Surgical Site Infection and the Routine Use of Perioperative Hyperoxia in a General Surgical Population: Title and subTitle BreakA Randomized Controlled Trial FREE

Surgical site infection (SSI) in the general surgical population is an important public health issue. In 2001, there were 5.3 million operations involving the gastrointestinal system in the United States.¹ Mean infection rates for several high-risk procedures exceed 10% and may exceed 20% in some centers.² Kirkland et al³ found that SSI resulted in a 225% increase in total direct costs per patient after laparotomy and a 77% increase after colon surgery. Thus, interventions that alter the incidence of SSI have significant public health and economic implications.

In recent years, there has been increased interest in the potential clinical benefits of supplemental perioperative oxygen administration.^{4 - 10} Historically, oxygen has been delivered during the perioperative period, with the primary goal of maintaining an adequate oxygen saturation of hemoglobin (SaO₂). Because additional oxygen adds little to the total arterial oxygen content of blood, there seemed little reason to advocate the administration of greater quantities of oxygen, especially given that a high fractional inspired concentration of oxygen (FIO₂) has been reported to cause atelectasis¹¹ and pulmonary toxicity.¹²

In an Austrian study published in 2000, Greif et al⁵ reported that the delivery of 80% oxygen in the perioperative period produced a 50% reduction in the incidence of surgical wound infection in patients undergoing open colorectal resection compared with the incidence for those who received 30% oxygen. Most independent commentators were cautious in their interpretation and did not advocate a broad application until further data were available.^{13 - 15} If a similar finding were true for the generalized surgical population, it could justify a significant shift in common practice.

The physiologic changes and interactions that occur after a substantial increase in the PaO₂ are multiple, complex, and difficult to study in vivo. A high oxygen partial pressure will increase the production of a number of derived reactive oxygen species,^{16 - 19} including the superoxide anion, hydroperoxyl radical, and hydrogen peroxide. A number of the reactions involving these reactive oxygen species are components of bactericidal host defenses. The reduced nicotinamide adenine dinucleotide phosphate oxidase enzyme,²⁰ located in the membrane of phagocytic vesicles, catalyzes the formation of superoxide in an oxygen-dependent process. This superoxide is reduced to hydrogen peroxide, which then combines with chloride ion to form bacteriotoxic hypochlorous acid in the myeloperoxidase reaction.²¹

However, reactive oxygen species are also involved in several processes that produce tissue injury and inhibit antibacterial mechanisms. Reactive oxygen species cause cellular dysfunction through damage to DNA,²² proteins,²³ and increased lipid peroxidation^{24 - 25} and can induce tissue cell death through apoptosis²⁶ or necrosis.²⁷ Additionally, the actions of hyperoxia on actin cause endothelial cell damage²⁸ and impair the antibacterial function of macrophages.²⁹

Because increased oxygen partial pressure can produce opposing physiologic consequences, the net effect of hyperoxia likely depends on the summation of microenvironment and systemic factors. Although there is good evidence that increased local tissue oxygen tension is associated with reduced infection in specific circumstances,^{5 ,30 - 31} it is unknown whether the routine administration of a high perioperative FIO₂ is an appropriate intervention in the general surgical population.

The aim of this study was to assess the effect of increased perioperative FIO₂ on infection in a heterogeneous surgical patient population in an academic setting. The study was designed as a double-blind, parallel-group randomized controlled trial to test the null hypothesis that increased perioperative FIO₂ will not alter the incidence of SSI in these patients.

Approval of the study protocol was obtained from the institutional review board of Weill Medical College of Cornell University (New York, NY). Written informed consent was obtained from all patients. Between September 2001 and May 2003, 165 patients older than 18 years who were undergoing major abdominal surgical procedures were studied; eligible procedures were prospectively defined and included colectomy (right, left, hemicolectomy, and sigmoid), low anterior resection, abdominoperineal resection, gastrectomy, pancreaticoduodenectomy, exploratory laparotomy, and large gynecologic staging/debulking procedures in which bowel or peritoneum was involved. Procedures that were laparoscopically assisted were eligible, provided that laparotomy was performed at some point during the surgery. Procedures that were fully laparoscopic were not eligible. Other exclusion criteria were patients whose respiratory status required an FIO₂ in excess of 0.35, patients with severe chronic obstructive pulmonary disease who were likely to experience respiratory depression at an FIO₂ of 0.80 (although milder chronic obstructive pulmonary disease was acceptable), patients who were hemodynamically unstable before surgery (as indicated by a systolic blood pressure <90 mm Hg or the use of vasopressors), patients who had received bleomycin at any time, and patients who had an American Society of Anesthesiologists physical status class 5 or 5E, which indicates a moribund patient not expected to survive 24 hours, irrespective of the surgery.

The treatment allocation method used an advance simple randomization without blocking or stratification. Before the recruitment phase of the study, 300 envelopes containing all protocol materials were prepared and numbered sequentially. A random-number table was used to assign each consecutively numbered envelope to either the 0.35 FIO₂ or the 0.80 FIO₂group so that each envelope had an independent 50% probability of being included in either group. A sheet indicating the allocated FIO₂ was then placed in the envelope, and the envelope was sealed. Throughout the course of the study, the sealed envelopes were removed and opened sequentially only after prospective patients had been screened and had consented to participation. During the entire protocol timeline, the only individuals aware of the randomization were the anesthesiologist(s) assigned to perform the surgical case, the postanesthesia care unit (PACU) or intensive care unit (ICU) nurse who treated the patient, and a research assistant.

The majority of elective patients were admitted on the morning of surgery. Most patients had undertaken a bowel preparation regimen the night before, according to surgeon instructions. Patients received intravenous antibiotics either immediately before arriving at or on arrival to the operating room, according to the surgeon's usual practice. Details of the type and timing of antibiotics given were recorded.

A number of relevant clinical characteristics that might influence wound healing and infection were recorded. These characteristics included basic physical characteristics, smoking history, American Society of Anesthesiologists physical class status, vital signs, several laboratory values (hematocrit levels, white blood cell count, and creatinine levels), and the presence of 8 major comorbidities: diabetes mellitus, asthma, hypertension, coronary artery disease, obesity (defined as a body mass index [BMI] >30),³² chronic obstructive pulmonary disease, end-stage renal disease, and organ transplantation or immunosuppression.

The anesthesiologist administered the assigned FIO₂ to the patient throughout the operative period. Inhalation of 100% oxygen was permitted during the preoxygenation and induction period and also during the emergence and extubation period. To ensure patient safety, anesthesiologists were also permitted to increase the FIO₂ as required to maintain an oxygen SaO₂ in excess of 94% at all times, as measured by pulse oximetry. In all instances, patients were to return to the assigned FIO₂ as tolerated.

The attending anesthesiologist determined all other aspects of intraoperative anesthetic care. Clinically relevant characteristics of the intraoperative anesthetic management were recorded, including the details of fluid status and therapy, anesthetic agents used, core temperature measurements, oxygenation variability, and any significant interventions.

Patients in the FIO₂ of 0.35 group were transported from the operating room with nasal cannula (Allegiance Healthcare Corp, McGaw Hill, Ill) at an oxygen flow rate of 4 L/min; patients in the group with FIO₂ of 0.80 were transported with either a closed reservoir bag-mask system (SIMS Portex Inc, Keene, NH) or a Jackson-Rees modified Mapleson E circuit (King Systems Corp, Noblesville, Ind), with an oxygen flow rate of 10 L/min and the mask loosely fitted over the patient's face. In the PACU, patients received oxygen via a high-flow, nonrebreathing, humidified, aerosol delivery system (all components by Allegiance Healthcare Corp), which incorporated a selector to provide a stable FIO₂ to the face mask.³³ Patients who remained intubated received the assigned FIO₂ through the ventilator. The assigned FIO₂ was maintained for 2 hours after arrival, after which the PACU/ICU team determined oxygen therapy.

The surgical team treating the patient was blinded to the FIO₂ assignment. View of flow meters was shielded by surgical drapes, and the anesthetic record was secured in a compartment of the patient's medical record not used by the surgical team. The attending surgeon, according to the usual assessment of optimal care, determined all surgical management. A number of relevant characteristics of the perioperative surgical management were recorded, including surgery duration, procedure type, closure technique, primary organ involvement, and the presence or absence of neoplastic pathology. For each patient, a National Nosocomial Infections Surveillance System (NNISS) risk index category was calculated.^{2 ,34} The NNISS uses this scoring system to stratify infection risk for surgical procedures. Scores are based on American Society of Anesthesiologists physical class status, wound class, duration of surgery, and use of a laparoscope (in certain procedures), with higher scores representing a larger number of risk factors for infection. In the postoperative period, the duration of immediate postoperative antibiotic therapy, time spent intubated or in the ICU, requirements for vasopressor therapy and further surgery, and total length of hospital stay after surgery were recorded. If a patient was readmitted during the evaluation period, the readmission was not included in the length-of-stay calculation unless it occurred within 24 hours of discharge.

The surgical team recorded the details of infection assessment at least once daily. The infection assessment was clearly documented, as was any change in management made as a result. The surgical team was not required to perform any additional evaluation or intervention that they did not consider necessary.

An investigator blinded to the randomization performed the evaluation for evidence of infection. The evaluation process occurred in 2 phases (Figure 1). The first phase was an assessment for evidence of infection during the hospitalization. An end point was unable to be established in this first phase if patients had been hospitalized for fewer than 14 days postoperatively and had no evidence of infection up to discharge. These patients were then entered into the second phase of evaluation, which was an assessment of the first postoperative visit with the surgeon, plus an assessment of any emergency department visit, telephone calls, or other contact made within the first 14 days after surgery. For a few patients, the first postoperative visit occurred before the 14 days had expired, and the assessment was extended to the second postoperative visit. One patient had postoperative care in another hospital system; in that instance, the patient was contacted directly and carefully questioned to confirm no evidence of infection.

The assessment for evidence of infection was conducted through a comprehensive review of all components of the patient's medical record, which included all physician and nursing documentation, laboratory and radiology reports, pharmacy records, telephone records, and any other relevant information. The criteria for a positive SSI end point were prospectively defined: (1) the surgical team clearly documented a clinical assessment of SSI; (2) the infection precipitated a management action, such as the initiation or changing of antibiotics, opening of the wound, aspiration, drain placement, or further surgery; and (3) the clinical assessment was supported by the presence of at least 3 of the following objective criteria: (a) a white blood cell count higher than 11 000/µL; (b) temperature higher than 38.5°C; (c) radiological evidence of infection; (d) extrusion of pus from the wound; (e) positive culture result from the infected site; and (f) documentation of wound erythema and induration on physical examination that resolved with treatment of infection.

For the initial power analysis, departmental records were used to predict the likely case mix for the study. The expected overall infection rate was calculated by prorating from rates reported in the literature. The analysis assessed an overall risk of .05 for type I error and of .20 for type II error and a detectable treatment effect of 40%. On this basis, a maximum of 300 patients were to be studied, with an interim analysis to be performed after 160 patients had been recruited. By using the Lan-DeMets α spending function approach, the study would be halted at the interim analysis if P≤.03. To preserve a cumulative type I error of .05, P≤.03 would be required to reject the null hypothesis at the final analysis of 300 patients.

An intention-to-treat analysis was used. Patients remained in their assigned group even if FIO₂ was increased to maintain an adequate SaO₂. All randomized patients were included in the analysis, unless they were removed from the study before the randomization was revealed and the treatment protocol commenced. Those randomization envelopes were retired, and the randomization was revealed only at the conclusion of the study. Statistical analysis followed Berger's recommendations for the use of permutation and approximate tests.³⁵ The null hypothesis tested was the "weak" null hypothesis in that the null hypothesis could still be true if there were patients who would respond to one treatment but not the other. Proportions and contingency tables were analyzed with the Fisher exact test or Pearson χ² ; means were analyzed with the Wilcoxon-Mann-Whitney test or the independent-samples t test in the following manner. Where the permutation test (PT) was clearly indicated, such as for small sample sizes or clear failures of normality, the PT is stated; otherwise, if the conservative PT was more powerful than the approximate test (AT), the PT was stated. If the AT was more powerful than the PT, then the AT is stated only if the power advantage was exclusively attributable to the conservatism of the PT. If p(AT) < P₀{T > t*}, where p(AT) is the P value for the AT and t* is the observed value of the test statistic T, then the PT is stated using the 2-sided value for the sum of small P values. A backward stepwise multivariate analysis was performed with a likelihood ratios logistic regression model. Variables with a P<.25 in the univariate analysis were included in the multivariate model and then removed stepwise with an exclusion threshold of P<.05. Data are presented as mean (SD) unless otherwise noted. Data analysis was performed with SPSS (version 11.0, SPSS Inc, Chicago, Ill) statistical software.

Between September 2001 and May 2003, 165 patients at the New York Presbyterian Hospital–Weill Cornell Medical Center (New York, NY) were enrolled and randomized. Five patients who received randomization envelopes were removed from the study before the randomization was revealed, either because there was a last-minute transfer of care to an anesthesiologist who was unaware of the patient's involvement (n = 3) or because alterations in the surgical plan negated the patient's eligibility (n = 2). The randomization envelopes for these 5 patients were retired, and the original randomization was not revealed until the conclusion of the study. Data were not collected from these 5 patients, and they do not appear in the final analysis. For the remaining 160 patients, there was complete protocol compliance and follow-up (Figure 2). An interim analysis was conducted after 160 eligible patients had been enrolled, at which time there was a statistically significant difference in the rate of surgical infection between the 2 groups. The null hypothesis was rejected at the interim analysis, and enrollment in the study was discontinued.

Three patients in the group with FIO₂ of 0.35 required a transient increase in FIO₂ during the operative period to maintain an adequate SaO₂; all were eventually returned to the assigned FIO₂. Two patients in the group with FIO₂ of 0.80 refused to wear the face mask in the PACU for the full 2 hours; in each case, supplemental oxygen was then provided by nasal cannula at the highest flow rate tolerated. All of these patients were included in the analysis because of the intention-to-treat design. None of these 5 patients subsequently developed infection.

The study groups were compared throughout a range of patient clinical characteristics, including demographics, American Society of Anesthesiologists physical status scores, laboratory values, comorbidities, surgical site, pathology, and vital signs (Table 1). The groups were also compared for a large number of operative anesthesia and surgical variables, including surgeon characteristics and NNISS risk index categories (Table 2 and Table 3). Less than 0.2% of the data sought was unavailable, and all missing values were records of one of patient weight, height, or preoperative hematocrit levels.

Although the BMI in the group with FIO₂ of 0.80 was slightly higher than in the group with FIO₂ of 0.35 (mean [SD], 27.1 [6.7] vs 25.1 [5.0]; P = .04) and there was a difference in the number of obese patients in the groups (patients with BMI >30: 23.8% vs 11.3%; P = .04), BMI and obesity were not predictors of infection in the multivariate analysis. Fifteen of 28 patients classified as obese had a BMI less than 35, and there was no significant difference between groups if obesity was defined as a BMI greater than 32 or a BMI greater than 35. During the operative period, patients in the group with FIO₂ of 0.80 had a slightly higher estimated blood loss (mean [SD], 230 [180] mL vs 200 [190] mL; P = .03) and received a greater volume of crystalloid (4.5 [2.1] L vs 3.8 [1.9] L; P = .02), but these values were also not significant on multivariate analysis. The group with FIO₂ of 0.80 received less nitrous oxide (5% [10%] vol/vol vs 21% [30%] vol/vol; P = .008). Because the concentration of nitrous oxide used was essentially inversely coupled to FIO₂, it was not included in the multivariate analysis. By rank-sum analysis, the group with FIO₂ of 0.80 had a statistically significant higher initial SaO₂ in the PACU (P = .005), but the values for both groups were well within the acceptable range (99% [1%] vs 98% [2%]).

A total of 29 patients (18.1%) developed SSIs. Nine infections (11.3%) occurred in the group with FIO₂ of 0.35 compared with 20 (25.0%) in the group with FIO₂ of 0.80 (P = .02). The odds ratio for the 2 groups was 2.63 (Wald 95% confidence interval [CI], 1.1-6.2), and the risk ratio was 2.22 (95% CI, 1.1-4.6). Of the 29 infections, 18 were superficial, 4 were in deep structures, and 7 involved superficial and deep structures (Table 4). The mean (SD) time to first detection of infection was 5.6 (2.4) days.

Positive cultures of purulent drainage were obtained in 16 of 18 patients and most were polymicrobial. The most commonly identified species were Escherichia coli (7 patients), Bacteroides fragilis (7 patients), Enterococcus faecalis (6 patients), and coagulase-negative staphylococcal species (4 patients).

Patients who developed an infection had a significantly longer length of hospitalization after surgery than those who did not develop infection (mean [SD], 13.3 [9.9] days vs 6.0 [4.2] days; P<.001). Throughout the entire study group, the length of hospitalization was longer in the group with FIO₂ of 0.80, but this did not attain statistical significance (8.3 [7.5] vs 6.4 [4.7] days; P = .07). Four patients who developed infection required reoperation; all were in the group with FIO₂ of 0.80. One patient in the group with FIO₂ of 0.35, who developed wound infection and later a deep abscess, experienced a postoperative myocardial infarction followed by a large middle cerebral artery stroke and later died on postoperative day 16. One patient in the group with FIO₂ of 0.80 experienced a pulmonary embolus on postoperative day 3 but recovered without further incident.

In the multivariate logistic regression analysis, FIO₂ remained predictive of infection (P = .03). A total of 5 patients in the group with FIO₂ of 0.80 and 1 patient in the group with FIO₂ of 0.35 remained intubated at the end of surgery, which was the only other significant predictor of infection. All other variables were not predictive in the multivariate model.

The aim of this study was to address the question of a potentially significant change in the care of general surgical patients. Because SSIs are an important cause of increased morbidity and mortality and can double or triple the total direct costs of hospitalization in this cohort,³ measures that reduce the incidence of SSI have significant public health implications. The study by Greif et al⁵ raised the issue of whether the delivery of high FIO₂ during the perioperative period could be routinely used to reduce the incidence of SSI. The potential for the expanded use of a high perioperative FIO₂ has been discussed by several authors,^{13 - 15 ,36} but commentary has stopped short of advocating its routine use, pending further data.

This study has yielded unexpected results. The administration of an FIO₂ of 0.80 during the perioperative period resulted in a doubling of the rate of SSI in patients undergoing a variety of major abdominal procedures. These infections were clinically significant, resulting in a mean duration of hospitalization more than double that of those who did not develop infection.

The protocol was designed to model the routine use of an elevated perioperative FIO₂ for abdominal surgery with a high infection risk in a typical academic practice setting. Because this protocol implied a degree of heterogeneity in surgical and anesthetic management, a number of clinically relevant variables were collected to evaluate for overall bias in the study arms. No bias was detected that could account for the results obtained. There was a slightly higher BMI in the group with FIO₂ of 0.80, with a corresponding higher incidence of obesity as defined by a BMI higher than 30, but these variables were not significant predictors of infection in the multivariate analysis. The difference between the groups with respect to obesity was principally due to patients with BMIs only slightly above the threshold; differences between the groups were not significant at higher, more clinically relevant levels of BMI. There was a slightly higher estimated blood loss in the group with FIO₂ of 0.80, but the absolute difference is not in the clinically relevant range. There was also a slightly higher crystalloid administration in the group with FIO₂ of 0.80. In that context, supplemental perioperative fluid administration has been reported to increase tissue oxygen tensions,³⁷ but the full significance of this finding is unknown. Neither estimated blood loss nor crystalloid administration was predictive in multivariate analysis. Because the routine use model did not control for the concentration of nitrous oxide, it was expected that the group with FIO₂ of 0.80 would be administered a significantly lower concentration of nitrous oxide. Because nitrous oxide was thus essentially inversely coupled to FIO₂ in this model, it was not included as an independent variable in the multivariate analysis. Nitrous oxide does not have a direct effect on bacterial killing by neutrophils,^{38 - 39} but because of its other physiologic effects, it would be of interest to investigate it independently. The groups were closely matched for NNISS risk index categories, a widely used and validated scoring system used to assess the risk for surgical wound infection.³⁴

Aside from the greater diversity of surgical procedures studied, there are a number of important differences between the study population described here and that reported in the study by Greif et al.⁵ Patients in that study were much more likely to receive transfusions of red blood cells (30% of patients vs 4% of patients) and, when transfused, to receive double the number of units (3.0 units vs 1.4 units). The mortality rate was double (1.3% vs 0.6%), and the mean length of hospitalization was nearly twice as long (12.1 days vs 7.4 days). However, the overall infection rate reported in that study was approximately half of that reported here (8.2% vs 18.1%). It is impossible to know the full significance of these differences, but they do suggest a fundamental dissimilarity in management paradigms.

The actions of oxygen at a cellular and tissue level are highly complex. There is no dispute that elevations in tissue oxygen tension exert a number of effects that are bacteriocidal. Certainly, the respiratory burst that forms the superoxide radical essential to oxidative killing is an oxygen-dependent process²⁰ and at least in vitro responds to changes in oxygen tension in the physiologic range.⁴⁰ Direct determination of oxygen tensions at the cellular level is impossible in vivo, but measures of oxygen tension at the subcutaneous tissue level suggest that elevated tissue oxygen correlates with reduced infection.^{30 - 31}

However, oxidative killing by neutrophils does not represent a complete picture of the effects of hyperoxia. A number of the deleterious effects of reactive oxygen species at the host cellular level have already been mentioned.^{22 - 29} The adaptive responses of bacterial pathogen species might also be influenced by oxygen tension. For example, the influence of oxygen on the expression of capsular polysaccharide can differ between phenotypic variants of the same species.⁴¹ Also, bacteria can exhibit rapid natural selection under stress through mutagenesis.⁴² The net effect of these and other factors on bacterial survival in any tissue environment cannot be measured.

These considerations are not offered as a definitive explanation for the findings in this study but rather to demonstrate that the physiology of hyperoxia is complex and that it should not be an a priori assumption that a high tissue oxygen partial pressure is protective in all circumstances. The balance of beneficial and deleterious effects may depend on tissue factors that can vary greatly in a heterogeneous surgical population.

This study was limited in its ability to identify subgroups of patients who might be especially advantaged or disadvantaged by the use of supplemental perioperative oxygen. It was also restricted in its ability to detect whether other elements of anesthetic management might affect SSI. In the univariate and multivariate analyses, no significant patient, anesthetic, or surgical technique subgroup was identified. This study deliberately permitted variation in anesthetic and surgical management to model routine practice by a representative and thus somewhat heterogeneous practitioner population. The identification of significant subgroups would require a protocol appropriately standardized to isolate the patients or practices of interest, which may be an area for future investigation.

In conclusion, the results of this study do not support the routine use of a high FIO₂ in patients undergoing major abdominal surgery to reduce the incidence of SSI. In fact, a high FIO₂ may have deleterious effects. It is possible that certain subgroups of patients may experience benefit or harm through this intervention, but no clear pattern was established in this study. Further investigation is required to elucidate which groups may benefit from supplemental perioperative oxygen. General surgical patients should continue to receive oxygen in the perioperative period with oxygen transport and cardiorespiratory physiology as principal determinants.