Intraoperative Driving Pressure and Postoperative Pulmonary Complications After Abdominal Surgery – A Posthoc Propensity Score–Weighted Cohort Analysis of the LAS VEGAS Study Comparing Open to Closed Surgery


 Background: It is uncertain whether associations between driving pressure (ΔP) during and occurrence of pulmonary complications after abdominal surgery depend on the surgical approach. Our primary objective was to test the time–weighted average ΔP (ΔPTW) association with postoperative pulmonary complications and our secondary objective was to test the association between ΔPTW and intraoperative Adverse Events. Methods: We realized a posthoc retrospective propensity score weighted cohort analysis of the ‘Local ASsessment of Ventilatory management during General Anesthesia for Surgery’ (LAS VEGAS) study including patients undergoing abdominal surgery from the study database including data from 146 hospitals across 29 countries. The primary endpoint was a composite of postoperative pulmonary complications. The secondary endpoint was the occurrence of intraoperative adverse events. Results: The analysis included 1,128 and 906 patients undergoing open or closed abdominal surgery repsectively. Absolute postoperative pulmonary complications rate was 5%. While driving pressure was lower in open abdominal surgery patients, time-weighted driving pressure was not different between groups. The association of ΔPTW with occurrence of postoperative pulmonary complications was significant in both groups, with a higher risk ratio in closed than in open abdominal surgery patients (1.11 [95%CI 1.10 to 1.20], P<0.001 vs. 1.05 [95% CI 1.05 to 1.05; P<0.001; risk difference 0.05: [95%CI 0.04 to 0.06], P<0.001). ΔPTW marginal effect estimation showed increased probability of pulmonary complication in both groups with a steeper increase in closed surgery patients at ΔPTW above 20 cmH2O∙hour-1. The association of ΔPTW with occurrence of intraoperative adverse events was also significant in both groups, with higher odds ratio in closed surgery (1.13 [95%CI 1.12 to 1.14]; P<0.001 vs. 1.07 [95%CI 1.05 to 1.10]; P<0.001; difference 0.05 [95%CI 0.03 to 0.07]; p<0.001). Conclusions: Our results show how driving pressure represents a marker for pulmonary complications and adverse events in abdominal surgery regardless of surgical approach.Trial registration: LAS VEGAS was registered at clinicaltrials.gov (trial identifier NCT01601223).


Introduction
The incidence of postoperative pulmonary complications varies between 9 and 40%, depending on de nitions and studied populations, [1] and their occurrence is associated with increased morbidity and mortality. [2,3] Postoperative pulmonary complications can be prevented by reducing lung strain using a low tidal volume (V T ) [4] and su cient positive end-expiratory pressure (PEEP). [5] Since the driving pressure, de ned as the difference between plateau pressure and PEEP, is also strongly associated with the development of postoperative pulmonary complications, [5,6] titrating V T and PEEP to minimise it could be an effective strategy to prevent pulmonary complications.
The respiratory system overall behaviour depends on its components properties, i.e., the arti cial and native airways, lung tissue, and the chest wall consisting of the rib cage and the diaphragm. A part of the force applied during invasive ventilation is used to expand the chest wall, and another fraction to in ate the lungs. [7] When the chest wall elastance increases, e.g., during pneumoperitoneum, the driving pressure increases even when V T is left unchanged. [8] This rise is interpreted as unharmful and may cause reluctance to target a low driving pressure in the presence of intraoperative pneumoperitoneum, i.e., during closed abdominal surgery. However, the cephalad shift of the diaphragm could induce or worsen atelectases, and the resulting increase in driving pressure is related with a rise in lung applied force. [9] Driving pressure effect during pneumoperitoneum can be thus mixed.
To determine and compare the exact associations between driving pressure and development of postoperative pulmonary complications in patients undergoing open abdominal surgery versus patients undergoing closed abdominal surgery, we reassessed the database of the 'Local ASsessment of Ventilatory management during General Anesthesia for Surgery' (LAS VEGAS) study. [10] The LAS VEGAS study was a prospective international observational study that showed a large proportion of surgery patients to be at an increased risk of pulmonary complications. It also showed intraoperative ventilation that consists of relatively high V T and low PEEP.
The primary hypothesis tested in this analysis was that the association between driving pressure and development of pulmonary complications is weaker in closed abdominal surgery patients than in open abdominal surgery patients. Our primary objective was to test the time-weighted average driving pressure (ΔP TW ) association with postoperative pulmonary complications, and our secondary objective was to test the association between ΔP TW and intraoperative adverse events.

Study design and setting
This is a posthoc analysis of the LAS VEGAS study database, [10] and was carried out in accordance to current guidelines and the recommendations of the statement for strengthening the reporting of observational studies in epidemiology (STROBE) (www.strobe-statemenent.org). The statistical analysis plan was prede ned, updated and nalised before data extraction and is presented as Additional File 1.
The LAS VEGAS study is a worldwide international multicentre prospective seven-day observational study describing intraoperative ventilation practice, occurrence of intraoperative complications and PPCs in the rst ve postoperative days, and hospital length of stay and mortality.
The study protocol was approved by the ethical committee of the Academic Medical Center, Amsterdam, the Netherlands (W12_190#12.17.0227). Each participating centre obtained approval from their institutional review board if needed, and patients were included after obtaining written informed consent when dictated by national or regional legislation. The LAS VEGAS study was partially funded and endorsed by the European Society of Anaesthesiology and registered at https://clinicaltrials.gov (NCT01601223, First posted date: 17/05/2012).

Inclusion and exclusion criteria
The LAS VEGAS study recruited patients undergoing general anaesthesia with mechanical ventilation for surgery consecutively during seven days in each participating centre between 14 January and 4 March 2013. Exclusion criteria of the LAS VEGAS study were: (1) age < 18 years, (2) invasive ventilation in the preceding month, (3) obstetric or ambulatory surgical interventions, and (4) cardiothoracic surgery cardiopulmonary bypass.
For the current posthoc analysis, the studied cohort is restricted to patients undergoing an abdominal intervention with su cient data to calculate driving pressure at least at one time point other than induction of anaesthesia. Also, to increase the homogeneity of the compared patient cohorts, patients who had received intraoperative ventilation through an airway device other than a tracheal tube and patients under assisted or spontaneous ventilation mode were excluded. Patients in whom laparoscopy only assisted the surgery, i.e., surgeries that could not be classi ed as mere open or mere closed abdominal surgery, were also excluded from the current analysis.

Data recording and calculations
Full details on data collection can be found in the original paper [10] and Additional File 2. In the study database, ventilatory parameters at every hour of surgery, from induction up to the last hour of surgery were recorded.
Using the data as collected in the LAS VEGAS database, the following calculations were done for the current analysis. Driving pressure was calculated by subtracting PEEP from plateau pressure or inspiratory pressure at every hour in volume-controlled and pressure-controlled ventilated patients, respectively. ΔP TW , i.e., the pressure that is proportional to the amount of time spent at each driving pressure in relation to the total time, was calculated by summing the mean values between consecutive time points multiplied by the time between those points and then dividing by the entire time. [11] Similarly, time-weighted average peak pressure and PEEP were determined. Details on calculations are provided in the Additonal File 2 Figure S1.

De nitions
The LAS VEGAS study had a rigid study protocol that was approved before data collection took place, as reported elsewhere. [10] The statistical analysis plan was written, updated, and nalised before data extraction. We used for this analysis severe postoperative pulmonary complications de ned in the protocol as a collapsed composite of the following events: unexpected postoperative invasive or noninvasive ventilation, acute respiratory failure, acute respiratory distress syndrome, pneumonia, and pneumothorax. The occurrence of each type of complication was monitored until hospital discharge but restricted to the rst ve postoperative days.
Intraoperative adverse events were prede ned and described in the protocol of the LAS VEGAS study as follows: any oxygen desaturation or lung recruitment manoeuvres performed to rescue from hypoxemia, any need for adjusting ventilator settings for reducing airway pressures or correction of expiratory ow limitation, any hypotension or need for vasoactive drugs, and any new cardiac arrhythmia. Since the simultaneous occurrence of various adverse events is frequent, we analysed them as an ordinal variable with a range spanning from zero to seven adverse events.
A detailed list of de nitions of the composites of postoperative pulmonary complications and intraoperative adverse events is provided in Additional File 2Table S1 and Table S2.

Endpoints
The primary endpoint was the composite of postoperative pulmonary complications. The secondary endpoint was the occurrence of one or more intraoperative adverse events.

Control of Bias
Bias controlling strategy is reported in Additional File 2 Statistics.

Analysis plan
The analysis plan was prespeci ed before data access, and we used data of all available patients in LAS VEGAS database without formal sample size calculation. Also, as the purpose of the analysis was exploring a physiological hypothesis, we did not specify any a priori effect size.
Continuous variables were reported as median and interquartile ranges; categorical variables expressed as n (%). Normality of distributions was assessed by inspecting quantile-quantile plots. If variables were normally distributed, the two-sample t-test was used, if not, the Wilcoxon rank sum test was used. For categorical variables, we used the Chi-square test or Fisher's exact test, or when appropriate as relative risks. Statistical uncertainty was expressed by showing the 95%-con dence intervals (CI).
We estimated whether ΔP TW was associated with the occurrence of postoperative pulmonary complications with a weighted mixed-effect logistic regression and whether ΔP TW was associated with intraoperative adverse events with a mixed ordinal regression. To t every model, centres as a random intercept and an inverse probability weighting (IBW) factor computed from the covariate-balancing propensity score (CBPS) method to simultaneously optimise prediction of treatment assignment, i.e. ΔP TW as a continuous variable, and confounders in uence were introduced. [ To build the CBPS to relate exposure variable, i.e. ΔP TW , with potential confounders, we included by clinical judgment the Assess Respiratory Risk in Surgical Patients in Catalonia (ARISCAT) risk class, [13,14] and the average intraoperative V T and then we performed feature selection with an augmented backward elimination selection method introducing 37 pre-and intraoperative variables (Additional File 2 Statistics for a detailed list). The selection was based on a sequential process where initially all variables entered the model and nally those pre-and intraoperative factors that yielded a change in the effect estimate > 0.1 and a signi cance criterion (alpha) < 0.1 were included. The algorithm stopped when all variables left in the model complied with both criteria. [15] We carried out a selection process of potential variables to avoid bias in the effect estimates using a comprehensive strategy to prevent the drawbacks of simple stepwise methods. [16] The internal validation of the model was assessed by bootstrap using ve hundred generated samples and estimating the Area Under Curve (AUC) of the full and stepwise-selected variables models.
To further unravel the role of surgical approach, i.e., closed versus open abdominal surgery, on postoperative pulmonary complications we performed a sensitivity analysis tting a mixed logistic regression with a random intercept for centre on a propensity score matched cohort. The propensity was used to match patients with a similar covariable structure using the R matchit package carrying out the matching with the nearest neighbour method with a caliper of 0.1 with a ratio of patients in the open surgery arm of 2 to 1. Full details on the covariables introduced in the propensity score matching procedure can be found in the Additional File 2 Statistics. To assess the type of surgery as an effect modi er, we carried out another sensibility analysis tting a weighted mixed logistic regression model an all data, i.e. both surgery cohorts, introducing type of surgery as an independent variable and an interaction term between ΔP TW and type of surgery.
Statistical signi cance was considered for two-tailed P<0.05. No imputation routine of missing values and no correction for multiple comparisons was prespeci ed; thus, all the ndings should be viewed as exploratory. All analyses were performed with R 3.5.2 (The R Foundation for Statistical Computing, www.r-project.org).  Table 2).

Patients
Primary and secondary outcome rates In 102 (5%) patients, one or more postoperative pulmonary complications occurred, and their prevalence was higher in open surgery patients than in patients who underwent a closed surgical procedure (7 vs. 3%; P<0.001). Hypotension or need for vasopressors was more frequently observed during open surgery, while the need for airway pressure reduction was more often needed during closed surgery (Table 3).

Propensity score estimation variables
The variables that nally entered the propensity core and covariate balance assessment are detailed in the Additional File 2 Statistics and Figure S2 and S3.
Association between driving pressure and occurrence of postoperative pulmonary complications by type of surgery on postoperative pulmonary complication occurrence probability is showed in Figure 3. A rise in ΔP TW was associated with increased probability of pulmonary complications in both types of surgery with a steeper increase in closed surgery patients for ΔP TW above 20 cmH 2 O • hour -1 .
After matching, the resulting cohort consisted of 344 open surgery patients, and 254 closed surgery patients. Baseline characteristics between groups were well balanced (Additional File 2Table S2 and S3).
Type of surgery at matched levels of driving pressure was not associated with either outcome. (Additional File 2Table S4 and S5).

Discussion
The analysis' main ndings can be summarised as follows: (i.) the intraoperative ΔP TW was not different between open and closed surgery groups, and (ii.) was associated with an increased risk of pulmonary complications occurrence of 10% and 5% for each additional cmH2O •hour -1 in closed and open surgery patients respectively; (iii.) was associated with the appearance of intraoperative adverse events, and (iv.) type of surgery has a modifying effect on the association between ΔP and postoperative pulmonary complications, although this was not con rmed in the matched cohort analysis.
This analysis uses the database of a worldwide international multicentre prospective observational study as a convenience sample, [10] strictly followed a plan, and was characterised by a robust method accounting for the multilevel data structure and allowing precise estimation and confounder control, even with seven or fewer events per confounder. [17,18] Also, the outcome of interest, i.e., severe postoperative pulmonary complications, was prede ned, well-described, and largely followed the European Perioperative Clinical Outcome (EPCO) group de nitions. [19] A recent metanalysis of individual trials' on protective ventilation during general anaesthesia including patients undergoing cardiac and thoracic surgery found a signi cant association between driving pressure and the occurrence of pulmonary complications (OR 1.16, 95% CI 1.13 to 1.19; p<0·0001). [5] We found an almost identical association between driving pressure and pulmonary complications in closed abdominal surgery patients. Our results, thus, con rm that driving pressure might be a promising target for preventative interventions aiming at reducing pulmonary complications also in patients undergoing closed surgery. While the sensibility analysis with type of surgery as interaction con rmed the main analysis' results, the propensity score matched sensitivity analysis did not, although the matching process lead to a decrease in sample size thus limiting statistical power.
Respiratory driving pressure is an indicator of the amount of strain delivered to the respiratory system during mechanical ventilation. [7] Several studies investigated the effect of pneumoperitoneum on respiratory mechanics. Pneumoperitoneum was consistently found to decrease chest wall compliance, whereas lung compliance seems mostly spared by it. [20][21][22][23][24][25][26][27] Thus, inferring the amount of lung strain from plateau pressure and PEEP during pneumoperitoneum is challenging, since the part of the rise in plateau pressure caused by chest wall stiffening should not intensify lung it. [28] Consequently, a higher driving pressure during closed abdominal surgery could be less harmful or even 'non-injurious'. The results of the current study reject this assumption, as a driving pressure rise was stronger associated with an increase in the occurrence of PPCs in closed abdominal surgery patients compared to open abdominal surgery.
Indeed, pneumoperitoneum can affect lung mechanics in several ways. [20][21][22][23][24][25][26][27] A cranial shift of the diaphragm during laparoscopic surgery increases alveolar collapse, especially in lung parts close to the diaphragm. This is particularly true in upper abdominal surgery, which was the most common surgical procedure in patients undergoing closed surgery in the here studied cohort. [29,30] PEEP may partially prevent this, and usually only when high PEEP is used. [31] In the patients studied here, mostly low PEEP was used, regardless of the group. Additional studies are needed to test how high PEEP affects the association between intraoperative driving pressure and postoperative pulmonary complications. Also, we found that driving pressure was higher in closed as compared to open abdominal surgery patients.
However, since open abdominal surgeries lasted longer, ΔP TW was remarkably similar in the two study groups, thus, at least in part, a higher absolute driving pressure was compensated by shorter duration of intraoperative ventilation, and vice versa. Indeed, the time-weighted parameter allows to estimate an exposure limit threshold to an injurious factor as in occupational health. The steeper increase in probability of pulmonary complications above a 20 cmH 2 O•hour -1 can be related with increased atelectasis at low PEEP in this cohort of patients.
As expected, postoperative pulmonary complications occurred more frequently in open abdominal surgery patients. This could be explained by an increased baseline risk for pulmonary complications due to typical differences in patient characteristics but also given the duration and the type of surgery. However, this nding that pulmonary complications occurred more often in open abdominal surgery patients strengthens the current analysis since we observed the association even in a cohort of patients, i.e. closed abdominal surgery, at low risk for postoperative pulmonary complications and even after controlling for confounding effects with propensity score analysis.
Several intraoperative ventilation approaches, like the use of recruitment manoeuvres and higher PEEP, may result in a lower driving pressure. [32,33] Findings of a metanalysis including clinical trials on intraoperative ventilation suggest that PEEP titrations that resulted in a driving pressure rise increased the risk of postoperative pulmonary complications. [5] One randomised clinical trial showed an intraoperative PEEP strategy targeting the best compliance to reduce occurrence of pulmonary complications, though this was only a secondary endpoint in that study. [34] Thus, the best approach to minimise postoperative pulmonary complications is still debated.
ΔP TW was associated with intraoperative adverse events in both closed and open surgery patients.
Among adverse events, airway pressure reduction was more frequently needed in closed surgery group underlining the need of ventilation strategies aimed at lowering peak and plateau pressures in this group of patients re ecting unacceptable high airway pressure during surgery.
Several limitations must be acknowledged. Our used PPCs de nitions are previous to those recently proposed, [1] although comparable. The LAS VEGAS study protocol did not include oesophageal pressure recording. Information regarding surgical positioning was not collected, and intra-abdominal pressure levels were not recorded in the database of the LAS VEGAS study. Both could in uence ΔP. [35][36][37] Also, the granularity of airway pressure data is limited to hourly time points and time and driving pressure values are assessed as a whole thus we cannot estimate the effect of pressure alone. Furthermore, we only included patients intubated and ventilated in controlled mode, thus limiting our focus on a speci c type of intraoperative airway device and ventilation mode, however still representative of most surgical patients. Of note, 25% of patients had a Body Mass Index (BMI) > 30. Extrapolation of the ndings of this analysis to obese or morbidly obese patients should be done with some caution. Also, the original LAS VEGAS study was performed 7 years ago. Since then, there could have been changes in clinical practice, e.g., in the use of 'Enhanced Recovery After Surgery' (ERAS) pathways, as well as muscle relaxant monitoring during and reversal at the end of surgery. Although the time gap between research ndings and practice changes usually lasts longer than a decade, [38][39][40] it still could be that more immediate changes may affect the here found associations. Finally, we did not set any a priori effect threshold nor multiple comparisons correction; hence the results' statistical signi cance and the exploratory nature of secondary outcome analysis must be con rmed in future trials.

Funding
The LAS VEGAS study was endorsed and partly funded by a restricted research grant from the European Society of Anesthesiology through their Clinical Trial Network.

Availability of Data and Materials
The data as well as the code used for analysis are available from the corresponding author upon reasonable request.

Ethics approval and consent to participate
The original study protocol was approved by the ethical committee of the Academic Medical Center, Data are presented as median [25 th -75 th percentile] or % (n/N). For binary and continuous variables risk difference and median difference with con dence intervals are reported respectively. Abbreviations: EtCO 2 , end-tidal CO 2 ; FiO 2 , fraction of inspired oxygen; SpO 2 , peripheral oxygen saturation, OR, Odds ratio.
*Difference between groups is signi cant but very small and masked by rounding process. PPC, postoperative pulmonary complications. Figure 1 Patients' inclusion owchart. Patients' inclusion owchart. represents the induction of general anaesthesia. Solid lines are means, and bandwidths is 95% bootstrapped con dence intervals. Gray boxes : More than 95% of data points represented.