Discussion

Material in this chapter has been reproduced with permission from Elfeky et al.2 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/. The text below includes minor additions and formatting changes to the original text.

The Royal College of Anaesthetists estimates that, in a given year, 1 in 20 people undergoes anaesthesia for surgery in the UK.3 As part of this procedure, patients will usually be given additional oxygen both intraoperatively and for a period postoperatively. Despite its daily and frequent use, there is significant clinical uncertainty about different aspects of perioperative oxygen therapy. There are mixed views surrounding the potential benefits and harms of using a higher concentration of oxygen to reduce SSIs in different populations, and around the use of HFNO therapy. A significant number of systematic reviews and meta-analyses looking at various aspects of perioperative oxygen therapy have been published, but clinical uncertainty remains because of the different scopes and focuses of the reviews, and consequently the varied coverage, synthesis and interpretation of evidence leading to apparently inconsistent or sometimes conflicting conclusions.

Our overview of systematic reviews provides a panoramic view of the best evidence on the topic and reconciles some of the discrepancies by bringing these reviews together, critiquing and selecting the most comprehensive ones and updating them with the latest evidence to aid clinical decision-making and the direction of future research.

Summary of main results and certainty of evidence

Our overview reveals several important findings:

1. Our panoramic meta-analysis including different strategies and methods/devices for oxygen delivery (use of high FiO₂, HFNO or NIV) indicates that giving higher levels of oxygen perioperatively has little or no effect on mortality compared with COT. This was based on a relatively small number of death events.
2. A high FiO₂ (80%) may result in a slight reduction in SSI compared with low FiO₂ (30–35%) when administered intraoperatively. Evidence from exploratory subgroup analyses suggests that reduction in SSI was more likely to be observed among patients under general anaesthesia; undergone emergency surgery; in trials rated to be low risk of bias; and had surgery that was not a caesarean section. These potential subgroup effects were based on exploratory analyses of trial-level data and are highly speculative.
3. Perioperative high FiO₂ (80%) may result in little to no difference in rates of 30-day mortality, PPCs, pneumonia, respiratory failure, unplanned ICU admission and length of hospital stay but may result in a large increase in the occurrence of atelectasis compared with low FiO₂ (30–35%).
4. HFNO and NIV may have little to no effect on mortality compared with COT when administered postoperatively.
5. Although pooled estimates indicate a potential effect of postoperative HFNO on reducing adverse respiratory events, the evidence is very uncertain as the findings were based on trials of relatively small sample sizes and outcome events, which showed a moderate level of heterogeneity.
6. Postoperative NIV may reduce PPCs and may slightly reduce or have little or no effects on other adverse respiratory events.

We included 59 reviews in this systematic overview, of which 5 anchoring reviews62,72,133,134,171 provided comprehensive and up-to-date evidence on the effectiveness of perioperative oxygen therapy were selected.

Our updated meta-analysis on the effectiveness of 80% versus 30–35% FiO₂ included 24 trials with 9725 participants. The pooled relative risk (0.91, 95% CI 0.78 to 1.05) suggests that a high FiO₂ during anaesthesia, surgery and recovery may result in a slight reduction in SSI, but this finding did not reach statistical significance at the conventional 5% level, and there is a moderate level of heterogeneity between included trials.

Our updated subgroup analyses highlighted a few potential effect modifiers, including methods of oxygen delivery (tracheal intubation vs. face masks or hoods), type of surgery (abdominal vs. caesarean section vs. other types of surgery), urgency of surgery (emergency vs. elective or mixed type) and risk of bias. While these analyses were prespecified, we consider them as exploratory because these analyses were based on trial-level data (rather than individual-level patient data) and therefore may be subject to ecological fallacy (associations observed at group level do not reflect actual associations at individual level). In addition, the number of subgrouping factors that we explored is relatively large in relation to the number of trials included in the analyses. More importantly, the potential effect modifiers identified above are inter-related, and are highly correlated in some cases (e.g. all trials in which patients received oxygen through face masks or hoods were trials for caesarean sections) and the findings of subgroup analyses were heavily confounded and required great caution in their interpretation. We planned multivariable meta-regression, but the number of available trials (24) means analyses would be underpowered to explore more than a couple of variables simultaneously. As a result, we were unable to provide any firm conclusion on effect modifiers based on current evidence; future research should seek to explain the heterogeneity observed among trials of perioperative oxygen therapy on SSI through within-trial comparison using individual-level patient data.

A rather unexpected finding from our subgroup analysis was that the reduction in SSI was more prominently observed among trials judged to be of low risk of bias (RR 0.80, 95% CI 0.66 to 0.96; I² = 35%) compared with the effect observed among trials of unclear/high risk of bias (RR 1.06, 95% CI 0.85 to 1.31). The veracity of this subgroup effect and the potential mechanism behind it may be worth further exploration, but we caution about over-interpreting this finding due to reasons mentioned above.

Trial sequential analysis using a random-effects model, showed only 71% of the required information size of 14,468 randomly assigned participants to detect or reject a 20% RRR with 90% power has been reached so far, and neither the conventional boundary (p = 0.2) nor the trial sequential monitoring boundaries for benefit, harm and futility have been crossed. In trials with low risk of bias, trial sequential analysis showed that only 47% of the required information size of 13,124 participants to detect or reject an a priori anticipated intervention effect of 20% RRR with 90% power has been reached. Although the cumulative Z-curve was found to cross the boundaries of conventional significance (p = 0.01), it failed to cross the trial sequential monitoring boundaries for benefit. The analysis indicated that approximately 7000 more patients will be needed to make a definitive conclusion on the benefits of supplemental oxygen. Although this is not a small number, such sample size would not be infeasible considering the large volume of surgery being carried out daily worldwide.

Our updated analyses suggest that a high FiO₂ during anaesthesia, surgery and recovery results in little to no difference in 30-day mortality (RR 1.21, 95% CI 0.80 to 1.85) and mortality within longest follow-up compared with low FiO₂ (RR 1.08, 95% CI 0.81 to 1.45). These findings are consistent with those documented recently.73,147,258 Interestingly, the largest trial in the meta-analysis reported that high FiO₂ results in a statistically significant increase in all-cause mortality in cancer patients within 1 year and within long-term follow-up with a median of 3.9 years.55 It is worth noting that the PROXI trial56 was originally powered to detect a reduction in SSI only, and therefore, these post hoc analyses should still be treated with some caution. It was postulated that high FiO₂ might have a negative effect on mortality possibly through formation of ROS259 and immunosuppression260 in cancer patients.

The trial sequential analysis showed that only 12% and 14% of the diversity-adjusted required information size has been accrued to detect or reject a RRR of 20% in a random-effects model for 30-day mortality and mortality up to longest follow-up, respectively. 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. Our GRADE certainty in evidence ratings were low, indicating that the ability to accurately determine the strength of the association between high FiO₂ and mortality may be limited. Further RCTs are still needed to determine whether high perioperative FiO₂ is associated with an increased risk of mortality, especially in patients undergoing cancer surgery.

The evidence suggests that high FiO₂ results in little to no difference in the incidence of pneumonia, respiratory failure, PPCs, unplanned ICU admission and length of hospital stay. High FiO₂, however, may result in a large increase in the incidence of atelectasis (RR 1.47, 95% CI 1.20 to 1.79; RD 6.5% higher, 2.8% higher to 10.9% higher, low-certainty evidence). Previous meta-analyses did not report any differences in the incidence of atelectasis between the high and low FiO₂ groups.71,73,258 Discrepancy between studies could be, at least in part, attributed to the inspired gas mixture used. N₂O promotes absorption atelectasis in the lung as effectively as breathing 100% oxygen.261 Therefore, its presence in the inspired gas mixture may have contributed to the increased incidence of atelectasis in the low FiO₂ group, resulting in a comparable incidence of atelectasis between the two groups.

Upon completion of our overview, there remains a lot of uncertainty about how much oxygen patients should receive while undergoing surgery. Given the knowledge of potentially harmful effects of high level of oxygen in cardiac and critical care settings,53,262 it is likely that the benefits/harms of oxygen do not increase/decrease in a linear fashion, and an optimal level of oxygen may need to be found and such a level may also need to be calibrated depending on the patient’s condition and type of surgery. An intraoperative FiO₂ of 50% currently represents standard intraoperative practice in the UK, with surgical patients often experiencing moderate levels of hyperoxaemia. This seems to be a rational practice considering the unknown trade-off between potential harms and benefits. Nevertheless, this differs from both WHO’s recommendation of using an FiO₂ of 80% intraoperatively, and the value most previous interventional oxygen therapy trials have used to represent standard care (typically 30% FiO₂). Future research should aim to explore whether giving less oxygen will generate less strain on parts of the body, particularly the lungs, and should identify the safest percentage of FiO₂. As of spring 2023, the Perioperative UtiLisation of SupplEmental Oxygen (PULSE Ox) UK feasibility trial (NCT03552627)263 is recruiting patients undergoing elective major surgery who will be randomised to receive 80% (WHO’s recommendation); 55% (current usual care) or 30% oxygen (intervention) throughout anaesthesia. The study aims to evaluate the feasibility of performing a large multicentre effectiveness study to define ‘safe’ oxygen administration concentrations during surgery.

Compared with COT, the evidence suggests that HFNO results in little to no difference in mortality (RR 0.78, 95% CI 0.27 to 2.25; low‐certainty evidence) reintubation (RR 0.78, 95% CI 0.29 to 2.07; low‐certainty evidence) and the length of hospital stay (MD 0.45 days lower, 95% CI 0.86 days lower to 0.03 days lower; low‐certainty evidence). HFNO may reduce the need to escalate respiratory support (RR 0.61, 95% CI 0.41 to 0.91) but the evidence is very uncertain given the heterogeneity observed among trials of relatively small sizes. Subgroup analysis by type of surgery, risk of postoperative respiratory complications and body mass index did not show credible subgroup effects. Since the included trials examined the effect of HFNO only in patients undergoing cardiothoracic and major abdominal surgery, the effect of postoperative use of HFNO in other surgical patient populations who may be at risk of respiratory failure (neurosurgery; vascular surgery) remains unknown and requires investigation.

Meta-analyses included in our anchoring review that examined the HFNO use in the postoperative period reported similar reductions in escalation of respiratory support and reintubation.94,95 However, since the publication of the most recent meta-analysis in 2020,62 three new RCTs180,181,183 reporting data on reintubation have been published and our updated meta-analysis suggests that HFNO results in little to no difference in reintubation. It is worth noting that all trials included in previous and our current meta-analyses were relatively small with few events (< 10) in individual treatment arms. The finding therefore remains uncertain. The trial sequential analysis for this outcome assuming a RRR of 20% with 90% power and an alpha error of 5% showed that the required information size (n = 21,142) was not met and, consequently, we downrated the certainty of evidence because of imprecision.

Compared with COT, postoperative NIV results in little to no difference in mortality (RR 0.91, 95% CI 0.62 to 1.33; high‐certainty evidence) and pulmonary aspiration (RR 1.11, 95% CI 0.54 to 2.27; high‐certainty evidence). NIV may decrease the incidence of PPCs (RR 0.62, 95% CI 0.44 to 0.87) based on evidence of low certainty. NIV may result in a slight reduction in the incidence of ARDS (RR 0.70, 95% CI 0.53 to 0.93; moderate-certainty evidence). NIV may also result in little to no difference in the incidence of pneumonia (RR 0.95, 95% CI 0.78 to 1.15; low-certainty evidence) and probably results in little to no difference in reintubation rate (RR 0.82, 95% CI 0.65 to 1.04; moderate-certainty evidence). The evidence suggests that NIV results in a slight reduction in unplanned ICU admissions (RR 0.73, 95% CI 0.50 to 1.06; low‐certainty evidence). Subgroup analysis based on the type of surgery did not show credible subgroup effects between abdominal and cardiothoracic surgery. The trial sequential analysis for all outcomes assuming a RRR of 20% with 90% power and an alpha error of 5% showed that the number of patients evaluated did not surpass the required information size.

In summary, our overview of systematic reviews has shown that high FiO₂ in the perioperative period may lead to a slight reduction in SSI but there may be important effect modifiers that remain to be confirmed. High FiO₂ is associated with increased atelectasis. Our top-level panoramic meta-analysis did not show a clear signal of benefit or harm in terms of mortality when a higher level of oxygen was given to patients perioperatively through different means, although the analysis was still very underpowered to detect an important difference. Postoperative HFNO and NIV may reduce some of the adverse respiratory events, but these findings were based on relatively small trials with some heterogeneity observed between their findings and therefore require further confirmation.

Strengths and limitations of this review

We performed a comprehensive systematic search of published systematic reviews and recent RCTs, conducted every step of the overview in duplicate, registered our overview methods prospectively and followed through our protocol.

The strength of our overview stems from it being the single, accessible and most comprehensive synthesis of evidence that combines all relevant reviews on the topic. It signposts clinicians and policy makers toward relevant systematic reviews to support clinical decision-making. There is a relatively large number of overlapping systematic reviews with small numbers of trials related to the effectiveness of perioperative oxygen strategies in improve patient outcomes following surgery. This can make interpretation of the evidence difficult. Although we did not select our anchoring reviews purely based on methodological quality, we took several steps to minimise the impact of the varied methodological quality of the reviews on treatment effect estimates. For example, we reviewed GRADE ratings for all comparisons or graded the certainty of evidence ourselves when a review did not assess the certainty of evidence. We provided GRADE evidence profiles and summary of findings tables for all comparisons. We phrased the findings and certainty of the evidence as suggested in the GRADE informative statements guidance.132 Our overview will play a key role in research prioritisation, promoting collaborative work toward shared priorities, ensuring effective use of resources and avoiding duplication.

There were several challenges associated with an overview of this nature. One challenge arose from the use of the ROBIS tool to assess risk of bias of included anchoring reviews. We found difficulty in evaluating and consequently in reaching agreements on the potential impact of sub-optimal search strategies on the risk of bias for a review in phase 2 of ROBIS. We also found that a final overall, potentially reductionist judgement on overall risk of bias of the review taking into account original authors’ interpretation of findings in phase 3 of ROBIS assessment less suitable for our purpose, as we updated meta-analyses from our anchoring reviews with new trial data and subsequently interpreted and graded the evidence by ourselves.

To overcome clinical uncertainty in the field, we have presented the most up-to-date, albeit conservative, interpretation of findings when compared with interpretations given in some other reviews and meta-analyses previously published.

We included additional RCTs and reanalysed the data of several reviews and conducted sensitivity and subgroup analyses. However, this does not detract from the fact that we did not have access to individual participant data from each study. Accordingly, we could not explore potential influence of patient characteristics such as age, underlying disease conditions, ethnicity, sex and smoking status on the effect of a high versus low FiO₂ used during anaesthesia and surgery. While access to individual-level patient data would have improved the precision of our findings and, importantly, would help to elucidate potential effect modifiers that contribute to heterogeneity present in different RCTs, it would require the investment of substantial time and effort, which is beyond the scope of the current overview. Nevertheless, we hope that our overview will provide some pointers of focus for such endeavour in the future.

The assessment of publication bias was challenging given that most comparisons had a limited number of trials. For example, only 13 of the 23 comparisons in our review had 10 or more trials, which is the minimum number of studies recommended when assessing publication bias with a funnel plot. The GRADE working group recognises the challenge of assessing publication bias.129 Most of our selected anchoring reviews searched clinical trial registries and/or Cochrane CENTRAL database, which covers records from trial registries, and therefore conferred some protection against publication bias. Nevertheless, the potential impact of publication and outcome reporting bias cannot be ruled out. This is particularly relevant for evidence related to HFNO and NIV, which are mainly based on data from relatively small trials that are more susceptible to these biases.

We conducted trial sequential analysis for all outcomes to control random errors due to repeated significance testing when meta-analyses are updated and to assess whether further trials need to be conducted. While trial sequential analysis has these advantages and is increasingly being adopted as part of evidence synthesis and knowledge translation process such as GRADE, there are also certain weaknesses to consider. It has been criticised for transferring a method for interim analyses in RCTs to a setting where it cannot impact decisions for already finished trials and, possibly, will also not impact decisions for ongoing trials or decisions to initiate trials. Trial sequential analysis has been faulted for being too conservative, since sceptical a priori assumed intervention effects can cause the required information size to become unreasonably large.122 This became apparent in some of the analyses that we conducted: where the baseline (control group) risk of an event of interest is low, a prespecified intervention effect in relative terms (e.g. RRR of 20%) could result in the calculation of an unrealistically large optimal information size that is aiming to detect a trivial intervention effect in absolute terms. These and other points of critique have been discussed and addressed by Wetterslev et al.122

The conclusion of our overview is necessarily restricted by the quantity, quality and nature of the underlying evidence base. We noted that existing studies have tended to focus on abdominal (including colorectal) and cardiothoracic surgeries and caesarean sections. Surgeries such as joint replacement, which represented a large share of common surgeries, seem to be underrepresented. The incidence of SSI for these surgeries is well under 1% in the UK264 and therefore a sufficiently powered trial would require a very large sample size and prohibitive from the costs point of view. Nevertheless, given the large volume of these surgeries and hence their contribution towards a significant number of SSIs in absolute terms, pragmatic trials adopting novel data collection methods or routine data that allow large sample sizes may be worth exploring. The same approach may also help answer the question on the effect of perioperative oxygen therapy on mortality, which is another important outcome with great uncertainty due to the low incidence of the event.

Reducing SSI has been the target of numerous clinical and quality improvement interventions, and as a result, its incidence has been decreasing over time.264 This raises a question of applicability of evidence for some of the trials conducted more than a decade ago. Nevertheless, we did not notice any clear trend of the effect of perioperative oxygen therapy having changed over time. The incidence of SSI varies substantially between different surgical procedures and between settings. Current evidence is dominated by studies conducted in developed countries; further research is needed, and insight may be gained, from studies carried out in developing countries where the incidence of SSI is much higher and represent a much more serious problem.

All identified systematic reviews focused on adult population; there is a lack of synthesis for evidence related to surgery carried out in children. There is a lack of evidence on the use of perioperative oxygen with different anaesthetic techniques. There is also no evidence on the use of HFNO and NIV intraoperatively for patients who do not require general anaesthesia.

Although we have updated the selected anchoring reviews where new trial evidence meeting their inclusion criteria was found, we have not been able to incorporate data from isolated trials of perioperative oxygen therapy in areas that have not been covered by existing systematic reviews due to time constraints. This limitation is highlighted in our recommendations for future research. Finally, perioperative practice continues to evolve with new anaesthetic techniques and minimally invasive surgeries being introduced in recent years. Their potential interaction with and impact on perioperative oxygen therapy need to be evaluated with further evidence.

Implications for practice

We could not exclude benefits or harms from use of a high FiO₂ compared with a standard approach. Our results suggest that evidence is insufficient to support the routine use of a high FiO₂ during and after surgery beyond what is needed to maintain normal arterial oxygen saturation. Available randomised data do not suggest that either strategy is to be preferred. Further targeted research may provide definitive guidance in this area.

Therefore, with no certain evidence, clinicians should judiciously administer oxygen to patients in their care and tailor oxygen administration to the individual patient’s risk profile and the type of surgery being performed. Concerns over the risk of hyperoxaemia and cancer recurrence after cancer surgery should be taken into account when titrating oxygen in the perioperative period. Clinicians should continue to use clinical signs, routine monitoring and near patient testing such as arterial blood gas sampling to adjust oxygen therapy in the intra- and postoperative period.

Similarly, current evidence does not support the routine postoperative use of NIV or HFNO to improve patient outcomes after surgery in adults. These interventions are resource intensive and require trained staff for their delivery. Feasibility and cost-effectiveness are therefore additional factors that need to be considered. Our evidence synthesis has shown a trend towards reduced need for reintubation/escalation of respiratory support when NIV/HFNO have been used but there is uncertainty in the estimate. Our analysis did not show any evidence of harm from these interventions. It is possible that these interventions would be beneficial in some groups of patients (high body mass index, suffering from obstructive sleep apnoea) and clinicians could consider using it in limited settings with the awareness of the limitations in the current evidence.

Implications for research

Further research is still required to establish high‐certainty evidence related to perioperative oxygen strategies to prevent postoperative complications in patients undergoing surgery. Care should be taken to ensure that both reviews and RCTs are of the highest possible methodological quality, with comprehensive reporting.

Future RCTs ought to be conducted with long-term follow-up and ought to stratify participants by type of surgery, anaesthesia technique (e.g. inhalation vs. total intravenous anaesthesia), surgical subspecialty (e.g. upper vs. lower gastrointestinal, open operations vs. minimally invasive) and documented risk factors for mortality and SSI such as the American Society of Anesthesiology (ASA) score, duration of surgery, body mass index (high vs. low body mass index) and comorbidity such as diabetes. Future trials should make available individual participant data, so that individual participant data meta-analyses of the effects of perioperative oxygen adjusted for relevant confounders can be conducted.

To avoid research waste and ensure that evidence relating to the effectiveness of perioperative oxygen strategies is easily accessible to end users, any new and updated reviews should be carefully planned. New reviews should be planned to synthesise evidence relating to interventions that have been identified as clinically relevant and important by key stakeholders. Careful consideration should be given to issues of lumping and splitting, ensuring that reviews address questions that are clinically useful and that the same evidence is not synthesised within multiple overlapping reviews. Arising from, but not limited to, the results of this overview, we believe that the following research is justified and important:

1. Additional large-scale, high‐quality RCTs are needed on the effects of different amounts of oxygen on mortality, SSI and PPCs either in a broad range of surgical patients allowing sufficient power to detect important potential effect modifiers such as type of surgery or methods of oxygen delivery within the trial or in a focused high-risk population, to provide a definitive answer through achievement of required information size.
2. A large-scale RCT of postoperative HFNO to confirm or refute its potential effect on reducing adverse respiratory events.
3. High‐quality head-to-head RCTs to establish effectiveness of HFNO use in the postoperative period compared with NIV alone or in combination with NIV. A targeted approach investigating the use of HFNO versus NIV in patients at higher risk for postoperative respiratory complications would be of value.
4. Living systematic reviews/overviews or prospective meta-analyses that allow continuous update of relevant evidence, which replaces the needs for multiple systematic reviews with overlapping efforts and potentially inconsistent conclusions.
5. High-quality systematic review to establish the benefits of perioperative oxygen therapy in paediatric populations, because there is an absence of synthesised evidence.

Equality, diversity and inclusion

We set out to explore whether oxygenation strategy in the perioperative period can benefit different groups of patients in different surgical specialties and settings (emergency and elective surgery). We encountered several equality, diversity and inclusion issues that should be considered:

• We have presented our findings in different surgical specialties and settings, but we found very little relevant evidence on how different strategies may affect patients with majority risk factors set out in our original proposal, including adults versus children versus neonates; body mass index (high vs. low)/obese versus non-obese; comorbidity: diabetes versus no diabetes and by ASA physical status classification. Future research should consider stratifying patient risk factors in developing complications and the impact of oxygenation strategy.
• Oxygen administration is often guided by pulse oximetry in the clinical setting, which we know is less accurate in darker skin pigmentation.265,266 It is possible that current evidence included in our overview did not consider these potential differences.

While medical oxygen is widely available in high-income countries, the availability and associated costs of oxygen in resource-limited settings will have impact on its use during perioperative period. Such inequity was highlighted during COVID-19 pandemic.267 It is possible that current evidence included in our overview did not consider these potential differences.

Patient and public involvement

The aim of our advisory panel was to provide oversight into the outcomes, interpretation of results and highlight any areas that require more research. Our panel consisting of three patients and a multidisciplinary group (doctors, nurses, advanced practitioners, physiotherapy) with expertise across the clinical areas of anaesthesia, respiratory medicine, critical care and surgery. A patient representative (MT) had lived experience of supplementary oxygen during their major surgery. They worked closely with the research team throughout the project, including drafting of the study proposal, review protocol and interpretation of results.

The members were kept informed with the progress throughout the project via project updates and email. The concept of overview of systematic reviews was difficult to explain to lay people and non-expert clinicians. The members also met virtually with our study team three times during the project, when they were provided with background information on methodology and could also pose any questions to the team. The input from our advisory panel was instrumental in the review of our mapping of studies, synthesis strategy and interpretation of evidence and helped to guide the writing of clinically relevant recommendations and review conclusions.