Scolaris Content Display Scolaris Content Display

Efectos de una alta fracción de oxígeno inspirado en el perioperatorio para pacientes quirúrgicos adultos

Collapse all Expand all

Resumen

available in

Antecedentes

La evidencia disponible de los efectos sobre la mortalidad y la infección del sitio quirúrgico de una alta fracción de oxígeno inspirado (FIO2) del 60% al 90% comparada con una fracción habitual de oxígeno inspirado del 30% al 40%, durante la anestesia y la cirugía no ha sido concluyente. Ensayos y metanálisis anteriores han dado lugar a conclusiones diferentes con respecto a si una alta fracción de oxígeno inspirado suplementario durante la anestesia puede disminuir o aumentar la mortalidad y las infecciones del sitio quirúrgico en los pacientes quirúrgicos.

Objetivos

Evaluar los efectos beneficiosos y perjudiciales de una FIO2 igual o mayor del 60% comparada con una FIO2 control del 40% o menos en el contexto perioperatorio en cuanto a mortalidad, infección del sitio quirúrgico, insuficiencia respiratoria, eventos adversos graves y estancia hospitalaria durante el ingreso índice para pacientes quirúrgicos adultos.

Se examinaron diversos desenlaces, se realizaron análisis de subgrupos y de sensibilidad, se examinó la función del sesgo y se aplicó el análisis secuencial de ensayos (ASE) para examinar el nivel de la evidencia que apoya o rechaza una alta FIO2 durante la cirugía, la anestesia y la recuperación.

Métodos de búsqueda

Se hicieron búsquedas en el Registro Cochrane Central de Ensayos Controlados (Cochrane Central Register of Controlled Trials, CENTRAL), MEDLINE, EMBASE, BIOSIS, International Web of Science, the Información Científica y Técnica en Salud de América Latina y el Caribe (LILACS), advanced Google y en Cumulative Index to Nursing and Allied Health Literature (CINAHL) hasta febrero 2014. Se verificaron las referencias de los ensayos incluidos y las revisiones en busca de ensayos relevantes no identificados y se realizaron nuevamente búsquedas en marzo de 2015. Se considerarán dos estudios de interés cuando se actualice la revisión.

Criterios de selección

Se incluyeron los ensayos clínicos aleatorizados que compararon una alta fracción de oxígeno inspirado con una fracción habitual de oxígeno inspirado durante la anestesia, la cirugía y la recuperación en personas con 18 años de edad o más.

Obtención y análisis de los datos

Dos autores de la revisión extrajeron los datos de forma independiente. Se realizaron metanálisis de efectos aleatorios y fijos, y para los desenlaces dicotómicos se calcularon las razones de riesgos (RR). Se utilizaron los datos publicados y los obtenidos mediante contacto con los autores de los ensayos.

Para minimizar el riesgo de error sistemático, se evaluó el riesgo de sesgo de los ensayos incluidos. Para reducir el riesgo de errores aleatorios causado por los datos escasos y la actualización repetitiva de los metanálisis acumulativos, se aplicaron análisis secuenciales de ensayos. Se utilizó el método Grades of Recommendation, Assessment, Development and Evaluation (GRADE) para evaluar la calidad de la evidencia.

Resultados principales

Se incluyeron 28 ensayos clínicos aleatorizados (9330 participantes); en los 21 ensayos que informaron desenlaces relevantes para esta revisión, 7597 participantes se asignaron al azar a una alta fracción de oxígeno inspirado versus una fracción habitual de oxígeno inspirado.

En los ensayos con un riesgo general de sesgo bajo, una alta fracción de oxígeno inspirado comparado con una fracción habitual de oxígeno inspirado no se asoció con mortalidad por todas las causas (modelo de efectos aleatorios: RR 1,12, Intervalo de confianza (IC) del 95%: 0,93 a 1,36; GRADE: calidad baja) en el seguimiento más largo y en los 30 días de seguimiento (odds ratio [OR] de Peto 0,99; IC del 95%: 0,61 a 1,60; GRADE: calidad baja): En un análisis secuencial de ensayos no se alcanzó el tamaño necesario de información y el análisis no pudo rechazar un aumento del 20% de la mortalidad. De manera similar, cuando se incluyeron todos los ensayos, una alta fracción de oxígeno inspirado no se asoció a mortalidad por todas las causas al seguimiento más largo (RR 1,07; IC del 95%: 0,87 a 1,33) o en los 30 días de seguimiento (OR de Peto 0,83; IC del 95%: 0,54 a 1,29), ambos de calidad muy baja según GRADE. La alta fracción de oxígeno inspirado tampoco se asoció a riesgo de infección del sitio quirúrgico en los ensayos con bajo riesgo de sesgo (RR 0,86; IC del 95%: 0,63 a 1,17; GRADE: calidad baja) o en todos los ensayos (RR 0,87, IC del 95%: 0,71 a 1,07; GRADE: calidad baja). Una alta fracción de oxígeno inspirado no se asoció con insuficiencia respiratoria (RR 1,25; IC del 95%: 0,79 a 1,99), eventos adversos graves (RR 0,96; IC del 95%: 0,65 a 1,43) ni con la duración de la estancia hospitalaria (diferencia de medias ‐0,06 días; IC del 95%: ‐0,44 a 0,32 días).

En los análisis de subgrupos de nueve ensayos que utilizaron antibióticos preoperatorios, una fracción alta de oxígeno inspirado se asoció a una disminución de las infecciones del sitio quirúrgico (RR 0,76; IC del 95%: 0,60 a 0,97; GRADE: calidad muy baja). Se observó un efecto similar en los cinco ensayos adecuadamente cegados para la evaluación de desenlaces (RR 0,79; IC del 95%: 0,66 a 0,96; GRADE: calidad muy baja). No se observó un efecto de una alta fracción de oxígeno inspirado sobre las infecciones del sitio quirúrgico en otros análisis de subgrupos.

Conclusiones de los autores

Como el riesgo de eventos adversos, incluida la mortalidad, puede aumentar con una fracción de oxígeno inspirado del 60% o mayor, y como falta evidencia consistente de un efecto beneficioso de una fracción de oxígeno inspirado del 60% o mayor sobre la infección del sitio quirúrgico, los resultados generales indican que no hay evidencia suficiente para apoyar la administración habitual de una alta fracción de oxígeno inspirado durante la anestesia y la cirugía. Debido al riesgo de sesgo de desgaste y de informe de desenlace, así como a otras deficiencias en la evidencia disponible, se justifica la realización de ensayos clínicos aleatorizados adicionales con bajo riesgo de sesgo en todos los dominios, incluido un tamaño de la muestra grande y un seguimiento a largo plazo.

PICOs

Population
Intervention
Comparison
Outcome

The PICO model is widely used and taught in evidence-based health care as a strategy for formulating questions and search strategies and for characterizing clinical studies or meta-analyses. PICO stands for four different potential components of a clinical question: Patient, Population or Problem; Intervention; Comparison; Outcome.

See more on using PICO in the Cochrane Handbook.

Resumen en términos sencillos

Los efectos de administrar una alta fracción de oxígeno inspiratorio alrededor del momento de la cirugía a pacientes adultos

Pregunta de la revisión

El aire normal que se respira contiene 21% de oxígeno. Esta revisión sistemática evalúa los efectos beneficiosos y perjudiciales de un porcentaje de oxígeno inspirado del 60% al 90% comparado con un porcentaje habitual del 30% al 40% administrado durante la anestesia, la cirugía y el período de recuperación inmediata sobre el número de muertes e infecciones del sitio quirúrgico informadas en pacientes quirúrgicos adultos.

Antecedentes

La reducción de la función pulmonar y circulatoria durante la cirugía puede dar lugar a una reducción en los niveles de oxígeno (hipoxia). Además, los niveles de oxígeno suelen ser bajos en las heridas al final de la cirugía. Este hecho puede deteriorar la actividad bacteriolítica y la cicatrización de la herida. Los ensayos y metanálisis anteriores han dado lugar a diferentes conclusiones acerca de si un porcentaje alto de oxígeno inspirado durante la anestesia puede reducir o aumentar el riesgo de muerte o de infecciones del sitio quirúrgico. Esta revisión sistemática utilizó la mejor metodología Cochrane para realizar revisiones sistemáticas para reevaluar la evidencia disponible derivada de ensayos clínicos aleatorizados.

Características de los estudios

Se identificaron 28 ensayos clínicos aleatorizados. Ocho ensayos con 4918 participantes informaron el riesgo de muerte y 15 ensayos con 7219 participantes informaron las infecciones del sitio quirúrgico en el transcurso de 14 a 30 días de la cirugía. Cuatro ensayos informaron los eventos adversos graves, tres ensayos la insuficiencia respiratoria, nueve ensayos la estancia hospitalaria durante el ingreso asociado con el procedimiento y un ensayo la calidad de vida. Todos los ensayos se realizaron sin financiación directo de la industria.

El número de participantes en cada ensayo varió de 38 a 2012. La media de la edad de los participantes fue de 50 años (rango: 15 a 92 años) y el 63% fueron mujeres. Los tipos de cirugía incluyeron cirugía abdominal (ocho ensayos), cesárea (cuatro ensayos), cirugía de mama (un ensayo), cirugía ortopédica (dos ensayos) y otros procedimientos quirúrgicos (cuatro ensayos).

Resultados clave

Un porcentaje alto de oxígeno inspirado no se asoció estadísticamente a un aumento del riesgo de muerte, ni a una disminución de las infecciones del sitio quirúrgico en todos los ensayos que midieron estos desenlaces, en los ensayos de calidad más alta y en los que tuvieron un seguimiento más largo.

No fue posible probar un aumento del riesgo de eventos adversos a favor o en contra de un porcentaje alto de oxígeno inspirado durante la anestesia y la cirugía.

Calidad y cantidad de la evidencia

Solamente cinco de los ensayos incluidos presentaba un bajo riesgo de sesgo. Los ensayos asignaron al azar a 9330 participantes, de los que solamente 7537 participantes proporcionaron datos para esta revisión. No se alcanzó el número de participantes necesario para detectar o rechazar una reducción del riesgo relativo del 20% en las muertes. Por lo tanto, los resultados observados eran inciertos.

Authors' conclusions

Implications for practice

We could not exclude benefits or harms from use of a high inspired oxygen fraction (0.6 to 0.9) compared with a standard approach (0.3 to 0.4). Available randomized data do not suggest that either strategy is to be preferred. Further targeted research may provide definitive guidance in this area. As the risk of adverse events, including mortality, may be higher, and robust evidence for a beneficial effect on surgical site infection is lacking, overall our results suggest that evidence is insufficient to support the routine use of a high fraction of inspired oxygen beyond what is needed to maintain normal arterial oxygen saturation.

Implications for research

Additional RCTs are needed on the effects of a high fraction of inspired oxygen of 60% to 90% compared with 30% to 40% oxygen on mortality and on surgical site infection in all kinds of surgical patients. Before definitive conclusions can be drawn, we need more large trials with low risk of bias in all bias domains, testing effects of a fraction of 60% to 90% inspired oxygen versus 30% to 40% on patient important outcomes of all‐cause mortality, surgical site infection and health‐related quality of life. Future RCTs ought to be conducted with long‐term follow‐up and ought to stratify participants by type of surgery and documented risk factors for mortality and surgical site infection. Future trials ought to be designed according to the SPIRIT guidelines (Chan 2013) and reported according to the CONSORT guidelines (www.consort‐statement.org). Future trials also ought to report individual participant data, so that proper individual participant data meta‐analyses of the effects of a high fraction of inspired oxygen of 60% to 90% compared with 30% to 40% oxygen adjusted for relevant confounders can be conducted. Further, in future updates of this review, we will seek to conduct subgroup analyses of the effects of a high inspiratory fraction of oxygen in trials using regional anaesthesia compared with trials using only general anaesthesia.

Summary of findings

Open in table viewer
Summary of findings for the main comparison. Summary of findings in randomized clinical trials with overall low risk of bias and in all trials

Fraction of inspired oxygen of 60% to 90% compared with 30% to 40% oxygen during anaesthesia, surgery and recovery

Patient or population: surgical patients with need for abdominal, caesarean section, orthopaedic or breast surgery

Settings: perioperative and postoperative

Intervention: high fraction of inspired oxygen of 60% to 90% during and after surgery and anaesthesia

Comparison: fraction of inspired oxygen of 30% to 40% during and after anaesthesia and surgery

Outcomes

Illustrative comparative risks (95% CI)

Relative effect
(95% CI)

Number of participants
(studies)

Quality of the evidence
(GRADE)

Comments

Assumed risk*

Corresponding risk#

Inspired fraction of oxygen (30% to 40%)

Inspired fraction of oxygen (60% to 90%)

Mortality within the longest follow‐up in trials with overall low risk of bias

(follow‐up: 30 days to a median of 3.9 years)

High‐risk population

RR 1.12

(0.93 to 1.36)

4758

(3)

⊕⊕⊝⊝
Lowa

TSA (Figure 1) shows that the required information size of 10,736 for a 20% RRI has not been achieved, and that no trial sequential monitoring boundaries have been crossed. The TSA adjusted CI for the RR is 0.73 to 1.60. The required information size is even greater for a lower effect on mortality than a 20% RRI. We therefore downgraded by 1 level for imprecision and by 1 level for attrition bias (Analysis 1.2; Analysis 1.3)

185 per 1000
(168 to 203)

208 per 1000
(190 to 226)

Mortality within the longest follow‐up, irrespective of risk of bias (14 days to median of 3.9 years)

164 per 1000
(150 to 180)

175 per 1000
(143 to 218)

RR 1.07

(0.87 to 1.33)

4525

(6)

⊕⊝⊝⊝

Very lowb

TSA (Figure 2) shows that the required information size for a 20% RRI has not been achieved, and that no trial sequential monitoring boundaries have been crossed. The required information size for a lower excess mortality than a 20% RRI is even greater. We therefore downgraded by 1 level for imprecision, by 1 level for high risk of attrition bias (Analysis 1.2; Analysis 1.3) and by 1 level for overall risk of bias

Mortality within 30 days of follow‐up with overall low risk of bias

(14 days to 30 days)

18 per 1000

(13 to 25)

18 per 1000

(13 to 25)

POR 0.99

(0.61 to 1.60)

4758

(3)

⊕⊕⊕⊝
Moderatec

The required information size for a 20% RRI has not been reached, and no trial sequential monitoring boundaries have been crossed

Mortality within 30 days of follow‐up, irrespective of risk of bias (14 days to 30 days)

20 per 1000

(15 to 27)

17 per 1000

(11 to 26)

POR 0.83

(0.54 to 1.29)

4525

(6)

⊕⊕⊝⊝
Lowd

TSA shows that the required information size for a 20% RRI has not been achieved, and that no trial sequential monitoring boundaries have been crossed. The required information size for a lower excess mortality than a 20% RRI is even greater. We therefore downgraded by 1 level for imprecision and by 1 level for overall risk of bias

Surgical site infection within 30 days in trials with low risk of bias

(14 days to 30 days)

140 per 1000

(126 to 155)

119 per 1000

(106 to 134)

RR 0.86

(0.63 to 1.17)

4201

(5)

⊕⊕⊝⊝
Lowa

TSA (Figure 3) shows that the required information size of 13,189 participants for a 20% RRR has not been achieved, and that no trial sequential monitoring boundaries have been crossed. The required information size is even greater for a lower effect on mortality than a 20% RRR. Best/worst and worst/best case scenarios indicate possible attrition bias. Therefore we downgraded by 1 level for imprecision and by 1 level for attrition bias

Surgical site infection within 30 days, irrespective of risk of bias (14 days to 30 days)

129 per 1000

(118 to 140)

112 per 1000

(92 to 138)

RR 0.87

(0.71 to 1.07)

7229

(15)

⊕⊕⊝⊝
Lowe

TSA (Figure 3) shows that the required information size of 13,189 participants for a 20% RRR of surgical site infection has not been reached, and that no trial sequential monitoring boundaries have been crossed. The required information size for a lower effect on surgical site infection than a 20% RRR is even greater. As the result of overall risk of bias and imprecision, we downgraded by 2 levels

Respiratory insufficiency

44 per 1000

(31 to 62)

55 per 1000

(35 to 88)

RR 1.25

(0.79 to 1.99)

1386

(1)

⊕⊕⊕⊝
Moderatec

Information size is too small to be conclusive. We downgraded by 1 level for imprecision

*The basis for the assumed risk (e.g. median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).
CI: Confidence interval; LCL: Lower confidence limit; MD: Mean difference; POR: Peto odds ratio; RR: Risk ratio; RRR: relative risk reduction; RRI: relative risk increase; TSA: trial sequential analysis; UCL: Upper confidence limit.

GRADE Working Group grades of evidence.
High quality: Further research is very unlikely to change our confidence in the estimate of effect.
Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate.
Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.
Very low quality: We are very uncertain about the estimate.

*Mean of mortality in groups of included trials with 30% to 40% inspiratory fraction of oxygen.

#Calculated from: RR × assumed risk (95% CL: LCL of RR × assumed risk to UCL of RR × assumed risk).

aDowngraded by 2 levels because of risk of attrition bias and imprecision.

bDowngraded by 3 levels because of risk of attrition bias, imprecision and overall risk of bias.
cDowngraded by 1 level because of imprecision.

dDowngraded by 2 levels because of overall risk of bias and imprecision.

eDowngraded by 2 levels because of overall risk of bias and imprecision.


Funnel plot of comparison: 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, outcome: 1.5 Surgical site infection stratified according to overall risk of bias in a random‐effects model.

Funnel plot of comparison: 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, outcome: 1.5 Surgical site infection stratified according to overall risk of bias in a random‐effects model.


Trial sequential analysis of the effect on mortality within the longest follow‐up in trials with overall low risk of bias. With an anticipated relative risk increase (RRI) of 20%, mortality in the control group of 15.7% with a type 1 error of 5% and a type 2 error of 20%, and diversity (D2) of 57%, the required information size is 10,736 participants. The cumulative Z‐curve does not cross the conventional boundary nor the trial sequential monitoring boundary for harm. The cumulative Z‐curve does not reach the futility area. Therefore evidence of an effect on all‐cause mortality based on trials with low risk of bias is lacking, and we cannot exclude a 20% RRI due to lack of data.

Trial sequential analysis of the effect on mortality within the longest follow‐up in trials with overall low risk of bias. With an anticipated relative risk increase (RRI) of 20%, mortality in the control group of 15.7% with a type 1 error of 5% and a type 2 error of 20%, and diversity (D2) of 57%, the required information size is 10,736 participants. The cumulative Z‐curve does not cross the conventional boundary nor the trial sequential monitoring boundary for harm. The cumulative Z‐curve does not reach the futility area. Therefore evidence of an effect on all‐cause mortality based on trials with low risk of bias is lacking, and we cannot exclude a 20% RRI due to lack of data.


Trial sequential analysis of all trials reporting all‐cause mortality. With an anticipated relative risk increase (RRI) of 20%, mortality in the control group of 15.7% with a type 1 error of 5% and a type 2 error of 20% and diversity (D2) of 65%, the required information size is 13,264 participants. The cumulative Z‐curve does not cross the conventional boundaries nor the boundaries for benefit or harm. The cumulative Z‐curve does not reach the futility area. Therefore evidence of both beneficial and harmful effects on all‐cause mortality is lacking for all trials regardless of risk of bias and with varying time to follow‐up. We cannot exclude a 20% RRI or RRR due to lack of data.

Trial sequential analysis of all trials reporting all‐cause mortality. With an anticipated relative risk increase (RRI) of 20%, mortality in the control group of 15.7% with a type 1 error of 5% and a type 2 error of 20% and diversity (D2) of 65%, the required information size is 13,264 participants. The cumulative Z‐curve does not cross the conventional boundaries nor the boundaries for benefit or harm. The cumulative Z‐curve does not reach the futility area. Therefore evidence of both beneficial and harmful effects on all‐cause mortality is lacking for all trials regardless of risk of bias and with varying time to follow‐up. We cannot exclude a 20% RRI or RRR due to lack of data.

Background

Trials by Greif et al (Greif 2000) and by Belda et al (Belda 2005) have suggested a significant reduction in the frequency of surgical wound infections within 30 days when 80% rather than 30% oxygen was given for inspiration during surgery and the first postoperative hours. On the other hand, the PROXI trial (Meyhoff 2009) found no significant differences and the trial of Mayzler et al (Mayzler 2005) was inconclusive because these studies had very low power. The trial by Gardella et al (Gardella 2008) was stopped early, as it was unlikely that investigators were going to show the anticipated difference if continued, and the trial by Pryor et al (Pryor 2004) was stopped prematurely because the frequency of wound infection was more than doubled in the high oxygen fraction group. High inspiratory oxygen concentrations have been associated with adverse outcomes in a variety of emergency medical conditions, including exacerbation of chronic obstructive pulmonary disease (Austin 2010), myocardial infarction (Cabello 2013), resuscitation after cardiac arrest (Kilgannon 2010) and traumatic brain injury (Brenner 2012).

The ENIGMA trial provided long‐term follow‐up data for 83% of participants (Leslie 2011) with an unadjusted hazard ratio (HR) for death with 80% oxygen compared with 30% oxygen of 1.09 (95% confidence interval (CI) 0.90 to 1.35). Recently Meyhoff et al (Meyhoff 2014) reported long‐term follow‐up (median 3.9 years) from the PROXI trial with similar rates of new and recurrent cancer with 80% versus 30% oxygen during anaesthesia and surgery within the observation time. However, the HR for death, new cancer or recurrent cancer was increased with a high fraction of inspired oxygen (HR 1.19, 95% CI 1.01 to 1.42), indicating a median of 99 days shorter time to death or new or recurrent cancer. Accordingly evidence suggests that a high fraction of inspired oxygen during anaesthesia and surgery may be associated with an increased number of adverse events, as well as beneficial effects on surgical site infection (Hovaguimian 2013).

At least one large trial (Myles 2007) reported a lower rate of major complications with nitrous oxide‐free anaesthesia. However, recent data indicate that the presence of adjuvant nitrous oxide in the inhaled gas mixture during anaesthesia does not affect clinically important outcomes after surgery (Fleischmann 2005; Imberger 2014; Myles 2014; Myles 2014b).

We undertook this review to investigate all available evidence on benefits and harms for mortality and surgical site infection with an inspiratory fraction of oxygen of 60% to 90% compared with a routine fraction of 30% to 40% oxygen during anaesthesia, surgery and recovery.

Description of the condition

According to the criteria of the Centers for Disease Control and Prevention (CDC), surgical site infection may consist of superficial or deep wound infection or intra‐abdominal organ or space infection (Mangram 1999). The condition may or may not be accompanied by a positive bacterial culture. The CDC criteria (CDC 2009) for diagnosing surgical site infection are listed in Appendix 1. Surgical site infection is a common and serious complication following surgery, especially after abdominal surgery (Coello 2005). Surgical site infection may or may not be defined according to the criteria developed by the CDC (CDC 2009). We evaluated the investigators' definitions of surgical site infection with respect to CDC criteria.

Description of the intervention

After induction of anaesthesia and tracheal intubation, participants randomly assigned to a high fraction of inspired oxygen (FIO2) were given an FIO2 equal to or above 60% until the end of surgery. Additionally, in the first two hours following extubation, these participants might have breathed an FIO2 equal to or above 60% as administered by means of a face mask with a reservoir and a high flow mixture of oxygen and air (around 16 L/min). Participants randomly assigned to the control FIO2 were given an FIO2 at or below 40% after tracheal intubation; after extubation, they received a high flow (around 16 L/min) mixture of oxygen and air through an identical face mask.

How the intervention might work

To prevent surgical site infection, it is essential to optimize perioperative conditions, as the first hours following bacterial contamination are pivotal for avoiding an established infection (Hopf 2008; Miles 1957). Oxygen tension is often low in wounds and in colorectal anastomoses at the end of surgery. This may reduce bacterial eradication, the body's defences against bacteria and tissue healing. Possible mechanisms occur via diminished oxidative killing by neutrophils and impaired tissue healing caused by reduced collagen formation, neovascularization and epithelialization (Allen 1997; Babior 1978; Hopf 1997; Hopf 2008; Niinikoski 1977). Perioperative arterial and wound oxygen tension may be increased by a higher inspiratory oxygen fraction (Greif 2000; Hopf 1997). However, wound oxygen tension is low in hypoperfused patients and may not increase much with a high fraction of inspiratory oxygen (Gottrup 1987; Jonsson 1987). Further, many of the antibiotics used perioperatively are oxygen‐dependent in their effect; it is therefore possible that patients receiving a high FIO2 will benefit more from these antibiotics than patients receiving antibiotics without an elevated oxygen fraction. Beneficial effects have also been reported for other outcomes such as improved healing of colorectal anastomoses (García‐Botello 2006) and reduced postoperative nausea and vomiting (Greif 1999; Turan 2006).

Hyperoxia throughout the perioperative period may result in pulmonary complications, but this important aspect or oxygenation has been studied in only 30 patients (Akca 1999b). In a subgroup of participants from the trial by Greif et al (Greif 2000), a statistically non‐significant trend towards larger areas of atelectasis (collapsed areas of the lung) in the 80% oxygen group was observed on computed tomography scans (Akca 1999b). The PROXI trial by Meyhoff et al (Meyhoff 2009) found no significant differences in proportions of atelectasis, pneumonia or respiratory failure, but a trend towards higher mortality was found in the 80% oxygen group. More specifically, a high inspiratory oxygen fraction has been associated with detrimental effects such as increased airway inflammation (Carpagno 2004); poor regulation of blood glucose, which may affect healing (Bandali 2003); and decreased cardiac output/index (Harten 2003).

Why it is important to do this review

Complications after surgery are numerous and range from minor problems to fatalities, especially after emergency and upper abdominal procedures. The proportion of patients developing surgical site infections varies across types of surgery and reaches 10% to 20% for open abdominal surgery (Meyhoff 2009; Pryor 2004). Surgical site infection is a significant complication for all surgical patients and increases the risk of sepsis, admission to an intensive care unit (ICU), incisional hernia and hospital stay (Gotzsche 2000). Moreover, surgical site infection causes a heavy workload and an economic burden for hospitals as well as society. A high FIO2 administered to surgical patients, equal to or greater than 60% oxygen, is an easily applied and inexpensive intervention with few known/unknown contraindications or adverse events. However, several trials have reported contradictory results, and the benefits and harms of an FIO2 equal to or greater than 60%, as compared with a control FIO2 at or below 40%, are presently uncertain. Although several meta‐analyses and reviews have been published (Al‐Niaimi 2009; Chura 2007) they have not included recent trials, they have emphasized the fixed‐effect model and they have not assessed bias control according to the methods described in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). Recently, several trials (Duggal 2013; Schietroma 2013; Scifres 2011; Stall 2013; Thibon 2012) that tested the effects of an FIO2 of 80% compared with 30% on the occurrence of surgical site infection have been published. Likewise, three meta‐analyses (Hovaguimian 2013; Klingel 2013; Patel 2012) have been published within the past two years, and none of these have used stringent Cochrane methodology to assess risk of bias or to perform Grades of Recommendation, Assessment, Development and Evaluation (GRADE) of the overall quality of the evidence. Further, these meta‐analyses have not determined the information size required to detect or reject a 20% relative risk reduction (RRR) or increase (RRI), and they have not used trial sequential analysis (TSA) to assess whether sufficient evidence was obtained before the required information size was reached (Wetterslev 2008; Wetterslev 2009). These meta‐analyses therefore are prone to increased risk of random error due to sparse data and repetitive testing because of multiple updating, as new trials are included in the cumulative meta‐analyses.

Objectives

To assess the benefits and harms of an FIO2 equal to or greater than 60% compared with a control FIO2 at or below 40% in the perioperative setting in terms of mortality, surgical site infection, respiratory insufficiency, serious adverse events and length of stay during the index admission for adult surgical patients.

We looked at various outcomes, conducted subgroup and sensitivity analyses, examined the role of bias and applied TSA (Brok 2008; Brok 2009; Thorlund 2009; Wetterslev 2008; Wetterslev 2009) to examine the level of evidence supporting or refuting a high FIO2 during surgery, anaesthesia and recovery.

Methods

Criteria for considering studies for this review

Types of studies

We included randomized clinical trials (RCTs) without consideration of publication status, blinding status or language. We contacted investigators and study authors to retrieve relevant data.

We included unpublished trials only if trial data and methodological descriptions were provided in written form or by direct contact with study authors.

We excluded trials using quasi‐randomization and observational studies for the study of benefits. However, we were not able to establish an appendix enumerating the findings from observational studies regarding adverse events, as we did not find any. We excluded cross‐over trials even if they compared high FIO2 with routine FIO2 during the perioperative period.

Types of participants

We included surgical patients 18 years of age or older who were undergoing elective or emergency surgery.

Types of interventions

We compared a high FIO2 of 60% or above with a control FIO2 of 40% or below during surgery or both during surgery and during time spent in the postanaesthetic care unit.

Types of outcome measures

Primary outcomes

  1. All‐cause mortality: assessed according to the longest follow‐up period for each trial.

  2. Surgical site infection within 30 days of follow‐up after surgery: defined by investigators of involved trials or according to the criteria of the CDC (Appendix 1) as superficial or deep wound infection or intra‐abdominal organ or space infection (Mangram 1999).

Secondary outcomes

  1. All‐cause mortality within 30 days of follow‐up.

  2. Respiratory insufficiency: defined as the need for respiratory assistance provided as ventilator therapy or non‐invasive ventilation within the longest follow‐up period.

  3. Serious Adverse events: with a serious adverse event defined, according to the International Conference on Harmonisation Guidelines and the European Directive (Directive 2001), as "any event that leads to death, was life‐threatening, required in‐patient hospitalisation or prolongation of existing hospitalisation, resulted in persistent or significant disability, and any important medical event, which may have jeopardised the patient or required intervention to prevent it". All other adverse events were considered to be non‐serious events.

  4. Duration of postoperative hospitalizations.

  5. Quality of life as measured by the included trials.

Search methods for identification of studies

Electronic searches

We searched the Cochrane Central Register of Controlled Trials (CENTRAL) (2014, Issue 2); SilverPlatter MEDLINE (WebSPIRS) (1950 to February 2014); SilverPlatter EMBASE (WebSPIRS) (1980 to February 2014); SilverPlatter BIOSIS (WebSPIRS) (1993 to February 2014); International Web of Science (1964 to February 2014); Latin American Caribbean Health Sciences Literature (LILACS via BIREME) (1982 to February 2014); the Chinese Biomedical Literature Database; advanced Google and the Cumulative Index to Nursing and Allied Health Literature (CINAHL via EBSCO host) (1980 to February 2014). We reran the searches in March 2015. We will consider studies of interest when we update the review.

We performed systematic and sensitive searches to identify relevant RCTs without language or date restrictions. These searches were conducted within six months of the date the draft review was emailed to the editorial office. For specific information regarding our search strategies, please see the Appendices (Appendix 2, MEDLINE; Appendix 3, EMBASE; Appendix 4, CENTRAL; Appendix 5, Web of Science; Appendix 6, CINAHL).

We searched for ongoing clinical trials and unpublished studies on the following Internet sites.

  1. Current Controlled Trials.

  2. ClinicalTrials.gov.

  3. www.centerwatch.com.

Searching other resources

We handsearched the reference lists of reviews, randomized and non‐randomized studies and editorials to look for additional studies. Moreover, we contacted the main authors of studies and experts in this field to ask about missed, unreported or ongoing trials.

Data collection and analysis

Selection of studies

Two review authors (JW and CSM) independently evaluated all relevant trials and provided a detailed description of included and excluded articles under the sections Characteristics of included studies and Characteristics of excluded studies, respectively. We also provided a detailed description of our search results.

Data extraction and management

We screened titles and abstracts to identify studies for eligibility. JW and CSM independently extracted and collected data on a standardized paper form (Appendix 7). We were not blinded to the study author, institution or publication source of trials. We resolved disagreements by discussion. We approached all corresponding authors of included trials to ask for additional information relevant to the review's outcomes measures and risk of bias domains. For more specific information, please see the section Contributions of authors.

Assessment of risk of bias in included studies

We evaluated the validity and design characteristics of each trial. To draw conclusions about the overall risk of bias for an outcome, it is necessary to evaluate the trials for major sources of bias, also defined as domains (random sequence generation, allocation concealment, blinding, incomplete outcome data, selective outcome reporting and other sources of bias). The tool recommended by The Cochrane Collaboration for assessing risk of bias is neither a scale nor a checklist but rather a domain‐based evaluation. Any assessment of the overall risk of bias involves consideration of the relative importance of the different domains (Higgins 2011).

Even the most realistic assessment of the validity of a trial may involve subjectivity, as it is impossible to know the extent of bias (or even the true risk of bias) in a given trial. Some domains affect the risk of bias across outcomes in a trial, for example, sequence generation and allocation sequence concealment; others such as blinding and incomplete outcome data may present different risks of bias for different outcomes within a trial. Thus, the risk of bias is not the same for all outcomes in a trial. We performed separate sensitivity analyses for patient‐reported outcomes (subjective outcomes) and for mortality (Higgins 2011).

We defined trials as having low risk of bias only if they adequately fulfilled the criteria listed in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011) and reproduced in Appendix 8; we performed summary assessments of the risk of bias for each important outcome (across domains) within and across studies and prepared a 'Risk of bias graph' and a 'Risk of bias summary figure' (Higgins 2011).

We presented results for all outcomes including adverse events in summary of findings Table for the main comparison(Higgins 2011).

As no sufficiently well‐designed formal statistical method was available by which to combine the results of trials with high and low risk of bias, the principal approach to incorporating risk of bias assessments into Cochrane reviews is to restrict meta‐analyses to trials with low (or lower) risk of bias (Higgins 2011). We used the risk of bias (ROB) table described in the Cochrane Handbook for Systematic Reviews of Interventions, section 8.5 (Higgins 2011), as a tool for assessing risk of bias in included trials. We assessed risk of bias in the different domains as described in Appendix 8.

Measures of treatment effect

We reported length of stay in hospital after surgery, in days, as a continuous outcome, and the intervention effect as the mean difference with 95% confidence interval. All other outcomes are dichotomous and were reported as risk ratios (RRs) with 95% confidence limits. For mortality, we calculated the Peto odds ratio (POR). We also calculated the risk difference (RD) with 95% confidence interval and subsequently numbers needed to treat for an additional beneficial outcome, when possible.

Unit of analysis issues

Numbers of events in all binary meta‐analyses and days in the meta‐analysis of length of stay in hospital after surgery.

Dealing with missing data

We contacted all first study authors and contact persons for trials with missing data to retrieve the relevant data. We performed a modified intention‐to‐treat (ITT) analysis, including, if possible, all randomly assigned participants who underwent surgery, or who did not withdraw their consent before surgery.

Intention‐to‐treat analysis is recommended to minimize bias in the design, follow‐up and analysis of efficacy of RCTs. It yields a pragmatic estimate of the benefit of a change in treatment policy rather than a measure of the potential benefit in patients who receive treatment exactly as planned (Hollis 1999). Full application of ITT is possible only when complete outcome data are available for all randomly assigned participants. Despite the fact that about half of all published reports of RCTs state that ITT analysis was used, handling of deviations from randomized allocation varies widely, and in many trials data for the primary outcome variable are missing. Methods used to deal with this are generally inadequate, potentially leading to bias (Hollis 1999).

Performing an ITT analysis in a systematic review is not straightforward, as review authors must decide how to handle outcome data that are missing from contributing trials (Gamble 2005). No consensus has been reached regarding how missing data should be handled in ITT analyses, and different approaches may be appropriate in different situations (Higgins 2011; Hollis 1999).

In cases of missing data, for our primary outcomes we used a 'complete‐case analysis' by simply excluding all participants for whom the outcome was missing from the analysis. Additionally, we conducted sensitivity analyses for our primary outcomes by applying best/worst and worst/best case scenarios.

Best case scenarios included the following: All participants lost to follow‐up in the high FIO2 group survived, and all participants lost to follow‐up in the FIO2 below 40% group died; all participants lost to follow‐up in the high FIO2 group did not have a surgical site infection, and all participants lost to follow‐up in the FIO2 below 40% group did have a surgical site infection. Worst case scenarios included these: All participants lost to follow‐up in the high FIO2 group died, and all participants lost to follow‐up in the FIO2 below 40% group survived; all participants lost to follow‐up in the high FIO2 group had a surgical site infection, and all participants lost to follow‐up in the FIO2 below 40% group did not have a surgical site infection.

Selective outcome reporting occurs when non‐significant results are selectively withheld from publication (Chan 2004). It is defined as selection, on the basis of results, of a subset of the original recorded variables for inclusion in the publication of a trial. The most important types of selective outcome reporting are selective omission of outcomes from reports; selective choice of data for an outcome; selective reporting of analyses using the same data; selective reporting of subsets of data and selective underreporting of data (Higgins 20118). Statistical methods developed to detect within‐study selective reporting are still in their infancy stage. We explored selective outcome reporting by comparing publications with their protocols, when the latter were available.

Assessment of heterogeneity

We quantified the degree of heterogeneity observed in the results by using diversity (D2) (Wetterslev 2009) and inconsistency factor (I2) statistics, which can be interpreted as the proportion of the total variation observed between trials that is attributable to differences between trials rather than to sampling error (chance) (Higgins 2002). P value ≤ 0.10 indicates significant heterogeneity, and suggested I2 statistic thresholds for low, moderate and high heterogeneity are 25% to 49%, 50% to 74% and ≥ 75%, respectively (Higgins 2003). If I2 = 0, we reported results using the random‐effects model only. In the case of I2 > 0, we reported results using both random‐effects and fixed‐effect models. However, we believed that using a fixed‐effect model offered little value in cases of substantial heterogeneity, which may be present in this review because of inclusion of various patient types, as well as variable adjuvant gases, definitions of surgical site infection and outcome reporting. So we emphasized the results from the random‐effects model analysis unless a few trials dominated the meta‐analysis (e.g. > 50% of the accumulated fixed weight percentage). Additionally, in cases of I2 > 0 (for mortality and surgical site infection outcomes), we sought to determine the cause of heterogeneity by performing meta‐regression analyses and relevant subgroup and sensitivity analyses. We aimed to combine trial results in a meta‐analysis only when clinical heterogeneity was low to moderate.

Assessment of reporting biases

Publication bias occurs when publication of research results depends on their nature and direction (Dickersin 1990). We examined this by providing funnel plots to detect publication bias or differences between smaller and larger studies (small study effects), expressed as asymmetry of the funnel plot (Egger 1997). In cases of asymmetry, we applied the Arcsine‐Thompson test as proposed by Rücker (Rücker 2008).

Funding bias is defined as bias in the design, outcome and reporting of industry‐sponsored research showing that a drug has a favourable outcome (Bekelman 2003). Relationships between industry, scientific investigators and academic institutions are widespread and often result in conflicts of interest (Bekelman 2003). We conducted a sensitivity analysis to examine the role of funding bias when relevant (see Sensitivity analysis).

Data synthesis

We used Review Manager software (RevMan 5.3.3) for statistical analysis. We calculated risk ratios (RRs) with 95% confidence intervals (CIs) for dichotomous variables (binary outcomes). We also calculated risk differences (RDs) (Keus 2009); if results were similar, we reported only RRs. Additionally, we calculated mean differences (MDs) as measures of absolute change with 95% CIs for continuous outcomes. We used D2 (Wetterslev 2009) and I2 statistics (Higgins 2002) to describe heterogeneity among the included trials. We explored causes of substantial heterogeneity by performing meta‐regression using Comprehensive Meta‐Analysis (CMA, version one) and Stata, version 13. We used the Chi2 test to provide an indication of heterogeneity between studies, with P value ≤ 0.10 considered significant.

Adverse events may be rare but serious, and hence important (Sutton 2002), when meta‐analysis is applied to combined results from several trials reporting binary outcomes (i.e. event or no event). First, we applied the Peto odds ratio (POR) in cases of small event proportions. Most meta‐analytical software packages do not include options for analyses to calculate RRs when included trials have 'zero events' in both arms (intervention vs control). Exempting these trials from calculations of RRs and CIs may lead to overestimation of a treatment effect, as the control event proportion may be overestimated. Thus we performed a sensitivity analysis by applying empirical continuity corrections to our zero event trials, as proposed by Sweeting et al (Keus 2009; Sweeting 2004), by applying an imaginary small mortality in both arms (Higgins 2011). We used trial sequential analysis (TSA) (Thorlund 2011).

Meta‐analyses may result in type 1 errors as the result of sparse data and repeated significance testing following updates with new trials (Brok 2008; Brok 2009; Thorlund 2009; Wetterslev 2008; Wetterslev 2009). Systematic errors from trials with high risk of bias, outcome reporting bias, publication bias, early stopping for benefit and small trial bias may result in spurious P values.

In a single trial, interim analysis increases the risk of type 1 errors. To avoid type 1 errors, group sequential monitoring boundaries (Lan 1983) are applied to show whether a trial could be terminated early because of a sufficiently small P value, that is, when the cumulative Z‐curve crosses the monitoring boundary. Sequential monitoring boundaries can be applied to meta‐analyses as well and are called trial sequential monitoring boundaries (TSMB). In TSA, the addition of each trial to a cumulative meta‐analysis is regarded as an interim meta‐analysis and reveals whether additional trials are needed (Higgins 2011; Thorlund 2011; Wetterslev 2008). So far several meta‐analyses and reviews have been published, providing an increasing quantity of trial results as new trials have been published (Al‐Niaimi 2009; Chura 2007; Qadan 2009). Therefore adjusting new meta‐analyses for multiple testing on accumulating data seems appropriate for controlling the overall type 1 error risk in cumulative meta‐analysis (Pogue 1997; Pogue 1998; Thorlund 2009; Wetterslev 2008).

In TSA if the cumulative Z‐curve crosses the boundary, a sufficient level of evidence is reached and no further trials may be needed. However, evidence is insufficient to allow a conclusion if the Z‐curve does not cross the boundary or does not surpass the required information size. To construct the TSMB, the required information size is needed and will be calculated as the least number of participants needed in a well‐powered single trial (Brok 2008; Pogue 1998; Thorlund 2011; Wetterslev 2008). We adjusted the required information size for heterogeneity by using the diversity adjustment factor (Wetterslev 2009). We applied TSA, as it prevents an increase in the risk of type 1 errors (< 5%) due to potential multiple updating and testing on accumulating data whenever new trial results are included in a cumulative meta‐analysis (Pogue 1997; Pogue 1998). This provided important information to permit estimation of the level of evidence on the experimental intervention (Pogue 1997; Pogue 1998; Thorlund 2009), and to determine the need for additional trials and their required sample sizes (Wetterslev 2008; Wetterslev 2009; Wetterslev 2010; Wetterslev 2012; Wetterslev 2012b). We applied TSMB according to an information size suggested by trials with low risk of bias (Wetterslev 2008; Wetterslev 2009) and an a priori 20% relative risk reduction (RRR) of surgical site infection, using a control event proportion suggested by large observational studies and by the pooled estimate of event proportions in the control groups of included trials. As mortality seemed low in trials conducted when the protocol was finished, and hence the ability to detect small intervention effects was lacking, we also performed a TSA using an information size estimated for an a priori 35% RRR of mortality (Wetterslev 2008; Wetterslev 2009).

Subgroup analysis and investigation of heterogeneity

We conducted the following subgroup analyses by assessing the benefits and harms of a high FIO2 for surgical site infection.

  1. Trials using inspiratory oxygen with or without nitrous oxide.

  2. Trials using inspiratory oxygen with an FIO2 of 80% or higher compared with trials using an FIO2 equal to or greater than 60% but lower than 80%.

  3. Trials using a high FIO2 only during surgery or during both surgery and postoperative care.

  4. Participants undergoing abdominal surgery according to type of surgery, that is, all kinds of abdominal surgery, procedures requiring laparotomy, upper laparotomy if possible, lower laparotomy if possible or laparoscopic procedures.

  5. Type of surgery (abdominal, orthopaedic, other).

  6. Whether follow‐up for surgical site infection lasted 14 or fewer days or more than 14 days.

We made inferences from subgroup analyses in terms of implications for clinical practice only if the overall analysis of one of the co‐primary outcomes became statistically significant. When analyses of co‐primary outcomes did not become statistically significant, we referenced them in Implications for research to provide hypotheses for future research.

We compared intervention effects in subgroups using a test of interaction (Altman 2003). P value < 0.05 was considered indicative of a significant interaction between the high FIO2 effect on surgical site infection and the subgroup category (Higgins 2011; Chapters 9.6.1 and 9.7).

We explored causes of moderate to high heterogeneity using meta‐regression, including mean age of the trial population at baseline. We planned to explore causes of moderate to high heterogeneity using meta‐regression with the following co‐variates: mean body mass index (BMI) of the trial population at baseline; fraction of participants with diabetes in the trial population at baseline; fraction of smokers in the trial population at baseline; and fraction of participants with a contaminated or dirty infected surgical field during surgery; however, because data were lacking in the included trials, this was not possible.

Sensitivity analysis

  1. We compared estimates of the pooled intervention effect in trials with low risk of bias versus estimates from trials with high risk of bias (i.e. trials with at least one unclear or high risk of bias component).

  2. We compared estimates of the pooled intervention effect in trials based on different components of risk of bias (random sequence generation, allocation concealment, blinding, follow‐up, intention to treat).

  3. We assessed benefits and harms of high FIO2 by conducting a continuity correction of trials with zero events.

  4. We assessed benefits and harms of high FIO2 by conducting sensitivity analyses excluding the smallest or the largest trial.

  5. We assessed benefits and harms of high FIO2 by excluding data from trials published only as abstracts.

  6. We assessed benefits and harms of high FIO2 by excluding data from trials with commercial funding.

We calculated RRs with 95% CIs and applied a complete case analysis, when possible, to sensitivity and subgroup analyses based on the co‐primary outcomes of mortality and surgical site infection.

Results

Description of studies

Results of the search

We identified a total of 2148 references of possible interest by searching the Cochrane Central Register of Controlled Trials (CENTRAL), MEDLINE, EMBASE, the Cumulative Index to Nursing and Allied Health Literature (CINAHL), the Chinese Database of Randomized Trials and reference lists. We identified two additional ongoing trials by searching databases of ongoing trials (Osvaldo 2011; Santa Clara Valley Health). We will include data from these two trials in future updates of this review. We excluded 403 duplicates and 1701 clearly irrelevant references upon reading the abstracts. Accordingly, we retrieved 59 references for further assessment. Of these, we excluded 26 references describing 26 studies because they were not randomized trials or did not fulfil the inclusion criteria of our review. We listed reasons for exclusion in the Characteristics of excluded studies table. We reran the search in March 2015 and found 40 citations, one of which was a duplicate from a previous search; we excluded 37 trials. We will consider two studies of interest when we update the review.

In total, 28 RCTs described in 33 references fulfilled our inclusion criteria (Figure 4). The included trials consisted of a total of 9330 participants; 7537 participants provided data for the outcomes analysed in this review.


Trial flow diagram. We reran the search in March 2015. We found two studies of interest, which we will consider when we update the review.

Trial flow diagram. We reran the search in March 2015. We found two studies of interest, which we will consider when we update the review.

We approached 28 corresponding or first study authors to request missing or unclear information and received answers from 10 of them.

We presented detailed descriptions in the Characteristics of included studies table and in the appendices. A summary overview follows.

Included studies

Trial characteristics

Eight of the 28 trials, with 4918 participants (Belda 2005; Greif 2000; Meyhoff 2009; Myles 2007; Pryor 2004; Schietroma 2013; Stall 2013; Williams 2013), reported on mortality; 15 trials reported on surgical site infections (Belda 2005; Bickel 2011; Duggal 2013; Gardella 2008; Golfam 2011; Greif 2000; Mayzler 2005; Meyhoff 2009; Myles 2007; Pryor 2004; Schietroma 2013; Scifres 2011; Stall 2013; Thibon 2012; Williams 2013); three reported on serious adverse events (SAEs) (Meyhoff 2009; Myles 2007; Pryor 2004); three on respiratory insufficiency (Kotani 2000; Meyhoff 2009; Zoremba 2010); seven on length of stay during the index admission (LOS) (Belda 2005; Bickel 2011; Gardella 2008; Greif 2000; Meyhoff 2009; Myles 2007; Pryor 2004); one on quality of life (Greif 1999); 10 only on outcomes that were not relevant to this review (Bhatnagar 2005; Goll 2001; Joris 2003; McKeen 2009; Purhonen 2003; Purhonen 2006; Simurina 2010; Turan 2006) ; eight only on nausea and vomiting (Bhatnagar 2005; Goll 2001; Joris 2003; McKeen 2009; Purhonen 2003; Purhonen 2006; Simurina 2010; Turan 2006; Mackintosh 2012; García‐Botello 2006); one on requirements for postoperative supplemental oxygen (Mackintosh 2012); and one on intestinal pH (phi) and carbon dioxide (CO2) gaps (García‐Botello 2006). All 28 trials used a parallel‐group design, and the 18 trials that provided data on the outcomes selected for this review used a two‐parallel group design. These trials were published from 1999 to 2013. All were conducted without direct funding by industry; however, two trials reported that a company was involved in their design (Belda 2005; Thibon 2012). Seven trials were conducted in Europe (Belda 2005; García‐Botello 2006; Greif 2000; Meyhoff 2009; Schietroma 2013; Thibon 2012; Zoremba 2010); seven in North America (Duggal 2013; Gardella 2008; Mackintosh 2012; Pryor 2004; Scifres 2011; Stall 2013; Williams 2013); three in the Middle East (Bickel 2011; Golfam 2011; Mayzler 2005); one in Australia, Hong Kong and the People's Republic of China (Myles 2007); and one in Japan (Kotani 2000). Twentysix trials were conducted in high‐income countries (Belda 2005; Bickel 2011; Duggal 2013; Gardella 2008; Greif 2000; Mayzler 2005; Meyhoff 2009; Myles 2007; Pryor 2004; Schietroma 2013; Scifres 2011; Stall 2013; Thibon 2012; Williams 2013;Kotani 2000; Zoremba 2010; Goll 2001; Joris 2003; McKeen 2009; Purhonen 2003; Purhonen 2006; Simurina 2010; Turan 2006; Mackintosh 2012; García‐Botello 2006), and two in a developing country (Golfam 2011; Bhatnagar 2005).

Participants

A total of 7597 participants were randomly assigned to an FIO2 of 60% or more versus 30% to 40% in the 18 trials reporting on outcomes for this review (Belda 2005; Bickel 2011; Duggal 2013; Gardella 2008; Golfam 2011; Greif 1999; Greif 2000; Kotani 2000; Mayzler 2005; Meyhoff 2009; Myles 2007; Pryor 2004; Schietroma 2013; Scifres 2011; Stall 2013; Thibon 2012; Williams 2013; Zoremba 2010), and three trials described as using an FIO2 of less than 80% (Duggal 2013; Myles 2007; Williams 2013). The number of participants in each trial ranged from 38 to 2012 (median 222; interquartile ratio (IQR) 147 to 564). One trial included a minor proportion of participants 15 to 17 years of age (Bickel 2011). The approximate weighted mean age of participants was 50 years (range 15 to 92 years). The approximate mean proportion of women was 63%.

Seven trials included mainly colorectal and abdominal surgical patients (Belda 2005; García‐Botello 2006; Greif 2000; Mayzler 2005; Meyhoff 2009; Pryor 2004; Schietroma 2013), and one trial included exclusively patients for appendectomy (Bickel 2011). Four trials included only women for caesarean section (Duggal 2013; Gardella 2008; Scifres 2011; Williams 2013); one trial included only breast surgery patients (Golfam 2011); two trials only patients for orthopaedic surgery (Kotani 2000; Stall 2013); and four trials patients undergoing a variety of surgical procedures (Myles 2007; Mackintosh 2012; Thibon 2012; Zoremba 2010).

Experimental interventions

Of the 28 trials, 25 used an FIO2 of 80% oxygen or greater in the group with a high FIO2 during anaesthesia; two of these used more than 90% oxygen (Kotani 2000; Mackintosh 2012), and three used an FIO2 of 60% to 80% (Duggal 2013; Myles 2007; Williams 2013). Fourteen trials also aimed for an FIO2 of 80% or more in the high FIO2 group during the recovery room period, and two trials used a high FIO2 only during surgery. Thirteen trials used non‐rebreathing or tight/sealed masks with high flow, and three trials used ordinary face masks. Two trials used oxygen at the discretion of the attending physicians during the recovery room period.

Comparator interventions

An FIO2 of 30% to 40% during surgery was applied in all 28 trials in the control group. Of these, 17 trials aimed to achieve an FIO2 of 30% to 40%, also during recovery room stay, and two trials used only supplemental oxygen during surgery. Fourteen trials used non‐rebreathing or tight masks with high flow, and three trials used a nasal catheter or cannula in the 30% to 40% oxygen group. Thirteen trials did not use nitrous oxide during surgery, and four trials used nitrous oxide with an FIO2 of 30% to 40% during surgery in the control groups (Bickel 2011; Mayzler 2005; Myles 2007; Pryor 2004). Use of nitrous oxide was not described in two trials (García‐Botello 2006; Golfam 2011).

Co‐interventions used in experimental and comparator intervention groups

An antibiotic was administered before surgical intervention to both intervention groups in 12 trials, whereas four trials with patients for caesarean section reported on antibiotics used after delivery at cord clamping in both intervention groups (Duggal 2013; Gardella 2008; Scifres 2011; Williams 2013). The different regimens of antibiotic prophylaxis are described in Characteristics of included studies. Bowel preparation on the day before surgery was described in two trials (Belda 2005; Greif 2000).

Ongoing studies

Two ongoing trials (Osvaldo 2011; Santa Clara Valley Health) identified at www.clinicaltrials.gov are seeking to include women for emergency caesarean section and elective caesarean section, respectively. The trial by Osvaldo 2011 aimed for 382 participants and, according to www.clinicaltrials.gov, planned to start in 2011. The Santa Clara Valley Health trial did not mention the aim for a sample size and, according to www.clinicaltrials.gov, planned to start in 2006. See additional details at Ongoing studies.

Awaiting classification

Two studies are awaiting classification (Gin 2013; Schietroma 2014). Please refer to Characteristics of studies awaiting classification for further details.

Excluded studies

We presented a detailed description of the characteristics of 24 excluded studies in Characteristics of excluded studies. We excluded seven articles for not reporting a randomized trial but instead a review or a meta‐analysis. We excluded seven articles because these trials used interventions outside the scope of this review. We excluded four articles because they were letters to the editor. We excluded three articles because of retractions for fraud and one article because the study included children only.

Risk of bias in included studies

Three trials had low risk of bias in all bias domains, including adequate outcome reporting on mortality (Greif 2000; Meyhoff 2009; Myles 2007). Five trials had low risk of bias in all bias domains, including adequate outcome reporting on surgical site infection (Gardella 2008; Greif 2000; Meyhoff 2009; Myles 2007; Williams 2013). Six trials reported on mortality; three of these had overall low risk of bias (Meyhoff 2009; Myles 2007; Greif 2000), and four had high or unclear risk of bias in one or more bias domains other than outcome reporting bias (Belda 2005; Greif 2000; Pryor 2004; Schietroma 2013) (Figure 5).


Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

The funnel plot (Figure 1) of all trials reporting surgical site infection showed slight asymmetry indicating bias;however, Egger's test was not statistically significant (P value = 0.63). As fewer than 10 trials were included in all other analyses. we did not construct a funnel plot for these outcomes.

Allocation

Generation of the allocation sequence

Ten trials reporting on the primary outcome of surgical site infection described adequate generation of the allocation sequence (Belda 2005; Duggal 2013; Gardella 2008; Greif 2000; Meyhoff 2009; Myles 2007; Pryor 2004; Stall 2013; Thibon 2012; Williams 2013); four trials did not describe the method of sequence generation and were considered as having unclear risk of bias (Bickel 2011; Golfam 2011; Schietroma 2013; Scifres 2011). One trial had high risk of bias, as investigators used tossing of envelopes for sequence generation (Mayzler 2005).

Allocation concealment

Twelve trials reporting on the primary outcome of surgical site infection described adequate allocation concealment (Belda 2005; Bickel 2011; Gardella 2008; Greif 2000; Mayzler 2005; Meyhoff 2009; Myles 2007; Pryor 2004; Schietroma 2013; Scifres 2011; Stall 2013; Williams 2013); three trials did not describe whether allocation concealment was adequate and thus were judged as having unclear risk of bias (Duggal 2013; García‐Botello 2006; Golfam 2011). One trial did not adequately conceal allocation, as the anaesthesiologist enrolling participants to the trial had access to the allocation list (Thibon 2012).

Blinding

Blinding of participants and personnel

Seven trials reporting on the primary outcome of surgical site infection described adequate blinding of participants and personnel (Belda 2005; Gardella 2008; Greif 2000; Meyhoff 2009; Myles 2007; Schietroma 2013; Williams 2013); five trials reported unclear blinding of participants and personnel (Duggal 2013; Golfam 2011; Mayzler 2005; Stall 2013; Thibon 2012). Three trials were described as inadequately blinded for participants and personnel and thus were considered at high risk of bias (Bickel 2011; Pryor 2004; Scifres 2011).

Blinding of outcome assessors

Twelve trials reporting on the primary outcome of surgical site infection described adequate blinding of outcome assessors (Belda 2005; Bickel 2011; Duggal 2013; Gardella 2008; Greif 2000; Mayzler 2005; Meyhoff 2009; Myles 2007; Schietroma 2013; Stall 2013; Thibon 2012; Williams 2013); two trials did not describe this bias risk and thus were considered as having unclear risk of bias (Golfam 2011; Scifres 2011). One trial had high risk of bias, as perioperative FIO2 could be assessed during the postoperative period and at wound evaluation, and as data on FIO2 or gas flow used during anaesthesia were available in the participant file (Pryor 2004).

Incomplete outcome data

Eleven trials reporting on the primary outcome of surgical site infection provided numbers and reasons for dropouts and withdrawals in the intervention groups, or described no dropouts or withdrawals (Belda 2005; Bickel 2011; Duggal 2013; Gardella 2008; Mayzler 2005; Meyhoff 2009; Myles 2007; Pryor 2004; Schietroma 2013; Thibon 2012; Williams 2013). We judged four trials as having unclear risk of bias, as they did not consistently report on dropouts or withdrawals (Golfam 2011; Greif 2000; Scifres 2011; Stall 2013).

Selective reporting

Fourteen trials reported adequately on surgical site infection, and one trial reported exclusively on anastomosis leakage ‐ not on all types of surgical site infection (García‐Botello 2006). Three trials did not report surgical site infection at all (Mackintosh 2012; Mayzler 2005; Zoremba 2010).

Six trials reported on mortality within 14 to 30 days after randomization (Belda 2005; Greif 2000; Meyhoff 2009; Myles 2007; Pryor 2004; Schietroma 2013). Two trials described long‐term mortality within a median of 3.9 years and 3.5 years, respectively (Meyhoff 2009; Myles 2007).

Three trials reported on SAEs (Meyhoff 2009; Myles 2007; Pryor 2004) ‐ two trials on respiratory insufficiency (Meyhoff 2009; Myles 2007) and one trial on postoperative pulse oximetry monitored desaturation (Zoremba 2010).

Nine trials reported on length of stay (LOS) (Belda 2005; Bickel 2011; Gardella 2008; Greif 2000; Mayzler 2005; Meyhoff 2009; Myles 2007; Pryor 2004; Schietroma 2013).

One trial reported on quality of life (Greif 1999).

Therefore we determined that 20 studies had low risk of reporting bias, and eight studies had unclear risk (García‐Botello 2006; Gardella 2008; Kotani 2000; Mackintosh 2012; Mayzler 2005; Purhonen 2006; Turan 2006; Zoremba 2010) as they did not present adequate information for assessment of selective reporting.

Other potential sources of bias

Two trials had unclear risk of industry bias, as a company was involved in planning and design of the trials (Belda 2005; Thibon 2012).

Seventeen trials reporting on outcomes included in this review had low risk of financial bias (Bickel 2011; Duggal 2013; García‐Botello 2006; Gardella 2008; Golfam 2011; Greif 1999; Greif 2000; Mackintosh 2012; Mayzler 2005; Meyhoff 2009; Myles 2007; Pryor 2004; Scifres 2011; Williams 2013; Zoremba 2010; Schietroma 2013; Stall 2013).

Effects of interventions

See: Summary of findings for the main comparison Summary of findings in randomized clinical trials with overall low risk of bias and in all trials

Primary outcomes

1. All‐cause mortality in all trials according to the longest follow‐up period for each trial.

Eight trials with 4229 participants provided data on all‐cause mortality (Belda 2005; Greif 2000; Meyhoff 2012; Myles 2007; Pryor 2004; Schietroma 2013; Stall 2013; Williams 2013). Two trials reported long‐term mortality with a median follow‐up of 3.9 years in 83% of randomly assigned participants and 3.5 years in 99% of participants, respectively (Myles 2007; Meyhoff 2009). Four trials reported mortality only within 30 days (Belda 2005; Greif 2000; Pryor 2004; Schietroma 2013). Overall, in all trials a high fraction of inspired oxygen with an FIO2 of 60% to 90% compared with 30% to 40% oxygen used during anaesthesia and surgery was not associated with all‐cause mortality within the longest follow‐up in a random‐effects model (RR 1.07, 95% CI 0.87 to 1.33; participants = 4918; trials = 8; I2 = 30%; Analysis 1.1; nor in a fixed‐effect model (RR 1.10, 95% CI 0.97 to 1.25; participants = 4918; trials = 8; I2 = 30%; Analysis 1.1).

All‐cause mortality according to overall risk of bias in the included trials

Within longest follow‐up

In trials with overall low risk of bias (Meyhoff 2009; Myles 2007; Greif 2000; Williams 2013), an FIO2 of 60% to 90% compared with 30% to 40% oxygen used during anaesthesia and surgery was not associated with mortality within the longest follow‐up in a random‐effects model (RR 1.12, 95% CI 0.93 to 1.36; P value = 0.23; I2 = 41; Analysis 1.1). In trials with overall high or unclear risk of bias (Belda 2005; Pryor 2004; Schietroma 2013; Stall 2013), a high fraction of oxygen during anaesthesia and surgery was not associated with all‐cause mortality within the longest follow‐up in a random‐effects model (RR 0.43, 95% CI 0.15 to 1.20; P value = 0.11; I2 = 0; Analysis 1.1). The test of interaction for a subgroup difference was not statistically significant (P value = 0.07; I2 = 69.5%).

All‐cause mortality in best/worst and worst/best case scenarios of participants lost to follow‐up

In a best/worst case scenario (Analysis 1.2): Assuming that all participants lost to follow‐up in the group randomly assigned to a high fraction of inspired oxygen did survive, and that all participants lost to follow‐up in the group randomly assigned to a 30% to 40% fraction of inspired oxygen died, 60% to 90% oxygen was not associated with all‐cause mortality in a random‐effects model (RR 0.56, 95% CI 0.25 to 1.25).

In a worst/best case scenario (Analysis 1.3): Assuming that all participants lost to follow‐up in the group randomly assigned to a high fraction of inspired oxygen died, and that all participants lost to follow‐up in the group randomly assigned to 30% to 40% inspiratory oxygen survived, 60% to 90% oxygen was associated with an increase in all‐cause mortality in a random‐effects model (RR 1.50, 95% CI 1.04 to 2.15; P value = 0.03; I2 = 66%).

Trial sequential analysis on all‐cause mortality

TSA (Figure 2) of the four trials (Greif 2000; Meyhoff 2009; Myles 2007; Williams 2013) with overall low risk of bias reporting all‐cause mortality showed that with an anticipated RRI of 20%, mortality in the control group of 15.7% and a type 1 error of 5% and a type 2 error of 20%, and diversity (D2) of 57%, the required information size was 10,736 participants. The cumulative Z‐curve did not cross any boundaries for benefit and harm nor TSMB for futility, indicating that evidence was insufficient to refute a 20% RRI or a 20% RRR for benefit or harm of a high fraction of inspired oxygen of 60% to 90%, in the light of sparse data and repetitive testing. The TSA adjusted confidence interval for the intervention effect measured in the three trials with low risk of bias was RR 1.12 with TSA adjusted CI 0.80 to 1.59.

TSA (Figure 3) of all eight trials (Belda 2005; Greif 1999; Meyhoff 2009; Myles 2007; Pryor 2004; Schietroma 2013; Stall 2013; Williams 2013) of the effect of a high fraction of inspired oxygen on all‐cause mortality within the longest follow‐up showed that with an anticipated RRI of 20%, mortality in the control group of 15.7% and a type 1 error of 5% and a type 2 error of 20%, and diversity (D2) of 67%, the required information size was 13,264 participants. The cumulative Z‐curve did not cross the conventional boundary (P value < 0.05) nor the TSMB for harm. Neither did the cumulative Z‐curve reach the futility area. Therefore evidence was insufficient for an effect on all‐cause mortality based on all trials disregarding risk of bias and time to follow‐up. We could not exclude a 20% RRI nor a 20% RRR for lack of data. The TSA adjusted

confidence interval for the intervention effect measured in all trials was RR 1.07, with TSA adjusted CI 0.73 to 1.60.

Using a fixed‐effect model for all trials reporting all‐cause mortality did not change the results noticeably, as the cumulative Z‐curve did not enter the futility area for a 20% RRI or RRR.

2. Surgical site infection within 14 to 30 days of follow‐up after surgery

Fifteen trials with 7219 participants provided data on surgical site infection. Overall, in all trials a high fraction of inspired oxygen of 60% to 90% compared with 30% to 40% oxygen used during anaesthesia and surgery was not associated with a change in the risk of surgical site infection in a random‐effects model (RR 0.87, 95% CI 0.71 to 1.07; P value = 0.18; I2 = 48%; Analysis 1.5). However, in the fixed‐effect model, an FIO2 of 60% to 90% compared with 30% to 40% oxygen used during anaesthesia and surgery was associated with a reduction in the risk of surgical site infection (RR 0.87, 95% CI 0.77 to 0.99; Analysis 1.6).

Surgical site infection according to overall risk of bias

In trials with overall low risk of bias (Gardella 2008; Greif 2000; Meyhoff 2009; Myles 2007; Williams 2013), an FIO2 of 60% to 90% compared with 30% to 40% oxygen used during anaesthesia and surgery was not associated with risk of surgical site infection in a random‐effects model (RR 0.86, 95% CI 0.63 to 1.17; participants = 4201; studies = 5; I2 = 60%; Analysis 1.5) and in the fixed‐effect model (RR 0.86, 95% CI 0.73 to 1.00; participants = 4201; studies = 5; I2 = 60%; Analysis 1.6). In trials with overall high or unclear risk of bias (Belda 2005; Bickel 2011; Duggal 2013; Golfam 2011; Mayzler 2005; Pryor 2004; Schietroma 2013; Scifres 2011; Stall 2013; Thibon 2012) in one or more bias domains, a high fraction of inspired oxygen during anaesthesia and surgery was not associated with risk of surgical site infection in a random‐effects model (RR 0.87, 95% CI 0.64 to 1.18; participants = 3018; studies = 10; I2 = 47%; Analysis 1.5) and in the fixed‐effect model (RR 0.91, 95% CI 0.74 to 1.11; Analysis 1.6). The test of interaction for a subgroup difference in effect between trials with low risk of bias and those with unclear or high risk of bias was not significant (P value = 0.96; Analysis 1.5).

Trial sequential analysis on surgical site infection

In the TSA of a high fraction of inspired oxygen of 60% to 90% versus 30% to 40% on surgical site infection within 14 to 30 days of follow‐up in trials with low risk of bias (RR 0.86, 95% CI 0.63 to 1.17; participants = 4201; studies = 5; I2 = 60%; Analysis 1.5), the required information size to detect or reject an a priori anticipated intervention effect of 20% RRR in a random‐effects model was estimated to be 18,399 participants using a control event proportion of 12.9% and the diversity of 74% found among included trials (I2 = 60%; 95% CI 34% to 75%; P value = 0.04). Five trials with 4211 participants provided data on surgical site infection. The conventional boundary for statistical significance was not crossed (P value = 0.33), and the TSMB for benefit was not crossed (RR 0.86, TSA adjusted CI 0.42 to 1.75), suggesting that a high fraction of inspired oxygen was not associated with risk of surgical site infection.

TSA (Figure 6) of a high fraction of inspired oxygen of 60% to 90% versus 30% to 40% oxygen on surgical site infection in all trials of surgical patients within 14 to 30 days of follow‐up shows that the required information size to detect or reject a 20% RRR in a random‐effects model was 13,189 participants using a control event proportion of 12.9% and the 63% diversity found among the included trials (I2 = 48%; 95% CI 30% to 62%; P value = 0.02). The conventional boundary for statistical significance was not crossed (P value = 0.18), nor was the TSMB for benefit (RR 0.87, 95% CI 0.65 to 1.17). However, in a fixed‐effect model (RR 0.87, TSA adjusted CI 0.77 to 0.998; P value = 0.033), the TSMB for benefit was crossed, suggesting that a high fraction of inspired oxygen may be associated with surgical site infection.


Trial sequential analysis (TSA) of high inspiratory supplemental oxygen fraction 60% to 90% vs 30% to 40% in surgical site infection for all trials of surgical participants within 14 to 30 days of follow‐up. The required information size to detect or reject a 20% relative risk reduction in a random‐effects model was estimated at 13,189 participants, using a control event proportion of 12.9% and diversity of 63% among included trials (I2 = 48%; 95% CI 30% to 62%; P value = 0.02). Fifteen trials with 7219 participants provided data on surgical site infection. The conventional boundary for statistical significance was not crossed (P value = 0.18) and the trial sequential monitoring boundary for benefit was not crossed, as the TSA adjusted CI for the risk ratio was as follows: RR 0.87, 95% CI 0.65 to 1.17).

Trial sequential analysis (TSA) of high inspiratory supplemental oxygen fraction 60% to 90% vs 30% to 40% in surgical site infection for all trials of surgical participants within 14 to 30 days of follow‐up. The required information size to detect or reject a 20% relative risk reduction in a random‐effects model was estimated at 13,189 participants, using a control event proportion of 12.9% and diversity of 63% among included trials (I2 = 48%; 95% CI 30% to 62%; P value = 0.02). Fifteen trials with 7219 participants provided data on surgical site infection. The conventional boundary for statistical significance was not crossed (P value = 0.18) and the trial sequential monitoring boundary for benefit was not crossed, as the TSA adjusted CI for the risk ratio was as follows: RR 0.87, 95% CI 0.65 to 1.17).

The information size required to detect or reject an intervention effect suggested by the point estimate of 13% RRR in the random‐effects model of all trials was 34,890. No boundaries were crossed.

In the fixed‐effect model (RR 0.87, 95% CI 0.77 to 0.99), assuming absence of heterogeneity, the required information size of 4840 was surpassed and the high inspired oxygen fraction may be associated with a reduction in surgical site infection (RR 0.87, TSA adjusted CI 0.77 to 1.00; participants = 7219; studies = 15; D2 = 63%).

Surgical site infection according to use or no use of nitrous oxide in the control group

In trials with no use of nitrous oxide in the control group, a high fraction of inspired oxygen of 60% to 90% compared with 30% to 40% oxygen used during anaesthesia and surgery was not associated with a change in the risk of surgical site infection (RR 0.88, 95% CI 0.71 to 1.11; Analysis 1.7); in trials with use of nitrous oxide in the control group, supplemental oxygen during anaesthesia and surgery was not associated with a change in the risk of surgical site infection (RR 0.85, 95% CI 0.45 to 1.61; Analysis 1.7) and the test of interaction was not significant (P value = 0.91). Results when a fixed‐effect model was used were not noticeably different.

Surgical site infection according to FIO2 of 80% or higher in the experimental intervention

In trials that used an FIO2 of 80% or higher in the experimental group compared with 30% to 40% oxygen during anaesthesia and surgery, a high fraction of inspired oxygen was not associated with risk of surgical site infection in a random‐effects model (RR 0.87, 95% CI 0.66 to 1.15; Analysis 1.8). Likewise, in trials that used an FIO2 between 60% and 80% in the experimental intervention during anaesthesia and surgery, the high fraction of inspired oxygen was not associated with risk of surgical site infection in a random‐effects model (RR 0.81, 95% CI 0.65 to 1.02; Analysis 1.8), and the test of interaction was not significant (P value = 0.71). Results when a fixed‐effect model was used were not noticeably different.

Surgical site infection according to use of high FIO2 during the postoperative recovery period in the experimental intervention group

In trials that used a high FIO2 during postoperative recovery compared with 30% to 40% oxygen during anaesthesia, surgery and recovery, a high fraction of inspired oxygen was not associated with risk of surgical site infection in a random‐effects model (RR 0.87, 95% CI 0.70 to 1.09; Analysis 1.9). However, in a fixed‐effect model, a high fraction of inspired oxygen during both surgery and postoperative recovery was associated with lower risk of surgical site infection (RR 0.87, 95% CI 0.77 to 0.99). In trials reporting use of a high fraction of inspired oxygen only during surgery, the high fraction of inspired oxygen was not associated with risk of surgical site infection in a random‐effects model (RR 0.88, 95% CI 0.45 to 1.73; Analysis 1.9) nor in a fixed‐effect model (RR 0.87, 95% CI 0.44 to 1.70). However, the test for subgroup differences was not statistically significant (P value = 0.91).

Surgical site infection according to type of surgery

In trials including colorectal and abdominal surgical patients, a high FIO2 compared with 30% to 40% oxygen during anaesthesia and surgery was not associated with risk of surgical site infection (RR 0.74, 95% CI 0.50 to 1.09; Analysis 1.10), but in the fixed‐effect model, a high fraction of inspired oxygen was associated with lower risk of surgical site infection (RR 0.84, 95% CI 0.70 to 0.99). In trials including only patients for caesarean section, a high FIO2 during anaesthesia and surgery was not associated with risk of surgical site infection in a random‐effects model (RR 1.21, 95% CI 0.91 to 1.60; Analysis 1.10), and the result did not change noticeably in the fixed‐effect model. In trials including only patients for orthopaedic surgery, a high FIO2 during anaesthesia and surgery was not associated with risk of surgical site infection (RR 0.71, 95% CI 0.37 to 1.34; Analysis 1.10), and the result did not change noticeably in a fixed‐effect model. In the trial including only patients for breast surgery, a high FIO2 during anaesthesia and surgery was not associated with a change in the risk of surgical site infection (RR 0.33, 95% CI 0.01 to 7.87; Analysis 1.10). Among patients undergoing various types of surgical procedures, use of a high FIO2 during anaesthesia and surgery was associated with a decrease in the risk of surgical site infection (RR 0.76, 95% CI 0.59 to 0.99; Analysis 1.10), but the test of interaction for a subgroup difference in effect was not statistically significant (P value = 0.10).

Surgical site infection according to follow‐up longer or shorter than 14 days

In trials with follow‐up longer than 14 days, a high FIO2 during postoperative recovery compared with 30% to 40% oxygen used during anaesthesia and surgery was not associated with risk of surgical site infection (RR 0.87, 95% CI 0.72 to 1.05; Analysis 1.11); this association also was not seen in trials with follow‐up of 14 or fewer days (RR 0.84, 95% CI 0.48 to 1.47; Analysis 1.11). The test for a subgroup difference was not statistically significant (P value = 0.91). Use of a fixed‐effect model did not change the results noticeably.

Surgical site infection according to preoperative use of antibiotics

In trials with preoperative use of antibiotics, a high FIO2 compared with 30% to 40% oxygen used during anaesthesia and surgery was associated with risk of surgical site infection in a random‐effects model (RR 0.76, 95% CI 0.60 to 0.97; P value = 0.04; I2 = 49%; Analysis 1.12), but this association was lacking in trials not reporting or not using preoperative antibiotics (RR 1.20, 95% CI 0.91 to 1.58; P value = 0.21; I2 = 0; Analysis 1.12). The test for a subgroup difference was statistically significant (P value = 0.02; I2 = 82.6%). Use of a fixed‐effect model did not change the results noticeably.

Surgical site infection according to low risk of bias compared with unclear or high risk of bias of specific bias components

Surgical site infection according to adequate allocation sequence generation

In trials with adequate allocation sequence generation, an FIO2 of 60% to 90% compared with 30% to 40% oxygen used during anaesthesia and surgery was not associated with surgical site infection in a random‐effects model (RR 0.89, 95% CI 0.71 to 1.11; Analysis 1.13) nor in trials with unclear or inadequate allocation sequence generation (RR 0.69, 95% CI 0.36 to 1.31; Analysis 1.13). The test for a subgroup difference was not statistically significant (P value = 0.46; I2 = 0).

Surgical site infection according to adequate allocation concealment

In trials with adequate allocation concealment, an FIO2 of 60% to 90% compared with 30% to 40% oxygen used during anaesthesia and surgery was not associated with risk of surgical site infection in a random‐effects model (RR 0.85, 95% CI 0.67 to 1.08; Analysis 1.14) nor in trials with unclear or inadequate allocation concealment (RR 0.96, 95% CI 0.66 to 1.40; Analysis 1.14). The test for a subgroup difference was not statistically significant (P value = 0.60; I2 = 45%).

Surgical site infection according to blinding of participants and personnel

In trials with adequate blinding of participants and personnel, an FIO2 of 60% to 90% compared with 30% to 40% oxygen used during anaesthesia and surgery was not associated with risk of surgical site infection in a random‐effects model (RR 0.79, 95% CI 0.61 to 1.03; Analysis 1.15) nor in trials with unclear or inadequate blinding of participants and personnel (RR 0.98, 95% CI 0.70 to 1.37; Analysis 1.15). The test for a subgroup difference was not statistically significant (P value = 0.33; I2 = 0).

Surgical site infection according to blinding of outcome assessors

In trials with adequate blinding of outcome assessors, an FiO2 of 60% to 90% compared with 30% to 40% oxygen used during anaesthesia and surgery was associated with a decrease in risk of surgical site infection in a random‐effects model (RR 0.79, 95% CI 0.66 to 0.96; Analysis 1.16), but in trials with unclear or inadequate blinding of outcome assessors, a high oxygen fraction was associated with an increase in the risk of surgical site infection (RR 1.57, 95% CI 1.04 to 2.37; Analysis 1.16). The test for a subgroup difference was statistically significant (P value = 0.003; I2 = 48%).

Surgical site infection according to completeness of outcome data

In trials with adequate completeness of outcome data, an FIO2 of 60% to 90% compared with 30% to 40% oxygen used during anaesthesia and surgery was not associated with risk of surgical site infection in a random‐effects model (RR 0.89, 95% CI 0.71 to 1.11; Analysis 1.17) nor in trials with unclear or inadequate completeness of outcome data (RR 0.77, 95% CI 0.41 to 1.43; Analysis 1.17). The test for subgroup differences was not statistically significant (P value = 0.67; I2 = 0).

Surgical site infection according to adequate outcome reporting

Six trials (Belda 2005; Greif 2000; Meyhoff 2009; Myles 2007; Pryor 2004; Schietroma 2013) reported adequately on all‐cause mortality within a follow‐up ranging from 14 days to a median of 3.9 years. Sixteen trials reported adequately on their predefined outcome of surgical site infection (RR 0.87, 95% CI 0.71 to 1.07; participants = 7264; studies = 16; I2 = 48%; Analysis 1.18), whereas one trial (García‐Botello 2006) reported only on pHi, CO2 and anastomotic dehiscence within seven days ‐ not on surgical site infections. Nine trials (Bhatnagar 2005; Joris 2003; Kotani 2000; Mackintosh 2012; McKeen 2009; Purhonen 2003; Purhonen 2006; Simurina 2010; Zoremba 2010) did not report any of the outcomes chosen for this review.

Surgical site infection according to the presence of other bias

In trials with no presence of other bias risk, an FIO2 of 60% to 90% compared with 30 to 40% oxygen used during anaesthesia and surgery was not associated with risk of surgical site infection in a random‐effects model (RR 0.89, 95% CI 0.71 to 1.13; Analysis 1.19 ), nor in trials with the presence of unclear or other bias (RR 0.72, 95% CI 0.48 to 1.07; Analysis 1.19). The test for subgroup differences was not statistically significant (P value = 0.36; I2 = 48%).

Surgical site infection in best/worst compared with worst/best case scenarios of participants lost to follow‐up

The best/worst case scenario (Analysis 1.20): Assuming that all participants lost to follow‐up in the group randomly assigned to a high fraction of inspired oxygen did not develop a surgical site infection, and that all participants lost to follow‐up in the group randomly assigned to 30% to 40% inspired oxygen developed a surgical site infection, the intervention effect was as follows: RR 0.71, 95% CI 0.54 to 0.93.

The worst/best case scenario (Analysis 1.21): Assuming that all participants lost to follow‐up in the group randomly assigned to a high fraction of inspired oxygen developed a surgical site infection, and that all participants lost to follow‐up in the group randomly assigned to 30% to 40% oxygen did not develop a surgical site infection, the intervention effect was as follows: RR 1.05, 95% CI 0.78 to 1.42.

Sensitivity analyses

Assessment of the benefits and harms of high FIO2 performed by conducting a continuity adjustment of trials with zero events did not change the results, as very few of the trials included in meta‐analyses had a zero event.

Excluding the largest trial did change the point estimate of all‐cause mortality (RR 0.61, 95% CI 0.26 to 1.41) but was not associated with a statistically significant intervention effect and did not change the effect on surgical site infection (RR 0.89, 95% CI 0.70 to 1.12).

Excluding the smallest trial did not change our results substantially regarding all‐cause mortality (RR 1.07, 95% CI 0.86 to 1.35) nor surgical site infection (RR 0.88, 95% CI 0.71 to 1.08).

Excluding data from trials published only as abstracts and from trials with commercial involvement did not change the results for all‐cause mortality (RR 1.07, 95% CI 0.87, 1.33) and surgical site infection (RR 0.89, 95% CI 0.71 to 1.13).

Secondary outcomes

1. All‐cause mortality within 30 days of follow‐up

Overall, in all trials a high fraction of inspired oxygen of 60% to 90% compared with 30% to 40% oxygen used during anaesthesia and surgery was not associated with all‐cause mortality within 30 days of follow‐up in a model using the Peto odds ratio (POR 0.83, 95% CI 0.54 to 1.29; P value = 0.41; I2 = 60%; 4229 participants; six trials; Analysis 1.4).

All‐cause mortality within 30 days of follow‐up according to overall risk of bias in the included trials

Within 30 days of follow‐upIn trials with overall low risk of bias (Greif 2000; Meyhoff 2009; Myles 2007; Williams 2013), an FIO2 of 60% to 90% compared with 30% to 40% oxygen used during anaesthesia and surgery was not associated with mortality within 30 days of follow‐up (POR 0.99, 95% CI 0.61 to 1.60; participants = 4058; trials = 4; I2 = 77%; Analysis 1.4). In trials with overall high or unclear risk of bias (Belda 2005; Pryor 2004; Schietroma 2013; Stall 2013), a high fraction of oxygen during anaesthesia and surgery was not associated with all‐cause mortality within 30 days of follow‐up in a random‐effects model (POR 0.37, 95% CI 0.13 to 1.05; participants = 860; trials = 4; I2 = 0%; Analysis 1.4 ). The test of interaction for a subgroup difference was not statistically significant (P value = 0.09; I2 = 64.7%).

2. Respiratory insufficiency

Respiratory insufficiency, defined as the need for respiratory assistance through ventilator therapy or non‐invasive ventilation within the longest follow‐up period, was explicitly reported in only three trials (Kotani 2000; Meyhoff 2009; Zoremba 2010) and could not be meta‐analysed, as two trials (Kotani 2000; Zoremba 2010) reported no events during 24 hours of observation; however, in the other trial (Meyhoff 2009), a high fraction of inspired FIO2 was not statistically significantly associated with respiratory insufficiency (RR 1.25, 95% CI 0.79 to 1.99; Analysis 1.22).

3. Serious adverse events (SAEs)

An SAE, defined (Directive 2001) as "any event that led to death, was life‐threatening, required in‐patient hospitalisation or prolongation of existing hospitalisation, resulted in persistent or significant disability, and any important medical event which might jeopardise the patient or required intervention to prevent it", was reported in only three trials (Meyhoff 2009; Myles 2007; Pryor 2004). A high FIO2 compared with 30% to 40% oxygen used during anaesthesia and surgery was not associated with the proportion of SAEs in a random‐effects model (RR 0.96, 95% CI 0.65 to 1.43; Analysis 1.23). In trials with overall low risk of bias, a high FIO2 compared with 30% to 40% oxygen used during anaesthesia and surgery was not associated with risk of an SAE in a random‐effects model (RR 0.91, 95% CI 0.63 to 1.31; Analysis 1.23) nor in trials with high risk of bias in a random‐effects model (RR 5.00, 95% CI 0.60 to 41.85; Analysis 1.23), and the test for subgroup differences was not statistically significant (P value = 0.12).

4. Duration of postoperative hospitalizations

Overall, in all trials, supplemental oxygen with an FIO2 60% to 90% compared with 30% to 40% oxygen used during anaesthesia and surgery was not associated with LOS in days in a random‐effects model (MD ‐0.06, 95% CI ‐0.44 to 0.32; P value = 0.77; I2 = 62%; 4702 participants; nine trials; Analysis 1.24).

Duration of postoperative hospitalization according to overall risk of bias

In trials with an overall low risk of bias, an FIO2 of 60% to 90% compared with 30% to 40% oxygen used during anaesthesia and surgery was not associated with LOS in a random‐effects model (MD ‐0.13 days, 95% CI ‐0.49 to 0.24; participants = 4041; studies = 4; I2 = 41%; Analysis 1.24) nor in trials with overall high or unclear risk of bias (MD 0.69 days, 95% CI ‐0.82 to 2.20; participants = 661; studies = 5; I2 = 80%; Analysis 1.24). The test for subgroup differences was not significant (P value = 0.30; I2 = 6.6%).

5. Quality of life

Quality of life was not reported in any of the articles reporting trial results. However, we obtained data on Short Form (SF)‐36 from the Grief et al trial (Greif 1999) through Daniel Sessler, who was a co‐author of this trial report. Accordingly, no meta‐analysis could be performed for this outcome, but the trial reporting this outcome did not show an association between a high fraction of inspired oxygen and 30% to 40% oxygen and quality of life, as SF‐36 scores on the various domains were as follows.

SF‐36 domain

80% oxygen

30% oxygen

Physical functioning

61 ± 31

67 ± 29

Role physical

78 ± 34

79 ± 37

Role emotional

84 ± 32

84 ± 33

Vitality

60 ± 16

61 ± 20

Bodily pain

72 ± 27

74 ± 29

General health

72 ± 20

79 ± 31

Social functioning

84 ± 21

80 ± 25

Mental health

72 ± 18

70 ± 20

Meta‐regression

Univariate meta‐regression did not show an association between the intervention effect, Log(RR), of a high inspired fraction of oxygen and the mean age of included trial populations (MD ‐0.007 years, 95% CI ‐0.016 to 0.0023 years; P value = 0.14). However, meta‐regression of characteristics on the trial level may reveal both false‐positive and false‐negative results, and may not be applicable for the inference of the impact of age on the effect of a high inspired fraction of oxygen on outcomes at the patient level.

As information was lacking for most trials, we were not able to perform meta‐regression analyses of the intervention effect of a high inspired fraction of oxygen on mean BMI of the trial population at baseline; the fraction of diabetic patients in the trial population at baseline; the fraction of smokers in the trial population at baseline; and the fraction of patients with a contaminated or dirty infected surgical field during surgery.

Discussion

Summary of main results

Our systematic review reveals several important findings. Analysis of the eight trials reporting on mortality suggests that no statistically significant association can be found between postoperative mortality and a high fraction of inspired oxygen (FIO2) during anaesthesia, surgery and recovery when we disregard risks of bias. Meta‐analysis of the four trials with low risk of bias in all bias domains (Analysis 1.1) did not show a statistically significant association between a high inspired fraction of oxygen and mortality (risk ratio (RR) 1.12, 95% confidence interval (CI) 0.93 to 1.36). Follow‐up was within 30 days in the trial of Greif et al (Greif 2000), and median time of follow‐up was longer than three years in two trials (Meyhoff 2012; Myles 2007). Meyhoff 2009 reported a statistically significant increase in all‐cause mortality in the high inspiratory oxygen group within one year and within long‐term follow‐up with a median of 3.9 years (Meyhoff 2014). During long‐term follow‐up of the Myles 2007 trial, Leslie et al (Leslie 2011) reported no statistically significant differences in all‐cause mortality within a median follow‐up of 3.5 years. However, Meyhoff 2009 and Greif 2000 did not use nitrous oxide in the control group, and these trials were specifically designed to investigate the effects of 80% oxygen versus 30% oxygen during surgery and recovery; Myles 2007 used nitrous oxide in the control group and the trial was actually designed to investigate the effects of nitrous oxide versus oxygen on mortality, surgical site infection and adverse events. In the group of trials with unclear or high risk of bias, which reported only mortality within 30 days of follow‐up (RR 0.43, 95% CI 0.15 to 1.20), no association was found between a high inspired fraction of oxygen and mortality in a random‐effects model, and the test for subgroup differences between trials with low and high or unclear risk of bias was not statistically significant (P value = 0.07; I2 = 69.5%). A fixed‐effect meta‐analysis does not noticeably change the results. The required information size to detect or reject a relative risk increase (RRI) of 35% is surpassed, and we can refute such an intervention effect. However, trial sequential analysis (TSA) (Figure 2; Figure 3) showed that only 36% of the diversity‐adjusted required information size has been accrued to detect or reject an RRI of 20% in a random‐effects model. The effect of a high fraction of inspiratory oxygen during surgical and recovery periods of 20% RRI or relative risk reduction (RRR) on mortality may be noted when further data arrive. The TSA of the four trials with low risk of bias (Figure 1) indicates that an RRI or RRR of 20% cannot be refuted in the light of sparse data and repetitive updating of the cumulative meta‐analysis. However, the estimate of the intervention effect is likely to change when future trial data become available and are added to the cumulative meta‐analysis. A post hoc analysis (POR 0.83, 95% CI 0.54 to 1.29; Analysis 1.4) showed that the effect of a high fraction of inspired oxygen versus routine use of oxygen was not statistically significantly associated with all‐cause mortality within 30 days of follow‐up; however, the required information size was far from reached, and effect as well as lack of effect cannot be excluded.

Overall, in 15 trials, an FIO2 of 60% to 90% compared with 30% to 40% oxygen used during anaesthesia and surgery was not associated with risk of surgical site infection in a random‐effects model (RR 0.87, 95% CI 0.71 to 1.07; Analysis 1.5), but in a fixed‐effect model (RR 0.87, 95% CI 0.77 to 0.99; Analysis 1.6), a high FIO2 compared with routine use of oxygen seemed to be associated with reduced risk of surgical site infection. Given the substantial clinical and statistical heterogeneity among all trials with I2 of 60%, and among trials with low risk of bias with I2 of 47%, a fixed‐effect model seems to lack robustness. Even though the point estimate (RR 0.87) of the intervention effect is the same in both models, a considerably wider confidence interval is seen in the random‐effects model compared with the fixed‐effect model. In the subgroup analysis of trials with overall low risk of bias compared with trials with unclear or high risk of bias, the intervention with high inspiratory oxygen was not associated with surgical site infection in any of the groups, and the test for subgroup differences was not statistically significant (P value = 0.96); however, heterogeneity was substantial in both groups. TSA (Figure 3), using a random‐effects model, showed that only 55% of the required information size of 13,189 randomly assigned participants has been reached so far, and neither the conventional boundary (P value = 0.05) nor the trial sequential monitoring boundaries for benefit, harm and futility have been crossed.

In the subgroup analysis of trials using adequate blinding of outcome assessment, use of a high FIO2 compared with routine use of oxygen during anaesthesia and surgery was associated with risk of surgical site infection in both fixed‐effect and random‐effects models, and the test for subgroup differences between trials with adequate blinding of outcome assessment and trials with unclear or inadequate blinding of outcome assessment was statistically significant (P value = 0.003). However, many of the trials using adequate blinding in outcome assessment had other types of unclear or high risk of bias, and the estimate of the intervention effect is likely to change if further trials with adequate blinding are included. The information size required to detect or reject an intervention effect of 20% on surgical site infection has not been reached.

In the subgroup of eight trials with colorectal or abdominal surgery, use of a high FIO2 compared with routine use of oxygen during anaesthesia and surgery was not associated with risk of surgical site infection in a random‐effects model, and the test for subgroup differences was not statistically significant (P value = 0.10). In a fixed‐effect model, the high inspiratory oxygen fraction was associated with a decrease in surgical site infection (P value = 0.04), but the fixed‐effect model seems inappropriate in the light of substantial clinical and statistical heterogeneity between trials included in this analysis (I² = 64%; P value = 0.010). Further, several trials had unclear or high risk of bias, and the information size required to detect or reject an RRR of 20% is far from being reached.

In the subgroup of trials including various types of surgery (Myles 2007; Thibon 2012), use of a high FIO2 compared with routine use of oxygen was associated with risk of surgical site infection in both fixed‐effect and random‐effects models, but the test for subgroup differences was not statistically significant (P value = 0.10).

In the post hoc subgroup analysis of trials using antibiotics administered before surgery commenced, use of a high FIO2 compared with routine use of oxygen was associated with risk of surgical site infection in both fixed‐effect and random‐effects models, and the test for subgroup differences between trials with preoperative use of antibiotics and trials not using preoperative antibiotics (or lack of reporting) was statistically significant (P value = 0.03). However, only two of the trials using preoperative antibiotics had overall low risk of bias in all bias domains, and in these trials a high FIO2 compared with routine use of oxygen was not associated with a change in the risk of surgical site infection.

We were not able to exclude a 20% excess in all‐cause mortality of a high FIO2 compared with routine use of oxygen in trials with overall low risk of bias (RR 1.12, TSA adjusted CI 0.80 to 1.59) nor in all trials (RR 1.07, TSA adjusted CI 0.73 to 1.60) because the cumulative Z‐curve in the TSA did not reach the futility area. Further, best/worst and worst/best case analyses show that the results are highly sensitive to the loss to follow‐up noted during long‐term follow‐up, especially in the ENIGMA trial (Leslie 2011; Myles 2007), as the effect in these analyses is estimated to be RR 0.56, with 95% CI 0.25 to 1.25, and RR 1.50, with 95% CI 1.04 to 2.15, respectively. We did not find consistent evidence that a high FIO2 compared with routine use of oxygen is associated with risk of surgical site infection, respiratory insufficiency, serious adverse events (SAEs) or length of stay during the index admission. The effect on quality of life was reported in only one trial (Greif 1999) and accordingly could not be meta‐analysed, but the trial did not report an association between a high fraction of inspired oxygen and quality of life as measured by Short Form (SF)‐36 score. Some of the subgroup analyses performed in both fixed‐effect and random‐effects models indicate that a high FIO2 compared with routine use of oxygen was associated with a reduction in surgical site infection in trials of various types of surgical procedures; using preoperative antibiotics; and using adequate blinding of the outcome assessment. However, the evidence is sparse in all of these analyses, not reaching the required information size, and all subgroup analyses have considerable drawbacks according to other types of confounders and bias risks;analysis of trials using antibiotics was a post hoc defined analysis, performed after the protocol had been published.

Overall completeness and applicability of evidence

The plan to analyse the effects of a high FIO2 compared with routine use of oxygen during anaesthesia and surgery followed our published protocol to a great extent. We included all eligible randomized clinical trials (RCTs) up to February 2014. All trials were conducted in high‐income countries. Both sexes were included (63% women, 37% men). Most participants were patients undergoing colorectal or abdominal surgery under general anaesthesia; the second largest group of participants consisted of women undergoing caesarean section, mainly under regional anaesthesia. All participants came from primary prevention trials. Moderate to substantial heterogeneity was found in all of our analyses, which unfortunately was not explained by our subgroup analyses. Although more than half of the participants were randomly assigned in trials considered to have overall low risk of bias, our analyses revealed that lack of outcome reporting could influence the effect of the estimate of the intervention on surgical site infection. Accordingly, our 'best/worst case' and 'worst/best case' analyses revealed that results were compatible with no association and a harmful association with mortality and a beneficial effect, as well as with no association with surgical site infection of a high FIO2 compared with routine use of oxygen. Although these extreme sensitivity analyses represent unlikely results, they reveal how few participants lost to follow‐up should have died to change our lack of findings into a potentially harmful effect. On the contrary, these sensitivity analyses also reveal how few participants lost to follow‐up should have developed surgical site infection to change our lack of finding an effect on surgical site infections into a potentially beneficial effect.

Two trials were stopped early for benefit (meta‐analytical estimate of effect: 65% RRR for surgical site infection) (Bickel 2011; Greif 1999), one trial was stopped early for harm (estimate of effect: 100% RRI for surgical site infection) (Pryor 2004) and three trials were stopped early for futility (meta‐analytical estimate of effect: 16% RRI) (Gardella 2008; Thibon 2012, Williams 2013). Three trials did not report a sample size estimation (meta‐analytical estimate of effect: 46% RRR) (Kotani 2000; Mayzler 2005; Schietroma 2013). All of these trials were stopped early on invalid stopping criteria (one‐sided P values, P < 0.05) and on criteria not accepted by the Food and Drug Administration (FDA) or the European Medicines Agency (EMA), increasing the risk of random errors, that is, the risk of finding a difference that is not true and the risk of overlooking a difference that actually is true (Tikkinen 2011).

All of our meta‐analyses fall short of the required information size of 13,264 participants (Thorlund 2011a) to detect or reject a 20% RRI or RRR for mortality and surgical site infection. In this respect, they are merely interim analyses performed to reach or not reach conclusive evidence of a 20% effect. When evaluating benefits and harms, we used trial sequential analysis to control for random errors, and we applied congruence (i.e. we requested the same thresholds for beneficial and harmful effects). This may not be ethically defensible in the case of harms, as we usually require more solid evidence for benefits than for harms. Moreover, many interventions that may offer benefits have been withdrawn from the market because some patients have died in association with the intervention. Accordingly, societies and regulatory agencies do not have congruent requirements for evidence on benefits and harms. Therefore, we may have been too stringent when dealing with risks of harm.

Quality of the evidence

A total of eight trials including 4918 participants reported on mortality. Fifteen trials with 7219 participants reported on surgical site infection.

A major drawback of some of the included trials is the proportion of participants lost to follow‐up. This opens up the potential for attrition bias, and our 'best/worst' and 'worst/best' intention‐to‐treat analyses demonstrate that the intervention effect of a high FIO2 compared with routine use of oxygen during anaesthesia and surgery may be either beneficial or harmful. This observation calls for more comprehensive meta‐analyses of individual participant data plus further large RCTs. We have abstained from conducting 'uncertainty' analyses (Gamble 2005), which accept the point estimate from the complete participant analysis, while assuming that the distribution of surgical site infections among participants lost to follow‐up is equal to the distribution of deaths among all participants. Actually, the distribution of surgical site infections among participants lost to follow‐up may indeed be different from the distribution of participants actually followed through the whole observation period, making the 'uncertainty' analyses themselves uncertain.

We conducted various subgroup analyses. We observed no statistically significantly differences in the intervention effect of a high FIO2 compared with routine use of oxygen on mortality and surgical site infections in subgroup analyses of trials with overall low risk of bias compared with trials with overall unclear or high risk of bias; nor of trials using nitrogen compared with no nitrous oxide in the control group; trials using a high inspiratory oxygen fraction in the recovery period compared with trials using the intervention only during surgery; trials using a supplemental inspiratory oxygen fraction equal to or greater than 80% during the intervention compared with trials using 60% to 80% oxygen during the intervention; trials using a follow‐up period shorter than 14 days compared with trials using follow‐up of 30 days or longer; and trials in colorectal and abdominal surgical patients compared with trials in other types of surgery. However, in a post hoc analysis of trials using preoperative antibiotics compared with trials not using or not reporting use of preoperative antibiotics, we observed subgroup differences in effects of the intervention. Furthermore, it may be worthwhile to analyse whether the effects of a high FIO2 compared with routine use of oxygen depend on the type of anaesthesia (general or regional anaesthesia) used during surgery. We did not plan such an analysis, and analysis may be complicated by the fact that some trials used both types of anaesthesia.

In addition to the six trials reporting mortality, nine trials reported on surgical site infection but not on mortality. Most of these trials were RCTs conducted to assess the effects of high FIO2 compared with routine use of oxygen on surgical site infection; two trials reported exclusively on surrogate outcomes. A high FIO2 compared with routine use of oxygen decreased surgical site infection on the fixed‐effect model meta‐analysis, but heterogeneity was substantial. The random‐effects model did not reveal a statistically significant effect of a high FIO2 compared with routine use of oxygen on surgical site infection. Accordingly, the decrease in the proportion of surgical site infections in a fixed‐effect model may be due to an assumption that the intervention effect is equal across included trials ‐ an assumption that probably is not valid because the incidence of surgical site infection and the bacteria involved differ substantially (e.g. in abdominal surgery vs other types of surgery).

We present our main results on all investigated outcomes in a summary of findings table (summary of findings Table for the main comparison) and have graded the quality of the evidence, including imprecision, according to Grades of Recommendation, Assessment, Development and Evaluation (GRADE) (Guyatt 2011; Guyatt 2013), stressing that results of meta‐analyses reflect data derived from trials of low or very low quality.

The finding of a lack of significant association between a high FIO2 of 60% to 90% and 30% to 40% oxygen and mortality and surgical site infection may be due to a type 2 error ‐ a false‐negative finding. This is exemplified by the fact that the cumulative Z‐curve did not cross the trial sequential monitoring boundaries for futility nor for benefit or harm nor for the boundaries of futility for an anticipated effect of 20% RRR. However, such an analysis cannot remove risks of bias ‐ detected or undetected. One should discuss, however, how much evidence one would require when dealing with potential benefit or harm. On the one hand, estimated beneficial or harmful effects can occur as the result of random errors. Therefore, sufficient information needs to be assessed to demonstrate benefit or harm beyond reasonable doubt. In this review, the TSA performed was based on anticipation of rather big intervention effects, undoubtedly clinically relevant but possibly not realistic. To detect or reject even smaller intervention effects, much larger information sizes are required.

Potential biases in the review process

We repeatedly searched several databases and contacted authors of trials investigating a high fraction of inspired oxygen of 60% to 90% compared with 30% to 40% oxygen used during anaesthesia and surgery. Therefore, we assume that we have not overlooked important RCTs. On the other hand, in general only about every second trial is reported (Gluud 2008), so we cannot exclude reporting biases, and inspecting our funnel plot for the effect of high supplemental oxygen on surgical site infection may suggest small trial bias. On the positive side, we managed to obtain more information on several trials than was revealed by any previously conducted reviews and meta‐analyses. However, this does not detract from the fact that we did not have access to individual participant data. Accordingly, we have no chance of analysing the effect of an FIO2 of 60% to 90% compared with 30% to 40% oxygen used during anaesthesia and surgery in patients receiving epidural analgesia, in patients kept normothermic during anaesthesia and surgery and in those with adequate fluid balance postoperatively. We selected all trials and extracted all data in duplicate, and we reached a high level of agreement. We did not conduct quality assessments nor data extractions while blinded to study authors and bias risks, and four of the authors of this review have been involved in one of the included trials (Meyhoff 2009). In this review, we present a conservative and, we believe, a more correct interpretation of findings compared with interpretations given in some other reviews and meta‐analyses previously published outside The Cochrane Collaboration.

Agreements and disagreements with other studies or reviews

Previous meta‐analyses of preventive trials of a high fraction of inspired oxygen with FIO2 60% to 90% compared with 30% to 40% oxygen have provided substantially less information and have not examined the separate influence of different ways of using the intervention (Brar 2009; Brar 2011; Chura 2007; Fakhry 2012; Hovaguimian 2013; Kao 2012; Klingel 2013; Patel 2012; Qadan 2009; Togioka 2012). None of these meta‐analyses or systematic reviews conducted a proper bias assessment according to the Cochrane method, investigated the effect of bias in the seven domains as recommended by The Cochrane Collaboration or conducted analyses of possible subgroup differences in the effects of a high fraction of inspired oxygen between trials with overall low risk of bias and trials with one or more domains having high or unclear risk of bias. We conducted a thorough review in accordance with The Cochrane Collaboration methodology (Higgins 2011), implementing findings of methodological studies (Kjaergard 2001; Lundh 2012; Moher 1998; Savović 2012; Schulz 1995; Wood 2008). Between‐trial heterogeneity was moderate to substantial in all of our meta‐analyses. This may emphasize the inconsistency of our findings but also the diversity of the populations and settings included in the different trials, making results more generalizable (Ioannidis 2006). Furthermore, all‐cause mortality generally should be connected with unbiased estimates (Savović 2012; Wood 2008). We also performed trial sequential analyses to control the risk of random errors in a cumulative meta‐analysis and to prevent premature statements of superiority of a high fraction of inspired oxygen compared with routine use of oxygen (Brok 2008; Brok 2009; Thorlund 2009; Thorlund 2011; Thorlund 2011a; Wetterslev 2008; Wetterslev 2009).

Several trials (Goll 2001; Greif 1999; Joris 2003; McKeen 2009; Purhonen 2003; Purhonen 2006; Simurina 2010; Turan 2006) using a high fraction of inspired oxygen during surgery did not report on mortality or surgical site infection but reported exclusively on nausea and vomiting or surrogate outcomes for respiratory complications.

The number of meta‐analyses conducted so far and the fact that meta‐analyses have been updated several times from 2007 until now underline the necessity to take into account multiple testing on accumulating data when evaluating cumulative meta‐analyses (Thorlund 2009; Wetterslev 2008). Further, when data are sparse and have not reached the required information size, the effect on, for example, mortality (Brar 2011) and surgical site infection (Qadan 2009) seems to be subject to change when data from further trials are included. However, none of the previously conducted meta‐analyses or systematic reviews estimated a required information size for a realistic intervention effect on all‐cause mortality or surgical site infection. Accordingly, they were not able to adjust the CI for increased risk of random errors when data were sparse and repetitive testing due to numerous updates was performed.

Despite discrepancies between our review and other meta‐analyses and systematic reviews, we agree strongly with recently published reviews reporting no firm evidence that a high fraction of inspired oxygen of 60% to 90% compared with 30% to 40% oxygen during surgery (and for some trials also during the recovery room period) reduces all‐cause mortality and surgical site infection. Further, meta‐analyses and reviews concluding that a high inspiratory fraction of oxygen during surgery has an effect on surgical site infection have not demonstrated this convincingly by using both fixed‐effect and random‐effects models. Moreover, our analyses did not refute a clinically relevant effect, and lack of information is not the same as information of lack of effect. Therefore, further trials with low risk of bias and many more randomly assigned participants may very well change the estimates of effects on mortality and surgical site infection of a high fraction of inspired oxygen during surgery and recovery.

Funnel plot of comparison: 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, outcome: 1.5 Surgical site infection stratified according to overall risk of bias in a random‐effects model.
Figures and Tables -
Figure 1

Funnel plot of comparison: 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, outcome: 1.5 Surgical site infection stratified according to overall risk of bias in a random‐effects model.

Trial sequential analysis of the effect on mortality within the longest follow‐up in trials with overall low risk of bias. With an anticipated relative risk increase (RRI) of 20%, mortality in the control group of 15.7% with a type 1 error of 5% and a type 2 error of 20%, and diversity (D2) of 57%, the required information size is 10,736 participants. The cumulative Z‐curve does not cross the conventional boundary nor the trial sequential monitoring boundary for harm. The cumulative Z‐curve does not reach the futility area. Therefore evidence of an effect on all‐cause mortality based on trials with low risk of bias is lacking, and we cannot exclude a 20% RRI due to lack of data.
Figures and Tables -
Figure 2

Trial sequential analysis of the effect on mortality within the longest follow‐up in trials with overall low risk of bias. With an anticipated relative risk increase (RRI) of 20%, mortality in the control group of 15.7% with a type 1 error of 5% and a type 2 error of 20%, and diversity (D2) of 57%, the required information size is 10,736 participants. The cumulative Z‐curve does not cross the conventional boundary nor the trial sequential monitoring boundary for harm. The cumulative Z‐curve does not reach the futility area. Therefore evidence of an effect on all‐cause mortality based on trials with low risk of bias is lacking, and we cannot exclude a 20% RRI due to lack of data.

Trial sequential analysis of all trials reporting all‐cause mortality. With an anticipated relative risk increase (RRI) of 20%, mortality in the control group of 15.7% with a type 1 error of 5% and a type 2 error of 20% and diversity (D2) of 65%, the required information size is 13,264 participants. The cumulative Z‐curve does not cross the conventional boundaries nor the boundaries for benefit or harm. The cumulative Z‐curve does not reach the futility area. Therefore evidence of both beneficial and harmful effects on all‐cause mortality is lacking for all trials regardless of risk of bias and with varying time to follow‐up. We cannot exclude a 20% RRI or RRR due to lack of data.
Figures and Tables -
Figure 3

Trial sequential analysis of all trials reporting all‐cause mortality. With an anticipated relative risk increase (RRI) of 20%, mortality in the control group of 15.7% with a type 1 error of 5% and a type 2 error of 20% and diversity (D2) of 65%, the required information size is 13,264 participants. The cumulative Z‐curve does not cross the conventional boundaries nor the boundaries for benefit or harm. The cumulative Z‐curve does not reach the futility area. Therefore evidence of both beneficial and harmful effects on all‐cause mortality is lacking for all trials regardless of risk of bias and with varying time to follow‐up. We cannot exclude a 20% RRI or RRR due to lack of data.

Trial flow diagram. We reran the search in March 2015. We found two studies of interest, which we will consider when we update the review.
Figures and Tables -
Figure 4

Trial flow diagram. We reran the search in March 2015. We found two studies of interest, which we will consider when we update the review.

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.
Figures and Tables -
Figure 5

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Trial sequential analysis (TSA) of high inspiratory supplemental oxygen fraction 60% to 90% vs 30% to 40% in surgical site infection for all trials of surgical participants within 14 to 30 days of follow‐up. The required information size to detect or reject a 20% relative risk reduction in a random‐effects model was estimated at 13,189 participants, using a control event proportion of 12.9% and diversity of 63% among included trials (I2 = 48%; 95% CI 30% to 62%; P value = 0.02). Fifteen trials with 7219 participants provided data on surgical site infection. The conventional boundary for statistical significance was not crossed (P value = 0.18) and the trial sequential monitoring boundary for benefit was not crossed, as the TSA adjusted CI for the risk ratio was as follows: RR 0.87, 95% CI 0.65 to 1.17).
Figures and Tables -
Figure 6

Trial sequential analysis (TSA) of high inspiratory supplemental oxygen fraction 60% to 90% vs 30% to 40% in surgical site infection for all trials of surgical participants within 14 to 30 days of follow‐up. The required information size to detect or reject a 20% relative risk reduction in a random‐effects model was estimated at 13,189 participants, using a control event proportion of 12.9% and diversity of 63% among included trials (I2 = 48%; 95% CI 30% to 62%; P value = 0.02). Fifteen trials with 7219 participants provided data on surgical site infection. The conventional boundary for statistical significance was not crossed (P value = 0.18) and the trial sequential monitoring boundary for benefit was not crossed, as the TSA adjusted CI for the risk ratio was as follows: RR 0.87, 95% CI 0.65 to 1.17).

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 1 Mortality within longest follow‐up stratified according to overall risk of bias.
Figures and Tables -
Analysis 1.1

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 1 Mortality within longest follow‐up stratified according to overall risk of bias.

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 2 Mortality within longest follow‐up best/worst case scenario of participants lost to follow‐up.
Figures and Tables -
Analysis 1.2

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 2 Mortality within longest follow‐up best/worst case scenario of participants lost to follow‐up.

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 3 Mortality within longest follow‐up worst/best case scenario of participants lost to follow‐up.
Figures and Tables -
Analysis 1.3

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 3 Mortality within longest follow‐up worst/best case scenario of participants lost to follow‐up.

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 4 Mortality within 30 days of follow‐up stratified according to overall risk of bias.
Figures and Tables -
Analysis 1.4

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 4 Mortality within 30 days of follow‐up stratified according to overall risk of bias.

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 5 Surgical site infection stratified according to overall risk of bias in a random‐effects model.
Figures and Tables -
Analysis 1.5

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 5 Surgical site infection stratified according to overall risk of bias in a random‐effects model.

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 6 Surgical site infection stratified according to overall risk of bias in a fixed‐effect model.
Figures and Tables -
Analysis 1.6

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 6 Surgical site infection stratified according to overall risk of bias in a fixed‐effect model.

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 7 Surgical site infection stratified according to use of nitrous oxide.
Figures and Tables -
Analysis 1.7

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 7 Surgical site infection stratified according to use of nitrous oxide.

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 8 Surgical site infection stratified according to FIO2 in the intervention group higher or lower than 80%.
Figures and Tables -
Analysis 1.8

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 8 Surgical site infection stratified according to FIO2 in the intervention group higher or lower than 80%.

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 9 Surgical site infection stratified according to use of high FIO2 during operation and the postoperative period.
Figures and Tables -
Analysis 1.9

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 9 Surgical site infection stratified according to use of high FIO2 during operation and the postoperative period.

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 10 Surgical site infection stratified according to type of surgery.
Figures and Tables -
Analysis 1.10

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 10 Surgical site infection stratified according to type of surgery.

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 11 Surgical site infection according to follow‐up longer or shorter than 14 days.
Figures and Tables -
Analysis 1.11

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 11 Surgical site infection according to follow‐up longer or shorter than 14 days.

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 12 Surgical site infection stratified according to use of preoperative antibiotics.
Figures and Tables -
Analysis 1.12

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 12 Surgical site infection stratified according to use of preoperative antibiotics.

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 13 Surgical site infection stratified according to sequence generation.
Figures and Tables -
Analysis 1.13

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 13 Surgical site infection stratified according to sequence generation.

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 14 Surgical site infection stratified according to allocation concealment.
Figures and Tables -
Analysis 1.14

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 14 Surgical site infection stratified according to allocation concealment.

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 15 Surgical site infection according to blinding of participants and personnel.
Figures and Tables -
Analysis 1.15

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 15 Surgical site infection according to blinding of participants and personnel.

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 16 Surgical site infection stratified according to outcome assessment.
Figures and Tables -
Analysis 1.16

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 16 Surgical site infection stratified according to outcome assessment.

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 17 Surgical site infection stratified according to completeness of outcome data.
Figures and Tables -
Analysis 1.17

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 17 Surgical site infection stratified according to completeness of outcome data.

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 18 Surgical site infection stratified according to outcome reporting.
Figures and Tables -
Analysis 1.18

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 18 Surgical site infection stratified according to outcome reporting.

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 19 Surgical site infection stratified according to presence of other bias.
Figures and Tables -
Analysis 1.19

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 19 Surgical site infection stratified according to presence of other bias.

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 20 Surgical site infection best/worst case scenario of participants lost to follow‐up.
Figures and Tables -
Analysis 1.20

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 20 Surgical site infection best/worst case scenario of participants lost to follow‐up.

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 21 Surgical site infection worst/best case scenario of participants lost to follow‐up.
Figures and Tables -
Analysis 1.21

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 21 Surgical site infection worst/best case scenario of participants lost to follow‐up.

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 22 Respiratory insufficiency stratified according to overall risk of bias.
Figures and Tables -
Analysis 1.22

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 22 Respiratory insufficiency stratified according to overall risk of bias.

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 23 Serious adverse events.
Figures and Tables -
Analysis 1.23

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 23 Serious adverse events.

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 24 Length of stay after surgery.
Figures and Tables -
Analysis 1.24

Comparison 1 60% to 90% oxygen vs 30% to 40% oxygen perioperatively, Outcome 24 Length of stay after surgery.

Summary of findings for the main comparison. Summary of findings in randomized clinical trials with overall low risk of bias and in all trials

Fraction of inspired oxygen of 60% to 90% compared with 30% to 40% oxygen during anaesthesia, surgery and recovery

Patient or population: surgical patients with need for abdominal, caesarean section, orthopaedic or breast surgery

Settings: perioperative and postoperative

Intervention: high fraction of inspired oxygen of 60% to 90% during and after surgery and anaesthesia

Comparison: fraction of inspired oxygen of 30% to 40% during and after anaesthesia and surgery

Outcomes

Illustrative comparative risks (95% CI)

Relative effect
(95% CI)

Number of participants
(studies)

Quality of the evidence
(GRADE)

Comments

Assumed risk*

Corresponding risk#

Inspired fraction of oxygen (30% to 40%)

Inspired fraction of oxygen (60% to 90%)

Mortality within the longest follow‐up in trials with overall low risk of bias

(follow‐up: 30 days to a median of 3.9 years)

High‐risk population

RR 1.12

(0.93 to 1.36)

4758

(3)

⊕⊕⊝⊝
Lowa

TSA (Figure 1) shows that the required information size of 10,736 for a 20% RRI has not been achieved, and that no trial sequential monitoring boundaries have been crossed. The TSA adjusted CI for the RR is 0.73 to 1.60. The required information size is even greater for a lower effect on mortality than a 20% RRI. We therefore downgraded by 1 level for imprecision and by 1 level for attrition bias (Analysis 1.2; Analysis 1.3)

185 per 1000
(168 to 203)

208 per 1000
(190 to 226)

Mortality within the longest follow‐up, irrespective of risk of bias (14 days to median of 3.9 years)

164 per 1000
(150 to 180)

175 per 1000
(143 to 218)

RR 1.07

(0.87 to 1.33)

4525

(6)

⊕⊝⊝⊝

Very lowb

TSA (Figure 2) shows that the required information size for a 20% RRI has not been achieved, and that no trial sequential monitoring boundaries have been crossed. The required information size for a lower excess mortality than a 20% RRI is even greater. We therefore downgraded by 1 level for imprecision, by 1 level for high risk of attrition bias (Analysis 1.2; Analysis 1.3) and by 1 level for overall risk of bias

Mortality within 30 days of follow‐up with overall low risk of bias

(14 days to 30 days)

18 per 1000

(13 to 25)

18 per 1000

(13 to 25)

POR 0.99

(0.61 to 1.60)

4758

(3)

⊕⊕⊕⊝
Moderatec

The required information size for a 20% RRI has not been reached, and no trial sequential monitoring boundaries have been crossed

Mortality within 30 days of follow‐up, irrespective of risk of bias (14 days to 30 days)

20 per 1000

(15 to 27)

17 per 1000

(11 to 26)

POR 0.83

(0.54 to 1.29)

4525

(6)

⊕⊕⊝⊝
Lowd

TSA shows that the required information size for a 20% RRI has not been achieved, and that no trial sequential monitoring boundaries have been crossed. The required information size for a lower excess mortality than a 20% RRI is even greater. We therefore downgraded by 1 level for imprecision and by 1 level for overall risk of bias

Surgical site infection within 30 days in trials with low risk of bias

(14 days to 30 days)

140 per 1000

(126 to 155)

119 per 1000

(106 to 134)

RR 0.86

(0.63 to 1.17)

4201

(5)

⊕⊕⊝⊝
Lowa

TSA (Figure 3) shows that the required information size of 13,189 participants for a 20% RRR has not been achieved, and that no trial sequential monitoring boundaries have been crossed. The required information size is even greater for a lower effect on mortality than a 20% RRR. Best/worst and worst/best case scenarios indicate possible attrition bias. Therefore we downgraded by 1 level for imprecision and by 1 level for attrition bias

Surgical site infection within 30 days, irrespective of risk of bias (14 days to 30 days)

129 per 1000

(118 to 140)

112 per 1000

(92 to 138)

RR 0.87

(0.71 to 1.07)

7229

(15)

⊕⊕⊝⊝
Lowe

TSA (Figure 3) shows that the required information size of 13,189 participants for a 20% RRR of surgical site infection has not been reached, and that no trial sequential monitoring boundaries have been crossed. The required information size for a lower effect on surgical site infection than a 20% RRR is even greater. As the result of overall risk of bias and imprecision, we downgraded by 2 levels

Respiratory insufficiency

44 per 1000

(31 to 62)

55 per 1000

(35 to 88)

RR 1.25

(0.79 to 1.99)

1386

(1)

⊕⊕⊕⊝
Moderatec

Information size is too small to be conclusive. We downgraded by 1 level for imprecision

*The basis for the assumed risk (e.g. median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).
CI: Confidence interval; LCL: Lower confidence limit; MD: Mean difference; POR: Peto odds ratio; RR: Risk ratio; RRR: relative risk reduction; RRI: relative risk increase; TSA: trial sequential analysis; UCL: Upper confidence limit.

GRADE Working Group grades of evidence.
High quality: Further research is very unlikely to change our confidence in the estimate of effect.
Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate.
Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.
Very low quality: We are very uncertain about the estimate.

*Mean of mortality in groups of included trials with 30% to 40% inspiratory fraction of oxygen.

#Calculated from: RR × assumed risk (95% CL: LCL of RR × assumed risk to UCL of RR × assumed risk).

aDowngraded by 2 levels because of risk of attrition bias and imprecision.

bDowngraded by 3 levels because of risk of attrition bias, imprecision and overall risk of bias.
cDowngraded by 1 level because of imprecision.

dDowngraded by 2 levels because of overall risk of bias and imprecision.

eDowngraded by 2 levels because of overall risk of bias and imprecision.

Figures and Tables -
Summary of findings for the main comparison. Summary of findings in randomized clinical trials with overall low risk of bias and in all trials
Comparison 1. 60% to 90% oxygen vs 30% to 40% oxygen perioperatively

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

1 Mortality within longest follow‐up stratified according to overall risk of bias Show forest plot

8

4918

Risk Ratio (M‐H, Random, 95% CI)

1.07 [0.87, 1.33]

1.1 Trials with overall low risk of bias

4

4058

Risk Ratio (M‐H, Random, 95% CI)

1.12 [0.93, 1.36]

1.2 Trials with overall unclear or high risk of bias

4

860

Risk Ratio (M‐H, Random, 95% CI)

0.43 [0.15, 1.20]

2 Mortality within longest follow‐up best/worst case scenario of participants lost to follow‐up Show forest plot

8

4918

Risk Ratio (M‐H, Random, 95% CI)

0.56 [0.25, 1.25]

3 Mortality within longest follow‐up worst/best case scenario of participants lost to follow‐up Show forest plot

8

4918

Risk Ratio (M‐H, Random, 95% CI)

1.50 [1.04, 2.15]

4 Mortality within 30 days of follow‐up stratified according to overall risk of bias Show forest plot

8

4918

Peto Odds Ratio (Peto, Fixed, 95% CI)

0.83 [0.54, 1.29]

4.1 Trials with overall low risk of bias

4

4058

Peto Odds Ratio (Peto, Fixed, 95% CI)

0.99 [0.61, 1.60]

4.2 Trials with overall high or unclear risk of bias

4

860

Peto Odds Ratio (Peto, Fixed, 95% CI)

0.37 [0.13, 1.05]

5 Surgical site infection stratified according to overall risk of bias in a random‐effects model Show forest plot

15

7219

Risk Ratio (M‐H, Random, 95% CI)

0.87 [0.71, 1.07]

5.1 Trials with low risk of bias

5

4201

Risk Ratio (M‐H, Random, 95% CI)

0.86 [0.63, 1.17]

5.2 Trials with high or unclear risk of bias

10

3018

Risk Ratio (M‐H, Random, 95% CI)

0.87 [0.64, 1.18]

6 Surgical site infection stratified according to overall risk of bias in a fixed‐effect model Show forest plot

15

7219

Risk Ratio (M‐H, Fixed, 95% CI)

0.87 [0.77, 0.99]

6.1 Trials with low risk of bias

5

4201

Risk Ratio (M‐H, Fixed, 95% CI)

0.86 [0.73, 1.00]

6.2 Trials with high or unclear risk of bias

10

3018

Risk Ratio (M‐H, Fixed, 95% CI)

0.91 [0.74, 1.11]

7 Surgical site infection stratified according to use of nitrous oxide Show forest plot

14

7159

Risk Ratio (M‐H, Random, 95% CI)

0.87 [0.71, 1.08]

7.1 No use of nitrous oxide

10

4739

Risk Ratio (M‐H, Random, 95% CI)

0.88 [0.71, 1.11]

7.2 Use of nitrous oxide

4

2420

Risk Ratio (M‐H, Random, 95% CI)

0.85 [0.45, 1.61]

8 Surgical site infection stratified according to FIO2 in the intervention group higher or lower than 80% Show forest plot

15

7219

Risk Ratio (M‐H, Random, 95% CI)

0.87 [0.71, 1.07]

8.1 FIO2 in the intervention group 80% or higher

12

4216

Risk Ratio (M‐H, Random, 95% CI)

0.87 [0.66, 1.15]

8.2 FIO2 in intervention group lower than 80% and higher than 60%

3

3003

Risk Ratio (M‐H, Random, 95% CI)

0.81 [0.65, 1.02]

9 Surgical site infection stratified according to use of high FIO2 during operation and the postoperative period Show forest plot

15

7219

Risk Ratio (M‐H, Random, 95% CI)

0.87 [0.71, 1.07]

9.1 Use of high FIO2 during both surgery and the postoperative period

13

6725

Risk Ratio (M‐H, Random, 95% CI)

0.87 [0.70, 1.09]

9.2 Use of high FIO2 only during surgery

2

494

Risk Ratio (M‐H, Random, 95% CI)

0.88 [0.45, 1.73]

10 Surgical site infection stratified according to type of surgery Show forest plot

15

7219

Risk Ratio (M‐H, Random, 95% CI)

0.87 [0.71, 1.07]

10.1 Colorectal and abdominal surgery

7

2761

Risk Ratio (M‐H, Random, 95% CI)

0.74 [0.50, 1.09]

10.2 Caesarean section

4

1719

Risk Ratio (M‐H, Random, 95% CI)

1.21 [0.91, 1.60]

10.3 Orthopaedic surgery

1

233

Risk Ratio (M‐H, Random, 95% CI)

0.71 [0.37, 1.34]

10.4 Breast surgery

1

60

Risk Ratio (M‐H, Random, 95% CI)

0.33 [0.01, 7.87]

10.5 Mixed types of surgery

2

2446

Risk Ratio (M‐H, Random, 95% CI)

0.76 [0.59, 0.99]

11 Surgical site infection according to follow‐up longer or shorter than 14 days Show forest plot

15

7219

Risk Ratio (M‐H, Random, 95% CI)

0.87 [0.71, 1.07]

11.1 Follow‐up longer than 14 days

8

6141

Risk Ratio (M‐H, Random, 95% CI)

0.87 [0.72, 1.05]

11.2 Follow‐up of 14 or fewer days

7

1078

Risk Ratio (M‐H, Random, 95% CI)

0.84 [0.48, 1.47]

12 Surgical site infection stratified according to use of preoperative antibiotics Show forest plot

15

7219

Risk Ratio (M‐H, Random, 95% CI)

0.87 [0.71, 1.07]

12.1 Use of preoperative antibiotics

10

5440

Risk Ratio (M‐H, Random, 95% CI)

0.76 [0.60, 0.97]

12.2 Lack of use or lack of reporting use of preoperative antibiotics

5

1779

Risk Ratio (M‐H, Random, 95% CI)

1.20 [0.91, 1.58]

13 Surgical site infection stratified according to sequence generation Show forest plot

15

7219

Risk Ratio (M‐H, Random, 95% CI)

0.87 [0.71, 1.07]

13.1 Adequate sequence generation

10

6155

Risk Ratio (M‐H, Random, 95% CI)

0.89 [0.71, 1.11]

13.2 Unclear/inadequate sequence generation

5

1064

Risk Ratio (M‐H, Random, 95% CI)

0.69 [0.36, 1.31]

14 Surgical site infection stratified according to allocation concealment Show forest plot

16

7264

Risk Ratio (M‐H, Random, 95% CI)

0.87 [0.71, 1.07]

14.1 Adequate allocation concealment

12

5894

Risk Ratio (M‐H, Random, 95% CI)

0.85 [0.67, 1.09]

14.2 Unclear/inadequate allocation concealment

4

1370

Risk Ratio (M‐H, Random, 95% CI)

0.96 [0.66, 1.40]

15 Surgical site infection according to blinding of participants and personnel Show forest plot

15

7219

Risk Ratio (M‐H, Random, 95% CI)

0.87 [0.71, 1.07]

15.1 Adequate blinding of participants and personnel

7

4668

Risk Ratio (M‐H, Random, 95% CI)

0.79 [0.61, 1.03]

15.2 Unclear/inadequate blinding of participants or personnel

8

2551

Risk Ratio (M‐H, Random, 95% CI)

0.98 [0.70, 1.37]

16 Surgical site infection stratified according to outcome assessment Show forest plot

15

7219

Risk Ratio (M‐H, Random, 95% CI)

0.87 [0.71, 1.07]

16.1 Adequate blinding of outcome assessment

12

6414

Risk Ratio (M‐H, Random, 95% CI)

0.79 [0.66, 0.96]

16.2 Unclear/inadequate blinding of outcome assessment

3

805

Risk Ratio (M‐H, Random, 95% CI)

1.57 [1.04, 2.37]

17 Surgical site infection stratified according to completeness of outcome data Show forest plot

15

7219

Risk Ratio (M‐H, Random, 95% CI)

0.87 [0.71, 1.07]

17.1 Complete outcome data

11

5841

Risk Ratio (M‐H, Random, 95% CI)

0.89 [0.71, 1.11]

17.2 Incompete outcome data

4

1378

Risk Ratio (M‐H, Random, 95% CI)

0.77 [0.41, 1.43]

18 Surgical site infection stratified according to outcome reporting Show forest plot

15

7219

Risk Ratio (M‐H, Random, 95% CI)

0.87 [0.71, 1.07]

18.1 Adequate outcome reporting

15

7219

Risk Ratio (M‐H, Random, 95% CI)

0.87 [0.71, 1.07]

18.2 Unclear/inadequate outcome reporting

0

0

Risk Ratio (M‐H, Random, 95% CI)

0.0 [0.0, 0.0]

19 Surgical site infection stratified according to presence of other bias Show forest plot

15

7219

Risk Ratio (M‐H, Random, 95% CI)

0.87 [0.71, 1.07]

19.1 No presence of other bias

13

6489

Risk Ratio (M‐H, Random, 95% CI)

0.89 [0.71, 1.13]

19.2 Unclear or definite presence of other bias

2

730

Risk Ratio (M‐H, Random, 95% CI)

0.72 [0.48, 1.07]

20 Surgical site infection best/worst case scenario of participants lost to follow‐up Show forest plot

15

7219

Risk Ratio (M‐H, Random, 95% CI)

0.71 [0.54, 0.93]

21 Surgical site infection worst/best case scenario of participants lost to follow‐up Show forest plot

15

7219

Risk Ratio (M‐H, Random, 95% CI)

1.05 [0.78, 1.42]

22 Respiratory insufficiency stratified according to overall risk of bias Show forest plot

3

1588

Risk Ratio (M‐H, Random, 95% CI)

1.25 [0.79, 1.99]

22.1 Trials with low risk of bias

1

1386

Risk Ratio (M‐H, Random, 95% CI)

1.25 [0.79, 1.99]

22.2 Trials with unclear or high risk of bias

2

202

Risk Ratio (M‐H, Random, 95% CI)

0.0 [0.0, 0.0]

23 Serious adverse events Show forest plot

3

3558

Risk Ratio (M‐H, Random, 95% CI)

0.96 [0.65, 1.43]

23.1 Trials with low risk of bias

2

3398

Risk Ratio (M‐H, Random, 95% CI)

0.91 [0.63, 1.31]

23.2 Trials with high risk of bias

1

160

Risk Ratio (M‐H, Random, 95% CI)

5.0 [0.60, 41.85]

24 Length of stay after surgery Show forest plot

7

4702

Mean Difference (IV, Random, 95% CI)

‐0.06 [‐0.44, 0.32]

24.1 Trials with low risk of bias

4

4041

Mean Difference (IV, Random, 95% CI)

‐0.13 [‐0.49, 0.24]

24.2 Trials with high or unclear risk of bias

3

661

Mean Difference (IV, Random, 95% CI)

0.69 [‐0.82, 2.20]

Figures and Tables -
Comparison 1. 60% to 90% oxygen vs 30% to 40% oxygen perioperatively