Intraoperative Mechanical Ventilation and Postoperative Pulmonary Complications After Cardiac Surgery

Michael R. Mathis, M.D.; Neal M. Duggal, M.D.; Donald S. Likosky, Ph.D.; Jonathan W. Haft, M.D.; Nicholas J. Douville, M.D., Ph.D.; Michelle T. Vaughn, M.P.H.; Michael D. Maile, M.D., M.S.; Randal S. Blank, M.D., Ph.D.; Douglas A. Colquhoun, M.B., Ch.B., M.Sc., M.P.H.; Raymond J. Strobel, M.D., M.S.; Allison M. Janda, M.D.; Min Zhang, Ph.D.; Sachin Kheterpal, M.D., M.B.A.; Milo C. Engoren, M.D.


Anesthesiology. 2019;131(5):1046-1062. 

In This Article

Materials and Methods

We obtained Institutional Review Board approval (HUM00132314) for this observational cohort study performed at our academic quaternary care center; the requirement for informed patient consent was waived. We adhered to the Strengthening the Reporting of Observational Studies in Epidemiology checklist for reporting observational studies. Study methods including data collection, outcomes, and statistical analysis were established prospectively and presented at an institutional peer-review committee on January 20, 2016; a revised finalized proposal was registered before accessing study data.[23]

Patient Population

Inclusion criteria for the study were adult (at least 18 yr old) patients who underwent elective or urgent cardiac surgical procedures with full CPB, limited to coronary artery bypass grafting, valve, and aortic procedures, performed in isolation or in combination. We reviewed patients over a continuous 11-yr study period from January 1, 2006 to June 1, 2017. Exclusion criteria were preoperative mechanical ventilation within 60 days of surgery, use of a double-lumen endotracheal tube or one-lung ventilation, American Society of Anesthesiologists class V or VI physical status, preoperative extracorporeal membrane oxygenation support, ventricular assist device implantation procedures (planned and unplanned), reoperative cardiac surgical procedures, transcatheter procedures, or procedures using partial- or left-heart bypass. At our institution, surgical techniques for the study cohort commonly included direct aortic cannulation via full sternotomy, and rarely, axillary or femoral cannulation or direct cannulation via mini-sternotomy. No robotic procedures or minimally invasive direct coronary artery bypass procedures were performed.

Data Collection

We collected study data from three sources: the Multicenter Perioperative Outcomes Group electronic anesthesia database, the Society of Thoracic Surgeons Adult Cardiac Surgery Database, and our hospital enterprise electronic health record. Within the Multicenter Perioperative Outcomes Group database, physiologic monitors including vital signs and ventilator settings and measurements are collected in automated fashion every 60 s and stored in an electronic intraoperative anesthesia record for all cases. Templated intraoperative script elements—including case times, medications and fluids administered, and anesthetic interventions such as airway management techniques—are additionally routinely recorded within the anesthesia record for all cases. Within the Society of Thoracic Surgeons database, patient history, surgical procedure, and outcome data are similarly stored as discrete concepts for all adult cardiac surgical procedures performed within our institution. To maintain high rates of interobserver agreement across cases, data are standardized using detailed prespecified definitions, and are collected (Society of Thoracic Surgeons database)[24] or validated (Multicenter Perioperative Outcomes Group database) by nurses with completed training in data definitions used. Detailed methods for data entry, validation, and quality assurance are described elsewhere,[25–27] and have been used for multiple published studies.[28–31] Within the Multicenter Perioperative Outcomes Group and Society of Thoracic Surgeons databases, local datasets were linked via unique codified surgical case and patient identifiers; data extraction and analysis were performed on a secure server. Finally, local electronic health record data (Epic Systems Corporation, USA) were used to determine postoperative arterial blood gas values and ICU ventilator data, as necessary for components of outcome variables described below; these data were similarly linked to the final analytic dataset. The quality of local electronic health record data used for this study was verified via manual review by an anesthesiologist investigator (M.R.M.) of all cases experiencing the primary outcome, all cases with outlier data, and 10% of cases not experiencing the primary outcome.

Clinical Processes of Care

Perioperative anesthetic management for all cases was at the discretion of the attending cardiac anesthesiologist, who directs an anesthesia care team of anesthesiology fellows and residents. Routinely, anesthetic agents included induction with midazolam, propofol, or etomidate; analgesia with fentanyl or morphine; neuromuscular blockade with rocuronium, vecuronium, or cisatracurium; and maintenance with isoflurane, transitioned to a propofol or dexmedetomidine infusion before transport to ICU. In addition to standard monitoring, intraoperative hemodynamic management was routinely guided by invasive arterial line, central venous pressure, and pulmonary artery catheter monitors, as well as transesophageal echocardiography and arterial/mixed venous blood gas measurements. Fluids, blood products, vasoactive infusions, and inotropic infusions were managed at the discretion of the attending anesthesiologist in communication with the cardiac surgeon, with typical hemodynamic targets including a mean arterial pressure greater than 65 mmHg, cardiac index greater than 2.2 l/min/m2, mixed venous oxygen saturation greater than 65%, hematocrit greater than 21%, and echocardiographic assessment of post-CPB ventricular systolic function unchanged to improved compared with pre-CPB function.

Ventilator settings in the operating room were managed by the attending anesthesiologist. Intubation was performed with a 7.5- or 8.0-mm-internal-diameter endotracheal tube. Mechanical ventilation was performed using Aisys CS2 anesthesia workstations (General Electric Healthcare, USA). Providers typically employed a pressure-controlled volume-guaranteed ventilation mode (default setting) throughout the entire study period, targeting normocapnia or mild hypocapnia, and avoiding hypoxemia. Of note, default settings on ventilators used included VT = 500 ml and PEEP = 0 cm H2O; the default PEEP setting was subsequently changed to PEEP = 5 cm H2O in March 2007. Ventilation was paused during CPB; the ventilator circuit remained connected to the patient, but with no application of PEEP. Before discontinuation of CPB, it was resumed after providing recruitment maneuvers. After transport to ICU, a structured handoff detailing intraoperative management, including final ventilator settings and plan for extubation, was communicated to an ICU team of intensivists, nurses, and respiratory therapists. Ventilator weaning, extubation, and management of complications were made at the discretion of the ICU team, as based on local protocols and targeting goals discussed during postoperative handoff.


The primary outcome was occurrence of a postoperative pulmonary complication, predefined as a composite of pulmonary complications recorded in the Society of Thoracic Surgeons database and adjudicated by nurses trained in outcome definitions, or recorded in our enterprise electronic health record and adjudicated by an anesthesiologist (M.R.M.). These included any one of the following: prolonged initial postoperative ventilator duration longer than 24 h (Society of Thoracic Surgeons database), pneumonia (Society of Thoracic Surgeons database), reintubation (Society of Thoracic Surgeons database), or postoperative partial pressure of oxygen to fractional inspired oxygen (PaO2/FIO 2) below 100 mmHg within 48 h postoperatively while intubated (local electronic health record, Appendix 1).

We selected a threshold of PaO2/FIO2 below 100 mmHg as a postoperative pulmonary complication component based on previously validated assessments of pulmonary dysfunction associated with mortality after cardiac surgery.[32–34] Given varied mechanisms of pulmonary injury, and the distinction between pneumonia versus other pulmonary complications as described in recent consensus guidelines,[35,36] each component of the postoperative pulmonary complication composite outcome was also separately analyzed as a secondary outcome. Additional predefined secondary outcomes included 30-day postoperative mortality, initial postoperative mechanical ventilation duration, minimum PaO2/FIO 2 within 48 h postoperatively while intubated (as a continuous variable), length of ICU stay, and length of hospital stay. All secondary outcomes were similarly adjudicated by trained Society of Thoracic Surgeons nurse reviewers with the exception of minimum PaO2/FIO 2 which was adjudicated by an anesthesiologist (M.R.M.).

Exposure Variables – Lung-protective Ventilation

The primary exposure variable studied was a bundled intraoperative lung-protective ventilation strategy, comprising median VT below 8 ml/kg predicted body weight and median driving pressure below 16 cm H2O and median PEEP at or above 5 cm H2O. Varying lung-protective cutoffs for each ventilator component are currently described in the literature, ranging from VT 6 to 10 ml/kg predicted body weight,[1,13,15] driving pressure 8 to 19 cm H2O,[16,18,37] and PEEP 3 to 12 cm H2O.[13,15] Given these ranges, our cutoffs were selected by inspection of previously collected ventilation practice institutional data, targeting upper quartiles (approximately 75% compliance for each component) to ensure class balance between cases with lung-protective ventilation versus non-lung-protective ventilation and to improve multivariable model discrimination.[5,13,28,38–40]

Predicted body weight (in kg) was calculated as: 50 + 2.3 • (height [in] – 60) for men; 45 + 2.3 • (height [in] – 60) for women.[41] Modified airway driving pressure was calculated as (peak inspiratory pressure – PEEP). As performed in previous studies,[42] we used modified driving pressure for all cases, given the lack of ventilator plateau pressure data available within our electronic medical record necessary for a true driving pressure calculation. To adjust for decisions to maintain normoxia rather than a lung-protective ventilation strategy (otherwise favoring lower FIO 2 and moderate PEEP), intraoperative oxygen saturation measured by pulse oximetry and FIO 2 were included as covariates. To summarize each ventilator variable on a per-case basis, median values while mechanically ventilated were calculated. Ventilator parameters while on CPB, during which ventilators were routinely paused, were excluded from the median value calculation. For descriptive purposes, ventilator parameters were additionally subdivided into median value pairs, separated into the pre-CPB and post-CPB periods. In cases with multiple instances of CPB, post-CPB ventilator parameters were analyzed after the final CPB instance.

Covariate Data

For descriptive purposes and to adjust for confounding variables potentially associated with the exposure variables or study outcomes, a range of perioperative characteristics were included as covariates within our study. Patient anthropometric, medical history, anesthetic, surgical, and laboratory testing/study variables were selected as available within the Multicenter Perioperative Outcomes Group and Society of Thoracic Surgeons databases. All variables used in several existing scores for calculating risk of complications including postoperative pulmonary complications after cardiac surgery were included (e.g., cardiac surgery type, bypass times, comorbidities, etc.), in addition to other relevant descriptive covariates (Table 1).[9,10,43] To evaluate for changes in practice and Society of Thoracic Surgeons database reporting over the study time period, the Society of Thoracic Surgeons Adult Cardiac Surgery Database version was included as a covariate; this resulted in four time periods for adjustment (1/1/2006–12/31/2007; 1/1/2008–6/30/2011; 7/1/2011–6/30/2014; 7/1/2014–5/31/2017) To account for variation in unmeasured intraoperative practices attributable to the attending anesthesiologist and potentially associated with postoperative pulmonary complications, we characterized attending anesthesiologists by tertiles of low/medium/high frequency of bundled intraoperative lung-protective ventilation use.

Statistical Analysis

All statistical analyses were performed using SAS version 9.3 (SAS Institute, USA). Normality of continuous variables was graphically assessed using histograms and Q-Q plots. Continuous data were presented as mean ± SD or median and interquartile range; binary data were summarized via frequency and percentage. Comparisons of continuous data were made using a two-tailed independent t test or a Mann–Whitney U test, and categorical data were compared by a Pearson chi-square or Fisher's exact test, as appropriate. Trend analyses of the components of the lung-protective ventilation bundle were completed using the Cochran–Armitage test. A P value less than 0.05 denoted statistical significance.

Before any multivariable analysis, collinearity among covariates was assessed using the variance inflation factor; variables with a variance inflation factor greater than 10 were excluded. To target development of a clinically usable reduced-fit postoperative pulmonary complication multivariable model avoiding overfitting, covariates meaningfully describing the study population but not used in existing cardiac surgery risk score models were additionally excluded from multivariable analysis. Missing data were handled via a complete case analysis. To further aid in covariate selection, we used the least absolute shrinkage and selection operators technique and restricted covariates to the number of outcomes divided by 10, while also accounting for the lung-protective ventilation bundle as well as lung-protective ventilation bundle components (VT, driving pressure, and PEEP). We chose this variable selection technique, given its ability to perform regularization and variable selection to improve model accuracy and interpretability, particularly among analyses with a relatively large number of covariates and modest number of outcomes.[44] Using a multivariable logistic regression model, we characterized the risk-adjusted association between the primary exposure of intraoperative lung-protective ventilation bundle and the primary outcome of postoperative pulmonary complication. Additionally, we repeated our multivariable analysis to assess independent associations between each lung-protective ventilation bundle component and postoperative pulmonary complications. Overall model discrimination of logistic regression models was assessed using the c statistic. Secondary outcomes were assessed using multivariable linear regression models. Goodness-of-fit for linear regression models was summarized via R-squared; such models were evaluated using varied distributional assumptions (i.e., linear versus logarithmic transformations) for continuous secondary outcomes. Multilevel modeling clustering at the provider level was not possible because of limited sample size per provider; instead, the previously mentioned fixed covariate of anesthesiology attending lung-protective ventilation frequency tertile was used.

In addition to analyzing independent associations between an overall lung-protective ventilation strategy and the postoperative pulmonary complication primary outcome, we performed several sensitivity analyses, including an analysis of lung-protective ventilation separated into component parts: VT below 8 ml/kg predicted body weight, driving pressure below 16 cm H2O, or PEEP at or above 5 cm H2O, and analysis of lung-protective ventilation strategies separately examined before and after CPB.

We also performed a sensitivity analysis, using a model that further restricted the number of covariates to the number of outcomes divided by 20.[45] Additionally, we compared our multivariable postoperative pulmonary complication model developed using least absolute shrinkage and selection operator for covariate selection with a multivariable postoperative pulmonary complication model including all noncollinear covariates with P less than 0.10. Finally, we performed subgroup analyses stratified by salient clinical characteristics.