Association between inspired oxygen fraction and development of postoperative pulmonary complications in thoracic surgery: a multicentre retrospective cohort study

Abstract

Background

Limited data exist to guide oxygen administration during one-lung ventilation for thoracic surgery. We hypothesised that high intraoperative inspired oxygen fraction during lung resection surgery requiring one-lung ventilation is independently associated with postoperative pulmonary complications (PPCs).

Methods

We performed this retrospective multicentre study using two integrated perioperative databases (Multicenter Perioperative Outcomes Group and Society of Thoracic Surgeons General Thoracic Surgery Database) to study adult thoracic surgical procedures using one-lung ventilation. The primary outcome was a composite of PPCs (atelectasis, acute respiratory distress syndrome, pneumonia, respiratory failure, reintubation, and prolonged ventilation >48 h). The exposure of interest was high inspired oxygen fraction (FiO₂), defined by area under the curve of a FiO₂ threshold > 80%. Univariate analysis and logistic regression modelling assessed the association between intraoperative FiO₂ and PPCs.

Results

Across four US medical centres, 141/2733 (5.2%) procedures conducted in 2716 patients (55% female; mean age 66 yr) resulted in PPCs. FiO₂ was univariately associated with PPCs (adjusted OR [aOR]: 1.17, 95% confidence interval [CI]: 1.04–1.33, P=0.012). Logistic regression modelling showed that duration of one-lung ventilation (aOR: 1.20, 95% CI: 1.03–1.41, P=0.022), but not the time-weighted average FiO₂ (aOR: 1.01, 95% CI: 1.00–1.02, P=0.165), was associated with PPCs.

Conclusions

Our results do not support limiting the inspired oxygen fraction for the purpose of reducing postoperative pulmonary complications in thoracic surgery involving one-lung ventilation.

Editor's key points.

• •The dose and duration of higher inspired oxygen fraction required during one-lung ventilation for thoracic surgery is poorly evidence-based.
• •The authors examined whether high intraoperative inspired oxygen fraction during lung resection surgery requiring one-lung ventilation may account for postoperative pulmonary complications, using a large mutlicentre database.
• •Postoperative pulmonary complications occurred in 141/2733 (5.2%) patients.
• •Logistic regression modelling failed to find any link with time-weighted average FiO₂ and postoperative pulmonary complications.
• •By considering the individual effects of duration of exposure and FiO_2, these results do not support limiting the inspired oxygen fraction for the purpose of reducing postoperative pulmonary complications.

Oxygen is one of the most widely used therapies during anaesthesia and mechanical ventilation but might also contribute to lung injury and postoperative pulmonary complications via a number of potential pathways.¹ In general, clinical evidence directly linking alveolar hyperoxia to postoperative pulmonary complications is mixed and there is a paucity of evidence in the thoracic surgical population.^2,3 Large observational studies of surgical patients have identified inspired oxygen fraction (FiO₂) as a risk factor for respiratory complications⁴ and mortality⁵ as well as acute respiratory distress syndrome (ARDS)⁶ in some but not all investigations.⁷ Recent meta-analyses of randomised trials^8,9 failed to demonstrate a link between higher FiO₂ and pulmonary complications; however, this previous work did not focus on the thoracic surgery population. Patients presenting for pulmonary resection surgery may serve as a sensitive clinical model for the genesis of hyperoxia-related complications for a number of reasons. These include pre-existing pulmonary disease, resection of functional lung parenchyma, the detrimental effects of one-lung ventilation including oxidative stress and the potential for ventilation-induced lung injury. Best-practice strategies for administration of intraoperative FiO₂ during thoracic surgery remain unknown.

To assess the association between high FiO₂ during thoracic surgery and postoperative pulmonary complications we performed a retrospective multicentre cohort study. By utilising both the Multicenter Perioperative Outcomes Group (MPOG) and Society of Thoracic Surgeons General Thoracic Surgery Databases (STS-GTSD), we were able to integrate intraoperative parameters with adjudicated registry outcomes. We hypothesised that intraoperative inspired oxygen fraction (assessed as area under the curve of FiO₂ above the 80% threshold) during one-lung ventilation is independently associated with postoperative pulmonary complications after lung resection surgery.

Methods

Study design

This retrospective observational cohort study was approved by the University of Virginia Institutional Review Board (21039) and the REporting of studies Conducted using Observational Routinely-collected health Data (RECORD) guidelines, an extension of the existing STrengthening the Reporting of OBservational studies in Epidemiology (STROBE), were followed.¹⁰ Informed patient consent was waived because no patient care interventions were involved in the conduct of the study. Additionally, the Institutional Review Board of each contributing organisation approved aggregation of this limited dataset. Study methods including data collection, outcomes, and statistical analyses were established a priori, reviewed and approved by the Multicenter Perioperative Outcomes Group (MPOG) peer-review committee (July 13, 2020).¹¹

Inclusion criteria

Adult patients (≥18 yr) who underwent a lung resection of at least 60 min duration and utilising one-lung ventilation between 2012 and 2020 were included in this observational cohort study. Additionally, included cases had to meet a previously validated Perioperative Research Standard for data completeness and quality (Supplementary Appendix S2)¹² and have an accompanying STS General Thoracic Surgical Database (STS-GTSD) record. In addition to patient characteristic elements (age, sex, ASA physical status) and validated case elements (anaesthesia start and end times), the Perioperative Research Standard requires submission of both laboratory results and discharge diagnoses, which not every MPOG institution submitted at all points in our study window (Figure 1).

Fig 1.

Derivation of the study cohort. MPOG, Multicenter Perioperative Outcomes Group; STS-GTSD, Society of Thoracic Surgeons General Thoracic Surgical Database.

Exclusion criteria

Exclusion criteria were (i) prior one-lung ventilation within 90 days, (ii) procedures requiring cardiopulmonary bypass or extracorporeal circulation, and (iii) lung transplantation. Pneumonectomies were also excluded from the analysis, as these were a major cause of the numerical imbalance and involved a drastically different set of procedural considerations (duration of surgery, severity of disease, and physiologic stressors) than the other classes. The inclusion/exclusion criteria for the study were designed to capture a wide sample of intraoperative oxygenation practices within the thoracic surgical population, and therefore included a diverse set of lung resection procedures to allow meaningful evaluation of clinical practice.

Data sources

We collected study data from two sources: the MPOG registry¹³ and the STS-GTSD¹⁴ that were integrated as previously described¹⁵ (Supplementary Tables S1 and S2). The STS data were used to obtain information for candidate covariates based on previously published thoracic surgery models16, 17, 18, 19 and for postoperative outcome data. GTSD records were linked to MPOG records using patient-level identifiers at each participating site as previously described.¹⁵

Primary outcome

The primary outcome was a composite of postoperative pulmonary complications from the STS Thoracic Database, including pneumonia, atelectasis, ARDS, respiratory failure, reintubation, and ventilator support for >48 h. Diagnostic criteria from the STS-GTSD Data Dictionary for each of the composite postoperative pulmonary complications can be found in Supplementary Appendix S3. These outcome variables were chosen based on known or suspected relationships to the exposure (high FiO₂) or to any of the other endpoints related to the exposure.

Because atelectasis differs markedly with regard to severity and clinical significance, when compared with the other outcomes (pneumonia, ARDS, respiratory failure, and prolonged ventilator support), post hoc sensitivity analysis was also performed using criteria for composite postoperative pulmonary complications that did not include the diagnosis of atelectasis.

Secondary outcomes

We also recorded non-pulmonary postoperative complications including death (Supplementary Table S1).

Exposure of interest

Inspired oxygen concentration

The primary metric for inspired oxygen concentration was defined a priori as area under the curve (AUC) of theFiO₂ above the 80% threshold (percentage∗minutes).⁵ AUC of FiO₂ above 80% was selected a priori, in an effort to most accurately capture the cumulative exposure experienced by a patient (dose times duration). This definition is analogous to that used in a recent large multicentre observational study of hyperoxia in surgeries excluding those utilising one-lung ventilation.⁵ The base unit for the AUC of FiO₂ above 80% logistic regression models is 1000%∗minutes, meaning FiO₂=90% for a 60-min case (minus the FiO₂ 80% threshold) would translate to a base unit of 10% times 60 min, or 600%∗minutes, or 0.6 in terms of 1000%∗minutes.

Because the primary exposure variable intrinsically contains both FiO₂ and duration of one-lung ventilation, we cannot directly resolve the independent effect of each; therefore, alternative definitions for high FiO₂, specifically, time in minutes with FiO₂ >80% and peripheral capillary oxygen saturation (SpO₂) >98% (time-discretionary high FiO₂), and time-weighted average FiO₂ were also tested in sensitivity analyses. The three definitions for the exposure variable: (i) AUC for FiO₂ above 80% threshold (Model 1), (ii) time-discretionary high FiO₂ (Model 2), and (iii) time-weighted average FiO₂ (Model 3) are illustrated in Figure 2.

Fig 2.

Exposure variables for hypothetical cases of desaturation during one-lung ventilation. (a) Illustration of exposure variables for a hypothetical case depicting marginal oxygen saturation at the initiation of one-lung ventilation followed by improvement and subsequent weaning of FiO₂. (b) Illustration of exposure variables for a hypothetical case depicting severe oxygen desaturation (such as might be due to lung isolation device malposition), followed by correction and subsequent less aggressive weaning of FiO₂. AUC, area under the curve; FiO₂, inspired oxygen fraction; SpO₂, peripheral capillary oxygen saturation.

Additionally, sensitivity analyses were performed using the primary high FiO₂ variable (AUC of FiO₂ above 80%) (i) during the entire period of mechanical ventilation (as opposed to just the period of one-lung ventilation) and (ii) excluding patients/cases who experienced intraoperative desaturation (SpO₂ <88% for more than three consecutive minutes). Within the high FiO₂ metric, discretionary high FiO₂ was calculated as time above the concurrent FiO₂ >80% and SpO₂ >98% threshold. Although Pao₂ data are generally only available for patients with indwelling arterial cannulae, we characterised the association between both (i) FiO₂ vs Pao₂ and (ii) SpO₂ vs Pao₂ in the subset of patients for whom Pao₂ data were available. To assist in visualisation of these relationships Pao₂ measurements were classified based on FiO₂ (80%) and SpO₂ (98%) threshold pairings.

Covariates

Patient characteristics, preoperative, anaesthetic, surgical, comorbidity, and exposure data were derived from the MPOG Database (Supplementary Table S2a) and STS Thoracic Database (Supplementary Table S2b). Additional comorbidities were derived using Elixhauser classifications²⁰ applied to International Classification of Diseases billing information obtained from the MPOG Database²¹ (Supplementary Table S2c). Additional details on how each covariate was defined (including specifications from the STS-GTSD Data Dictionary and standardised MPOG phenotypes) can be found in Supplementary Appendix S4, including how missing forced expiratory volume in 1 s (FEV₁) data were handled.22, 23, 24, 25, 26

Statistical analyses

It should be noted that 806 (29.5%) of the cases in our cohort have been previously reported in a 2021 study on low tidal volume during one-lung ventilation.¹⁵ There is no patient overlap with an earlier 2016 study.¹⁹ Perioperative and intraoperative characteristics were summarised using medians and interquartile ranges for continuous variables and counts and percentages for categorical covariates. Comparisons of continuous data were made using Mann–Whitney U tests and categorical data were compared using Pearson chi-squared tests. P-value <0.05 was selected a priori to denote statistical significance. Logistic regression modelling was performed for multivariable assessment of the primary outcomes, with additional covariates selected a priori using all data available in MPOG (Supplementary Fig. S1). All analyses were conducted using SAS 9.4 (SAS Institute, Cary, NC, USA) or R version 3.6.0 (R Core Team, Vienna, Austria).

Multivariable logistic regression: inspired oxygen fraction (primary model)

We assessed the association between intraoperative FiO₂ and postoperative pulmonary complications using logistic regression modelling (Model 1: AUC of FiO₂ above 80%). Association of the outcome with this exposure was evaluated for the period of one-lung ventilation, and for the entire period of mechanical ventilation (AUC of FiO₂ > 80%). Model discrimination is reported using the c-statistic. Measures of effect for model covariates are reported as adjusted odds ratios with 95% confidence intervals. With regard to treatment of missing values at the modelling stage, no imputation was performed and only patients with complete data for all variables used in a particular model were included in that model. Details on the missing number (and percentage) of cases for each of the variables included in our final regressions is provided in Supplementary Table S3. FEV₁ was included as a risk predictor in our models; however, as absence of these data is likely to be non-random, missing FEV₁ data was included as a covariate in regression models when FEV₁ data were unavailable. We handled missing FEV₁ as previously described.¹⁹ This approach leaves the FEV₁ coefficient identical to the condition in which data were restricted to only patients with non-missing FEV₁ and separately models risk for those patients with missing FEV₁ data. In each model, medical centre was treated as a fixed effect with multiple categories rather than a random effect due to the limited number of centres (n=4) available in our study.

Sensitivity analyses

Preplanned sensitivity analyses included evaluation of the impact of time-discretionary high FiO₂ (Model 2) and evaluation of a time-weighted average (Model 3) plus four further models (Models 4–7), including a post hoc sensitivity analysis during the review process, are provided in the Supplementary material.

Results

Study population characteristics

A total of 2733 procedures were included in the final analysis (Fig. 1), undertaken in 2716 patients (Table 1). Sixty cases were included within the categorical grouping of ‘missing’ as a result of the absence of recorded/available BMI data.²⁷ Patients were ventilated with a median dynamic driving pressure¹² of 16.7 cm H₂O (interquartile range [IQR] 13.9–19.6) and median PEEP 5.0 cm H₂O (4.5–5.2). Further ventilatory details are provided in Supplementary Table S4.

Table 1.

Comparison of preoperative covariates. AUC, area under the curve; FEV₁, forced expiratory volume in 1 s; FiO₂, inspired oxygen fraction; MPOG, Multicenter Perioperative Outcomes Group.

Variable	Level	n	Overall n=2716	0–25% of AUC n=679	25–50% of AUC n=678	50–75% of AUC n=680	75–100% of AUC n=679	P-value
Total area (%∗min) above FiO2=80%: primary exposure variable		2716	814.0 (287.5, 1862.2)	65.5 (3.6, 168.3)	537.8 (416.5, 680.5)	1227.4 (993.5, 1549.2)	2722.5 (2248.0, 3310.3)	-
Age(yr)		2716	66.0 (57.0, 73.0)	65.0 (55.0, 73.0)	65.0 (55.0, 72.0)	66.0 (58.0, 73.0)	68.0 (60.0, 74.0)	<0.001
Sex	Female	2716	1504 (55.4%)	397 (58.5%)	384 (56.6%)	363 (53.4%)	360 (53.0%)	0.128
	Male		1212 (44.6%)	282 (41.5%)	294 (43.4%)	317 (46.6%)	319 (47.0%)
Race	American Indian or Alaska Native	2716	5 (0.2%)	2 (0.3%)	1 (0.1%)	0 (0.0%)	2 (0.3%)	<0.001
	Asian or Pacific Islander		147 (5.4%)	48 (7.1%)	44 (6.5%)	35 (5.1%)	20 (2.9%)
	Black, not of Hispanic origin		113 (4.2%)	38 (5.6%)	23 (3.4%)	25 (3.7%)	27 (4.0%)
	Hispanic, Black		1 (0.0%)	0 (0.0%)	0 (0.0%)	1 (0.1%)	0 (0.0%)
	Hispanic, White		55 (2.0%)	17 (2.5%)	16 (2.4%)	15 (2.2%)	7 (1.0%)
	Middle Eastern		1 (0.0%)	0 (0.0%)	0 (0.0%)	1 (0.1%)	0 (0.0%)
	Unknown race		139 (5.1%)	42 (6.2%)	33 (4.9%)	36 (5.3%)	28 (4.1%)
	White, not of Hispanic origin		2255 (83.0%)	532 (78.4%)	561 (82.7%)	567 (83.4%)	595 (87.6%)
MPOG Institution	MPOG Institution 1	2716	708 (26.1%)	110 (16.2%)	160 (23.6%)	177 (26.0%)	261 (38.4%)	<0.001
	MPOG Institution 2		39 (1.4%)	8 (1.2%)	9 (1.3%)	10 (1.5%)	12 (1.8%)
	MPOG Institution 3		315 (11.6%)	138 (20.3%)	65 (9.6%)	47 (6.9%)	65 (9.6%)
	MPOG Institution 4		1654 (60.9%)	423 (62.3%)	444 (65.5%)	446 (65.6%)	341 (50.2%)
Year of surgery	2012	2716	121 (4.5%)	15 (2.2%)	33 (4.9%)	29 (4.3%)	44 (6.5%)	<0.001
	2013		182 (6.7%)	18 (2.7%)	52 (7.7%)	52 (7.6%)	60 (8.8%)
	2014		150 (5.5%)	30 (4.4%)	22 (3.2%)	41 (6.0%)	57 (8.4%)
	2015		498 (18.3%)	113 (16.6%)	118 (17.4%)	120 (17.6%)	147 (21.6%)
	2016		638 (23.5%)	163 (24.0%)	157 (23.2%)	159 (23.4%)	159 (23.4%)
	2017		400 (14.7%)	136 (20.0%)	93 (13.7%)	89 (13.1%)	82 (12.1%)
	2018		316 (11.6%)	93 (13.7%)	90 (13.3%)	72 (10.6%)	61 (9.0%)
	2019		390 (14.4%)	107 (15.8%)	110 (16.2%)	110 (16.2%)	63 (9.3%)
	2020		21 (0.8%)	4 (0.6%)	3 (0.4%)	8 (1.2%)	6 (0.9%)
ASA physical status	1	2716	4 (0.1%)	2 (0.3%)	2 (0.3%)	0 (0.0%)	0 (0.0%)	0.007
	2		401 (14.8%)	92 (13.5%)	110 (16.2%)	123 (18.1%)	76 (11.2%)
	3		2181 (80.3%)	551 (81.1%)	532 (78.5%)	528 (77.6%)	570 (83.9%)
	4		130 (4.8%)	34 (5.0%)	34 (5.0%)	29 (4.3%)	33 (4.9%)
Height (cm)		2657	167.0 (160.0, 175.0)	166.5 (160.0, 174.0)	167.0 (160.0, 175.1)	167.0 (160.0, 175.3)	165.5 (159.0, 175.3)	0.653
Weight (kg)		2672	77.0 (65.0, 90.0)	70.8 (62.0, 84.5)	76.9 (64.7, 90.0)	79.2 (66.2, 91.4)	80.3 (68.1, 94.6)	<0.001
BMI (kg m−2)		2656	27.3 (24.0, 31.1)	25.8 (22.6, 29.1)	27.4 (24.1, 30.8)	27.7 (24.4, 31.3)	28.7 (25.3, 33.1)	<0.001
Predicted body weight (kg)		2656	59.6 (52.4, 69.7)	59.3 (52.4, 68.7)	60.1 (52.4, 69.6)	60.5 (52.4, 70.7)	59.3 (52.4, 70.7)	0.703
Tobacco use	1 Non-smoker	2715	845 (31.1%)	244 (36.0%)	244 (36.0%)	202 (29.7%)	155 (22.8%)	<0.001
	2 Prior smoker		1653 (60.9%)	368 (54.3%)	389 (57.4%)	424 (62.4%)	472 (69.5%)
	3 Current smoker		217 (8.0%)	66 (9.7%)	45 (6.6%)	54 (7.9%)	52 (7.7%)
Preoperative comorbidities	N	2714	2128 (78.4%)	552 (81.4%)	539 (79.6%)	535 (78.7%)	502 (73.9%)	0.007
	Y		586 (21.6%)	126 (18.6%)	138 (20.4%)	145 (21.3%)	177 (26.1%)
FEV1 missing	N	2716	2466 (90.8%)	599 (88.2%)	595 (87.8%)	619 (91.0%)	653 (96.2%)	<0.001
	Y		250 (9.2%)	80 (11.8%)	83 (12.2%)	61 (9.0%)	26 (3.8%)
FEV1 predicted		2466	92.0 (78.0, 105.0)	93.0 (80.0, 105.0)	93.0 (79.0, 106.0)	93.0 (81.0, 106.0)	90.0 (73.0, 103.0)	<0.001
Prior chemotherapy or radiation	N	2715	2376 (87.5%)	576 (85.0%)	604 (89.1%)	611 (89.9%)	585 (86.2%)	0.017
	Y		339 (12.5%)	102 (15.0%)	74 (10.9%)	69 (10.1%)	94 (13.8%)
Procedure	Bilobectomy	2716	69 (2.5%)	16 (2.4%)	11 (1.6%)	11 (1.6%)	31 (4.6%)	<0.001
	Lobectomy		1302 (47.9%)	289 (42.6%)	252 (37.2%)	265 (39.0%)	496 (73.0%)
	Segmentectomy		223 (8.2%)	67 (9.9%)	56 (8.3%)	55 (8.1%)	45 (6.6%)
	Wedge resection		1122 (41.3%)	307 (45.2%)	359 (52.9%)	349 (51.3%)	107 (15.8%)
Thoracotomy	N	2716	2171 (79.9%)	551 (81.1%)	582 (85.8%)	556 (81.8%)	482 (71.0%)	<0.001
	Y		545 (20.1%)	128 (18.9%)	96 (14.2%)	124 (18.2%)	197 (29.0%)
Total fluid input (L)		2716	1.0 (0.7, 1.4)	1.0 (0.7, 1.4)	0.8 (0.6, 1.2)	0.9 (0.7, 1.2)	1.3 (0.9, 1.8)	<0.001

Distribution of exposure of interest: FiO₂ >80%

The median of AUC of FiO₂ > 80% (primary exposure variable, Model 1) across all cases was 814.0%∗minutes (IQR: 287.5–1862.2%∗minutes). Illustrative perioperative traces (Fig. 2) and the distribution of the exposure variables used in each model are shown in Supplementary Figure S2.

Very high FiO₂ levels were used early in one-lung ventilation and FiO₂ declined over the case duration (Fig. 3a). Mean FiO₂ at reinflation following completion of one-lung ventilation was 82.7%. Discretionary high FiO₂ was common and frequent during lung resection, with 60% of cases using FiO₂ >0.80 (despite SpO₂ >98%) at the start of one-lung ventilation (Fig. 3b). Intraoperative Pao₂ values were only available in 198 cases.

Fig 3.

Trends in FiO₂ administration as a function of time after the start of one-lung ventilation. (a) Discretionary FiO₂ administration (25%, median, and 75%) as a function of time after the start of one-lung ventilation. (b) Percentage of cases using discretionary high FiO₂ (SpO₂ >98% at different FiO₂ thresholds) as a function of time after the start of one-lung ventilation. For any given time point on the x-axis, the height of the lines (25%, median, and 75%) utilises the measurements of only those patients whose one-lung ventilation period lasted to that point or beyond. Therefore, any downward trends may be due to FiO₂ levels decreasing over time for individual patients as one-lung ventilation continues, or the one-lung ventilation period ending for those patients receiving higher FiO₂, or both. FiO₂, inspired oxygen fraction; SpO₂, peripheral capillary oxygen saturation.

Primary outcome

A postoperative pulmonary complication was experienced by141/2733 (5.2%) (Table 2).

Table 2.

Outcomes table. Tracheal reintubation was originally proposed as a separate pulmonary metric in the composite outcome; however, reintubation was included within the respiratory failure outcome on STS versions 2.3 and 2.41. For v2.2, respiratory failure and reintubation were defined as separate outcomes, and both were included within the primary outcome composite. The outcomes included in each composite are not mutually exclusive. For example: the total number of postoperative pulmonary complications is 141 (however, the number of atelectasis (43) + pneumonia (72) + acute respiratory distress syndrome (ARDS) (10) + respiratory failure (42) + prolonged ventilator support (7) = 174). This is because a single patient can only qualify for the composite outcome once, despite having multiple qualifying complications.

Complication	Numerator	Population (n=2733)	Percentage
		Denominator
Postoperative pulmonary complications
Atelectasis requiring bronchoscopy	43	2731	1.6
Pneumonia	72	2733	2.6
ARDS	10	2731	0.4
Respiratory failure	42	2731	1.5
Initial ventilator support >48 h	7	2731	0.3
Composite postoperative pulmonary complication	141	2733	5.2
Major morbidity
Postoperative events	820	2733	30.0
Atrial arrhythmia (requiring treatment)	186	2731	6.8
Ventricular arrhythmia (requiring treatment)	5	2731	0.2
Myocardial infarction	5	2731	0.2
Deep venous thrombosis (requiring treatment)	9	2731	0.3
Ileus	14	2731	0.5
Empyema (requiring treatment)	11	2731	0.4
Surgical site infection	15	2601	0.6
Sepsis	8	2603	0.3
Other infection (requiring i.v. antibiotics)	13	2731	0.5
New central neurological event	9	2731	0.3
Delirium	22	2731	0.8
Renal failure (RIFLE criteria)	26	2731	1.0
Readmission (within 30 days of discharge)	159	2723	5.8
Mortality
Death at discharge	10	2733	0.4
30-day mortality	17	2733	0.6
Composite major morbidity	854	2731	31.3

Secondary outcomes

Major morbidity was experienced by 854//2733 (31.3%) and 17/2733 (0.6%) died within 30 days of surgery (Table 2; Supplementary Fig. S3a).

Univariate analyses

The incidence of postoperative pulmonary complications (P=0.005) and major morbidity (P<0.001) increased with each quartile of higher FiO₂ exposure (Supplementary Table S4). The highest FiO₂ exposure quartile featured higher driving pressure and tidal volumes, compared with the lowest quartile. There were no differences in PEEP utilisation between FiO₂ exposure quartiles (P=0.198). The time above 80% FiO₂ and SpO₂ of 98% during one-lung ventilation (P<0.001) and time-weighted average FiO₂ also increased with increasing AUC above the FiO₂ 80% threshold (Supplementary Table S4). A detailed comparison between the cohort developing pulmonary complications and those not developing pulmonary complications can be found in Supplementary Table S5a (preoperative covariates) and Supplementary Table S5b (intraoperative events and outcomes).

Multivariable logistic regression models

Primary definition for inspired oxygen concentration (area under the curve above FiO₂=80% threshold during one-lung ventilation) (Model 1)

Exposure to supplemental oxygen (AUC above FiO₂ 80% threshold) was associated with postoperative pulmonary complications (adjusted OR [aOR]: 1.17, 95% confidence interval [CI]: 1.04–1.33, P=0.012; Table 3). ASA Physical Status Classification System (base unit: 1-point increase in ASA status) (aOR: 2.27, 95% CI: 1.40–3.68, P<0.001), past smoking history (aOR: 1.97, 95% CI: 1.16–3.35, P=0.012), thoracotomy (aOR: 2.47, 95% CI: 1.63–3.74, P<0.001), FEV₁ in litres) (aOR: 0.98, 95% CI: 0.97–0.99, P<0.001), and patient's comorbidities (aOR: 1.55, 95% CI: 1.04–2.32, P=0.032) were also associated with postoperative pulmonary complications. The AUC of FiO₂ above 80% model had good discrimination (c-statistic = 0.769).²⁸

Table 3.

Multivariable logistic regression for primary definition for alveolar hyperoxia (Model 1). C-statistic: 0.769. Note: the adjusted odds ratio for the variable: ‘Presence of missing FEV₁ data’ is relative to those with ‘FEV₁ predicted’ both non-missing and equal to 0. FEV₁, forced expiratory volume in 1 s; FiO₂, fraction inspired oxygen concentration; MPOG, Multicenter Perioperative Outcomes Group; SpO₂, peripheral capillary oxygen saturation.

	Odds ratio	95% confidence interval		P-value
Area under the curve above FiO2=80% (1000%∗min)	1.17	1.04	1.33	0.012
MPOG Institution 1 (reference: 4)	0.59	0.369	0.95	0.029
MPOG Institution 2 (reference: 4)	0.47	0.096	2.26	0.344
MPOG Institution 3 (reference: 4)	0.50	0.256	0.97	0.040
Age (yr)	1.01	0.99	1.03	0.198
Female sex	0.98	0.67	1.41	0.894
BMI (kg m−2)	0.99	0.96	1.02	0.379
ASA physical status	2.27	1.40	3.68	<0.001
Past smoking history (reference: non-smoker)	1.97	1.16	3.35	0.012
Current smoker (reference: non-smoker)	1.58	0.71	3.50	0.258
Prior history of chemotherapy, radiation, or both	1.00	0.60	1.70	0.986
Bilobectomy/lobectomy (reference: wedge resection)	1.58	1.01	2.48	0.045
Segmentectomy (reference: wedge resection)	1.16	0.54	2.46	0.708
Thoracotomy (reference: video-assisted thoracoscopic surgery)	2.47	1.63	3.74	<0.001
Presence of preoperative comorbidities	1.55	1.04	2.32	0.032
FEV1 predicted	0.98	0.97	0.99	<0.001
Presence of missing FEV1 data	0.211	0.070	0.64	0.006
Total fluid input (L)	1.16	0.93	1.46	0.186

Sensitivity analyses

There was no association between time above the FiO₂=80% and SpO₂=98% thresholds during one-lung ventilation (Model 2; Supplementary Table S6) or time-averaged FiO₂ (Model 3; Supplementary Table S7). When the primary definition for high FiO₂ was extended to the entire period of mechanical ventilation as opposed to just the period of single-lung ventilation (Model 4), AUC above FiO₂ 80% threshold) remained significant (Supplementary Table S8). Excluding any cases with intraoperative desaturation (defined as SpO₂ <88% for more than three consecutive minutes) demonstrated results similar to that of the primary analysis (Model 5; Supplementary Table S9). AUC above FiO₂ 80% threshold remained associated with postoperative pulmonary complications (Model 6), using a revised definition for postoperative pulmonary complications that did not include atelectasis (Supplementary Table S10) with similar discrimination characteristics to Model 1 (c-statistics = 0.769). Post hoc analyses requested during the review process to further clarify the relationship between FiO₂ and duration of exposure are detailed in Model 7 (Supplementary material).

Discussion

We found that AUC above FiO₂ 80% threshold is associated with postoperative pulmonary complications. Because this exposure variable intrinsically contains both FiO₂ and duration of one-lung ventilation, we could not directly resolve the independent effects of each based upon the positive association shown in the primary model. However, by considering the individual effects of duration of exposure and FiO_2, our results do not support limiting the inspired oxygen fraction for the purpose of reducing postoperative pulmonary complications.

Sensitivity models were designed to better understand the impact of time exposure (Model 2) and FiO₂ (Model 3). Because average FiO₂ is independent of time (unit: percentage); duration of ventilation could be controlled for in Model 3. This sensitivity analysis revealed duration of ventilation, but not average FiO₂ was associated with postoperative pulmonary complications. Additionally, examination of discretionary high FiO₂—that is, the combination of FiO₂ levels >80% concomitant with SpO₂ values >98%—revealed no significant association with the primary outcome. Overall, these results are consistent with those of prior studies in thoracic surgery which demonstrated that duration of exposure to mechanical ventilation and duration of one-lung ventilation³ were independent risk factors for postoperative pulmonary complications,^3,29 but do not unambiguously support a relationship between inspired oxygen concentration and postoperative pulmonary complications in this clinical context.

Discretionary high FiO₂ was also not associated with postoperative pulmonary complications, further suggesting that prolonged ventilation, baseline pulmonary dysfunction or other disease states (and not inspired oxygen concentration) may play a larger role in the pathogenesis of postoperative pulmonary complications. Because atelectasis and the other outcomes (pneumonia, ARDS, respiratory failure, and prolonged ventilator support) differ markedly with regard to severity and clinical significance, inclusion of atelectasis within the composite definition remains controversial. Our findings were confirmed in a post hoc sensitivity analysis using a definition that excluded cases classified as having a postoperative pulmonary complication solely upon diagnosis of atelectasis.

Very high FiO₂ levels are used early in one-lung ventilation and, to a moderate extent, decline gradually during the one-lung ventilation period in lung resection surgery. This likely reflects the need to compensate for the obligatory intrapulmonary shunt resulting from one-lung ventilation and is perhaps an intentional effort by practitioners to facilitate absorption atelectasis in the nonventilated lung. Subsequent decreases in FiO₂ likely reflect the effects of hypoxic pulmonary vasoconstriction and perhaps surgical factors leading to decreased shunt. However, high FiO₂ levels are used, even when not strictly necessary based on the degree of systemic oxygenation (Fig. 3b); 60% of cases use FiO₂ >0.80 (despite SpO₂ >98%) at the start of one-lung ventilation and high SpO₂ is typically maintained during the period of one-lung ventilation (Fig. 3b). The detailed characterisation of practice patterns for oxygen administration during lung resection surgery are of possible utility in alerting providers to the potential for FiO₂ reduction.

Use of high FiO₂ during one-lung ventilation has been previously reported in a very small sub-cohort of thoracic surgical cases,³⁰ although practice in this regard has not been previously well characterised. Practice patterns related to inspired oxygen are of interest in thoracic surgery because of the tension between an increased oxygen requirement because of intrapulmonary shunting and the known or suspected relationships between high inspired oxygen concentration and oxidative stress, ischaemia reperfusion injury, and alveolar damage. High oxygen concentrations are known to be toxic to lung tissue,³¹ increasing pulmonary capillary permeability³² and contributing to oxidative protein damage in patients undergoing lung resection.³³ Oxidative stress is related both to the duration of one-lung ventilation and to the risk of postoperative complications.³⁴ Further, experimental evidence supports the contention that oxidative stress is exacerbated by hyperoxia¹ suggesting that minimising excessive oxygen exposure might attenuate this stress response and associated injury.

In general, clinical evidence directly linking alveolar hyperoxia to postoperative pulmonary complications is mixed. A paucity of evidence in the thoracic surgical population (which involves unique stressors including one-lung ventilation, direct surgical trauma, and loss of functional lung parenchyma) limits direct comparison with other studies. Randomised trials comparing FiO₂ 30% vs 80% in abdominal³⁵ and colorectal surgery³⁶ did not demonstrate a difference in the incidence of postoperative pulmonary complications. In a study in the nonthoracic surgery population, Staehr-Rye and colleagues⁴ utilised a definition of hyperoxia similar to that used in our secondary analysis and found that median FiO₂ was associated with respiratory complications in a dose-dependent manner, even when adjusted for duration of anaesthesia. Another prospective study demonstrated an association between time-weighted average FiO₂ during one-lung ventilation and postoperative pulmonary complications.² This contrasts with our primary finding in the thoracic population that, although duration of ventilation was neither time-weighted average FiO₂ nor time above FiO₂, thresholds were independently associated with postoperative pulmonary complications.

The study of higher vs lower FiO₂ in the thoracic surgical population is, of course, limited by need for generally higher FiO₂ during one-lung ventilation because of the obligatory right to left intrapulmonary shunt which in turn limits practice variation. Consequently, present study findings should be viewed in light of the somewhat restricted range of clinical practice related to oxygen administration. Although substantial experimental and circumstantial evidence implicate inspired oxygen as a risk factor and potential aetiologic factor in the genesis of lung injury after major thoracic surgery, the clinical significance remains incompletely elucidated. Nonetheless, substantial expert opinion supports efforts to minimise this exposure, particularly for lung cancer surgery.³⁷ Expert consensus recommendations for protective ventilation of the surgical patient include the use of the lowest inspired oxygenation concentration compatible with adequate oxygenation,³⁸ consistent with guidelines of the British Thoracic Society.³⁹ Results of the present study indicate that further reduction of FiO₂, consistent with optimal systemic oxygenation during one-lung ventilation for lung resection, is achievable. We have demonstrated that the incidence of discretionary high FiO₂ during one-lung ventilation remains high, particularly during the later stages of lung resection, including during lung reinflation (mean FiO₂ at time of re-inflation=82.7%). However, evidence from this study also suggests that practitioners are making an effort to reduce FiO₂ during one-lung ventilation. Overall, during lung resection surgery in this study, the mean FiO₂ decreased from 89.3% at the start of one-lung ventilation to 79.4% at the 3-h time point. The reported incidence of discretionary high FiO₂, that is high FiO₂ relative to the level of systemic oxygenation, may have utility in alerting anaesthesiologists to the potential for FiO₂ reduction.

Our study has notable strengths, as well as several limitations. This large multicentre study was well positioned to assess the impact of FiO₂ on postoperative pulmonary complications across a range of practice patterns and geographically diverse set of institutions. Furthermore, outcome data were obtained from well-validated, abstractor-adjudicated national databases utilising standardised metrics and definitions.⁴⁰ The retrospective nature of the study has inherent limitations, including the possibility of yet unidentified confounding variables. As the study approach does not enable us to account for how different clinical situations may influence FiO₂ selection, we attempted to address this concern via two approaches: (i) modelling discretionary high FiO₂ and (ii) performing a sensitivity analysis where cases with desaturation were excluded (defined as SpO₂ <88% for more than three consecutive minutes). To minimise bias, we have excluded cases from institutions not meeting a minimum research standard. Further, we have controlled for a number of variables which might otherwise introduce bias, including patient characteristic variables, comorbidities, physiologic trespass of surgery (resection type), institution, and additional risk predictors. Additionally, the decision to use a missing indicator for FEV₁ pulmonary function data, as opposed to a technique such as multiple imputation, can introduce bias into the analysis.⁴¹ Finally, SpO₂ measurements may introduce racial biases based on skin pigmentation, a variable not controlled in the present study. Black patients exhibited much higher rates of occult hypoxemia not detected by pulse oximetry compared with White patients.⁴²

In summary, in this multicentre observational cohort study, alveolar hyperoxia was not unambiguously associated with postoperative pulmonary complications after lung resection surgery. Our results do not support limiting the inspired oxygen fraction for the purpose of reducing postoperative pulmonary complications. Further study is necessary to better elucidate the relationship between oxygen exposure and postoperative pulmonary complications.

Authors’ contributions

Study conception: NJD, MES, RSB

Study design: NJD, MES, BIN, MRM, DAC, SK, SRC, LWM, WMP, NLP, RSB

Data analysis: MES

Data interpretation: NJD, MES, BIN, MRM, DAC, SK, SRC, LWM, WMP, NLP, RSB

Writing the manuscript: NJD, RSB

Critical revision of the manuscript: NJD, MES, BIN, MRM, DAC, SK, SRC, LWM, WMP, NLP, RSB

Acknowledgements

The authors acknowledge Shelley Vaughn and Tomas Medina Inchauste (Department of Anesthesiology, University of Michigan Medical School, Ann Arbor, MI, USA) for their contributions in data acquisition and electronic search query programming for this project. Additionally, we thank MPOG contributors Marcel E. Durieux (Department of Anesthesiology, University of Virginia School of Medicine, Charlottesville, VA, USA) and Stephen Patrick Bender (Department of Anesthesiology and Critical Care Medicine, The University of Vermont Medical Center, Burlington, VT, USA). We thank Elizabeth Jewell from the Department of Anesthesiology at Michigan Medicine for creative and technical support on study figures.

Declaration of interest

The authors declare that they have no conflicts of interest beyond those described in the funding statement.

Funding

Foundation for Anesthesia Education and Research (FAER) – Mentored Research Training Grant (MRTG-02-15-2020-Douville) (to NJD). National Institute of Diabetes and Digestive and Kidney Diseases (K08 DK131346 to NJD). National Heart, Lung, and Blood Institute Grants (K01HL141701 to MRM, K08HL159327 to DAC). This project was supported by departmental funding from University of Virginia and University of Michigan Anesthesiology departments. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

MPOG funding statement

Funding was also provided by departmental and institutional resources at each contributing MPOG site. In addition, partial funding to support underlying electronic health record data collection into the Multicenter Perioperative Outcomes Group registry was provided by Blue Cross Blue Shield of Michigan/Blue Care Network as part of the Blue Cross Blue Shield of Michigan/Blue Care Network Value Partnerships program. Although Blue Cross Blue Shield of Michigan/Blue Care Network and Multicenter Perioperative Outcomes Group work collaboratively, the opinions, beliefs and viewpoints expressed by the authors do not necessarily reflect the opinions, beliefs, and viewpoints of Blue Cross Blue Shield of Michigan/Blue Care Network or any of its employees.

Data availability statement

The dataset is governed by the MPOG Data Use Agreement, which allows it only to be shared with other MPOG DUA holders. A limited dataset would be available to other parties after publication upon execution of a Data Use Agreement and fulfilment of other regulatory requirements (including IRB approval).

Handling Editor: Gareth Ackland

Appendix A. Supplementary data

The following are the Supplementary data to this article: