine differences in tight normoxic and mild hyperoxic conditions, assuming that any oxygen titration benefit may be gleaned in this more physiologic window.^40–43  Other cardiac surgical studies have employed continuous arterial blood gas analysis to allow closer real-time normoxic titration of nonpulsatile Pao₂.¹⁶ 

In the broader context of identifying the ideal perioperative oxygenation strategy, there is currently substantial debate. There is not a consensus definition of hyperoxia, leading to difficulties in advocating for a specific titration strategy as well as significant heterogeneity in both clinical trials and clinical practice. Currently, both the World Health Organization (Geneva, Switzerland) and the Centers for Disease Control and Prevention (Atlanta, Georgia) recommend high concentrations of perioperative oxygen, largely based on a subgroup analysis of a single trial of high Fio₂ to reduce surgical site infection.^44,45  These recommendations have elicited criticism because of the inconclusive nature of the evidence and ignore evidence of harm from hyperoxia.^46–48  It is likely that lower oxygenation targets can still provide adequate tissue oxygenation in the perioperative period, even for high-risk patients. In fact, a series of recent larger studies has demonstrated that lower oxygen targets could be applied safely during CPB without detrimental cardiac and renal outcomes.^16,17,36  Our findings are in congruence with a large retrospective study of 1,018 patients having cardiac surgery with CPB that failed to demonstrate a relationship between arterial hyperoxia and neurocognitive function 6 weeks after surgery.¹⁸  Our study did not find any significant differences in adverse outcomes with the use of a lower oxygenation target, although interpretation of these results in support of the safety of a lower oxygenation target must be made with caution. Although evidence suggests that liberal oxygen supplementation and hyperoxia may lead to neurotoxicity in the context of ischemia–reperfusion injury, ours and other studies aiming to prevent such complications by simply reducing the amount of oxygen administered have yet to show consistent benefit, suggesting additional factors may be at play. It should be noted, however, that further focus upon delirium severity in this population may be warranted. Higher delirium severity perhaps confers a higher risk of long-term cognitive decline.⁴⁹ 

This study has several limitations, some or all of which may have contributed to these findings. Our intervention was over a broad period of oxygenation focus rather than specifically at a critical time point such as myocardial reperfusion. The inclusion of only CABG patients with relatively brief bypass times and thus at lower risk for ischemia–reperfusion injury than patients having more extensive surgery may have limited our exposure to injury and thus baseline risk. Interestingly, the patients in this study were relatively low-risk, with few baseline risk characteristics or postoperative events that might put them at risk for postoperative decline. This could potentially contribute to the lack of differences we observed. Our intervention was based on Fio₂ rather than Pao₂ targets; therefore, it is possible that different exposure definitions could have resulted in changes. However, this is unlikely given the large separation we observed between groups. Additionally, the Telephonic Montreal Cognitive Assessment is a relatively new tool to evaluate neurocognitive function in this population. Previous studies have shown that the Telephonic Montreal Cognitive Assessment is able to reliably identify mild cognitive impairment, and it has been employed in surgical populations.^23,50,51  In our study, we identified relatively low variability, which could suggest either that our patients were very homogenous or that the instrument is perhaps not sensitive enough to detect small cognitive differences. Although it does address memory, attention, language, abstraction, recall, and several other important components of neurocognitive function, we are unable to comment on visuospatial or executive cognitive domains. We did not evaluate the individual domains because of our sample size; however, this would be an interesting avenue for future research in this patient population. Furthermore, scores may improve with repeat testing during short intervals. These issues could be mitigated by a wider testing battery with multiple individual tests, nonsurgical comparator groups, and/or factor analysis in future studies. It should also be noted that our conclusions for longer-term follow-up should be interpreted with extreme caution, in large part because of loss to follow-up. That is, only 55% of patients could be contacted at the 6-month mark. Despite our attempts to contact participants, it is possible that this missing data biased our interpretation of longer-term neurocognition. Although this only occurred for a few patients in our study, the impact of prolonged intubation or delirium may also make it hard to interpret values for our primary outcome. Further limiting our study, the sample size for this trial was based on expected cardiac surgical cognitive changes seen postoperatively using the Mini-Mental State Examination,²⁵  and subsequent extrapolation to our primary endpoint using a validated crosswalk between the Mini-Mental State Examination and Telephonic Montreal Cognitive Assessment²⁶ ; however, the effect size we observed is not as pronounced as anticipated. Additionally, assumptions were made using a parametric distribution; however, we found that neurocognitive scores were not normally distributed. Because nonparametric analyses require larger sample sizes, we are therefore potentially underpowered to detect a difference if one truly exists. Despite these potential limitations, we were able to show no difference in short-term postoperative cognition among older cardiac surgical patients undergoing differential titration of intraoperative oxygen therapy.

In conclusion, this trial demonstrated that the titration of intraoperative oxygenation resulted in no significant differences in postoperative cognition after cardiac surgery. These results suggest that a varied intraoperative oxygen strategy may be safely employed without impairing postoperative neurocognitive function.

The authors acknowledge support for protocol development, execution, and adherence by the Center for Anesthesia Research Excellence within the Department of Anesthesia, Critical Care and Pain Medicine at Beth Israel Deaconess Medical Center (Boston, Massachusetts).

Supported by the Foundation for Anesthesia Education and Research (American Society of Anesthesiologists, Schaumburg, Illinois) in the form of a Mentored Training Research Grant to support principal investigator time and effort and by additional funds through the Beth Israel Deaconess Medical Center (Boston, Massachusetts) Chief Academic Officer Pilot Award for supporting study staff salary, statistical support, and regulatory compliance consulting.

Dr. Shaefi received a speaking honorarium for a University of North Carolina (Chapel Hill, North Carolina) Visiting Professorship lecture. Ms. Mueller receives statistical consulting fees from the University of Chicago (Chicago, Illinois). Dr. O’Gara receives consulting fees from Sedana Medical (Sweden). Dr. Bagchi receives consulting fees from Lungpacer Medical Inc. (Canada). Ms. Banner-Goodspeed received salary support from several National Institutes of Health (Bethesda, Maryland) and Department of Defense (Arlington, Virginia) grants, unrelated to this project. Dr. Subramaniam receives grant support from Mallinckrodt Pharmaceuticals (United Kingdom) and Edward Lifesciences (Irvine, California). The remaining authors declare no competing interests.

Full protocol available at: sshaefi@bidmc.harvard.edu. Raw data available at: sshaefi@bidmc.harvard.edu.