Review methods

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.

We followed the Preferred Reporting Items for Overviews of Reviews (PRIOR).97 The full protocol for this review has been previously published1 and registered with PROSPERO (CRD42021272361).

Methodological approaches to this evidence synthesis

We opted to undertake an overview of systematic reviews because of the abundance of systematic reviews that already exist on the effectiveness of perioperative oxygen therapy (see Existing literature). Overviews of systematic reviews have evolved to address a growing need to filter the information overload, improve access to targeted information and inform clinical decision-making in a more timely fashion.98 Overviews can provide an effective entry point for clinicians, policy makers and guideline developers. They summarise broad issues and current knowledge where evidence relating to a specific topic exists but is conflicting.99

In designing and conducting this overview of systematic reviews, we have paid attention to a number of methodological issues and incorporated some design features to address them:

1. To tackle the (anticipated) large number of existing systematic reviews and address substantial overlap in evidence between them, we selected a few most recent and comprehensive systematic reviews as ‘anchoring reviews’ (see Chapter 3, Study mapping) for our overview. We have chosen this term given the connotations of a stable, well-situated base associated with anchoring. These anchoring reviews were the well-characterised, starting points in the evidence space from which we could add data from more recent RCTs to provide up-to-date, best available evidence in particular subject areas.
2. In addition to providing a summary of best available evidence on perioperative oxygen therapy for different perioperative stages and using different oxygen delivery devices to directly answers clinical questions, we also planned to explore the scientific hypothesis that delivery of higher levels of oxygen perioperatively reduces SSI by undertaking ‘panoramic meta-analyses’. The term ‘panoramic meta-analysis’ was originally used to describe a meta-analysis that pools effect estimates of an intervention across different disease types or clinical scenarios.100,101 For this overview, a panoramic meta-analysis would include data comparing different oxygenation strategies and/or different oxygen delivery devices across different types of surgery.

The rationale behind a panoramic meta-analysis is that while the effect of an intervention may vary by magnitude in different clinical scenarios (e.g. for different types of surgery), which are usually examined and estimated in separate systematic reviews, a generic hypothesis concerning the effect of an intervention underpinned by the same mechanism (e.g. supplying higher levels of oxygen around the time of surgery) on an outcome (SSI) can be tested by collating all relevant pieces of evidence and examining them under one meta-analysis.102 Panoramic meta-analyses may allow a general conclusion to be drawn when evidence for an intervention varies in volume (and hence is underpowered for some clinical scenarios, e.g. for certain diseases or surgery types) but the observed effects are broadly consistent across different clinical scenarios. On the other hand, panoramic meta-analyses may flag up inconsistency in the evidence across different clinical scenarios and, thus, challenge the validity of the hypothesis being tested. Either way, a panoramic view of all relevant evidence may be obtained and some additional insights may be gained by quantitatively examining a broad set of evidence guided by a scientific hypothesis. This approach fits the research question to be answered by this overview very well, and hence was adopted here. Further details on the panoramic meta-analyses that we planned and carried out are described in Chapter 3, Data synthesis.

1. We followed the widely adopted GRADE methodology to grade the overall quality/certainty of evidence. The methodology is continuously being developed and refined, including how the certainty domain of imprecision should be rated and how different levels of certainty should be worded in the textual description of findings. To facilitate a shared understanding of the findings presented in this report, we wish to highlight that the rating of imprecision accordingly to the latest GRADE guidance is determined by the CI of an effect estimate relative to the thresholds for health outcomes that correspond to trivial or none, small, moderate or large effects. For clinical events, the thresholds and CIs are expressed as absolute risks and therefore the baseline risk (control group event rate) of a clinical outcome has a major impact on the rating of imprecision. As a result, the rating for imprecision may be different for the same level of uncertainty when expressed using a relative measure, for example relative risk. For example, for an estimated relative risk of death of 1.2 (95% CI 0.8 to 1.7) for an intervention compared with a comparator, the uncertainty of the intervention effect based on the 95% CI would translate into ‘20 fewer to 70 more deaths per 1000 people treated’ when the baseline/comparator mortality rate is 10%; whereas the same 95% CI would translate into ‘2 fewer to 7 more deaths per 1000 people treated’ when the baseline/comparator mortality rate is 1%. The level of imprecision (and hence uncertainty of evidence) may be perceived as very high in the former case but may be considered low in the latter case. Given the significant impact of SSI on patients, we have adopted a threshold of 1%, 2.5% and 5% difference in absolute risk as small, moderate and large difference, respectively.

Study selection criteria

Details of the inclusion criteria are highlighted in Table 1.

Information sources and search strategy

A sensitive search was developed iteratively by our information specialist (RC), with feedback from the review team and with reference to previously published reviews,65,70,108 and recommendations for conducting overviews of systematic reviews.109 Index terms and text words relating to oxygen therapies, surgery and systematic reviews were used to search the following key electronic databases: MEDLINE (via Ovid), EMBASE (Ovid), the Cochrane Database of Systematic Reviews (Wiley), Epistemonikos,110 PROSPERO,111 the International Network of Agencies for Health Technology Assessment International Health Technology Assessment Database, and the Database of Abstracts of Reviews of Effects archives (via CRD, Centre for Reviews and Dissemination) and ECRI Guidelines Trust database. These databases were searched from inception to September 2021. Searches for reviews were not limited by date or publishing language. Non-English language articles were translated into English. The final search strategy is included in Appendix 1, Table 13. We also checked the reference lists of relevant articles and used citation search facilities provided by the Web of Knowledge to check records of papers citing our anchoring reviews. In addition, targeted searches of the Internet using the Google.co.uk search engine were used to identify unpublished materials and work in progress.

To ensure that emerging trial evidence was covered, we also searched for recently published or ongoing/planned RCTs in the Cochrane Central Register of Controlled Trials (CENTRAL) database, which includes records from major clinical trial registries. This search was undertaken in March 2022 and limited to records added in 2018 or later (see Appendix 1, Table 14). This date was determined by when the searches were performed in recent, relevant systematic reviews identified in the first set of searches. Searches for both systematic reviews and RCTs were updated in June 2022 to enable the team to maintain awareness of recently published literature to inform the discussion section.

Study selection and mapping

Initial review selection

Titles and abstracts of records retrieved were screened by two reviewers independently (AE, AW, AH, ST), disagreement was resolved by discussion or if needed with the input of a third senior reviewer (JY, KC, YC, AG). Full-text articles considered potentially meeting inclusion criteria were assessed for inclusion by two reviewers independently and disagreements resolved as above. Figure 1 illustrates the overview schema. Records retrieved from the Cochrane CENTRAL database and clinical trial registries went through the same screening and selection process as above. We used EPPI-Reviewer 4 (EPPI Centre, University College London, London, UK) software to manage records and data throughout the review.112

FIGURE 1 Overview schema. 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 figure includes minor additions and formatting changes to the original text. (Figure not reproduced here; see the original article.)

Study mapping

Following confirmation of eligibility, each review was mapped according to types of surgery, stages of perioperative care and comparisons made (e.g. between different oxygenation strategies or delivery device). We also mapped all RCTs included in the systematic reviews to gauge the extent of overlap between reviews and to ensure that no double counting of evidence occurred when undertaking further quantitative analyses. Where reviews covering largely overlapping topic areas and RCTs are identified, a single review with the most comprehensive coverage of literature and/or the highest methodological quality, as judged by the Risk of Bias in Systematic Reviews (ROBIS) tool,113 was selected (through deliberation within the review team) as the anchoring review. Where reviews overlap only partially in their scope or evidence coverage, multiple reviews were retained as anchoring reviews. RCTs published since the anchoring systematic review searches that were judged to be eligible were added to the RCT map, which is a list of all eligible RCTs identified either in relevant systematic reviews that we found or through the additional searches of recent RCTs that we conducted.

Then chosen anchoring reviews were checked against the completed RCT list to see if any important RCTs not included in the anchoring reviews warrant to be added, thus updating relevant analyses in the anchoring review (where this was considered necessary and feasible). We were aware of previous RCTs having been retracted48,114 and have taken these into account when updating the meta-analyses of the anchoring reviews by excluding trials published by the research team.

Data extraction

Data were extracted by two independent reviewers (AE, AW, AH, ST), from each chosen anchoring systematic review. This included:

1. review characteristics: year published, country, number of RCTs included, number of patients, summary of intervention and comparator
2. patient population: type of surgery, reported patient characteristics
3. all clinical outcomes and adverse events reported
4. reported pooled results for the primary and secondary outcomes listed above, results of subgroup and sensitivity analysis
5. certainty of evidence
6. risk of bias assessments and publication bias.

In the case of missing data or data queries, authors of the original review and RCT were contacted for clarification. A reminder email was sent 2 weeks after the initial contact if no member of the team responded.

Risk of bias assessment

Risk of bias for each of the systematic reviews was assessed using the ROBIS tool.113 ROBIS is completed in three phases, with the first assessing relevance, identifying concerns with the review process and judging the risk of bias in the review. The second phase assesses four domains to help assess specific concerns about potential biases within the review. These are described below.

1. Study eligibility criteria (i.e. review eligibility criteria were clear, appropriate and prespecified).
2. Identification and selection of studies (i.e. all relevant primary studies should have been identified and included in the review).
3. Data collection and study appraisal (i.e. judgement relating to bias that may have been introduced through the data collection and assessment of risk of bias of included studies).
4. Synthesis and findings (i.e. appropriate methods have been used for any meta-analyses, and syntheses of results).

Assessment of each domain involved firstly answering some signalling questions, followed by making a judgement on the level of concerns about risk of bias for that domain based on reviewer’s response to the signalling questions. We used the rating guidance published with the ROBIS tool in answering all signalling questions.

The final phase would have involved a judgement of the overall risk of bias of each review (which could be low, high or unclear), following assessment of three further signalling questions related to how the original review authors presented and interpreted their findings. We did not proceed to phase 3 of the assessment as we updated the meta-analyses of the anchoring reviews and made our own interpretation of the updated findings. We therefore focused on and presented findings related to concerns identified in phase 2 of the assessment.

New RCTs used to update the anchoring reviews were assessed using the appropriate Cochrane RoB2 tool for each relevant outcome.115 Where an anchoring review used different methodological approaches to assess the risk of bias that could impact on the comparability of findings between the reviews, the risk of bias for included RCTs was reassessed for each relevant outcome using the appropriate RoB2 tool. All assessments were conducted by two reviewers independently (ST, AE) with conflicts resolved through discussion or consultation with a third reviewer (YC). Risk of bias assessments are presented in Methodological quality of included reviews and Risk of bias in newly identified RCTs.

Data synthesis

The data synthesis served four purposes for this overview, each with a corresponding set of analyses:

1. panoramic meta-analyses for individual comparisons to explore the effectiveness of perioperative oxygen therapy in different patient populations and settings
2. top-level panoramic meta-analysis to test the scientific hypothesis that perioperative oxygen therapy reduces the risk of SSI100
3. top-level panoramic meta-analysis to evaluate the overall benefit/risk of perioperative oxygen therapy in terms of mortality across different clinical conditions and settings
4. meta-regression to explore potential trial-level effect modifiers for SSI among perioperative oxygen therapy trials.

Analyses were performed in Stata 17 (StataCorp LP, College Station, TX, USA).116 Each of these sets of analysis are described in further details below.

Panoramic meta-analysis for individual comparisons

Panoramic meta-analyses were planned for individual comparisons (e.g. high vs. low FiO₂ strategy intraoperatively; HFNO vs. COT postoperatively). Each analysis was planned to be stratified by one subgroup feature (e.g. type of surgery). We planned to use pooled estimates from anchoring reviews for the panoramic meta-analyses where possible. However, we carried out the panoramic meta-analysis using trial-level data instead, for two reasons: (1) we updated the meta-analyses for most of the anchoring reviews with data from new RCTs and (2) the scopes of the anchoring reviews and subgrouping of trials within these reviews did not align well with our intended subgrouping for the panoramic meta-analyses. RRs with 95% CIs were reported and presented in forest plots. The I² statistic and chi-squared test were used for assessing heterogeneity. We also assessed heterogeneity using the H² statistic.117 A value of H² = 1 indicates perfect homogeneity among the studies and values larger than one indicate higher level of heterogeneity.

The following subgroups were selected by the project team in consultation with the advisory panel, before the panoramic meta-analyses began:

• type of surgery: abdominal surgery versus caesarean section versus other surgery
• oxygen delivery through tracheal intubation versus face masks or nasal cannula
• use of N₂O as carrier gas for anaesthesia in the control group versus no use of N₂O
• trials with low risk of bias versus trials with high or unclear risk of bias
• urgency of the surgery (elective vs. emergency vs. mixed).

We accounted for data availability, and theoretical plausibility of the subgroup feature acting as a potential effect modifier when selecting these subgroups.

For all panoramic meta-analyses, we planned to use data judged to be of low risk of bias in the main analysis and included additional data judged to be of various levels of risk of bias to be included in sensitivity analyses. However, it became apparent during our data extraction that none of the published systematic reviews limited their primary analysis to only low risk of bias trials. Therefore, if we had restricted it to low risk of bias trials at least half of available trials included in our chosen anchoring reviews would have been excluded from our main analysis. Consequently, we retained all eligible trials irrespective of their risk of bias rating in our main analyses and included risk of bias rating as one of the subgrouping factors to be explored. A random-effects model was used to account for heterogeneity that cannot readily be explained.118 It involves an assumption that the effects being estimated in the different studies are not identical but follow some distribution. The ratio of between-study variance over total variance across studies within individual subgroups, and the ratio of between-subgroup variance over total variance across subgroups were estimated using the I² statistic.119

The panoramic meta-analyses were planned to be exploratory in nature. Where appropriate, we examined the adequacy and conclusiveness of the available evidence with the aid of trial sequential analyses.120,121

Top-level panoramic meta-analyses for surgical site infection and 30-day all-cause mortality

Two top-level, exploratory panoramic meta-analyses were planned. These top-level panoramic meta-analyses further aggregate the data from individual panoramic meta-analysis across oxygenation strategies and surgical care stages, with pooled data from individual panoramic meta-analysis as the unit of analysis.

The first top-level panoramic meta-analysis would explore the effect of perioperative oxygen therapy on SSI. We anticipated both clinical and statistical heterogeneity among the diverse evidence. We did not aim to generate pooled effect estimate as this would be difficult to interpret. Instead, the purpose was to enable a quantitative inspection of estimated treatment effects and level of heterogeneity across different stages of surgical care and oxygenation strategies. We wanted to explore the compatibility of existing evidence against the hypothesis that perioperative exposure to higher levels of oxygen reduces SSI.

A similar top-level panoramic meta-analysis was planned to evaluate the overall benefit/risk of perioperative oxygen therapy on postoperative 30-day all-cause mortality. This analysis mirrors the procedure and analysis described above (further aggregation of data from individual panoramic meta-analysis).

We aimed to proceed with these top-level panoramic meta-analyses only if the levels of heterogeneity within and between individual panoramic meta-analyses were acceptable (i.e. not clearly showing opposite effects). Analyses were planned to be performed using a random-effects model with heterogeneity between individual panoramic meta-analyses quantified using I² statistic.

Meta-regression

The panoramic meta-analyses only allowed a single subgrouping factor to be explored at a time. To further explore potential effect modifiers simultaneously, a meta-regression analysis of SSI data was planned and carried out.

Variables included in meta-regression were chosen from important subgrouping variables explored in individual panoramic meta-analyses above. The list of variables was determined a priori before the meta-regression was carried out to ensure that the analysis is theory rather than data-driven. As the number of studies reporting mortality outcome and the number of death events was expected to be substantially smaller than those of SSI, no meta-regression analysis was planned for this outcome as available data would be unlikely to provide sufficient statistical power.

Trial sequential analysis

Trial sequential analysis was performed to address the risk of random error associated with sparse data and/or repeated significance testing which can affect cumulative meta-analysis.122,123 We used the Trial Sequential Analysis software 0.9.5.10 (Copenhagen Trial Unit, Centre for Clinical Intervention Research, Copenhagen, Denmark).124 To control the risk of type 1 error, the thresholds for the z-values were adjusted using two-sided 5% symmetrical O’Brien–Fleming α-spending function, allowing the type 1 error risk to be restored to the desired maximum risk. Crossing the O’Brien–Fleming α-spending boundaries by a Z-curve indicates statistical significance. The risk of type 2 error was controlled using the β-spending function and futility boundaries. Adjusted thresholds for non-superiority and non-inferiority (or no difference) were constructed. Together, adjusted non-superiority and non-inferiority boundaries make up what is referred to as futility boundaries. When the Z-curve crosses the futility boundaries, this indicates that the two interventions do not differ more than the anticipated intervention effect.

The required information size was estimated assuming a relative risk reduction (RRR) of 20% with a 90% power and an alpha error of 5%. RRR was calculated by the estimate using the event rates of both arms with model variance-based heterogeneity correction.

Publication bias/issue of selective reporting

We explored relevant systematic reviews which have been registered but not subsequently published by contacting authors to clarify reasons for non-publication.

For anchoring reviews that needed to be updated, we drew funnel plots to examine the possibility of publication bias/small-study effects for outcomes with at least 10 included comparisons. Findings from this assessment were taken into account when assessing GRADE certainty of evidence (as described below).

Assessment of the certainty in evidence

The GRADE approach was used to assess the certainty of evidence for each outcome within each (updated) anchoring reviews.40 GRADE uses five considerations (study limitations, consistency of effect, imprecision, indirectness and publication bias) to assess the certainty of the body of evidence for each outcome. The evidence can be downgraded from ‘high certainty’ by one, two or three levels for serious, very serious or extremely serious concerns, depending on assessments for risk of bias,125 indirectness of evidence,126 inconsistency,127 imprecision of effect estimates128 or potential publication bias.129

Grading of recommendations assessment, development and evaluation now suggests aligning approach that relies on thresholds and CIs of the absolute effect as a primary criterion for imprecision rating. In the partially contextualised approach128 one would judge whether the effect of an intervention on a specific outcome (expressed in absolute terms) falls in a category of magnitude of effect (i.e. trivial or none, small, moderate or large effects).130 When rating the certainty of evidence under this approach, the ratings represent the certainty that the true effect lies within the thresholds of one of these four categories of magnitude of effect or beyond the threshold for large effects. The effects are a result of combining absolute estimates and the importance (value) of these outcomes. Specifically, when assessing imprecision, the extent of rating down depends on the number of thresholds being crossed by the CI. Conceptually, we rated down by one level when the CI for the absolute effect crossed one of the thresholds across the range of categories of magnitude of effect. We rated down by two or three levels when the CI for the absolute effect crosses two or three of the thresholds across the range of categories of magnitude of effect, respectively. For example, the RR for PPCs was 1.06 (95% CI 0.77 to 1.46) and the corresponding absolute risk difference (RD) of 11 more per 1000 (95% CI from 41 fewer to 82 more). The CI for the absolute effect included the possibility of a moderate beneficial effect (25/1000) and large harm, thereby crossing more than three thresholds. The certainty of evidence was downgraded by three levels for imprecision accordingly.

Grading of recommendations assessment, development and evaluation’s judgements were taken directly from the reviews if already reported in the anchoring reviews for outcomes which did not require updating; otherwise, the GRADE assessment for the updated anchoring reviews was undertaken by two researchers independently. We used GRADEpro GDT (Guideline Development Tool) software (McMaster University, Hamilton, ON, Canada) to create a ‘summary of findings’ table.131 We phrased the findings and certainty of the evidence as suggested in the informative statement guidance.132

Grading of recommendations assessment, development and evaluation informative statements to communicate the findings of the overview

We followed GRADE’s informative statements to communicate results of the review.132 Two aspects of a result were used to communicate the review’s findings: the size or magnitude of the effect and the level of certainty of the evidence. Two factors are crucially important when determining the size of effect. The first involves calculating and using absolute effects instead of using relative effects that can often be misleading. For instance, consider a RR of 0.82, or 18% relative reduction in SSI. If on the one hand, the baseline risk of SSI is 40/1000, the RR 0.82 would translate into 7 fewer infections per 1000 patients in absolute terms, which might be considered a trivial effect. If on the other hand, the baseline risk is much higher, at 400/1000, the same RR would translate into 72 fewer infections per 1000 patients, which would be considered a large effect. The second is identifying threshold values in absolute terms which would define the clinical importance of the outcome. We have defined small, moderate and large absolute RD to be 1% (10/1000), 2.5% (25/1000) and 5% (50/1000), respectively, in consultation with our advisory panel.

The statements communicate the size of the effect based on the point estimate in a meta-analysis, as it is the most likely value of true effect given available data. CIs are taken into account when determining the certainty of evidence. CIs are calculated based on factors such as sample sizes and variance between or within studies. Whether the estimate is sufficiently precise is determined by comparing the CIs again the thresholds for large, moderate and small effects. Wider CIs spanning cross one or more of these thresholds indicate that the estimate may not be sufficiently precise and would result in downgrading of the certainty of evidence. In addition to imprecision, GRADE assessment factors in the risk of bias in the included studies; inconsistency and indirectness of evidence, and the risk of publication bias. The final list of informative statements as recommended by GRADE132 is shown in Table 2.

We recognise that readers and clinical communities may be looking for more definitive statements and may feel some of the wording recommended by GRADE and used by us in this report is ambiguous. However, it is important to emphasise that the GRADE informative statements deliberately use the recommended wording to convey different levels of (un)certainty related to the evidence in a consistent way, which was also our motivation to adopt these informative statements.

Presentation of findings

A characteristics of anchoring reviews table was presented for all selected anchoring reviews. Information presented covered patient population characteristics, intervention including timing, comparator, type of surgery, number of studies, number of patients and outcomes reported. For clinical outcomes and adverse events, the overall pooled effect size and CIs and results of subgroup and sensitivity analyses (where available) were presented. A list of outcomes reported other than those prespecified in the protocol was made; data were also extracted for any extra outcomes that were not prespecified but deemed to be relevant by the advisory panel (postoperative nausea and vomiting).

Grading of recommendations assessment, development and evaluation evidence profiles (see Tables 7, 11 and 12) the certainty in the body of evidence for each outcome. Forest plots of effect sizes for comparisons of effectiveness of oxygen therapy stratified by different subgroups were produced as part of the updated meta-analyses. We created a combined forest plot (see Figure 39, Chapter 5) of the identified perioperative oxygen strategies to prevent postoperative complications. The strategies versus outcome relations derived from the updated meta-analyses are mapped out alongside each other. The plot shows the potential impact that each strategy might have on several outcomes. The plot further indicates the grading of the evidence of each key outcome. The summary tables and combined forest plot provided an overview of the evidence and allow identification of evidence gaps to highlight priority for future research.

Advisory panel and patient and public involvement

Our advisory panel represented key stakeholders involved in the use of perioperative oxygen, including patients and a multidisciplinary group of clinical specialists with expertise in anaesthesia, critical care, surgery and physiotherapy. Panel members reviewed mapping of studies, synthesis strategy and interpretation of evidence to guide the production of clinically relevant recommendations and conclusions. In addition, a patient representative (MT) worked closely with the research team throughout the project. We worked with advisory panel members on three occasions. Each meeting included pre-meeting background information, presentation from the study team and minutes were circulated to ensure comments were fully captured from members. Group members were not involved in data extraction or quality assessment of the reviews. The advisory panel provided advice on selection of outcomes, study protocol and interpretation of the findings, but it was the study team who finalised the review and conclusions.

Advisory Panel first meeting: January 2021

During our first meeting, we introduced background and what we currently know about perioperative oxygen use. We discussed our research questions on FiO₂ and oxygen delivery devices. With the help of advisory group members, we were able to finalise inclusion, exclusion criteria, intervention, comparison and outcome measures. Postoperative nausea and vomiting was considered important by advisory panel and included in our review.

Advisory Panel second meeting: July 2022

We presented our findings to advisory panel members. We also undertook the following discussions:

• What is the most sensible to categorise different types of surgery? We achieved consensus on timing of surgery: elective, emergency surgery; surgical specialties: intra-abdominal surgery, cancer surgery.
• What types of surgery might have been missed? There were no specific types of surgery identified.
• Proposed subgroup analyses was reviewed. What are other subgroups that maybe worth exploring?

Patient-related factors

• Adults versus children versus neonates.
• Body mass index (high vs. low)/obese versus non-obese.
• Comorbidity: diabetes versus no diabetes.
• Cancer versus non-cancer.
• Preventive (preventing complications in ‘normal patients’) versus therapeutic use (treating patients who already developed hypoxaemia).
• Elective versus emergency surgical patients.
• By American Society of Anesthesiologists Physical Status Classification.

Perioperative care-related factors

• Delivery of oxygen: intubation versus face mask.
• Use of N₂O versus no use.
• Use of prophylactic antibiotics versus not.
• General anaesthesia versus regional anaesthesia versus sedation.

Study related factor

• Low risk of bias versus unclear/high risk of bias.
• Advisory panel would like the team to undertake subgroup analyses according to patient-related factors if possible, especially children and at risk patients for example obese patients.
• We also explored what clinically important difference would be considered worthwhile. Advisory panel members agreed that this may depend on the risk of each complication and should be explored accordingly.

Advisory Panel third meeting: February 2023

Preliminary findings of our overview were presented to the group. Advisory panel members agreed with the findings of mapping and clinical effectiveness evidence synthesis. Recommendations included highlighting research gaps and that future research should be more focused on key patient outcomes.