A Lower Tidal Volume Regimen during One-lung Ventilation for Lung Resection Surgery Is Not Associated with Reduced Postoperative Pulmonary Complications

The Multicenter Perioperative Outcomes Group (MPOG) at the University of Michigan obtained institutional review board (IRB) approval for this observational cohort study (University of Michigan, Ann Arbor, Michigan, IRB MED HUM00024166, HUM00033894). Each participating site additionally obtained IRB approval for submission of a limited data set to the Multicenter Perioperative Outcomes Group database. The requirement for written informed consent was waived by the IRB at participating centers. This site IRB approval includes provision for submission of Society of Thoracic Surgeons registry data to the Multicenter Perioperative Outcomes Group from each center. In keeping with the Multicenter Perioperative Outcomes Group bylaws, this study protocol was presented to the Multicenter Perioperative Outcomes Group Perioperative Clinical Research Committee and was approved on March 28, 2017. After data acquisition, an unanticipated imbalance between the protective versus nonprotective cohorts was discovered. We revised the protocol twice to address this as well as unmeasured confounding caused by excess population heterogeneity. The plan for statistical analysis was revised, circulated, and approved by the Perioperative Clinical Research Committee on July 11, 2018, and January 29, 2020. After approval, a data analysis and statistical plan was written and filed with a private entity (Multicenter Perioperative Outcomes Group Perioperative Clinical Research Committee) before data were accessed or revised analysis conducted (Supplemental Digital Content 1, https://links.lww.com/ALN/C543). During the peer review process, additional changes as requested by editors were incorporated. Final methods are presented below. We followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) checklist in developing this manuscript.

The Multicenter Perioperative Outcomes Group database, as well as methods for data entry, validation, and quality assurance, have been previously described¹⁷  and have been used for multiple published observational studies.^18,19  Multicenter Perioperative Outcomes Group data are drawn from cases documented in the Electronic Health Record at participating sites. These data are extracted; standardized; joined to additional laboratory, billing, and diagnosis coding data; and de-identified with the exception of date of service, producing a limited dataset.

Five large academic medical centers that submit Society of Thoracic Surgeons General Thoracic Surgery Database and Multicenter Perioperative Outcomes Group data were included in this study. The General Thoracic Surgery Database is managed by each site and uses standard definitions and data elements captured by the data collection form (https://www.sts.org/registries-research-center/sts-national-database/general-thoracic-surgery-database/data-collection; accessed February 10, 2020).

Data are gathered and aggregated by trained data managers who review medical records of patients undergoing surgical procedures by participating thoracic surgeons at each institution to capture demographics, comorbidities, details of preoperative evaluation, intraoperative course, and postoperative outcomes. The Society of Thoracic Surgeons training manual is the common reference for all data managers who receive annual training at the Advances in Quality Outcomes Seminar hosted by the Society of Thoracic Surgeons (https://www.sts.org/meetings/calendar-of-events/advances-quality-outcomes-data-managers-meeting-0; accessed February 10, 2020). These data are externally and independently audited and are known to be greater than 95% accurate.²⁰ 

General Thoracic Surgery Database records were linked to Multicenter Perioperative Outcomes Group records using patient-level identifiers at each participating site. These identifiers were removed before data upload to the Multicenter Perioperative Outcomes Group Coordinating Center (University of Michigan, Ann Arbor, Michigan). At the Coordinating Center, the patient-matched records from both databases were linked using case start date and time.

Patients undergoing one-lung ventilation between January 1, 2012, and December 31, 2016, for pneumonectomy, bilobectomy, lobectomy, segmentectomy, or wedge resection/metastasectomy with available General Thoracic Surgery Database and Multicenter Perioperative Outcomes Group records were included. We had originally intended to include procedures through May 31, 2017, but data were not available across all sites for this time period, and thus the study period was restricted to December 31, 2016. Time of one-lung ventilation initiation and termination (where available) was defined based on use of a structured data element in the anesthesia record. Cases were excluded if one-lung ventilation was used for less than 15 min, if either height or weight data were unavailable, if lung transplantation was performed, or if surgery represented a reoperation within 30 days of a previous included surgery.

Values of V_T, PEEP, airway pressures (mean, peak [P_MAX] and plateau pressures), end tidal-carbon dioxide, fraction of inspired oxygen, respiratory rate, and calculated modified driving pressure (P_MAX – PEEP) were derived for use in this study. These variables are stored in the Multicenter Perioperative Outcomes Group database at 1-min intervals. Consistent with our previous work, we used a sampling methodology for evaluation of ventilation parameters.^18,19  We calculated the median value for the time period 5 to 15 min after the time-stamped documentation of initiation of one-lung ventilation for each case.

Criteria for protective ventilation were based upon expert opinion and guidelines for optimal practice during one-lung ventilation.^8–10  Cases were considered to have been conducted with protective ventilation only if both of the following criteria were met: median V_T was less than or equal to 5 ml/kg predicted body weight, and median PEEP was greater than or equal to 5 cm H₂O. Ventilation variables were subsequently expressed and analyzed as means of the individual case median values.

Modified driving pressure was used as a surrogate of driving pressure in this investigation, since plateau airway pressure data, required for the calculation of driving pressure, were not available from all participating institutions. This modification of driving pressure has been previously reported.²¹ 

In construction of the statistical models used in this manuscript, we included data from the Multicenter Perioperative Outcomes Group database and General Thoracic Surgery Database (appendix  1).

Data from the Multicenter Perioperative Outcomes Group database are institution, presence of blood product transfusion (as a binary variable), fluid balance (volume of input [crystalloids + colloids + blood products] – volume of fluid output [urine + gastric tube output + estimated blood loss + chest tube] as documented on the anesthetic record), and American Society of Anesthesiologists (ASA; Schaumburg, Illinois) physical status.

Data from the General Thoracic Surgery Database are forced expiratory volume in 1 s (FEV₁), presence of missing FEV₁ data, preoperative renal dysfunction, preoperative steroid therapy, Zubrod Performance Classification score, current smoking status, preoperative chemotherapy and/or radiation, and major preoperative comorbidity (defined as coronary artery disease, congestive heart failure, peripheral vascular disease, or diabetes). Procedure type was categorized as pneumonectomy, bilobectomy, lobectomy, segmentectomy, or wedge resection/metastasectomy (which acted as the reference value in our models). Additionally, we classified surgical approach as thoracotomy or video-assisted thoracoscopic surgery (which acted as the reference value in our models).

Demographic variables for age, sex, and body mass index were preferentially extracted from the General Thoracic Surgery Database; however, if not available or invalid, they were derived from the Multicenter Perioperative Outcomes Group database.

The primary outcome was a composite of major postoperative pulmonary complications drawn from the General Thoracic Surgery Database. Pulmonary complications were defined as one or more of the following: initial ventilator support greater than 48 h, reintubation, pneumonia, atelectasis requiring bronchoscopy, acute respiratory distress syndrome (ARDS), air leak greater than 5 days, bronchopleural fistula, respiratory failure, tracheostomy, pulmonary embolism, or empyema requiring treatment. Two progressively more comprehensive secondary outcomes were (1) major morbidity—pulmonary complications (as defined above) or one or more of the following: unexpected return to the operating room (during same hospital stay), atrial or ventricular dysrhythmias requiring treatment, myocardial infarction, sepsis, renal failure, central neurologic event, unexpected intensive care unit admission, or anastomotic leak; and (2) major morbidity (defined above) and/or mortality. All outcomes were drawn from the General Thoracic Surgery Database record and followed the definitions at time of data entry (https://www.sts.org/registries-research-center/sts-national-database/general-thoracic-surgery-database/data-collection; accessed February 10, 2020).

A complete case analysis was conducted. Data were presented as mean ± SD or frequencies with percentages. Univariate comparisons between groups were assessed using Pearson chi-square or Fisher exact tests for categorical data and Student’s t or Mann–Whitney U tests for continuous variables, as appropriate. Absolute standardized difference percentages are reported.

The final statistical analysis plan included the use of propensity score matching to adjust for differences between the protective and nonprotective ventilation groups. A nonparsimonious regression model was used to estimate each participant’s propensity to receive the protective ventilation exposure. The propensity score model contained age, sex, body mass index, FEV₁, presence of missing FEV₁ data, ASA physical status, preoperative renal dysfunction, preoperative steroid therapy, Zubrod score, current smoking status, preoperative chemotherapy and/or radiation, institution, and major preoperative comorbidity. Protective ventilation patients were propensity score–matched 1:1 to those not receiving protective ventilation using the “onetomanymtch” greedy matching algorithm.²²  Residual covariate imbalance after the match was assessed by computing standardized differences. Variables with an absolute standardized difference less than 10% were considered a strong match. Within the matched cohort, univariate differences between those with and without protective ventilation were assessed using McNemar test for categorical variables and paired t tests or Wilcoxon signed-rank tests for continuous variables, as appropriate.

Before regression models were constructed, all variables under consideration for model inclusion were assessed for collinearity using the condition index. If the condition index was greater than 30, a Pearson’s correlation matrix was developed. Those variable pairs with a correlation of greater than or equal to 0.70 were combined into a single concept, or the variable with the larger univariate effect size was selected for inclusion. All other variables were considered fit for model entry.

To evaluate the primary aim in the matched cohort, a conditional logistic regression model was used to assess the relationship between protective ventilation status and outcome with the covariates of blood product transfusion, fluid balance, surgical procedure (wedge resection, segmentectomy, lobectomy, bilobectomy, pneumonectomy), and surgical approach (video-assisted thoracoscopic surgery vs. open). Measures of effect for model covariates were reported as conditional adjusted odds ratios with 95% CIs. The model predictive capability was reported using the area under the receiver operating characteristic curve c-statistic. Any covariate found to be statistically significant was considered an independent predictor of the outcome of interest. These models were also constructed for the secondary outcomes (morbidity; morbidity and mortality).

The full study cohort was used for analysis of optimal V_T and PEEP combinations and examination for any relationship between airway pressures and outcome. Traditional logistic regression models were used for these analyses. Measures of effect for model covariates were reported for logistic regression as adjusted odds ratios with 95% CIs. Any covariate found to be statistically significant after adjustment was considered an independent predictor of the outcome of interest.

To assess if an alternative combination of V_T and PEEP was associated with a lower risk of pulmonary complications, a matrix of adjusted odds ratios was constructed with the reference category of V_T between 4 and 6 ml/kg predicted body weight and PEEP between 4 and 6 cm H₂O.

To assess if modified driving pressure was associated with primary or secondary outcomes, three multivariable logistic regression models were constructed, adjusted for the covariates specified above. A similar analysis was conducted for P_MAX.

All analyses were conducted using SAS 9.4 (SAS Institute, USA) and SPSS 24 (IBM Corp., USA). Two-tailed hypothesis testing was conducted, and a P value of 0.05 was considered statistically significant for all analyses. Additional information regarding aim-specific analyses can be found in appendix  2.

An a priori sample size calculation was performed using a two-sided Z test with unpooled variance. A sample size of 1,315 unmatched cases in each group (total study N = 2,630) provided 90% power at an alpha = 0.05 to detect a 5% difference (deemed to represent a clinically significant difference) in the rate of pulmonary complications, assuming a 22% rate of events in the nonprotective ventilation group.

Of 3,721 cases that were eligible for analysis, 489 were excluded for missing data required for model construction. A total of 3,232 cases from five institutions were available for the final analysis (fig. 1). Baseline cohort characteristics are shown in table 1. It should be noted that some cases from one institution have been previously reported (693 cases from 2012 to 2014; 194 cases that are included in the current matched cohort).¹² 

In the unmatched cohort, a primary pulmonary complication outcome occurred in 427 (13.2%) of cases; secondary outcomes—major morbidity and major morbidity and/or mortality—occurred in 659 (20.4%) and 676 (20.9%) cases, respectively (table 2). In 2012, mean ± SD V_T was 6.7 ± 1.61 ml/kg; in 2016, mean ± SD V_T was 6.0 ± 1.25 ml/kg (P < 0.0001), while mean ± SD PEEP was 4 ± 2 cm H₂O in 2012 and 5 ± 2 cm H₂O in 2016 (P < 0.0001; table 1; fig. 2). The proportion of cases meeting the definition of lung-protective ventilation was 5.7% in 2012 and 17.9% (P < 0.001) in 2016 (fig. 2). The prevalence of the primary outcome and major morbidity did not change significantly during the study period (pulmonary complications: 11.4 to 15.7%, P = 0.147; major morbidity: 18.5 to 22.9%, P = 0.088). However, there was a significant increase in secondary outcome of major morbidity and/or mortality from 2012 to 2016 (18.6 to 23.8%, P = 0.039; Supplemental Digital Content 2 [https://links.lww.com/ALN/C544] and 3 [https://links.lww.com/ALN/C545]).

Propensity score matching addressed differences between the baseline characteristics of the protective and nonprotective ventilation populations (table 1). Of the 388 cases that met the protective ventilation definition, 381 (98.2%) were propensity score–matched to nonprotective ventilation cases, resulting in a primary aim study population of 762 patients. In our conditional logistic regression model, protective ventilation was not found to be associated with differential risk of pulmonary complications (adjusted odds ratio, 0.86; 95% CI, 0.56 to 1.32; P = 0.480), major morbidity (adjusted odds ratio, 0.81; 95% CI, 0.55 to 1.19; P = 0.283), or morbidity and mortality (adjusted odds ratio, 0.81; 95% CI, 0.55 to 1.19; P = 0.281).

Given the lack of association between this definition of protective ventilation and outcome, we attempted to derive an alternative definition of protective ventilation associated with lower risk for pulmonary complications. We used a matrix of odds ratios to determine if an alternative combination of V_T and PEEP was associated with a lower risk of pulmonary complications. We did not find a combination of these parameters that predicted a lower risk of pulmonary complications compared to the reference definition (data not shown). When V_T or PEEP was analyzed in isolation as categorical ranges—per 1 ml/kg for V_T and 1 cm H₂O for PEEP—we found no significant relationship with predicted probability of pulmonary complications (Supplemental Digital Content 4 [https://links.lww.com/ALN/C546] and 5 [https://links.lww.com/ALN/C547]).

Consistent with previous work, modified airway driving pressure was used as a proxy for airway driving pressure. Using the subjects for which both values were available, we plotted the relationship between them (Supplemental Digital Content 6, https://links.lww.com/ALN/C548). The correlation between modified airway driving pressure and airway driving pressure was 0.87 (95% CI, 0.86 to 0.88; P < 0.001). In multivariable regression models, neither modified airway driving pressure nor P_MAX was associated with a significant increase in the odds of pulmonary complications for each 5 cm H₂O increase in pressure (modified airway driving pressure: adjusted odds ratio, 0.93; 95% CI, 0.84 to 1.04; P = 0.145; P_MAX: adjusted odds ratio, 0.94; 95% CI, 0.85 to 1.05; P = 0.304; Supplemental Digital Content 7, https://links.lww.com/ALN/C549; fig. 3).

In this study, we examined the relationship between ventilation variables, including V_T, PEEP, and airway pressures, and the subsequent development of postoperative complications in patients undergoing one-lung ventilation for thoracic surgery. We draw several conclusions. First, use of recommended ventilation parameters increased during the study period. Second, this definition of protective ventilation was not independently associated with a lower prevalence of pulmonary complications. Third, the development of postoperative complications was not associated with either modified driving pressure or P_MAX.

The use of a conventional definition of protective one-lung ventilation was not associated with a difference in the prevalence of pulmonary complications (primary outcome), major morbidity or major morbidity and/or mortality (secondary outcomes) between protective and nonprotective ventilation subcohorts after propensity score adjustment for population differences. Our study demonstrates a practice trend of increasing use of recommended protective ventilation parameters consistent with that from previous reports.^18,19,23  Despite the decrease in V_T (6.7 to 6.0 ml/kg predicted body weight), and an increase in use of protective ventilation, the prevalence of pulmonary complications and major morbidity did not change significantly during the study period. However, there was a significant increase in the prevalence of major morbidity and/or mortality from 2012 to 2016 (18.6 to 23.8%, P = 0.039; Supplemental Digital Content 2 [https://links.lww.com/ALN/C544] and 3 [https://links.lww.com/ALN/C545]).

Our chosen target values for V_T (5 ml/kg predicted body weight) and PEEP (5 cm H₂O) in the definition of protective one-lung ventilation are based on published expert opinion.^8–10  Although these parameters are generally considered to be “protective,” the former reflects a supraphysiologic V_T, and the latter (PEEP) may be insufficient to maintain an open lung state that prevents atelectasis and atelectrauma during one-lung ventilation.^24,25  The notion that low V_T in the setting of low PEEP is not intrinsically protective is supported by the previously demonstrated inverse relationship^11,12  or lack of relationship²⁶  between V_T and the risk of adverse outcomes in both two- and one-lung ventilation surgical settings.

These findings are consistent with results of trials that have evaluated putative protective regimens, combining lower V_T and higher levels of PEEP compared to conventional regimens combining supraphysiologic V_T with minimal PEEP.^2,4–7  Such protective regimens may minimize both volutrauma and atelectrauma by limiting distending stress and volume loss/atelectasis, respectively, and have been demonstrated to decrease airway driving pressure and mechanical energy delivery.^27,28  In a meta-analysis of multiple protective ventilation trials, protective ventilation differed from conventional ventilation most markedly on the basis of PEEP (greater than sixfold difference), whereas “protective” V_T was only 32% lower than that of the conventional groups.²⁹  Thus, the primary difference between protective and conventional ventilation may be the use of an open lung strategy, which includes sufficient PEEP to minimize volume loss, atelectasis, and the risk of atelectrauma rather than lower V_Tper se. This view is further supported by recent trials that demonstrated no outcome improvements in patients randomized to receive lower V_T.^30,31 

In our analysis, the primary outcome is a composite of 11 distinct postoperative pulmonary complications, rather than a single outcome more directly related to lung injury (e.g., ARDS). It should be noted that the individual outcome events contributing to a composite outcome vary greatly on the basis of severity (i.e., ARDS vs. atelectasis) and frequency (range, 0.3 to 3.1%). Despite its multicenter design and relatively large sample size, our study did not have sufficient power to determine if specific outcome events were associated with different V_T and PEEP combinations.

Previous studies have demonstrated an association between airway driving pressure and complications in patients ventilated for ARDS and surgery, although very little information is available for procedures involving one-lung ventilation.^13,14  In the current study, we found that neither modified driving pressure nor P_MAX was associated with significantly increased odds of pulmonary complications when analyzed as continuous variables in fixed-effects logistic regression models controlling for other risk predictors. These findings are not consistent with those of previous studies.^12,13 

The current study differs with regard to the use of a surrogate measure, modified driving pressure (P_MAX – PEEP). Despite being shown to predict ARDS in a large cohort of general surgical patients, its specific utility as a predictor of pulmonary complications in a thoracic surgical population receiving one-lung ventilation is not yet established.²¹  Despite its very close correlation with driving pressure (0.87; 95% CI, 0.86 to 0.88; P < 0.001), it is conceivable that this modification is less useful than driving pressure as a surrogate marker of dynamic strain. It is also conceivable that the dramatic elevation of lung elastance associated with one-lung ventilation in the lateral position could confound the relationship between airway driving pressure and dynamic strain.³²  Finally, it is also possible that the contribution to the overall postoperative pulmonary complication rate from specific pulmonary complications emerging from elevated dynamic strain (e.g., ARDS) is dwarfed by complications from other injurious processes (e.g., atelectasis).

Park et al. recently reported a randomized trial of thoracic surgical patients who were randomized to receive one-lung ventilation (V_T, 6 ml/kg) with either fixed PEEP 5 cm H₂O or an individualized PEEP based on an increment trial to the lowest driving pressure.¹⁴  In this study, PEEP titration was associated with a reduction in the incidence of pulmonary complications from 12.2 to 5.5%. However, both the delivered PEEP (5 vs. 3 cm H₂O) and resultant driving pressure (10 vs. 9 cm H₂O) differences between groups were small. The contribution of driving pressure, if any, to the observed findings remains unclear.

Although intraoperative ventilation exposures from the Multicenter Perioperative Outcomes Group database are detailed and accurate, available data are limited by relative practice homogeneity in ventilation management during the study period. While V_T differences between groups are similar to those seen in modern protective ventilation trials,²⁹  the smaller difference in PEEP between the protective and nonprotective groups may be insufficient to elicit detectable differences in outcome.

Although recruitment maneuvers have been advocated by some authors as a component of protective ventilation,^8,9  they were not included in our definition for two reasons. First, recruitment maneuvers cannot be accurately derived from physiologic data with one-minute temporal resolution. Second, there are no evidence-based standardized criteria for their use. Recruitment maneuvers represent a heterogeneous group of practices. Further, they neither constitute a universal feature of protective ventilation nor are required for the outcome benefits^5–7  or necessarily to maintain an open lung state avoiding atelectasis.^25,33  Further, they may have the potential to cause harm,^34,35  and recent guidelines for protective one-lung ventilation do not unambiguously support their use.¹⁰  While our study did not include them and is unable to account for them, the possibility that recruitment maneuvers contribute to the variance in patient outcome remains and may need to be addressed in future work.

We were not able to assess changes in ventilation management that may have occurred during the course of the anesthetic in response to hypoxemia because our sampling methodology focused on the start of one-lung ventilation. We have previously demonstrated that the ventilator data from this early period very closely match those used for the entire period of one-lung ventilation.¹⁹  Furthermore, hypoxemia typically occurs early in the one-lung ventilation period and is thought to be a very infrequent occurrence in modern thoracic anesthesia practice.³⁶ 

Finally, included data were derived from five academic medical centers, which exhibited variation in the prevalence of complications. The integration of both the Multicenter Perioperative Outcomes Group database and the General Thoracic Surgery Database allowed us to combine the advantages of the automatically gathered, detailed, annotated dataset from Multicenter Perioperative Outcomes Group with the highly accurate and validated outcome data derived from the General Thoracic Surgery Database. Limitations of the latter database derive from the fact that participation is voluntary. As participants are typically general thoracic surgeons, results may not be generalizable to those of other surgeons or institutions performing similar procedures. Our approach leverages the advantages and strengths of each data source, which improves the validity and generalizability of our findings.

This multicenter study demonstrates an increase in adoption of a ventilation regimen including V_T less than or equal to 5 ml/kg predicted body weight in combination with PEEP greater than 5 cm H₂O during one-lung ventilation. However, this lower V_T regimen was not associated with reduced odds of major pulmonary complications. Furthermore, in this study cohort, neither increasing P_MAX or modified airway driving pressure was associated with increased odds of major pulmonary complications.

The authors would also like to recognize the following members of the Multicenter Perioperative Outcomes Group Perioperative Clinical Research Committee: Robert M. Craft, M.D., Department of Anesthesiology, University of Tennessee, Knoxville, Tennessee, and William Hightower, M.D., Department of Anesthesiology, Henry Ford West Bloomfield Hospital, Detroit, Michigan. The authors recognize Linda W. Martin, M.D., Department of Surgery, University of Virginia, Charlottesville, Virginia, for her assistance with study design and manuscript review. The authors recognize Genevieve Bell, B.S., University of Michigan, Ann Arbor, Michigan, for her help with data acquisition. The authors recognize the assistance of Graciela Mentz, Ph.D., Department of Anesthesiology, University of Michigan, for her assistance in statistical support during the peer review process.

Research reported in this publication was supported by the National Institute for General Medical Sciences of the National Institutes of Health (Bethesda, Maryland) under award No. T32GM103730 (to Dr. Colquhoun) and by the National Heart, Lung and Blood Institute of the National Institutes of Health under award No. K01HL141701 (to Dr. Mathis). Additional funding is attributed to the participating institutions. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Dr. Colquhoun declares research grant support paid to the institution from Merck & Co., Inc. (Kenilworth, New Jersey), unrelated to the current work. Dr. Kheterpal declares research support paid to the institution from Merck & Co., Inc., Becton Dickinson (Franklin Lakes, New Jersey), BCBS of Michigan (Detroit, Michigan), and Apple, Inc. (Cupertino, California), all unrelated to the current work. Dr. Chang declares travel reimbursement from the American Board of Thoracic Surgery (Chicago, Illinois), expert witness fees from defendant attorneys, and peer review services/reimbursement from the Department of Defense (Washington, D.C.), all unrelated to the current work. Dr. Schonberger declares research grant support paid to the institution from Merck & Co., Inc., unrelated to the current work. Dr. Schonberger reports having an equity stake in Johnson and Johnson (New Brunswick, New Jersey) unrelated to the current work. Dr. Blank declares research support paid to the institution from the Association of University Anesthesiologists (San Francisco, California) unrelated to the presented work. The other authors declare no competing interests.

The matched cohort was used for this analysis. Univariate comparisons between lung-protective ventilation group status and the rate of each outcome were computed using McNemar test. A Cochran-Armitage test for trend was used to determine if there was an increase in documented use of lung-protective ventilation over time, where time is defined in quarters.

The full study cohort was used for this analysis. To determine the most beneficial combination of positive end-expiratory pressure (PEEP) and tidal volume (Vt) to reduce pulmonary complications, a matrix of adjusted odds ratios was constructed with the reference category of PEEP between 4 and 6 cm H₂O and Vt between 4 and 6 ml/kg predicted body weight. The logistic regression model was adjusted for age, gender, body mass index, forced expiratory volume in 1 s (FEV₁), presence of missing FEV data, American Society of Anesthesiologists (ASA) status, preoperative renal dysfunction, preoperative steroid therapy, Zubrod score, current smoking status, preoperative chemotherapy and/or radiation, major preoperative comorbidity, institution, presence of blood product transfusion, fluid balance, segmentectomy (vs. wedge resection), lobectomy (vs. wedge resection), bilobectomy (vs. wedge resection) or pneumonectomy (vs. wedge resection), and thoracotomy (vs. video-assisted thoracoscopic surgery).

The full study cohort was used in this analysis. Two multivariable logistic regression models were constructed as above to evaluate the impact of ventilator parameters on the primary outcome of pulmonary complications. In addition to the previously mentioned covariates, model 1 contained the variable for modified airway driving pressure (per 1 cm H₂O). Model 2 contained the variable P_MAX. If modified airway driving pressure or P_MAX were statistically significant after adjusting for other significant predictors, they were considered independent predictors of pulmonary complications. Similar models were constructed for the secondary outcomes. Nonlinear trends were not assessed.

The full study cohort was used in this analysis. To determine whether patients known to be at higher risk for receiving high V_T/kg predicted body weight were more likely to be subjected to ventilator regimens associated with higher levels of modified airway driving pressure, three bivariable linear regression models were constructed for the dependent variable of modified airway driving pressure. The first model contained the fixed effect of body mass index, the second model contained the fixed effect of height (cm), and the third model contained the fixed effect of gender.

Next, three nonparsimonious logistic regression models were constructed to evaluate whether patients known to be at higher risk for receiving high V_T were at higher risk of postoperative pulmonary complications. The covariates of body mass index and sex were removed from the model previously specified, to be entered separately. The first model contained the additional fixed effect of body mass index, the second model contained the additional fixed effect of height, and the third model contained the additional fixed effect of gender. A similar set of models was be constructed for all secondary outcomes. If the additional fixed effect for each model was found to be statistically significant, that characteristic was considered an independent predictor of the outcome of interest. If all three were independent predictors, then those at high risk for receiving high V_T were said to be at higher risk for postoperative complications.

Patrick J. McCormick, M.D., M.Eng., Department of Anesthesiology and Critical Care Medicine, Memorial Sloan Kettering Cancer Center, New York, New York

William Peterson, M.D., Department of Anesthesiology, Sparrow Health System, Lansing, Michigan.