Surgical site infection, including postpartum endometritis and wound infection, after cesarean delivery after the onset of labor is a common problem with significant impact at both an individual and societal level. Postpartum endometritis complicates more than 15% of cesarean deliveries performed after the onset of labor and can be associated with wound infection and breakdown.1 Infection usually requires treatment with intravenous antibiotics and prolongs hospitalization an average of 3 days, significantly increasing the cost of delivery.2 Although most surgical site infections can be prevented with strict attention to sterile technique and appropriate use of prophylactic perioperative antibiotics, the incidence of postcesarean surgical site infection remains unacceptably high.
The etiology of postcesarean surgical site infection is polymicrobial, mainly by bacterial seeding of the uterus and pelvis from the lower genital tract. Approximately 60% of microbes isolated from women with postpartum endometritis are anaerobic.3 Bacterial vaginosis, in which anaerobic bacteria predominate, is associated with a markedly increased risk of postpartum endometritis.4,5 Thus, the contamination that occurs in cesarean delivery may be analogous to that of colorectal surgery.
Oxidative killing is an important defense against surgical infections and depends on the level of oxygen in contaminated tissue.6 In patients undergoing colorectal surgery, a 50% reduction in surgical site infection was observed in patients given 80% inspired oxygen during general anesthesia and for 2 hours thereafter; surgical site infection occurred in 15% of the control group and 7% of the treatment group.7 After these findings were replicated,8 the American College of Surgeons recommended that 80% inspired oxygen be used for patients undergoing colorectal surgery to prevent surgical site infection.9 Based on these findings, we hypothesized that high-concentration supplemental oxygen delivered during cesarean delivery and for 2 hours postoperatively would decrease the rate of postcesarean surgical site infection.
We conducted a double blind randomized controlled trial in which women undergoing cesarean delivery after the onset of labor had equal odds of receiving high-concentration inspired oxygen or low-concentration inspired oxygen. For the initial power calculation, we used departmental records and determined the baseline surgical site infection rate to be 15% among women with cesarean delivery after onset of labor. Thus, 225 women per study arm were needed to detect a 50% decrease in surgical site infection with 80% power. Surgical site infection, the primary study outcome, was defined clinically as administration of intravenous antibiotics for postpartum endometritis or oral or intravenous antibiotic for wound infection during the initial hospital stay or within 14 days of surgery.
House staff received guidelines for diagnosis of postpartum endometritis, including fever equal to or greater than 38.5°C within the first 24 hours postpartum or greater than 38.0°C for at least 4 hours after the first 24 hours postpartum associated with uterine tenderness greater than expected without other sources of fever identified.10 Wound infection included cellulitis as well as deeper incisional infections that required the wound to be opened. Clinical diagnoses were extracted from the medical chart by study personnel. If information was missing at the time of the 2-week postpartum check, the participant was contacted and interval history obtained to determine whether surgical site infection occurred. House staff and study personnel were blind to treatment arms.
The study population consisted of women undergoing cesarean delivery at the University of Washington after the onset of labor or rupture of membranes. Women were recruited for study participation from the Antepartum service and from the Labor and Delivery unit before the onset of active labor. Only women who underwent a clinically indicated cesarean delivery after the onset of labor or rupture of membranes were randomized and followed. Study exclusion criteria for randomization were cesarean delivery before the onset of labor or rupture of membranes, emergent cesarean delivery, general endotracheal anesthesia (those who started with regional anesthetic and converted to general anesthesia were not excluded), clinical chorioamnionitis, and human immunodeficiency virus infection. All study participants underwent routine surgical preparation with betadyne scrub, and all but one received prophylactic antibiotics.
Before the recruitment phase of the study, envelopes containing all protocol materials were prepared and sequentially numbered. A random-number table was used to assign each consecutively numbered envelope to either the treatment or control group in blocks of 10. A card indicating the assigned group was placed in the envelope, and the envelope was sealed by a research assistant unrelated to the study. During the entire protocol timeline, the only individuals aware of the treatment assignment for a given participant were the anesthesiologists who provided care during the cesarean delivery.
Oxygen delivery required devising a system that could deliver high-concentration oxygen without endotracheal intubation in a manner that would maintain blinding to the study group assignment. A covered oxygen blender was set by the anesthesiologist to deliver a predetermined mixture of oxygen and air to an adult nonrebreathing mask (Hudson RCT, Hudson RCI, Temecula, CA) during the operation and for 2 hours after. At least 80% oxygen was delivered to the mask in the high-oxygen group, and 30% oxygen was delivered to the mask in the control group. The flow rate was 15 L/min in both groups. To verify higher blood oxygen tension in the high-oxygen group, a venous blood gas was collected from the dorsum of the foot in a subset of participants. The sample was collected from the dorsum of the foot to minimize patient discomfort. Venous blood gas measurements from peripheral veins approach arterial blood gas values during sympathectomy-induced vasodilation resulting from spinal anesthesia.11
We planned and performed an interim analysis using a formal sequential stopping rule for both efficacy and futility based on the O’Brien Fleming criterion.12 With 25% of planned data accrued, we determined that stopping for futility was appropriate if the P value for a test of association between surgical site infection rate and treatment arm was greater than .11, suggesting differences between treatment groups would be unlikely to reach statistical significance with continued recruitment. P values were determined by Fisher exact test for dichotomous variables, χ2 for polytomous variables, and Wilcoxon rank sum for continuous variables. Risk ratios for outcomes by treatment arm and associated confidence intervals (CIs) were obtained using poisson regression, allowing for extra-poisson variation or overdisperson.13 Intent-to-treat analysis principles were applied. The University of Washington Institutional Review Board approved this study, and all participants provided written informed consent.
Between October 2001 and April 2007, 143 participants were enrolled and randomized to the two study groups. An interim analysis of surgical site infection rate by treatment arm was performed after 143 eligible participants were randomized, at which time the criterion for stopping the trial for futility was met and study enrollment was discontinued (P=.13, Table 1).
Eight documented protocol deviations occurred in each group, most of which were intermittent mask use due to nausea and vomiting during the surgery or patient request postpartum to facilitate infant bonding (Fig. 1). All participants were followed throughout their hospital stay; of those who did not have surgical site infection during initial hospitalization, seven in the control group and two in the treatment group had no 2-week postpartum data. Those lost to follow-up remained in the analysis and were considered not to have surgical site infection. Participants with a deviation from protocol that changed the amount of oxygen received remained in their assigned study group for the analyses.
The study groups were similar for a large number of clinical variables, including body mass index, maternal diabetes, group B streptococcus colonization, gestational age at delivery, duration of surgery, surgical blood loss, and prophylactic antibiotic use, as displayed in Table 2. A relatively large portion of participants had complicated pregnancies, reflecting the population served by the University of Washington. The only statistically significant difference in baseline characteristics of the two groups was the prevalence of gestational or chronic hypertension—42% in the control group and 24% in the high-oxygen group (P=.02).
Participants assigned to high-concentration oxygen had a significantly higher median partial pressure of venous oxygen (177 mm Hg) than those in the control group (122 mm Hg), confirming the efficacy of the oxygen-delivery system (P<.001). However, considerable overlap occurred in the range of oxygen levels between the two groups (Fig. 2). Duration of oxygen use was similar in the two study groups, approximately 130 minutes (Table 2).
Other factors analyzed but not included in Table 2 that did not differ between the groups included multiple gestation, urinary tract infection in pregnancy, number of digital exams in labor, any internal monitoring in labor, betamethasone administration for prematurity, and Pfannensteil incision.
Surgical site infection occurred in 19% of participants—17 (25%, 95% CI 15–35%) of 69 participants assigned to the high-oxygen group and 10 (14%, 95% CI 6–22%) of 74 participants assigned to the control group (relative risk [RR] 1.8, 95% CI 0.9–3.7), P=.13). Although the primary comparison in a randomized trial needs to be unadjusted, a secondary analysis was performed to examine the effect of gestational or chronic hypertension on infection because this was more likely in the low-oxygen group. In a regression analysis adjusting for gestational or chronic hypertension, the RR of infection by treatment arm was unchanged (adjusted RR 1.8, 95% CI 0.9–3.8, P=.12), and no association between gestational or chronic hypertension and infection was detected (RR 1.2, 95% CI 0.6–2.6, P=.62). Intravenous antibiotics to treat surgical site infection were given to 14% of women in the high-oxygen group and 7% of women in the control group (P=.17) (Table 1).
Table 3 presents a subanalysis of factors associated with surgical site infection in our study cohort. The only factor associated with surgical site infection was increased surgical blood loss. Other factors not associated with surgical site infection were maternal gestational or chronic hypertension, tobacco use in pregnancy, preterm labor, number of digital examinations in labor, and use of any internal monitor.
The aim of this study was to evaluate whether high-concentration perioperative oxygen could decrease the risk of postcesarean surgical site infection as it did after colorectal surgery. If efficacious, high-concentration perioperative oxygen could represent a simple and inexpensive improvement in obstetrical practice that would reduce a major cause of postpartum morbidity. Unfortunately, our results suggest that high-concentration oxygen did not prevent surgical site infection among women undergoing intrapartum cesarean delivery.
The study was stopped after the interim analysis showed futility of further enrollment. The rate of surgical site infection in the high-oxygen group was nearly twice (25%) that of the control group (14%), counter to our hypothesis. The use of intravenous antibiotics for infection, rates of cellulitis, postpartum endometritis, and wound separation all were higher in the high-oxygen group than in the control group. The findings are similar to a study of perioperative hyperoxia in a general surgical population published after the first colorectal study. The infection rate in the group that received high inspired oxygen during general surgery and postoperatively was 25% compared with 11% in the control group (P=.02), prompting discontinuation of enrollment at interim analysis.14 The authors concluded that routine use of high inspired oxygen in patients undergoing major abdominal surgery did not reduce the incidence of surgical site infection and that high inspired oxygen, in fact, may be harmful.14 Pending further data, supplemental oxygen is not recommended for surgical patients undergoing abdominal procedures that do not include colorectal procedures. Why oxygen might be harmful in abdominal procedures but beneficial in colorectal procedures as determined by two large, randomized controlled trials is unclear.7,8–15 In our cohort, the rate of infection among women with venous partial pressure of oxygen greater than 200 mm Hg (75th percentile) was not higher than among those with lower values, suggesting that the level of oxygen achieved in this study was not overtly detrimental, although power to detect this is limited.
Given the biologic plausibility that high-concentration supplemental oxygen would help prevent postcesarean surgical site infection and its success in colorectal patients, why did our study fail to find a treatment benefit? High oxygen levels may not be a major factor in preventing infection among pregnant women compared with patients undergoing colorectal surgery because of the increased blood volume, cardiac output, and oxygen demands characteristic of pregnancy. Also, regional anesthesia, compared with general anesthesia, may improve capillary perfusion in some tissues.16 Thus, although the median partial pressure of venous oxygen was higher among women in the high-oxygen group, the oxygen levels in the tissue may have been similar between the two groups. Tissue oxygen levels, which are inversely correlated with infection risk,17 were not measured to determine whether more oxygen was delivered to the uterus and abdominal skin in the high-oxygen group.
It is possible that our oxygen delivery method failed to achieve a therapeutic oxygen level. The median partial pressure of oxygen achieved in the study of patients undergoing colorectal surgery was 348 mm Hg7 compared with 177 mm Hg in the treatment arm of the current study. This difference can be explained partially by arterial sampling in the colorectal study as opposed to venous sampling in this study. However, given the large difference in the partial pressure of oxygen between studies and the considerable overlap in the levels achieved in our treatment and control groups as demonstrated in Figure 2, the oxygen level achieved in this study may not have reached a therapeutic threshold. The delivery of high-concentration supplemental oxygen to women undergoing cesarean delivery was difficult because of the use of a face mask rather than intubation, nausea and vomiting that often accompany cesarean delivery making face-mask use intermittent, and difficulty in wearing the nonrebreathing mask for 2 hours postpartum in the setting of infant bonding. By comparison, colorectal patients undergo general anesthesia with intubation, making high-oxygen delivery efficient during the surgery, and are relatively quiescent for several hours postoperatively, making high-oxygen delivery more feasible in the postoperative period.
Our study had limitations that may have contributed to the failure to find a treatment effect. The outcome measure, although consistent with Centers for Disease Control guidelines,18 relied on clinical diagnosis of surgical site infection by hospital staff as documented in the patient chart and extracted by a reviewer blinded to study arm. It is possible that surgical site infection was over- or under-diagnosed by house staff. However, because house staff and study personnel remained blinded to study assignment throughout the duration of the study, this bias seems unlikely. More participants were lost to follow-up after hospital discharge in the control group than in the high-oxygen group. This occurred because once surgical site infection was diagnosed, the participant was considered to have met the outcome criteria and further follow-up was not required. Because fewer participants in the control group had surgical site infection diagnosed during hospitalization, more were at risk for being lost to follow-up. It is possible that surgical site infection occurring after hospital discharge was missed among the seven participants lost to follow-up in the control group. However, even if all of the participants lost to follow-up had surgical site infection, the risk of surgical site infection in the two groups would appear equivalent. Possibly, a baseline characteristic of the study participants differed significantly between the two groups, which confounded our results. The only known characteristic that varied significantly between the study groups was the prevalence of hypertensive disorders. Hypertensive disorders were more prevalent in the control group and would be expected to put this group at increased risk of infection because of the peripheral vasoconstriction characteristic of this disorder, thus biasing toward finding a treatment effect, opposite to our findings. Although not statistically significant, more participants with diabetes were randomized to the high-oxygen group, perhaps increasing the risk of infection. However, the number of women with insulin-dependent diabetes was similar between the groups. Depth of subcutaneous fat is a known risk factor for surgical site infection19 and was not measured in this study. Body mass index was similar between the study groups and may be a reasonable proxy for depth of subcutaneous fat.
The mechanisms of hyperoxia are complex, and the balance of beneficial and harmful effects may vary depending on bacterial and host-tissue factors.14 Therefore, study of the effects of hyperoxia in different patient populations undergoing surgical procedures is prudent before adopting its routine use. Difficulties delivering high oxygen levels during regional anesthesia and the failure to find a significant treatment effect in this study, which approximated “real world” use, suggest that high-concentration supplemental perioperative oxygen is unlikely to be clinically useful in preventing surgical site infection after cesarean delivery.
1. Duff P. Pathophysiology and management of postcesarean endomyometritis. Obstet Gynecol 1986;67:269–76.
2. Henderson E, Love EJ. Incidence of hospital-acquired infections associated with caesarean section. J Hosp Infect 1995;29:245–55.
3. Hillier S, Watts DH, Lee MF, Eschenbach DA. Etiology and treatment of post-cesarean-section endometritis after cephalosporin prophylaxis. J Reprod Med 1990;35:322–8.
4. Watts DH, Krohn MA, Hillier SL, Eschenbach DA. Bacterial vaginosis as a risk factor for post-cesarean endometritis. Obstet Gynecol 1990;75:52–8.
5. Watts DH, Hillier SL, Eschenbach DA. Upper genital tract isolates at delivery as predictors of post-cesarean infections among women receiving antibiotic prophylaxis. Obstet Gynecol 1991;77:287–92.
6. Babior BM. Oxygen-dependent microbial killing by phagocytes (first of two parts). N Engl J Med 1978;298:659–68.
7. Greif R, Akça O, Horn EP, Kurz A, Sessler DI. Supplemental perioperative oxygen to reduce the incidence of surgical-wound infection. Outcomes Research Group. N Engl J Med 2000;342:161–7.
8. Belda FJ, Aguilera L, García de la Asunción J, Alberti J, Vicente R, Ferrándiz L, et al. Supplemental perioperative oxygen and the risk of surgical wound infection: a randomized controlled trial [published erratum appears in JAMA 2005;294:2973]. JAMA 2005;294:2035–42.
9. Brasel K, McRitchie D, Dellinger P, EBRS Group. Canadian Association of General Surgeons and American College of Surgeons Evidence Based Reviews in Surgery. 21: the risk of surgical site infection is reduced with perioperative oxygen. Can J Surg 2007;50:214–6.
10. Ernest JM, Mead PB. Postpartum endometritis. In: Mead PB, Hager WD, Faro S, editors. Protocols for infectious diseases in obstetrics and gynecology. Malden (MA): Blackwell Science; 2000. p. 44–50.
11. Kim JM, Reed K. PvO2 changes in cutaneous veins during regression of spinal anesthesia. Can J Anaesth 1987;34:358–61.
12. Jennison C, Turnbull BW. Group sequential methods: applications to clinical trials. Boca Raton (FL): Chapman & Hall/CRC; 2000.
13. Diggle PJ, Heagerty PJ, Liang KY, Zeger SL. Analysis of longitudinal data. Oxford (UK): Oxford University Press; 2002.
14. Pryor KO, Fahey TJ 3rd, Lien CA, Goldstein PA. Surgical site infection and the routine use of perioperative hyperoxia in a general surgical population: a randomized controlled trial. JAMA 2004;291:79–87.
15. Dellinger EP. Increasing inspired oxygen to decrease surgical site infection: time to shift the quality improvement research paradigm. JAMA 2005;294:2091–2.
16. Greene NM, Brull SJ. Physiology of spinal anesthesia. 4th ed. Baltimore (MD): Williams & Wilkins; 1993. p. 111.
17. Hopf HW, Hunt TK, West JM, Blomquist P, Goodson WH 3rd, Jensen JA, et al. Wound tissue oxygen tension predicts the risk of wound infection in surgical patients. Arch Surg 1997;132:997–1004.
18. Horan TC, Gaynes RP, Martone WJ, Jarvis WR, Emori TG. CDC definitions of nosocomial surgical site infections, 1992: a modification of CDC definitions of surgical wound infections. Infect Control Hosp Epidemiol 1992;13: 606–8.
19. Vermillion ST, Lamoutte C, Soper DE, Verdeja A. Wound infection after cesarean: effect of subcutaneous tissue thickness. Obstet Gynecol 2000;95:923–6.