Supplemental Intraoperative Oxygen and Long-term Mortality: Subanalysis of a Multiple Crossover Cluster Trial

We conducted a post hoc analysis of a large, single-center multiple crossover cluster trial in which adults having colorectal surgery were allocated to receive either 30% or 80% inspired oxygen (Fio₂) during general anesthesia. In the original trial, we tested the primary hypothesis that supplemental oxygen (80% vs. 30% as tolerated) reduces the risk of a 30-day collapsed composite (one or more) of surgical site infection, healing-related wound complications, and mortality. The underlying trial was approved by the Cleveland Clinic Institutional Review Board (12-891; Cleveland, Ohio) with a waived requirement for informed consent and was registered at ClinicalTrials.gov (NCT01777568) on January 29, 2013. The principal investigator for registration was Andrea Kurz.

The current subanalysis was also approved with a waived requirement for informed consent by the Cleveland Clinic Institutional Review Board. A detailed statistical analysis plan was developed a priori and approved by the Cleveland Clinic Institutional Review Board. The primary outcome and several subanalyses from the original trial have been published, all focused on short-term outcomes.^22–24  The statisticians were not blinded to treatment group assignment.

In brief, all patients who had surgery in a designated operating room suite at the Cleveland Clinic Main Campus (Cleveland, Ohio) over a 39-month period from 2013 to 2016 participated in the underlying trial. However, the analysis was restricted to patients who had colorectal surgery lasting at least 2 h. The amount of intraoperative inspired oxygen was randomly assigned for the initial 2 weeks, and then alternated every 2 weeks between 30% and 80%. No other aspects of ventilation were controlled, nor was postoperative Fio₂. As a safety measure, clinicians were instructed to give enough oxygen to maintain oxygen saturation measured by pulse oximetry at or above 95%. During anesthetic induction and emergence, 100% inspired oxygen was allowed.

In the underlying trial, patients with missing oxygen data or baseline covariables and reoperations during the same hospitalization were excluded from analysis. We further restricted our primary analysis to patients with a home address in Ohio because we could obtain reliable vital status data from the Ohio Department of Health (Columbus, Ohio) for them. Ohio Death Index data are maintained by Ohio Department of Health and were updated to December 12, 2018, when we accessed the registry. Vital status and follow-up dates were identified using Cleveland Clinic Epic system and Ohio Death Index: (1) from the Epic system, we extracted patient death information and the latest office visit date; and (2) for patients with a home address in Ohio, we also determine their vital status from the Ohio Death Index. When death was not recorded in the Ohio index, statistical analyses were censored at the latest office visit date.

We descriptively compared the 80% and 30% oxygen patients based on demographic, baseline, and procedural variables using standard descriptive statistics and the absolute standardized difference. The absolute standardized difference was calculated as the absolute difference in means or proportions divided by the pooled SD, and any imbalanced factor with absolute standardized difference greater than 0.10 was adjusted for in the analyses.

We assessed the effect of 80% versus 30% target-inspired oxygen on long-term mortality using a Cox proportional hazards regression model incorporating patient as a random effect to account for correlation within some patients across multiple surgeries. Kaplan–Meier survival estimates were calculated. When patients were included more than once, previous surgeries were censored the day before the date of the subsequent surgery. Since some patients who participated several times received different treatments, we conducted sensitivity analyses by retaining only the first or last surgery or all surgeries to assess the robustness of our statistical methods. In the primary analysis, we used several methods to assess the proportional hazards assumption, including testing the interaction between treatment and log (time), a correlation test based on the weighted Schoenfeld residuals, a Supremum test, and a visual display of log (hazards) versus time.

We assessed heterogeneity of the treatment effect across levels of selected baseline variables using Cox proportional hazards regression and testing the treatment-by-variable interaction. Baseline factors of interest, assessed in separate models, included type of surgery (colorectal resection vs. others), age (less than 60 vs. greater than or equal to 60 yr), sex (female vs. male), American Society of Anesthesiologists (ASA; Schaumburg, Illinois) physical status (I to II vs. III to V), primary diagnosis (cancer vs. other), body mass index (less than 30 kg/m²vs. greater than or equal to 30 kg/m²), current smoking (yes vs. no), and laparoscopic surgery (yes vs. no). The treatment effect was assessed within levels of each factor.

As this study represents a post hoc analysis using available data from the original trial, we did not conduct an a priori power calculation. In a post hoc power calculation, we assumed that a hazard ratio of 1.2 or stronger was clinically important, indicating that patients in the treatment arm (or control group, since the test was two-tailed) had 20% higher chance of dying at any time during follow-up versus control. Using the estimated survival of 77% at 6 yr in our control group as reference, and assuming survival times followed an exponential distribution, with the attained sample size of 3,471, we could only detect a hazard ratio of 1.24 or stronger with 80% power.

Results are reported as hazard ratios. A two-sided significance level of 0.05 was used for the overall assessment. The significance criterion for each interaction was set to 0.05/8 = 0.0062. SAS 9.4 software (SAS Institute, USA) was used.

The underlying trial was conducted at the Cleveland Clinic Main Campus from January 28, 2013, to March 11, 2016. The original trial analyzed 4,088 patients who had 5,167 surgeries and assessed the treatment effect on a 30-day composite of deep tissue or organ-space surgical site infection, healing-related wound complications, and mortality. For the primary analysis of this post hoc analysis, a total of 3,471 qualifying colorectal surgeries performed in 2,801 Ohio patients were analyzed, including 1,753 (51%) surgeries in 1,577 patients assigned to 80% oxygen and 1,718 surgeries in 1,551 patients assigned to 30% oxygen (fig. 1). Among the 2,801 patients qualifying for the current analysis, 457 were enrolled twice and 101 were enrolled three or more times; 327 patients received different treatments.

The 80% and 30% oxygen groups were well balanced on most of demographic, baseline, and procedural variables (absolute standardized difference less than 0.10, table 1) except smoking (absolute standardized difference, 0.110) and time-weighted average intraoperative heart rate (absolute standardized difference, 0.119). Overall mortality was 13.8% after a median (25th, 75th percentiles) follow-up period of 3.2 (0.5, 4.9) yr. We did not find an effect of 80% versus 30% oxygen on long-term mortality, with an estimated hazard ratio of 0.94 (95% CI, 0.78 to 1.13; P = 0.493; table 2, fig. 2). After 5 yr, the estimated survival rate was 80% (95% CI, 78 to 83%) in patients assigned to 80% oxygen and 80% (95% CI, 77 to 82%) in those assigned to 30% oxygen. For the primary analysis, Kaplan–Meier curves were close to each other and nonoverlapping. While the treatment-by-time interaction was significant (P = 0.019), the other two proportional hazards tests were not (correlation test based on the weighted Schoenfeld residuals, P = 0.202; and supremum test, P = 0.233). We thus concluded that there was no clinically meaningful violation of the proportional hazard assumption.

There were also no differences in the relative effects of oxygen in cancer and noncancer patients (fig. 3). The effect of oxygen on mortality also did not depend on the type of surgery, age, body mass index, smoking status, ASA physical status, primary diagnosis, or surgical approach (fig. 3).

Results were similar when the analysis was restricted to only the first or last surgery or including patients from all states (appendix table A1). Demographic, baseline, and procedural variables are shown in appendix table A2 for the first surgery of Ohio patients, and in appendix table A3 for the last surgery of Ohio patients.

We conducted a post hoc analysis of a single-center crossover cluster trial. The original trial tested the hypothesis that supplemental oxygen decreases the incidence of postoperative wound infections and infection-related complications.¹²  The use of a novel crossover cluster design allowed rapid enrollment of approximately 5,100 surgeries on 4,088 patients. As might be expected for a subset with about 2,800 patients who had more than 3,400 operations, there was remarkably good balance between the groups on a long list of observed potential confounding factors. Long-term mortality hazard was nearly identical in patients assigned to 30% and 80% intraoperative oxygen. Our results are therefore inconsistent with the theory that high intraoperative inspired oxygen promotes long-term mortality after colorectal surgery.

Our findings are consistent with those of Podolyak et al., who analyzed long-term mortality in more than 900 patients who participated in various randomized trials conducted more than a decade ago.²¹  Supplemental oxygen did not increase long-term mortality, with an overall site-stratified hazard ratio of 0.93 (95% CI, 0.72 to 1.20).²¹  Our current results and those of Podolyak et al.²¹  contrast with those reported by the PROXI investigators, who randomized 1,400 patients to 30% or 80% inspired intraoperative oxygen. The primary outcome was surgical site infection, which did not differ. In a secondary analysis, mortality was substantially increased in patients given supplemental intraoperative oxygen (hazard ratio, 1.31 [95% CI, 1.03 to 1.66]).¹⁴ 

Why the PROXI results are discrepant remains unclear. Mortality in the PROXI trial was evaluated over a median of 2.3 yr versus 4.7 yr in our current analysis, and a range between 3.6 and 13.6 yr in the analysis of Podolyak et al. However, longer follow-up periods in the more recent articles²¹  do not explain reported mortality differences since the hazard ratios were similar throughout the follow-up periods. There were substantive differences among the three analyses in terms of patient age, the fractions of laparoscopic and emergency procedures, and the types of procedures. However, subgroup analyses in our current study showed no significant interaction with type of surgery, age, body mass index, smoking, ASA physical status, or surgical approach. It therefore seems unlikely that any of these factors greatly influences the relative effect of inspired oxygen concentration on long-term mortality or explains why mortality differed with oxygen concentration in PROXI but not in subsequent reports.

In the PROXI trial, supplemental oxygen increased long-term mortality specifically in a subgroup of 352 cancer patients. The investigators proposed that high Fio₂ increased long-term mortality in cancer patients by promoting the growth of circulating tumor cells after potentially curative surgery. In a subsequent substudy of the PROXI trial, however, the authors could not demonstrate that 80% oxygen increases the incidence of new cancer or cancer recurrence. Although new or recurrent cancers occurred 100 days earlier in patients receiving supplemental oxygen, earlier recurrence was insufficient to explain the excess mortality in cancer patients.²⁰  In our study, 80% inspired oxygen did not augment long-term mortality in 995 cancer patients, nor did it in 451 cancer patients in the study by Podolyak et al.²¹  Data from 1,446 patients, therefore, do not support the theory that supplemental oxygen has long-term toxicity specific to cancer surgical patients.

We used the Ohio Death Index from the Ohio Department of Health to evaluate mortality. The Ohio Death Index has high validity for vital status.²⁵  However, 31% of patients in the underlying trial lived elsewhere. The analysis population was thus restricted to 2,801 patients. Nonetheless, we report by far the largest trial of inspired intraoperative oxygen and mortality.

A partial limitation of our analysis is that approximately 20% of patients had more than one operation. For these patients, previous surgeries were censored the day before the date of the next surgery since some of them might receive different treatments. We adjusted for the potential within-patient correlation by including patient as a random effect. An advantage of retaining all cases during the study period is that we gained power and perhaps generalizability. However, we also conducted sensitivity analyses restricted to the first or last surgery; the results were similar, indicating that our statistical methods were robust. An additional limitation is that the inspired oxygen concentration in patients assigned to receive 30% was actually given with a median of 39% [Q1, Q3: 35%, 47%]. However, the higher concentration was not a protocol deviation because clinicians were instructed to give sufficient oxygen to maintain an oxygen saturation of at least 95%. Many of our patients needed more oxygen because they were obese and/or required Trendelenburg positioning, which decreased their saturation.

We assumed that a hazard ratio less than or equal to 0.83 or greater than or equal to 1.2 was clinically important. The 95% CI of the hazard ratio for the primary analysis was within 0.78 and 1.13, which was fairly narrow, and the upper limit of 1.13 was not beyond what the study was designed to detect (a hazard ratio of 1.2 or more). Therefore, it allows us to make a fairly strong negative conclusion.

In summary, we performed a post hoc analysis of a large crossover cluster trial and evaluated whether intraoperative supplemental oxygen causes long-term mortality. Supplemental oxygen did not increase postoperative mortality overall, or in cancer patients, and can be safely used when deemed appropriate.

Support was provided solely from institutional and/or departmental sources.

The authors declare no competing interests.