Results: clinical effectiveness

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.

Panoramic meta-analyses

This section presents the findings from each of the three comparisons (high vs. low FiO₂, HFNO vs. COT and NIV vs. COT) for which we undertook panoramic meta-analyses. As described in the methods section, for each comparison we used trial-level data obtained from the relevant anchoring review, added data from new trials that we identified and then carried out a panoramic meta-analysis showing findings with and without additional data from the new trials. We then conducted subgroup analyses using all available trial data, stratified by subgrouping factors that were considered as potentially important effect modifiers. Table 5 provides a summary of panoramic meta-analyses that were conducted. Further descriptions of results for each comparison follow.

Oxygenation strategy: what is the clinical effectiveness of higher versus lower fraction of inspired oxygen or targets of arterial oxygenation?

Three anchoring reviews assessed the effectiveness of high (80%) versus low FiO₂ (30–35%) in patients undergoing surgery.72,133,134 de Jonge et al. performed a systematic literature search from January 1990 to April 2018 for RCTs comparing the effect of 80% versus 30–35% FiO₂ on the incidence of SSI.72 Lim et al. conducted a systematic review and meta-analysis to determine whether high FiO₂ compared with low FiO₂ has more deleterious postoperative clinical outcomes in patients undergoing non-thoracic surgery under general anaesthesia.133 The review reported on several outcomes including mortality within 30 days, atelectasis, pneumonia, respiratory failure, PPCs, ICU admissions and length of hospital stay. Markwei et al. assessed the effect of 80% versus 30% FiO₂ on the incidence of postoperative nausea and/or vomiting.134

Effectiveness of 80% versus 30–35% fraction of inspired oxygen on the incidence of surgical site infection

Our updated meta-analysis included 24 trials with 9725 participants. Seven of these were newly identified since the publication of the review in 2019.72 Sample sizes ranged from 38 to 2012 patients with a wide range of international locations. Age groups were wide-ranging between trials, but comparable between intervention and comparator groups. Of the 24 RCTs, a broad range of surgical procedures was studied; these included colorectal surgery (n = 8), unclassified abdominal surgery (n = 5), caesarean section (n = 5), appendicectomy (n = 2), orthopaedic surgery for fractures (n = 1) and mixed surgery (n = 3). The mean procedure duration ranged from 30 minutes up to 4 hours; 13 studies included participants 7 studies included both types of surgery;56,172,174,175,188–190 22 studies were limited to adult participants and 2 included participants of 13 years and older.178,186

All studies administered 80% FiO₂ in the intervention group; 21 studies administered 30% FiO₂ in the control group, 1 study used 35% FiO₂,190 1 study used 33% FiO₂176 and 1 used 40% FiO₂.175 Four studies used N₂O in the anaesthetic gas mixture. The other studies used nitrogen (N₂) or room air. In 19 studies, patients were under general anaesthesia. In the remaining five trials, patients were anaesthetised but awake and breathing spontaneously, with the allocated gas mixture delivered via a face mask or nasal cannula.188,191–194 Fourteen studies used the CDC definition of SSI. The remaining studies used extensive clinical descriptions that overlapped with the CDC definition.

The updated meta-analysis of all included trials does not show a statistically significant difference for perioperative administration of high (80%) FiO₂ on the prevention of SSI compared with standard (30–35%) FiO₂, but a potentially small reduction cannot be ruled out: (RR 0.91, 95% CI 0.78 to 1.05, p = 0.20; RD 1.2% lower, 2.9% lower to 0.7% higher, low-certainty evidence).

The forest plot is presented in Figure 5. The funnel plot did not reveal an asymmetrical appearance suggesting no clear sign of publication bias (see Appendix 5, Figure 41). Furthermore, regression-based Egger’s test for small-study effects was not statistically significant (p = 0.22).

FIGURE 5 Effect of 80% FiO2 on SSI when compared with 30–35% FiO2. 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.)

Subgroup analyses

We planned our subgroup analyses a priori. Our planned subgroups were type of surgery (abdominal including colorectal, caesarean section, other types of surgical procedure), urgency of surgery, method of delivery of intervention intraoperatively (via endotracheal tube or other airway adjunct vs. face mask or nasal cannulae in awake patients), use of N₂O or not and risk of bias of the trial. Table 6 summarises all subgroup analyses.

Surgical site infection according to type of surgery

There was a statistically significant subgroup difference in the effect of administration of high FiO₂ between trials including colorectal and abdominal surgical patients, trials including patients undergoing caesarean section, and patients undergoing various types of surgical procedures (test for subgroup differences: p = 0.05), with the reduction in SSI not seen among patients undergoing caesarean section. The forest plot is presented in Figure 6. An appreciable level of heterogeneity remained in the subgroup of intra-abdominal surgery (I² = 52%).

FIGURE 6 Effect of 80% FiO2 on SSI when compared with 30–30% FiO2 by type of surgery. 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.)

Surgical site infection according to method of delivery of the intervention

When comparing methods of delivery of the intervention, we found evidence of statistically significant subgroup differences between patients under general anaesthesia, and mechanical ventilation and patients who were awake and breathed spontaneously via a face mask or nasal cannulae (test for subgroup differences: p = 0.03). The analysis showed an indication of a beneficial effect of high FiO₂ in intubated patients (RR 0.85, 95% CI 0.72 to 1.00), although an appreciable level of heterogeneity remained within this subgroup (I² = 43%).

The forest plot is presented in Figure 7.

FIGURE 7 Effect of 80% FiO2 on SSI when compared with low FiO2 by method of oxygen delivery. 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.)

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

There was no evidence of a subgroup difference in effect of administration of high FiO₂ during anaesthesia and surgery between trials that used N₂O in the control group and trials that used N₂ or room air in the control group (test for subgroup differences: p = 0.86). The forest plot is presented in Figure 8.

FIGURE 8 Effect of 80% FiO2 on SSI when compared with 30–30% FiO2 according to use or no use of N2O in the control group. 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.)

Surgical site infection according to overall risk of bias

There was a statistically significant subgroup difference in the effect of high FiO₂ between trials with low risk of bias and those with unclear or high risk of bias (test for subgroup differences: p = 0.05), with an overall reduction in SSI observed among low risk trials but not trials of unclear or high risk of bias. An appreciable level of heterogeneity remained within both subgroups. The forest plot is presented in Figure 9.

FIGURE 9 Effect of 80% FiO2 on SSI when compared with 30–35% FiO2 by overall risk of bias. 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.)

Surgical site infection according to urgency of surgery

There was an indication of a subgroup difference in the observed effect between trials including patients undergoing surgeries of different levels of urgency. Reduction in SSI was most pronounced among trials of emergency surgery compared with trials of elective surgery and trials including surgery with mixed levels of urgency but the test for subgroup differences was not statistically significant (p = 0.06). However, the subgroup of emergency surgery included only four trials of limited sample sizes. An appreciable level of heterogeneity remained among trials of elective surgery and mixed types of surgery. The forest plot is presented in Figure 10.

FIGURE 10 Effect of 80% FiO2 on SSI when compared to 30–35% FiO2 by urgency of surgery. 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.)

Trial sequential analysis on surgical site infection

The trial sequential analysis for SSI 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 (n = 14,468). The conventional boundary for statistical significance was not crossed (p = 0.2), nor was the trial sequential monitoring boundary for benefit (RR 0.9, 95% CI 0.78 to 1.05; see Figure 11).

FIGURE 11 Trial sequential analysis of SSI in all trials included in the meta-analysis. 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.)

The trial sequential analysis for SSI in trials at low risk of bias, the number of patients did not reach the required information size (n = 13,124) as shown in Figure 12. 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 and the analysis indicated that approximately 7000 more patients will be needed to make a definitive conclusion on the benefits of supplemental oxygen.

FIGURE 12 Trial sequential analysis of SSI in low risk of bias trials included in the meta-analysis. 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.)

Grading of recommendations assessment, development and evaluation assessment

Overall, the certainty in the evidence was assessed as low. GRADE evidence profile and summary of findings table are presented in Table 7.

Sensitivity analysis

Using sensitivity analyses, we examined the effect of including the 5700-patient alternating intervention-controlled trial published by Kurz et al. in 2018.202 The study was conducted in a pragmatic setting, where 80% or 30% FiO₂ during major abdominal procedures was allocated to an operation room section at random, and thereafter alternated every 2 weeks on a non-randomised basis; clinicians were not blinded to the FiO₂ received by the patient. Controversy still exists as to whether the trial could be considered an RCT with limited blinding and allocation concealment. The results from our sensitivity analysis were similar to the main analysis (see Appendix 6, Table 19).

Meta-regression

As set out in our plan, we further explored the potential effects of selected subgrouping factors using meta‐regression. In a series of univariable meta-regression analyses, only the risk of bias rating of the included studies was statistically significantly associated with the effect estimates (ratio of RR 1.34, 95% CI 1.02 to 1.77; p = 0.04 for high/unclear risk of bias vs. low risk of bias). In low-risk-of-bias trials, 80% FiO₂ reduced the incidence of SSI (RR 0.80, 95% CI 0.66 to 0.96; Ƭ²=0.03; I² =35%). There was no evidence of a benefit of the intervention in studies at high or unclear risk of bias (RR 1.06, 95% CI 0.85 to 0.1.31; Ƭ²=0.05; I² =33%). In addition, the risk of bias in the included studies appeared to explain almost half of the variations in SSI effect estimates (R² = 44%).

However, in our full meta-regression model including both risk of bias and the method of oxygen administration (i.e. under general anaesthesia with airway adjuncts and mechanical ventilation versus oxygen administration via a face mask or nasal cannula without intubation) which also appeared to be a potential effect modifier based on univariable analysis, the effect modification was reduced and no longer statistically significant. Table 8 summarises the findings from meta-regression analyses.

Effectiveness of 80% versus 30–35% fraction of inspired oxygen on the incidence of mortality and postoperative complications

We selected one anchoring review (Lim et al. 2021)133 that provided comprehensive evidence on the effectiveness of high versus low FiO₂ on postoperative clinical outcomes in patients undergoing surgery under general anaesthesia; 26 trials with a total 4991 patients were included in this review.

Our updated meta-analyses included 30 trials with 6624 participants. Four were newly identified since the publication of the Lim et al. review in 2021.174,175,177,178

Mortality within 30 days

The evidence from 10 studies including 4702 patients suggests that high FiO₂ results in little to no difference in 30-day mortality (RR 1.21, 95% CI 0.80 to 1.85, p = 0.37; RD 0.4% higher, 0.4% lower to 1.5% higher).

The forest plot is presented in Figure 13. GRADE certainty in the evidence was low. There was some evidence of funnel plot asymmetry (see Appendix 5, Figure 42). Furthermore, regression-based Egger’s test for small-study effects was statistically significant (p = 0.02).

FIGURE 13 Effect of 80% FiO2 on mortality when compared with 30–35% FiO2. 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.)

The trial sequential analysis for this outcome assuming a RRR of 20% with 90% power and an alpha error of 5% showed that the number of patients evaluated (n = 4.702) did not surpass the required information size (n = 46,433). The cumulative Z-curve crossed neither the conventional nor trial sequential monitoring boundaries of significance.

The result suggests that there is insufficient information to support the conclusion and further studies are still required.

Mortality within longest follow-up

A total of 14 studies including 5551 patients reported on mortality within longest follow-up (see Figure 14). Five trials reported long-term mortality with median follow-up ranging from 6 months173,175,203 to 3.9 years.56 Given the varied duration of follow-up, hazard ratio may have been a more appropriate effect measure but this was not reported in most of the included trials. The evidence based on RR is very uncertain about the effect of high FiO₂ on mortality up to longest follow-up (RR 1.08, 95% CI 0.81 to 1.45, p = 0.61; RD 0.7%, higher, 1.7% lower to 3.9% higher). GRADE certainty in evidence was very low.

FIGURE 14 Effect of 80% FiO2 on mortality within longest follow-up when compared with 30–35% FiO2. (Figure not reproduced here; see the original article.)

The funnel plot revealed an asymmetrical appearance as shown in Appendix 5, Figure 43. However, regression-based Egger’s test for small-study effects was not statistically significant (p = 0.09).

The trial sequential analysis for this outcome showed that the required information size (n = 38,001) was not met. The cumulative Z-curve crossed neither the conventional nor trial sequential monitoring boundaries of significance.

Incidence of atelectasis

Seven studies with a total of 1948 patients used chest radiography to examine atelectasis. Two studies with a total of 90 patients used chest computerised tomography scan and one study with a total of 630 patients used pulmonary ultrasound to examine atelectasis. Overall, high FiO₂ may result in a large increase in the incidence of atelectasis (RR 1.47, 95% CI 1.20 to 1.7, p = 0.01; RD 6.5% higher, 95% CI 2.8% to 10.9% higher), based on low-certainty evidence. Forest plot is presented in Figure 15.

FIGURE 15 Effect of 80% FiO2 on the incidence of atelectasis when compared with 30–35% FiO2. (Figure not reproduced here; see the original article.)

The trial sequential analysis for this outcome showed that the number of patients evaluated (n = 2668) did not surpass the required information size (n = 16,597). The cumulative Z-curve did not cross the trial sequential monitoring boundaries of significance. The result suggests that there is insufficient information to support the conclusion and further studies are still required.

Pneumonia

Six studies including 3042 patients reported on pneumonia. The evidence is very uncertain about the effect high FiO₂ on the incidence of pneumonia (RR 1.13, 95% CI 0.79 to 1.61, p = 0.50; RD 0.9% higher, 95% CI 1.5% lower to 4.3% higher). Forest plot is presented in Figure 16.

FIGURE 16 Effect of 80% FiO2 on the incidence of pneumonia when compared with 30–35% FiO2. (Figure not reproduced here; see the original article.)

The trial sequential analysis for this outcome indicated that the number of patients evaluated (n = 3042) did not surpass the required information size (n = 25,237). The cumulative Z-curve crossed neither the conventional nor trial sequential monitoring boundaries of significance.

Respiratory failure

Four studies including 2479 patients reported on respiratory failure. The evidence is very uncertain about the effect high FiO₂ on the incidence of respiratory failure (RR 1.03, 95% CI 0.52 to 2.04, p = 0.93; RD 0.1% higher, 95% CI 1.6% lower to 3.4% higher). Forest plot is presented in Figure 17.

FIGURE 17 Effect of 80% FiO2 on respiratory failure when compared with 30–35% FiO2. (Figure not reproduced here; see the original article.)

The trial sequential analysis for this outcome showed that the required information size (n = 72,431) was not met.

Postoperative pulmonary complications

Four studies of a total 1236 patients reported on PPCs. The evidence is very uncertain about the effect of high FiO₂ on the incidence of PPCs (RR 1.06, 95% CI 0.77 to 1.46, p = 0.71; RD 1.1% higher, 95% CI 4.1% lower to 8.2% higher). The forest plot is presented in Figure 18.

FIGURE 18 Effect of 80% FiO2 on PPCs when compared with 30–35% FiO2. (Figure not reproduced here; see the original article.)

The trial sequential analysis for this outcome revealed that the number of patients evaluated (n = 1236) did not surpass the required information size (n = 9648). The cumulative Z-curve crossed neither the conventional nor trial sequential monitoring boundaries of significance.

The result suggests that there is insufficient information to support the conclusion and further studies are still required.

Unplanned intensive care unit admissions

Five studies including 2677 patients reported on ICU admissions with 73/1331 events (5.5%) in the high FiO₂ group and 75/1346 events (5.6%) in the low FiO₂ group. The evidence suggests that high FiO₂ results in little to no difference in unplanned ICU admissions (RR 0.98, 95% CI 0.69 to 1.39, p = 0.91; RD 0.1% lower, 95% CI 1.7% lower to 2.2% higher). The forest plot is presented in Figure 19. GRADE certainty in the evidence was low.

FIGURE 19 Effect of 80% FiO2 on ICU admissions when compared with 30–35% FiO2. (Figure not reproduced here; see the original article.)

The trial sequential analysis for this outcome showed that the number of patients evaluated (n = 2677) did not surpass the required information size (n = 19,866).

Length of hospital stay

Ten studies of total 2863 patients reported on length of hospital stay as shown in Figure 20. The pooled mean length of hospital stay was 9.4 days in the high FiO₂ and 9.3 days in the low FiO₂ group. The evidence suggests that high FiO₂ results in little to no difference in the length of hospital stay [mean difference (MD) 0.27 days, 95% CI −0.28 to 0.81; p = 0.34]. GRADE certainty in the evidence was low.

FIGURE 20 Effect of 80% FiO2 on length of hospital stay when compared with 30–35% FiO2. (Figure not reproduced here; see the original article.)

There was no evidence of funnel plot asymmetry (see Appendix 5, Figure 44). Furthermore, regression-based Egger’s test did not show evidence of small-study effects (p = 0.31). The required information size for this outcome was not met (n = 32,212).

Nausea and vomiting

We selected one anchoring review (Makwei et al. 2023)134 that assessed the effect of 80% versus 30% oxygen concentration on the incidence of PONV. This outcome was not prespecified in our protocol but was considered as an important outcome for patients by our advisory panel following our identification of this anchoring review. Consequently, this outcome was included in this overview. We did not update meta-analyses from the anchoring review given its recency.

The meta-analysis included 10 trials representing 6773 patients. All the studies randomised individual patients except one,134 which used a cluster crossover design. Meta-analysis of all included trials showed little evidence of a benefit of perioperative administration of 80% FiO₂ on the prevention of postoperative nausea and vomiting compared with the 30% FiO₂ group (RR 0.97, 95% CI 0.86 to 1.08; RD 1.0% fewer, 3.3% fewer to 1.2% more).The forest plot is presented in Figure 21.

FIGURE 21 Effect of 80% FiO2 on postoperative nausea and vomiting when compared with 30% FiO2. Source: Reproduced with permission from Markwei et al.134 URL: https://doi.org/10.1097/ALN.0000000000004428. (Figure not reproduced here; see the original article.)

There was evidence of asymmetry in the funnel plot, suggesting possible publication bias for the treatment effect on PONV. The p-value for Egger’s test for the PONV outcome was not significant (p = 0.08) but the test was underpowered.

There was no evidence that surgery type influenced the effect of intraoperative supplemental oxygen (80% vs. 30%) on the incidence of postoperative nausea and vomiting, with interaction test p = 0.17. The relative risk was 0.87 (0.72 to 1.05) for abdominal and 1.06 (0.87 to 1.28) for non-abdominal surgeries.

Grading of recommendations assessment, development and evaluation certainty in evidence was low. The trial sequential analysis for this outcome assuming a RRR of 20% with 90% power and an alpha error of 5% showed that the number of patients evaluated (n = 6773) did not surpass the required information size (n = 7431). The cumulative Z-curve crossed neither the conventional nor trial sequential monitoring boundaries of significance. The Z-curve crossed the futility boundaries suggesting that 80% FiO₂ is not superior to 30% FiO₂.

In summary, high FiO₂ may result in a slight reduction in the incidence of SSI compared with standard FiO_2. The evidence suggests that high FiO₂ results in little to no difference in 30-day mortality, unplanned ICU admissions, PNOV and length of hospital stay compared to low FiO_2.

High FiO₂ may result in a large increase in the incidence of atelectasis. The evidence is very uncertain about the effect of high FiO₂ on PPCs, pneumonia and respiratory failure.

Delivery device: what is the clinical effectiveness of non-invasive respiratory support such as continuous positive airway pressure or high-flow nasal oxygen compared to conventional oxygen therapy?

High-flow nasal oxygen versus conventional oxygen therapy in the immediate postoperative period

We selected one anchoring review (Chaudhury et al. 2020)62 that examined the effectiveness of HFNO in the immediate postoperative period compared to COT or NIV. Among our prespecified outcomes of interest, the review reported on in-hospital mortality and hospital length of stay. In addition, the review reported findings on reintubation rate, escalation of respiratory support, ICU length of stay and postoperative hypoxemia.

Our updated meta-analysis included 16 RCTs enrolling 2201 patients; 5179–183 were newly identified since the publication of the review by Chaudhury et al. in 2020.

Only one RCT compared HFNO with NIV.219 The remaining trials compared HFNO with COT. Nine of the 16 RCTs were conducted in patients after cardiac surgery;180,181,183,219–224 of the remaining seven, four were conducted in patients after thoracic surgery,225–228 one was conducted in patients after bariatric surgery,179 one was conducted in patients after major thoracic and abdominal surgery229 and one was conducted in patients after major head and neck surgery.182

Six of the included studies examined patients at moderate to high risk of postoperative respiratory complications.180,181,220,223,224,228 Three trials examined patients who were obese exclusively,179,220,223 whereas two RCTs specifically excluded people who were obese.227,229 All trials, except for the RCT that used NIV as a comparator,219 used HFNO prophylactically rather than as a treatment for respiratory failure.

Fourteen RCTs179–183,219,220,222–224,226–230 used one device (Optiflow; Fisher and Paykel Healthcare, Auckland, New Zealand), and one trial used another device (MaxVenturi; Maxtec, LLC, Salt Lake City, UT, USA).225 Another trial221 did not specify the type of HFNO device used.

All HFNO devices provided heated and humidified nasal oxygen with flow rates titrated between 25 and 60 l/minute, with the goal of keeping the patient comfortable and aiming for an oxygen saturation as measured by pulse oximetry (SpO₂ target > 90%).

Panoramic meta-analyses of trials comparing high-flow nasal oxygen to conventional oxygen therapy

The review did report on some of our prespecified outcomes including SSI, PPCs and unplanned ICU admission. Below we present the results of our updated analyses for prespecified outcomes first followed by outcomes reported in the review.

Mortality

Five trials including 853 patients reported on mortality as shown in Figure 22. The evidence suggests that HFNO use in the immediate postoperative period results in little to no difference in mortality compared with COT (RR, 0.78, 95% CI 0.27 to 2.24; RD 0.4% lower, 95% CI 1.4% lower to 2.4% higher). GRADE certainty in evidence was low.

FIGURE 22 Effect of HFNO on mortality when compared to COT. 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.)

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 = 46,433) was not met.

Length of hospital stay

Twelve studies of total 1367 patients reported on length of hospital stay (see Figure 23). The evidence suggests that HFNO use in the immediate postoperative period results in little to no difference in the length of hospital stay compared with COT (MD 0.45 days lower, 95% CI 0.86 days lower to 0.03 days lower). The funnel plot did not reveal an asymmetrical appearance (see Appendix 5, Figure 45). Furthermore, regression-based Egger’s test for small-study effects was not statistically significant (p = 0.85).

FIGURE 23 Effect of HFNO on hospital length of stay when compared with COT. (Figure not reproduced here; see the original article.)

Grading of recommendations assessment, development and evaluation certainty in evidence was assessed as low. The trial sequential analysis for this outcome was not met (n = 3092).

Reintubation

Nine trials including 1056 patients reported on reintubation rate as shown in Figure 24. Compared with COT use, the evidence is very uncertain about the effect of HFNO in the immediate postoperative period (RR 0.78, 95% CI, 0.29 to 2.07; RD 0.9% reduction, 95% CI 2.9% reduction to 4.4% increase). GRADE certainty in evidence was very low.

FIGURE 24 Effect of HFNO on reintubation when compared with COT. (Figure not reproduced here; see the original article.)

The trial sequential analysis for this outcome showed that the required information size (n = 21,142) was not met, and, consequently, we rated down the certainty for this outcome because of imprecision.

Subgroup analyses

Subgroup analysis based on the type of surgery (test of group differences, p = 0.41), risk of postoperative respiratory complications (p = 0.69) and body mass index (p = 0.09) did not show credible subgroup effects for reintubation (see Table 9).

Escalation of respiratory support

Nine trials including 1309 patients reported on escalation of respiratory support (see Figure 25). Compared with COT use, HFNO use in the immediate postoperative period may reduce the need for escalation of respiratory support, but the evidence is very uncertain (RR 0.61, 95% CI 0.41 to 0.91; RD 7.8% lower, 95% CI 1.8% to 11.7% lower).

FIGURE 25 Effect of HFNO on the need for escalation of respiratory support when compared with COT. (Figure not reproduced here; see the original article.)

The trial sequential analysis for this outcome indicated that the required information size (n = 16,461) was not met, and, consequently, we rated down the certainty for this outcome because of imprecision.

Subgroup analyses

Subgroup analysis based on the type of surgery (test of group differences, p = 0.75), risk of postoperative respiratory complications (p = 0.35) and obesity (p = 0.37) did not show credible subgroup effects for escalation of respiratory support (see Table 10).

Postoperative hypoxaemia

Five trials including 829 patients reported on postoperative hypoxaemia. HFNO use in the immediate postoperative period may reduce the incidence of postoperative hypoxaemia (see Figure 26) but the evidence is very uncertain (RR 0.79, 95% CI 0.58 to 1.06; RD 10.4% lower, 95% CI 20.8% lower to 3% higher).

FIGURE 26 Effect of HFNO on postoperative hypoxaemia when compared to COT. (Figure not reproduced here; see the original article.)

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 = 6421) was not met.

Length of intensive care unit stay

Eleven studies of total 1341 patients reported on length of ICU stay displayed in Figure 27. HFNO is likely to result in little to no difference in length of ICU stay compared with COT (MD 0.03 days higher, 95% CI 0.1 days lower to 0.17 days higher). The funnel plot also did not reveal an asymmetrical appearance (see Appendix 5, Figure 46).

FIGURE 27 Effect of HFNO on ICU length of stay when compared to COT. (Figure not reproduced here; see the original article.)

Furthermore, regression-based Egger’s test for small-study effects was not statistically significant (p = 0.85). GRADE certainty in evidence was assessed as moderate.

Grading of recommendations assessment, development and evaluation assessment

Grading of recommendations assessment, development and evaluation evidence profile and summary of findings table for important outcomes is presented in Table 11.

In summary, the data suggest that HFNO oxygen may reduce the rate of postoperative hypoxaemia and subsequent need for invasive respiratory support, but the evidence is very uncertain. The evidence suggests that HFNO may result in little to no difference in mortality, reintubation rate and length of hospital stay.

Postoperative non-invasive ventilation compared with conventional oxygen therapy

We selected one anchoring review (Hui et al. 2021)171 that provided comprehensive evidence on the effectiveness of postoperative non-invasive respiratory support including CPAP, NIV or HFNO against standard postoperative care in the adult population.

The review included 38 trials consisting of 9782 patients. Pneumonia was observed in 214/4403 (4.9%) patients receiving non-invasive respiratory support compared with 216/3937 (5.5%) patients in the standard care group (RD −0.01; 95% CI −0.02 to 0.00; I² = 8%; p = 0.23). The rate of PPCs was 393/1379 (28%) in patients receiving non-invasive respiratory support compared with 280/902 (31%) in the standard care group (RD −0.11; 95% CI −0.23 to 0.01; I² = 79%; p = 0.07). There was no evidence of subgroup difference in the effect of CPAP, NIV or HFNO in preventing pneumonia.

To avoid the overlap between this review and Chauduri et al. 2020,62 we focused on trials comparing NIV (including CPAP) with standard postoperative care. We did not identify additional trials from updated search results.

The review did not report on some of our prespecified outcomes including SSI. Below, we present the results of our updated analyses for prespecified outcomes first followed by other outcomes reported in the review.

Meta-analyses of trials comparing non-invasive ventilation with standard postoperative care

Mortality

Nineteen trials including 7876 patients reported on mortality (see Figure 28). NIV results in little to no difference in mortality compared with COT in the immediate postoperative period (RR 0.91, 95% CI 0.62 to 1.32; RD 0.1% lower, 95% CI 0.6% lower to 0.5% higher). The was no evidence of funnel plot asymmetry (see Appendix 5, Figure 47). Furthermore, regression-based Egger’s test for small-study effects was not statistically significant (p = 0.66).

FIGURE 28 Effect of NIV on mortality when compared with COT. 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 table includes minor additions and formatting changes to the original text. (Figure not reproduced here; see the original article.)

Grading of recommendations assessment, development and evaluation certainty in evidence was high. The trial sequential analysis for this outcome showed that the number of patients evaluated (n = 7876) did not surpass the required information size (n = 62,194). However, this required information size was calculated assuming a RRR of 20% with 90% power and an alpha error of 5%. With a low baseline mortality risk of 1.5% (or 15 per 1000), the required information size would be sufficiently powered to detect a RD between intervention and comparison groups of 0.3% (20% × 1.5%) or 3 per 1000. This may be considered as an unnecessarily large sample size that aims to detect a trivial difference. Therefore, we did not downgrade the certainty of the evidence for imprecision. The high certainty rating is supported by the CI for the pooled RD on mortality lying completely within the range of trivial or no benefit and harm: the threshold for a small (minimally important) beneficial or harmful effect was assumed to be 10/1000, or 1% absolute RD, and the pooled RD was one fewer per 1000 (95% CI from six fewer to five more) or −0.1% (95% CI −0.6% to 0.5%).

Subgroup analyses

Subgroup analysis based on the type of surgery did not show credible subgroup effects for mortality (see Figure 29). Risk of mortality did not differ between patients who underwent abdominal surgery compared to cardiothoracic or mixed surgery, and risk of mortality. The test of subgroup differences was not statistically significant (p = 0.36).

FIGURE 29 Effect of NIV on mortality when compared with COT. Studies are grouped by type of surgery. (Figure not reproduced here; see the original article.)

Postoperative pulmonary complications

Ten trials including 2142 patients (see Figure 30) reported on PPCs. Compared with COT, NIV use in the immediate postoperative period may decrease the incidence of PPCs (RR 0.62, 95% CI 0.44 to 0.87; RD 12.2% lower, 95% CI 4.2% to 18% lower). Although the CI for the absolute effect includes clinically meaningful differences, GRADE certainty in evidence was low due to inconsistency and imprecision (the CI crosses the threshold of large effect). The funnel plot did not reveal an asymmetrical appearance (see Appendix 5, Figure 48). Furthermore, regression-based Egger’s test for small-study effects was not statistically significant (p = 0.54).

FIGURE 30 Effect of NIV on PPCs when compared to COT. (A) Late extubation. (B) Early extubation. (Figure not reproduced here; see the original article.)

The trial sequential analysis for this outcome assuming a RRR of 20% with 90% power and an alpha error of 5% showed that the number of patients evaluated (n = 2142) did not surpass the required information size (n = 14,394). The cumulative Z-curve crossed the conventional significance boundary but not the trial sequential monitoring boundaries. The result suggests that there is insufficient information to support the conclusion and further studies are still required.

Pneumonia

Nineteen trials including 7848 patients reported on pneumonia (see Figure 31). NIV use in the immediate postoperative period may result in little to no difference in the incidence of pneumonia compared with COT (RR 0.95, 95% CI 0.78 to 1.15; RD 0.3% lower, 95% CI 1.2% lower to 0.8% higher). The funnel plot showed an asymmetrical appearance (see Appendix 5, Figure 49). Furthermore, regression-based Egger’s test for small-study effects was statistically significant (p = 0.03).

FIGURE 31 Effect of NIV on pneumonia when compared with COT. (A) Late extubation. (B) Early extubation. (Figure not reproduced here; see the original article.)

Grading of recommendations assessment, development and evaluation certainty in evidence was low. The trial sequential analysis for this outcome assuming a RRR of 20% with 90% power and an alpha error of 5% showed that the number of patients evaluated (n = 7848) did not surpass the required information size (n = 16,343). The cumulative Z-curve crossed neither the conventional nor trial sequential monitoring boundaries of significance.

Subgroup analysis

Risk of pneumonia did not differ between patients who underwent abdominal surgery versus cardiothoracic surgery or mixed (see Figure 32). The test of subgroup differences was not statistically significant (p = 0.47).

FIGURE 32 Effect of NIV on pneumonia when compared with COT. Studies are grouped by type of surgery. (A) Late extubation. (B) Early extubation. (Figure not reproduced here; see the original article.)

Reintubation

Seventeen trials including 7782 patients reported on reintubation (see Figure 33). Compared with COT, NIV use in the immediate postoperative period probably results in little to no difference in reintubation rate (RR 0.82, 95% CI 0.65 to 1.04; RD 0.7% lower, 95% CI 1.4% lower to 0.2% higher). There funnel plot seems to show some asymmetry (see Appendix 5, Figure 50). However, regression-based Egger’s test for small-study effects was not statistically significant (p = 0.09).

FIGURE 33 Effect of NIV on reintubation when compared with COT. (A) Late extubation. (B) Early extubation. (Figure not reproduced here; see the original article.)

Grading of recommendations assessment, development and evaluation certainty in evidence was moderate. The trial sequential analysis for this outcome showed that the number of patients evaluated (n = 7782) did not surpass the required information size (n = 22,791). The cumulative Z-curve crossed neither the conventional nor trial sequential monitoring boundaries of significance. The result suggests that there is insufficient information to support the conclusion and further studies are still required.

Unplanned intensive care unit admission

Nine trials including 2382 patients reported on unplanned ICU admission (see Figure 34). Compared with COT, NIV in the immediate postoperative period may reduce unplanned ICU admission slightly (RR 0.73, 95% CI 0.50 to 1.06; RD 1.6% lower, 95% CI 3% lower to 0.4% higher). The funnel plot did not reveal an asymmetrical appearance as shown in Appendix 5, Figure 51. Furthermore, regression-based Egger’s test for small-study effects was not statistically significant (p = 0.81).

FIGURE 34 Effect of NIV on unexpected ICU admission when compared with COT. (A) Late extubation. (B) Early extubation. (Figure not reproduced here; see the original article.)

Grading of recommendations assessment, development and evaluation certainty in evidence was low. The trial sequential analysis for this outcome assuming a RRR of 20% with 90% power and an alpha error of 5% showed that the number of patients evaluated (n = 2382) did not surpass the required information size (n = 15,178). The cumulative Z-curve crossed neither the conventional nor trial sequential monitoring boundaries of significance. The result suggests that there is insufficient information to support the conclusion and further studies are still required.

Acute respiratory distress syndrome

Eight trials including 6617 patients reported on acute respiratory distress syndrome (ARDS) (see Figure 35). Compared with COT, postoperative use of NIV likely results in a slight reduction in the incidence of ARDS (RR 0.70, 95% CI 0.53 to 0.93; RD 1.2% lower, 95% CI 1.9% to 0.3 lower). GRADE certainty in evidence was moderate.

FIGURE 35 Effect of NIV on ARDS when compared with COT. (Figure not reproduced here; see the original article.)

The trial sequential analysis for this outcome showed that the number of patients evaluated (n = 6617) did not surpass the required information size (n = 30,476). The cumulative Z-curve crossed the conventional but not trial sequential monitoring boundaries of significance. The result suggests that there is insufficient information to support the conclusion and further studies are still required.

Pulmonary aspiration

Four trials including 6041 patients reported on pulmonary aspiration (see Figure 36). Compared with COT use, NIV in the immediate postoperative period results in little to no difference in the incidence of pulmonary aspiration (RR 1.11, 95% CI 0.54 to 0.2.27; RD 0.1% higher, 95% CI 0.2% lower to 0.6 higher). GRADE certainty in evidence was high. The trial sequential monitoring boundaries for this outcome could not be constructed due to too little information use. We did not downgrade the certainty of the evidence for imprecision because the CIs for the absolute estimate effects included only the range for trivial or no benefit and harm (the threshold for a small beneficial or harmful effect assumed to be larger than 10/1000). The absolute RD was one more per 1000 (95% CI from two fewer to six more) based on a baseline risk of 0.5% in the comparison group.

FIGURE 36 Effect of NIV on pulmonary aspiration when compared with COT. (Figure not reproduced here; see the original article.)

Length of hospital stay

Nineteen studies of total 8157 patients reported on length of hospital stay (see Figure 37). Compared with COT use, NIV use in the immediate postoperative period probably results in little to no difference in the length of hospital stay (MD 1.12 days lower, 95% CI 1.69 days lower to 0.55 days lower). There was evidence of funnel plot asymmetry as shown in Appendix 5, Figure 52. Furthermore, regression-based Egger’s test for small-study effects was statistically significant (p = 0.02). GRADE certainty in evidence was moderate.

FIGURE 37 Effect of NIV on length of hospital stay when compared with COT. (Figure not reproduced here; see the original article.)

Grading of recommendations assessment, development and evaluation assessment

Grading of recommendations assessment, development and evaluation evidence profile and summary of findings table for important outcomes are presented in Table 12.

Top-level panoramic meta-analyses

We had planned to carry out two top-level panoramic meta-analyses (one on SSI and the other on mortality) across all comparisons of oxygenation strategies/devices. For these analyses, we included comparisons where the intervention was conceptually aiming to deliver more oxygen to the patient perioperatively irrespective of the means to achieve it.

Unfortunately, we were unable to perform the top-level panoramic meta-analysis for SSI as this outcome was examined and reported only among trials of high versus low FiO₂ and was not reported in trials comparing different oxygenation devices with COT.

Our top-level panoramic meta-analysis on mortality (see Figure 38) showed that oxygenation strategy (high FiO₂/HFNO/ NIV) did not influence mortality (RR 1.01, 95% CI 0.77 to 1.32), although the total number of deaths included in the analysis was still relatively small.

FIGURE 38 Top-level panoramic meta-analysis on mortality. 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.)

Evidence map

To provide an accessible quantitative overview of the effect of perioperative oxygen therapy on important postoperative outcomes, we created a combined forest plot, or ‘evidence map’, that summarises findings across the three comparisons. The evidence map is presented in Figure 39.

FIGURE 39 Evidence map of systematic reviews and meta-analyses. 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.)

The evidence map shows the pooled effect estimates obtained from the most up-to-date analyses (either reported in the anchoring reviews or conducted by us through adding data from new trials to those reported by the anchoring reviews) for individual key outcomes in each of the comparisons. The pooled effect estimates are presented both in relative and absolute terms (i.e. relative risk and RD, respectively), with the centre of the bubbles corresponding to the point estimate of the pooled relative risks. The size of the bubble reflects the volume of evidence (total sample size) for each pooled estimate. The overall quality or certainty of evidence is rated according to the GRADE framework, and is colour-coded for the bubbles (green – high certainty; yellow – moderate certainty; amber – low certainty; red – very low certainty).

Overall, the evidence map indicates that:

1. The GRADE certainty of the evidence is generally low or very low.
2. Compared with low FiO₂, high FiO₂ is unlikely to have a large beneficial or harmful effect (RD of ≥ 5% point), except for a potentially large effect of increased atelectasis.
3. Evidence is insufficient to confirm or rule out a moderate to small effect of high versus low FiO₂ for various outcomes.
4. Compared with COT, NIV has little or no effect on mortality and pulmonary aspiration, but evidence is insufficient to confirm or refute a potentially large effect in reducing PPCs.
5. Evidence is insufficient to confirm or rule out a potentially large beneficial effect or a moderate harmful effect on various respiratory outcomes for HFNO compared with COT.