Association of Intraoperative Ventilator Management with Postoperative Oxygenation, Pulmonary Complications, and Mortality

Abstract

Background:

“Lung-protective ventilation (LPV)” describes a ventilation strategy involving low tidal volumes (V_T) and/or low driving(ΔP)/plateau pressures (P_plat), and has been associated with improved outcomes after mechanical ventilation. We evaluated the association between intraoperative ventilation parameters (including PEEP, ΔP, and V_T) and three postoperative outcomes: (i) PaO₂/FiO₂, (ii) postoperative pulmonary complications (PPCs), and (iii) 30-day mortality.

Methods:

We retrospectively analyzed adult patients who underwent major non-cardiac surgery and remained intubated postoperatively from 2006–2015 at a single US center. Using multivariable regressions, we studied associations between intraoperative ventilator settings and lowest postoperative PaO₂/FiO₂ while intubated, pulmonary complications identified from discharge diagnoses, and in-hospital 30-day mortality.

Results:

Among a cohort of 2,096 cases, the median PEEP, was 5 cmH₂O, interquartile range=4–6), median delivered tidal volume was 520 mL (IQR=460–580), and median driving pressure was 15 cmH₂O (13–19). Following multivariable adjustment, intraoperative median PEEP (B=−6.04, 95% CI=−8.22- −3.87, p<0.001), median FiO₂ (B=−0.30, 95% CI=−0.50- −0.10, p=0.003), and hours with driving pressure >16 cmH₂O (B=−5.40, 95% CI=−7.2- −4.2, p<0.001) were associated with decreased postoperative PaO₂/FiO₂. Higher postoperative PaO₂/FiO₂ ratios were associated with a decreased risk of pulmonary complications (adjusted odds ratio [aOR] for each 100 mmHg =0.495, 95% CI=0.331 – 0.740, p=0.001, model C-statistic of 0.852) and mortality (aOR=0.495, 95% CI= 0.366 – 0.606, p<0.001, model C-statistic of 0.820). Intraoperative time with tidal volume greater than 500 mL was also associated with an increased likelihood of developing a postoperative pulmonary complication (aOR=1.06 /hour, 95% CI=1.00–1.20, p=0.042).

Conclusions:

In patients requiring postoperative intubation after noncardiac surgery, increased median FiO₂, increased median PEEP, and increased time duration with elevated driving pressure predict lower postoperative PaO₂/FiO₂. Intraoperative duration of tidal volume greater than 500 mL was independently associated with increased postoperative pulmonary complications. Lower postoperative PaO₂/FiO₂ ratios were independently associated with pulmonary complications and mortality. Our findings suggest that postoperative PaO₂/FiO₂ may be a potential target for future prospective trials investigating the impact of specific ventilation strategies for reducing ventilator-induced pulmonary injury.

Introduction

Positive pressure ventilation exposes the lungs to mechanical stress, which contributes to ventilator-associated lung injury.¹ In severe cases, lung injury may progress to acute respiratory distress syndrome (ARDS), with morbidity and mortality as high as 40–50%.^2,3 Lung-protective ventilation (LPV) strategies reduce mechanical stress by employing low tidal volumes (V_T), driving pressures (ΔP), and positive end-expiratory pressures (PEEP) and have improved survival in critically ill patients with acute lung injury.^4,5 In contrast, studies of LPV in the operating room have yielded mixed conclusions with respect to postoperative pulmonary complications.^6–15 This lack of clarity has been in part attributed to incomplete adoption as recent estimates suggest that up to 14% of patients do not receive LPV intraoperatively.¹³ Varying definitions of LPV and evolving research methods may also have played a role as earlier studies dichotomized ventilation strategies, preventing separate evaluation of each ventilator parameter.⁶ In addition, control arms of existing studies have at times used high V_T, zero PEEP, and no recruitment maneuvers, have been criticized for not representing standard care.¹⁴ Finally, when compared to critically ill patients, a lower incidence of pulmonary complications, ARDS, and mortality in perioperative patients has limited the ability to adequately power studies of intraoperative ventilation strategies.

Intermediate laboratory values may offer an approach to clarifying the effect of intraoperative ventilator strategies. PaO₂/FiO₂ is frequently obtained in patients remaining intubated following major surgery. Although PaO₂/FiO₂ is a component of the definition of ARDS,¹⁶ associations between PaO₂/FiO₂, lung injury, and mortality, are not clear.^2,3,16

To investigate the relationship between intraoperative ventilation practices, postoperative PaO₂/FiO₂ and PPCs, we performed this observational study in a major non-cardiac surgical population. We hypothesized that non-use of LPV is associated with (i) worse postoperative PaO₂/FiO₂, (ii) more PPCs, and (iii) higher 30-day mortality, following adjustment for known risk factors.^17,18 We further hypothesized that lower postoperative PaO₂/FiO₂ is independently associated with PPCs and mortality.

Materials and Methods

Patient Population

Inclusion criteria for the study were adult patients (≥ 18 years) who underwent non-cardiac surgery with general anesthesia, tracheal intubation, case duration greater than 180 minutes (to ensure clinically significant exposure to intraoperative mechanical ventilation, an independent risk factor for PPCs)¹⁹ and who remained intubated or were re-intubated within two postoperative days. We studied cases between January 1, 2006 and September 30, 2015. Additional details about our study methodology including full exclusion criteria can be found in Supplementary Information, Appendix 1.

Data collection

Study data were collected via combined queries of the electronic perioperative anesthesia database (Centricity®, General Electric Healthcare, Waukesha, WI) and the hospital electronic medical records (Epic, Verona, WI). Perioperative anesthesia data were extracted from the local University of Michigan Multicenter Perioperative Outcomes Group (MPOG) dataset. Within the MPOG database, data are stored, validated, and extracted for quality improvement and research purposes.²⁰ We collected information on procedural class from Anesthesia current procedural terminology (CPT) coding and patient comorbidities using groupings based upon ICD-9/10 classifications based upon the Elixhauser Comorbidity Index.²¹ Details on ventilators used at our institution can be found in Supplementary Information, Appendix 2.

We studied seven intraoperative ventilator parameters for each patient: V_T, adjusted V_T using predicted body weight (PBW), driving pressure, PEEP, FiO₂, hypoxia (SpO₂ < 90%), and hyperoxia (SpO₂ > 97%). As the intraoperative provider may have adjusted PEEP and FiO₂ in response to SpO₂, we adjusted for this possibility by including both intraoperative FiO₂ and SpO₂ covariates in our models. Additionally, intraoperative FiO₂ was included as an independent ventilator parameter in our model, to account for variability in a patient’s postoperative PaO₂/FiO₂ when titrating FiO₂ despite presumed unchanged lung function.²² Furthermore, the ideal oxygenation level remains controversial, as intraoperative hyperoxia has been associated both with increased mortality and complications,^23–25 and a lower rate of surgical site infections.²⁶ PBW was calculated as: PBW for males = 50 kg + 2.3 kg x (Height [in] – 60); PBW for females = 45.5 kg + 2.3 kg x (Height [in] – 60).^5,27 Physiologically, driving pressure is commonly defined as: P_plat – PEEP. In situations in which P_plat was not routinely recorded, as with the interface between our anesthesia ventilators and electronic medical record, peak inspiratory pressure (PIP) was substituted for P_plat to calculate the driving pressure.²⁸

We first obtained all ventilator data collected from the anesthesia machine ventilator and gas analyzer every 60 seconds in the local MPOG database, then determined minimum, 5%, 25%, 50%, 75%, 95%, and maximum values for each patient’s FiO₂, PEEP, driving pressure, tidal volume, adjusted tidal volume using PBW, and SpO₂. We used SpO₂ as a surrogate for oxygenation as many patients did not have intraoperative blood gases. As we hypothesized that clinical injury might be impacted by a combination of overall ventilation strategy (characterized by the 50% values), extremes of intraoperative ventilation (characterized by minimum, 5%, 25%, 75%, 95%, and maximum values), and total duration of non-protective ventilation; we also determined the total amount of time the patient spent above “non-protective” thresholds. We proposed initial thresholds based upon previous literature (V_T - PBW > 8 mL/kg PBW⁵ and driving pressure > 16 cmH₂O¹²) and considered additional thresholds expanded at regular intervals (Duration with driving pressure > 16 cmH₂O, 18, 20, 22, 24; V_T > 500 mL, 600, 700, 800; V_T using PBW > 8 mL/kg, 10, 12, 14) (See Supplementary Information, Table S1 for full list of variables considered and Supplementary Information, Figure S1 for additional details on our variable selection methodology). Because use of duration exceeding a threshold has not been previously published, we included both descriptive and threshold variables for our two tidal volume parameters: 50% V_T and duration with V_T > 500 mL; 50% V_T using PBW and duration with V_T using PBW > 8 mL/kg. The descriptive and threshold variables considered can be found in the Supplementary Information, Table S1.

Next, we performed univariate analyses for all ventilator parameters – Mann-Whitney U test for non-normally distributed, continuous variables – prior to selecting either one or two parameters from each class to use for multivariable regressions.

Outcomes

The primary outcome was the minimum PaO₂/FiO₂ value recorded while mechanically ventilated during the first two postoperative days. Secondary outcomes were composite PPCs and 30-day in-hospital mortality. Composite PPCs were divided into three classes based upon discharge ICD-9/10 diagnoses. Class 1 PPCs were narrowly defined based upon the Agency for Healthcare Research and Quality Definitions for Respiratory Complications and included diagnoses for transfusion related lung injury (TRALI), ventilator associated pneumonia, and post-procedural aspiration pneumonia (full ICD 9 and 10 diagnoses included with our definition for each class of PPC can found in Supplementary Information, Table S2).²⁹ Class 2 PPCs comprised an expanded subset of diagnoses consistent with European Perioperative Clinical Outcome (EPCO) definitions.³⁰ These include pneumonia, respiratory failure, and pneumothorax. Class 3 PPCs were defined as all pulmonary or respiratory diagnoses, regardless of causative mechanism, and were expanded to include diagnoses as varied as pulmonary embolism and surgical subcutaneous emphysema. While other pulmonary complications may not have the high individual risk of death of ARDS, they are more common,²⁸ and cumulatively, they may have a high attributable mortality.³¹

Power Analysis

Based on previous studies,^23,24 we defined a PaO₂/FiO₂ change of 10% as clinically meaningful. A sample size of 1044 in each group will have 80% power to detect a difference in means of 20 (group 1 mean = 200, group 2 mean = 180), assuming a common standard deviation of 163 and using a two group two-sided t-test with alpha =0.05 significance level. This power analysis was performed using nQuery + nTerim 4.0 prior to data extraction.

Statistical Analysis

Normality of all covariates and ventilator parameters was assessed using the Kolmogorov-Smirnov test of normality. Perioperative characteristics were summarized using means and standard deviations for normally distributed continuous covariates, medians and interquartile range (IQR) for non-normally distributed continuous variables, and counts and percentages for categorical covariates. To determine factors associated with pulmonary complications and mortality, we first compared covariates using chi-squared for categorical variables, t-test for continuous variables following a normal distribution, and Mann-Whitney U test for continuous variables not-following a normal distribution. A missing data analysis was performed to assess ways that the subset of patients with missing data (subsequently excluded from additional analysis) differed from the cohort of patients with full data (included in the full analysis). Statistical analysis was performed in R version 3.5.1.³²

We used multivariable regression models to determine associations between ventilator settings and our three outcomes: minimum PaO₂/FiO₂, PPCs, and mortality. To analyze the continuous primary outcome, minimum PaO₂/FiO₂ we performed a multivariable linear regression with variable selection by least absolute shrinkage and selection operator (LASSO) to identify which demographics, comorbidities, and ventilator measurements were independently associated with minimum PaO₂/FiO₂. We chose LASSO for covariate reduction in the logistic regression models as LASSO performs both variable selection and regularization to improve the accuracy and interpretability of the model, particularly when there are many variables and few outcomes.³³ Additionally, year of surgery was included as a covariate to control for changes in practice patterns over the 9 year study period.

We used LASSO using the glmnet package in R to select variables for inclusion in our final models.³⁴ This method used 10-fold cross-validation with a sequence of 100 lambda values, which were automatically generated by the function. Each lambda value corresponded to a percentage of the deviance that was explained by the variables included at lambda. The cross-validation estimated the largest lambda that keeps the mean squared error (MSE) within one standard error of the minimum MSE of the model. The variables included in our final model were chosen with this lambda. This approach generates a parsimonious model while keeping the error relatively small. LASSO uses L1 regularization with a penalty which improves performance when the number of possible variables is large for the number of outcomes.³³

Following variable selection, we performed multivariable logistic regressions to evaluate the risk-adjusted relationship between ventilator measurements and our two binary outcomes, mortality and PPCs. To evaluate the suitability of our primary outcome (minimum PaO₂/FiO₂) as an intermediate outcome associated with either secondary outcome (PPC or mortality), minimum PaO₂/FiO₂ was included along with all other covariates in both subsequent logistic regressions. All multivariable models were adjusted for demographics, ASA status, comorbidities, operative procedure, intraoperative crystalloid, colloid, and individual blood component administered volumes, and urine output. Discrimination was measured as the receiver operator characteristic area under the curve (C-statistic). These models allow for evaluation of the independent association of the ventilator measurements with both minimum postoperative PaO₂/FiO₂and long-term outcomes of PPCs and mortality. We summarize our analytic plan in the Supplementary Information, Figure S1.

Results

A total of 3,406 cases initially met our inclusion criteria. Following removal of cases with missing data not able to be described as a categorical variable or imputed, 2,096 cases were included in our final analysis (Figure 1). Cases missing minimum PaO₂/FiO₂, median FiO2, PEEP, V_T, V_T PBW, or driving pressure were excluded from the analysis. The most common surgeries were: upper abdominal surgery (n=773, 37%), neurosurgery (n=294, 14%), interventional radiology procedure (n=244, 12%), lower abdominal surgery (n=211, 10%), otolaryngology/head and neck surgery (n=159, 8%), and spine/spinal cord (n=139, 7%).

Figure 1.

Derivation of Study Cohort

When compared to cases with incomplete data, those with complete data tended to be older (60 ± 14 years versus 58 ± 17 years, p < 0.001) and had more comorbidities. Patients with lower PaO₂/FiO₂ were older (60 ± 14 versus 58 ± 16, p = 0.001), higher percentage male (60.6 versus 54.9, p < 0.001), more likely to be smokers (24.3 versus 17.6, p < 0.001), and had a larger BMI (31.8 ± 10.4 versus 29.5 ± 10.7, p < 0.001) than those in the higher PaO₂/FiO₂ group (Table 1).

Table 1.

Baseline Patient and Procedural Characteristics

	Minimum PaO2/FiO2 ≥ 200 mmHg	Minimum PaO2/FiO2 < 200 mmHg	t-test
	N=1055 Mean (SD)	N=1041 Mean (SD)
Age at Surgery (years)	58 (16)	60 (14)	0.010
Body mass index (kg/m2)	29.5 (10.7)	31.8 (10.4)	<0.001
Preoperative SpO2 (%)	96.8 (4.4)	95.4 (6.6)	<0.001
	n (%)	n (%)	χ 2
Female Sex	476 (45.1)	410 (39.4)	0.009
ASA Status
ASA 1:	13 (1.2)	5 (.05%)	0.103
ASA 2:	203 (19.2)	122 (11.7)	<0.001
ASA 3:	493 (46.7)	544 (52.3)	0.013
ASA 4:	346 (32.8)	368 (35.4)	0.235
Emergent Surgery	261 (24.7)	336 (32.3)	<0.001
Tobacco Use
Current:	186 (17.6)	253 (24.3)	<0.001
Former:	362 (34.3)	387 (37.2)	0.186
Non-smoker:	393 (37.3)	289 (27.8)	<0.001
Missing/Unknown:	114 (10.8)	112 (10.8)	1.000
Preoperative LVEF:
Normal (>55%):	177 (16.8)	194 (18.6)	0.290
Borderline (40–55]:	28 (2.7)	25 (2.4)	0.819
Low (25–40]:	11 (1.0%)	12 (1.2)	0.974
Severe <=25%:	9 (0.9%)	7 (.07)	0.823
No Preop Echo:	830 (78.7)	803 (77.1)	0.427
Comorbidities
AIDS/HIV:	6 (0.6)	0 (0.0)	0.043
Alcohol Abuse:	93 (8.8)	112 (10.8)	0.154
Cardiac Arrhythmias:	549 (52.0)	637 (61.2)	<0.001
Chronic Pulmonary Disease:	324 (30.7)	373 (35.8)	0.015
Congestive Heart Failure:	179 (17.0)	232 (22.3)	0.003
Diabetes:	382 (36.2)	370 (35.5)	0.312
Fluid Electrolyte Disorders	429 (40.7)	604 (58.0%)	<0.001
Liver Disease:	303 (28.7)	321 (30.8)	0.312
Metastatic Cancer:	321 (30.4)	258 (24.8)	0.005
Procedure
Neurosurgical/ENT:	336 (31.8)	256 (24.6)	<0.001
Chest/Abdomen/Pelvis:	516 (48.9)	589 (56.6)	<0.001
Extremities:	81 (7.7)	62 (6.0)	0.140
Other:	122 (11.6)	134 (12.9)	0.397

Histograms showing the distribution of the variables ultimately selected from our univariate analysis are included in Figure 2. As all ventilator parameters were non-normally distributed (p<0.05), we reported descriptive statistics on our ventilator parameters as median and interquartile range (IQR). We observed median V_T was 520 mL (IQR=401–580) or 8.3 mL/kg PBW (7.4–9.3) and FiO₂ 0.54 (0.45–0.79). Seventy-five percent (n=1,571) of patients received a median PEEP between 4–6 cmH₂O. Only 151 patients (7.2%) received low median PEEP, defined as ≤ 2 cm H₂O.¹⁵

Figure 2.

Distribution of Ventilation Parameters (A) Histogram of Median Tidal Volumes, (B) Duration of Time with Tidal Volume > 500 mL, (C) Median Tidal Volume as a Function of PBW, (D) Duration of Time with Tidal Volume > 8 mL/kg, (E) Median Driving Pressures, (F) Duration with Driving Pressure > 16 cmH₂O, and (G) Median FiO₂, (H) Duration with SpO₂ < 90%, (I) Median PEEP.

Variable Selection (Results of Univariate Analysis)

Following inspection of ventilator data and hand review of a sample of intraoperative records among 5% of cases analyzed, we determined that minimum and maximum values were prone to artifact³⁵– such as during a recruitment maneuver or if the ventilator was briefly disconnected or occluded. For this reason, we removed minimum and maximum values from the analysis. We observed that per-case 5th and 25th percentile values had high correlation. Analogously, 75th and 95th percentile values were highly correlated (the correlation coefficient between 75% V_T and 95% V_T was 0.92). We therefore selected 25%, 50%, and 75% thresholds for ventilator variables as providing the most useful descriptive information. Full detailed results of our univariate analysis across 25%, 50%, and 75% values for each ventilation parameter included in our initial univariate analysis can be found in the Supplementary Information, Table S1. As these ventilator parameters were non-normal, continuous variables we performed the Mann-Whitney U test to assess significance.

Based upon results of univariate analyses we selected nine ventilator variables to be include in multivariable regressions: Duration with SpO2 < 90% (hypoxia), Duration with SpO₂ > 97% (hyperoxia), Median FiO₂, Median PEEP, Duration with ΔP > 16 cmH₂O, Duration with V_T > 500 mL, Median delivered V_T, Duration with adjusted V_T using PBW > 8 mL/kg, and Median V_T using PBW. Full results of the univariate analysis are shown in Supplementary Information, Table S1.

The variables considered and ultimately selected at each stage of our analysis, as well as the integration of our multiple part analytical plan are detailed in Supplementary Information, Figure S1.

Primary Outcome: Minimum postoperative PaO₂/FiO₂

The median postoperative PaO₂/FiO₂ was 201 mmHg (IQR=125–278). Seventy-nine percent of our cases had a PaO₂/FiO₂ < 300 mmHg (Full Distribution of Primary Outcome, PaO₂/FiO₂ is shown in Figure 3). Using multivariable linear regression to adjust for other factors that may impact our outcome (such as: year of surgery, patient age, preoperative SpO₂, and Tobacco Use), we found that each additional hour with driving pressure > 16 cmH₂O (B = −5.4 mmHg, 95% CI = −7.2 to −4.2, p < 0.001), higher PEEP (B= −6.04 mmHg/cmH₂O, 95% CI = −8.22 to −3.87, p < 0.001), each 0.10 increase in median intraoperative FiO₂ (B= −3.00 mmHg, 95% CI = −5.00 to −1.00, p = 0.003), and each additional hour intraoperative SpO₂ < 90% (B= −9.6, 95% CI = −16.8 to −3.0, p = 0.005) were all independently associated with lower minimum postoperative PaO₂/FiO₂, with R-square of 0.235 (Table 2).

Figure 3.

Distribution of Primary Outcome - Minimum Postoperative PaO₂/FiO₂

Table 2.

Multivariable Linear Regression Model for Primary Outcome - Minimum Postoperative PaO₂/FiO₂ value recorded (while mechanically ventilated) on post-operative day 0, 1, or 2.

	Minimum Postoperative PaO2/FiO2
	B	95% CI	P Value
Year of Surgery	4.77	2.27 to 7.28	<0.001
Patient Weight (kg)	−0.14	−0.34 to 0.05	0.144
Patient Age (years)	−0.89	−1.19 to −0.60	<0.001
ASA Class 2	4.74	−7.51 to 17.00	0.448
Surgery to Chest/Abdomen/Pelvis (based upon CPT Code)	−12.10	−21.39 to −2.81	0.011
Prior History of Coagulopathy	−13.49	−23.42 to 3.56	0.008
Prior Fluid or Electrolyte Disorders	−27.99	−36.49 to −19.50	<0.001
Prior Peripheral Vascular Disease	−10.60	−20.04 to −1.16	0.028
No Prior Tobacco Use	11.54	2.19 to 20.88	0.016
Current Tobacco Use	−22.99	−33.97 to −12.02	<0.001
Preoperative SpO2 (%)	1.36	0.62 to 2.10	<0.001
Normal Preoperative White Blood Cell Count (range: 4.0–10.0 K/μL)	12.33	3.95 to 20.71	0.004
Low Preoperative Hemoglobin ( <12.0 g/dL)	16.9	8.08 to 25.71	<0.001
Preoperative Supplemental Oxygen Requirement	−36.59	−48.05 to −25.12	<0.001
Intraoperative Transfusion of Fresh Frozen Plasma (unit)	−0.56	−1.46 to 0.34	0.222
Intraoperative Transfusion of Platelets (5-pack unit)	−2.35	−8.04 to 3.35	0.419
Intraoperative Transfusion of Cryoprecipitate (vs. no transfusion)	−15.59	−36.11 to 4.92	0.136
Intraoperative Albuterol use	−3.83	−7.11 to −0.56	0.022
Intraoperative Furosemide use	−11.69	−27.56 to 4.18	0.149
Intraoperative urine (L)	7.41	1.82 to 13.00	0.009
Intraoperative Extubation	−50.93	−61.85 to −40.01	<0.001
Median Delivered Tidal Volume (mL)	−0.05	−0.11 to 0.00	0.068
Median PEEP (cmH 2 O)	−6.04	−8.22 to −3.87	<0.001
Median FiO2 (0.01)	−0.30	−0.50 to −0.10	0.003
Duration of Time with Driving Pressure > 16 cmH 2 O (hours)	−5.4	−7.2 to −4.2	<0.001
Duration of Time with SpO2 < 90% (hours)	−9.6	−16.8 to −3.0	0.005
Duration of Time with SpO2 > 97% (hours)	0.6	−0.6 to 1.8	0.257

30-Day Mortality

Of the 2,096 cases, 112 (5.3%) died within 30 days of their procedure. We found that each 100 mmHg increase in minimum postoperative PaO₂/FiO₂ was independently associated with a halving of the odds of death (aOR = 0.495, 95% CI = 0.366–0.606, p < 0.001). Age, reduced cardiac ejection fraction on pre-operative echocardiography, comorbidities, and intraoperative transfusion of packed red blood cells were also independently associated with mortality (see Table 3, c-statistic = 0.820). No ventilation parameter was independently associated with mortality.

Table 3.

30-Day Mortality and Class 1 Pulmonary Complications Multivariate Regressions

	30-Day Mortality			Pulmonary Complications
	OR	95% CI	P Value	OR	95% CI	P Value
Minimum PaO2/FiO2 (100 mmHg)	0.495	0.366–0.606	<0.001	0.495	0.331–0.740	0.001
Median FiO2 (0.01)				1.013	0.998–1.028	0.095
Time with VT > 500 mL (hours)				1.062	1.000–1.197	0.042
Patient Age (years)	1.030	1.014–1.047	<0.001
Surgery Year				0.647	0.523–0.802	<0.001
ASA Class 4	1.297	0.836–2.014	0.246
Surgery class: Burn, Radiologic, Other (based upon CPT)	2.000	1.172–3.413	0.011
Prior Congestive Heart Failure	1.413	0.895–2.229	0.138
Prior Cardiac Arrhythmias				3.285	1.508–7.155	0.003
Prior Coagulopathy				0.514	0.212–1.243	0.140
Prior Anemia				0.336	0.041–2.748	0.309
Prior Depression				0.556	0.206–1.499	0.246
Prior Fluid or Electrolyte Disorders	2.429	1.512–3.900	<0.001
Prior Neurological Disorders	2.065	1.292–3.299	0.002
Preoperative Platelets (150–400 K/μL)				2.406	1.042–5.555	0.040
Intraoperative Transfusion of packed Red Blood Cells	1.024	1.004–1.044	0.016
Intraoperative Nitric Oxide				1.037	1.006–1.069	0.017
Last Train of Four Measured 3/4				2.241	0.699–7.182	0.174
Patient extubated in PACU				3.792	0.779–18.445	0.099
Low Cardiac Function on Preoperative Echocardiography	3.093	1.013–9.446	0.047	15.692	3.177–77.511	0.001

Composite Postoperative Pulmonary Complications (PPCs)

We found that 44 of the 2,140 cases (2.1%) experienced a Class 1 Pulmonary Complication, 789 (36.9%) a Class 2 Pulmonary Complication, and 839 (39.2%) a Class 3 Pulmonary Complication. The most common diagnoses for Class 2 and 3 were Acute respiratory failure following trauma and surgery [ICD-9: 518.51] (12.6%), Respiratory insufficiency [ICD-9: 786.09] (10.1%), and Pulmonary insufficiency following trauma and surgery [ICD-9: 518.5] (6.0%) (Supplementary Information, Table S3).

After accounting for covariates likely associated with pulmonary complications, our multivariable logistic regression with LASSO variable selection demonstrated that each 100 mmHg higher minimum PaO₂/FiO₂ was associated with a lower likelihood of developing pulmonary complications (Class 1: aOR = 0.495, 95% CI = 0.331 – 0.740, p = 0.001, C-statistic = 0.852; Class 2: aOR=0.819, 95% CI = 0.740 – 0.905, p < 0.001, C-statistic = 0.618; and Class 3: aOR=0.819, 95% CI = 0.740 – 0.819, p < 0.001, C-statistic = 0.621). Additionally, time (hours) with V_T > 500 mL (aOR = 1.06, 95% CI = 1.00 – 1.20, p = 0.042), prior history of cardiac arrhythmia, intraoperative nitric oxide use, low cardiac ejection fraction on preoperative echocardiography, and earlier year of surgery (Table 3) were associated with a higher likelihood of developing a Class 1 pulmonary complication. Multivariable logistic regressions with LASSO variable selection for each of the more expansive composite pulmonary complications also showed that more recent year of surgery, prior peripheral vascular disease, and lower preoperative hemoglobin were associated with higher rates of both Class 2 and Class 3 Pulmonary Complications (Supplementary Information, Tables S4 and Table S5). LASSO variable selection excluded all ventilation parameters from the Class 2 and 3 Pulmonary Complications models.

Discussion:

In patients undergoing major non-cardiac surgery with postoperative intubation and mechanical ventilation, we retrospectively observed that ventilator settings not consistent with LPV were associated with a lower postoperative PaO2/FiO2 ratio, and that a lower postoperative PaO₂/FiO₂ was associated with increased PPCs and mortality. We also observed that increased PEEP, FiO₂, and time with ΔP > 16 cmH₂O are associated with lower postoperative PaO₂/FiO₂. Our analysis demonstrated that following adjustment for SpO₂, higher intraoperative FiO₂ remained associated with PPCs, in agreement with a retrospective study.³⁶

Our findings are generally consistent with an extensive literature finding a beneficial effect of LPV on mortality and pulmonary complications in patients receiving mechanical ventilation.^6–8,12 However, our observation that ΔP was not associated with mortality differs from a 2015 analysis of ICU patients with ARDS, which found that lower ΔP was the ventilation parameter most strongly associated with improved survival.³⁷ Potential reasons for this discrepancy include a lower overall mortality in our patients as compared to those in the 2015 study,³⁷ a decreased duration of ventilation exposure in the OR, and stronger link between threshold-based pressure variable than ΔP.

Our finding that PEEP was not associated with mortality is similar to a 2015 analysis of randomized trials in ARDS.³⁷ and suggests that PEEP is the component of LPV least likely to affect survival. The use of low PEEP control cohorts not representative of clinical practice, and changes in practice regarding routine PEEP, have confounded analyses of the contribution of PEEP to overall LPV.^6,11 In our study, a narrow range of PEEP values (IQR 4 −6) and small number of low PEEP patients (7.2% receiving PEEP ≤ 2 cmH₂O) may have limited our ability to find associations with adverse outcomes. Additionally, our study is limited because other reasons, such as elevated intracranial pressure, may have affected PEEP management. While we found that higher PEEP was associated with a lower minimum postoperative PaO₂/FiO₂, we cannot determine whether the anesthesiologists adjusted PEEP based upon patient factors, intraoperative conditions, or to treat intraoperative hypoxia which persisted into the postoperative period.

Our analysis raises the possibility that postoperative PaO₂/FiO₂ ratio may be a direct consequence of intraoperative ventilator settings and be an independent predictor of postoperative outcomes. Our patients naturally represent a group at high risk for PPCs and mortality as they all required postoperative intubation and mechanical ventilation. PaO₂/FiO₂ ratio may thus represent a comorbidity finding unrelated to intraoperative ventilator settings. Such a possibility may expand the ability to study effects of mechanical ventilation on perioperative outcomes. Because the frequency of ARDS is lower in this patient population and variability in surgical procedure may contribute to perioperative outcomes, endpoints such as mortality may not be sufficiently specific to allow an accurate assessment of the effect of LPV on outcome.^17,36,38

Our study has multiple limitations. As a single center effort, our results may not be generalizable to other institutions or populations. Our patient population also noted high rates of baseline comorbidities. Because liver transplantation and intracranial neurosurgery (10%, 14% of our population, respectively) are routinely intubated postoperatively, both procedures were disproportionately represented in our dataset. Nevertheless, our data suggests that in this patient population ventilator settings may play an important role in postoperative pulmonary outcomes. In addition, although we adjusted for several non-ventilation related confounders, including fluid and blood administration and urine output, it is possible that we may have missed a confounding variable potentially relevant to outcome, such as, atelectasis from laparoscopic versus open surgery.

Another limitation was the large number of patients with missing data and that the missing data cohort was younger and healthier than the analyzed population. Additionally, the link between ARDS, PaO₂/FiO₂, V_T, and mortality is incompletely understood.³⁹ Although PaO₂/FiO₂ is a diagnostic criterion for ARDS,¹⁶ studies in patients with ARDS demonstrate improved PaO₂/FiO₂ (on days 1 and 3), but also increased mortality in groups treated with higher V_T.⁵ To address this concern, we confirmed an independent association between PaO₂ and mortality (and pulmonary complications). We caution that our data do not prove a causal relationship between oxygenation and mortality, but that oxygenation might have potential use as an intermediate outcome. For example, we could not distinguish disease processes associated with both higher driving pressures and PPCs (e.g. poor lung compliance secondary to pre-existing lung injury) from higher ΔP causing PPCs via barotrauma. We attempted to control for underlying pulmonary disease processes associated with PPCs by including FiO₂ and SpO₂ in our models, however, unmeasured markers of pre-existing pulmonary dysfunction may still be present. As our results are hypothesis generating, future prospective studies implementing PaO₂/FiO₂ are necessary to validate the utility of PaO₂/FiO₂ as an intermediate outcome.

Although risk adjustment for multiple comorbidities is a strength of our analyses, there are limitations associated with use of the Elixhauser Comorbidity Index.²¹ ICD-9/10 based diagnoses are often non-specific and reflect conditions with markedly different severity and definitions. Additionally, clinicians may choose to code postoperative mechanical ventilation as either respiratory insufficiency or failure, possibly confounding our analysis of pulmonary outcomes. To minimize potential for error, we performed a limited review to confirm validity of administrative data, however, we did not formally review records and diagnostic imaging at a patient-level. A final limitation is that our analysis is confined to ventilation management in the operating room and does not assess postoperative ventilation.

In summary, we demonstrate that lower postoperative PaO₂/FiO₂ was associated with increased PPCs and mortality following non-cardiac surgery. Time with V_T > 500 mL and higher median FiO₂ were also associated with Class 1 PPCs, and their effects may be assessed using postoperative PaO₂/FiO₂. Among intraoperative parameters, median PEEP, median FiO₂, time with ΔP > 16 cmH₂O, and time with SpO₂ < 90% were associated with decreased postoperative PaO₂/FiO_2. The use of the intermediate outcome, PaO₂/FiO₂, thus offers a promising target for future prospective studies investigating the impact of intraoperative LPV strategies on postoperative outcomes.

Supplementary Material

Supplementary Information, Table S1

P-values were determined using Mann-Whitney U Test. ΔP = driving pressure; V_T = Tidal Volume; and V_T – PBW = Tidal Volume as a function of Predicted Body Weight. We determined 25%, 50%, and 75% values for each ventilator variable provided the most useful descriptive information. In practice, SpO₂ – 50% is the median value for all intraoperative SpO₂ values for a given patient, SpO2 – 25% is the 25^th percentile value of all intraoperative SpO₂ values for a given patient. We also selected key threshold variables (for example: V_T < 8 mL/kg PBW and ΔP < 16 cm H₂O) based upon literature and then expanded the threshold at regular intervals. Table S1 shows the full list of descriptive and threshold variables tested in our initial univariate analysis, as well as median and intra-quartile range for each variable in the cohort with and without each of our clinical outcomes (class 1 PPC, class 2 PPC, class 3 PPC, and 30-Day Mortality). Ventilation variables selected for inclusion in the full multivariable analysis are listed in bold in Table S1. Because the threshold variable was more significant in some clinical outcomes, while the descriptive variable was more significant in others, we ultimately decided to include both descriptive and threshold variables for our two tidal volume parameters: 50% V_T and duration with V_T > 500 mL; 50% V_T – PBW and duration with V_T - PBW > 8 mL/kg.

Supplementary Information, Table S2

Note: the list of Class 2 complications includes the full list of Class 1 complications, and the list of Class 3 complications includes all Class 1 and Class 2 diagnoses.

Supplementary Information, Table S3

Some patients have multiple qualifying diagnoses (for example: acute respiratory failure following trauma and surgery and respiratory insufficiency), however, they would only be counted in the composite analysis as having a single Class 2 (and Class 3) respiratory complication.

Supplementary Information, Figure S1

V_T = Tidal Volume, V_T = PBW = Tidal Volume as a Function of Predicted Body Weight, ΔP = driving pressure. (1) Full descriptive and threshold variables considered. (2) 25%, 50%, and 75% descriptive values for: V_T, V_T – PBW, ΔP, PEEP, and FiO₂ were selected for univariate analysis. All threshold variables were included in univariate analysis. (3) Based upon the results of univariate analysis, we ultimately selected 50% values for descriptive values and the following threshold values for inclusion in multivariable analysis: SpO₂ < 90%, SpO₂ > 97%, V_T > 500 mL, V_T-PBW > 8 mL/kg, and ΔP > 16 cm H₂O. (4) Significant0 values from the primary outcome: minimum PaO₂/FiO₂ and clinical outcomes: 30-Day Mortality and PPCs.

Supplementary Information, Table S4

C-Statistic=0.618. Bold signifies p < 0.05.

Supplementary Information, Table S5

C-Statistic=0.621. Bold signifies p < 0.05.

Key Points:

Question:

In patients requiring postoperative intubation after non-cardiac surgery, is lung protective ventilation associated with higher postoperative PaO₂/FiO₂ ratios and improved postoperative outcomes?

Findings:

After adjustment, intraoperative PEEP, FiO₂, and time with driving pressure > 16 cmH₂O were associated with decreased postoperative PaO₂/FiO₂, and lower postoperative PaO₂/FiO₂ was associated with more postoperative pulmonary complications and higher mortality.

Meaning:

The intermediate laboratory finding, PaO₂/FiO₂ is a potential target for future prospective studies investigating the impact of intraoperative lung protective ventilation strategies on postoperative outcomes.