This study was a retrospective cohort analysis conducted using a subset of data previously described [8]. In brief, Institutional Review Board approval was obtained for this cohort study at The University of Michigan Health System (IRB-MED, Ann Arbor, MI), a large, quaternary care facility. All data were de-identified prior to analysis, and a waiver of consent was obtained for this study. All cases recorded in the anesthesia information management system (Centricity, General Electric Healthcare, Waukesha, WI) from 06/01/2004 to 06/01/2009 were screened for inclusion. Cases on the cardiac, thoracic, transplant, trauma, and vascular surgery services were excluded, as were cases with no recorded service.
Between February and March 2007, 63 anesthesia machines at the University of Michigan Hospital were exchanged in 8-10 room blocks from Drager Narkomed IIb to GE Aisys. Hence, all cases from February 1, 2007 to March 31, 2007 were excluded. No Narcomed devices were in use starting April 1, 2007. We then compared ventilation parameters on a subset for the two months before (December 1, 2006 to January 31, 2007) to the two months after (April 1, 2007 to May 30, 2007) to examine a possible immediate effect on ventilation. In order to determine if the change was durable and if there was an impact on development of ARDS, the two years before (February 1, 2005 to January 31, 2007) vs. the two years after (April 1, 2007 to March 31, 2009) the ventilator change were examined.
Preoperative data were prospectively collected from routine clinical documentation that was entered into the anesthesia information management system. The record includes a structured preoperative history and physical examination. Data abstracted from the preoperative history included basic demographic and comorbidity information necessary for the management of the critically ill or suggestive of the need for critical care services. A detailed description of variable definitions is included in previous publications [8]. Free text entries were hand-coded by the research team for analysis.
Unique hospital admissions were considered as the base unit for analysis. Admissions containing multiple anesthetic cases were analyzed from the last case of the admission or the last case prior to development of ARDS, as appropriate. Records from the final anesthetic of each admission were also used to determine preoperative comorbidities and ASA status. The ASA status recorded for this final case was collapsed into a binary variable reflecting whether a patient was considered ASA 1-2 or ASA 3-5.
Intraoperative physiologic and ventilator data were acquired using an automated, validated electronic interface from the anesthesia machines and physiologic monitors (Solar 9500; General Electric Healthcare). Fraction inspired oxygen (FiO2), peak inspiratory pressure (PIP), exhaled Vt, PEEP, oxyhemoglobin saturation (SpO2), drive pressure (∆P), and respiratory rate were obtained and analyzed for median values to eliminate spurious and isolated values. The number of ten-minute epochs of median PIP >30 cmH2O, and number of ten-minute epochs of median Vt > 12 cc/kg PBW from the time of incision to the end of anesthesia were examined as markers of continued high pressure and/or high volume ventilation.
Case times were validated using electronically documented heart rate from electrocardiogram or electronically documented start and end. Only cases with positive times were included. Cases from patients graded as ASA classification 6 were excluded.
When available, arterial blood gases that were manually entered by the anesthetic team into the anesthesia information management system were examined. From the recorded intraoperative PaO2 values and FiO2 the P/F ratio was calculated for each available blood gas. Volumes of crystalloid, colloid, units of packed erythrocytes (PRBC), units of fresh frozen plasma, and units of platelets were also obtained from the electronic anesthetic record.
The subpopulation of patients who went on to develop ARDS was identified from a prospectively collected research dataset of all adult critical care patients on ventilators at the University of Michigan Medical Center who were screened for entry into ARDS studies. Only patients receiving mechanical ventilation after their anesthetic were screened for ARDS. ARDS was diagnosed through analysis of the patient’s ventilator status, arterial blood gases, chest x-ray, and clinical documentation. Patients were deemed positive for the primary outcome of ARDS if they met the Berlin criteria for ARDS between postoperative days 0 and 7, inclusive. Charts of patients who developed ARDS on the day of their operation were examined by one of the authors (JMB), and those with a diagnosis of ARDS prior to their anesthetic were excluded. Finally, mortality data were collected from an institutional death database to compare the mortality of the risk-matched groups both with and without ARDS in order to determine the risk presented to patients who develop ARDS. This database is constructed using multiple resources including in-hospital mortality, failed follow-up at clinic visits and the social security death master file.
Statistical Analysis
Statistical analysis was performed using R version 2.15.2 (R Foundation for Statistical Computing, Vienna, Austria). Descriptive statistics were used to summarize the before and after ventilator change groups. Differences between groups were tested using the chi-square test, Fisher’s exact test, t-test, or Mann-Whitney U test, as appropriate, with p-values < 0.05 indicating statistical significance.
To reduce confounding and determine if changes in population characteristics may have impacted the development of postoperative ARDS, we performed matching on the likelihood for developing postoperative ARDS within both the two-month and two-year cohorts. These scores were calculated using logistic regression on variables that likely increase the risk of postoperative ARDS, from our prior work [8]. This included preoperative patient characteristics as well as intraoperative use of blood products. Before-ventilator-change cases were then matched 1:1 with after-ventilator-change cases using a nearest neighbor match, with a caliper of .001 on the risk score. Standardized mean differences were calculated to assess balance in the variables included in the risk score, with values less than 10% indicating good balance. Appropriate descriptive statistics and tests were again used to compare variables between groups after matching.