You are on page 1of 10

ORIGINAL ARTICLE

Prolonged Air Leak After Pulmonary Resection Increases Risk


of Noncardiac Complications, Readmission, and Delayed
Hospital Discharge
A Propensity Score-adjusted Analysis
Adam Attaar, MD,  James D. Luketich, MD,  Matthew J. Schuchert, MD,  Daniel G. Winger, MS,y
Downloaded from https://journals.lww.com/annalsofsurgery by 01UGrXh3ipqzR4DKqW7bOJtSxsRVOheLV9OzOeHq2POX2t1GrKUD6m1aBQdlm0lX7bZxxCFBpuxAM3exuxdXARVaPXROhrEIELxLuiE5dAoxEXL1EbBXOmtN5qOe2K0cl7bPi36/SOM4st4r1MMZ7w== on 03/16/2021

Inderpal S. Sarkaria, MD,  and Katie S. Nason, MD, MPH  z§

complications, unexpected admission to the ICU, and 30-day mortality are


Objective: The aim of the study was to determine whether prolonged air leak
not after propensity score adjustment.
(PAL) is associated with postoperative morbidity and mortality following
pulmonary resection after adjusting for differences in baseline characteristics Keywords: air leak, boosted regression, clinical care pathways,
using propensity score analysis. complications, fast tracking, inverse probability treatment weighting, lung
Summary Background Data: Patients with PAL after lung resection have cancer, persistent air leak, prolonged air leak, pulmonary resection
worse outcomes than those without PAL. However, adverse postoperative
(Ann Surg 2021;273:163–172)
outcomes may also be secondary to baseline risk factors, such as poor lung
function.
Methods: Patients who underwent pulmonary resection for lung
cancer/nodules (1/2009 –6/2014) were stratified by the presence of PAL P rolonged air leak (PAL) is defined by the Society of Thoracic
Surgeons General Thoracic Surgery Database (STSGTSD) as a
parenchymal air leak lasting >5 days. PAL complicates 6% to 18%
[n ¼ 183 with/1950 without; defined as >5 d postoperative air leak;
n ¼ 189 (8.3%)]; probability estimates for propensity for PAL from of lung resections.1– 7 Various surgical centers have reported that
31 pretreatment/intraoperative variables were generated. Inverse proba- PAL increases postoperative morbidity and mortality with a length-
bility-of-treatment weights were applied and outcomes assessed with ened hospital stay,4,5,7 –18 and increased incidence of empy-
logistic regression. ema9,10,14,19 commonly reported. Data conflict over higher
Results: Standardized bias between groups was significantly reduced incidences of pneumonia,17,19 mortality,5,20 readmission to the
after propensity weighting (mean ¼ 0.18 before vs 0.08 after, P < 0.01). ICU,5,17 30-day readmission,4 and cardiac complications.16 Numer-
After propensity weighting, PAL was associated with increased odds of ous studies have investigated the risk factors of PAL; consistently
empyema (OR ¼ 8.5; P < 0.001), requirement for additional chest tubes reported risk factors include older age,1,14 male sex,2,6,21 lower body
for pneumothorax (OR ¼ 7.5; P < 0.001), blood transfusion (OR ¼ 2; mass index,2,3,6,7,21 reduced pulmonary function,1 –7,21 upper lobec-
P ¼ 0.03), pulmonary complications (OR ¼ 4; P < 0.001), unexpected tomy,1,2,5,22 pleural adhesions,1– 3 and undergoing anatomic lung
return to operating room (OR ¼ 4; P < 0.001), and 30-day readmission resection,2,4,7 among others. Thus, we recognize the PAL population
(OR ¼ 2; P ¼ 0.009). Among other complications, odds of cardiac is distinct in demographics, medical history, and surgical treatment.
complications (P ¼ 0.493), unexpected ICU admission (P ¼ 0.156), and These differences confound the relationship of PAL and other
30-day mortality (P ¼ 0.270) did not differ. Length of hospital stay was postoperative complications.
prolonged (5.04 d relative effect, 95% confidence interval, 3.77 – 6.30; Baseline differences that confound the association between
P < 0.001). PAL and outcomes of interest (eg, mortality) can be balanced using
Conclusions: Pulmonary complications, readmission, and delayed hospital propensity score analysis, allowing ‘‘apples to apples’’ compari-
discharge are directly attributable to having a PAL, whereas cardiac son.23,24 Two published studies14,16 have used propensity score
methods to assess the complications of air leak, but were limited
by differing conclusions on cardiopulmonary risk, small sample size,
and use of logistic regression, which can be prone to less stable, less
From the Department of Cardiothoracic Surgery, Division of Thoracic and precise, and more biased estimates in moderate-sized datasets.25
Foregut Surgery, Pittsburgh, PA; yClinical and Translational Science Institute, We hypothesized that the previously reported worse postop-
University of Pittsburgh, Pittsburgh, PA; zInstitute for Healthcare Delivery and
Population Science, University of Massachusetts Medical School-Baystate, erative course of PAL patients can primarily be attributed to differ-
Springfield, MA; and §Department of Surgery, University of Massachusetts ences in baseline characteristics rather than PAL occurrence, and
Medical School-Baystate, Springfield, MA. investigated the relationship of PAL and postoperative morbidity and
Award Numbers K07CA151613 (to KSN) and UL1-TR-001857 from the National mortality after adjustment with propensity score weighting.
Cancer Institute and National Institutes of Health supported the study. The
content is solely the responsibility of the authors, and does not represent the
official views of the funding source. METHODS
The authors report no conflicts of interest.
Supplemental digital content is available for this article. Direct URL citations Patient Population and Data Definitions
appear in the printed text and are provided in the HTML and PDF versions of
this article on the journal’s Web site (www.annalsofsurgery.com). We collected data using variables defined by the STSGTSD
Reprints: Katie S. Nason, MD, MPH, 2 Medical Center Dr, Suite 205 Medical (http://www.sts.org/national-database) in versions 2.081 and 2.2.
Office Building, Springfield, MA 01060. Trained personnel abstracted patient data 4 to 6 weeks after operation
E-mail: Katie.NasonMD@baystatehealth.org. for real-time quality monitoring and national benchmarking via
Copyright ß 2019 Wolters Kluwer Health, Inc. All rights reserved.
ISSN: 0003-4932/19/27301-0163 biannual data submission to the STSGTSD National Data Center.
DOI: 10.1097/SLA.0000000000003191 Our Institutional Review Board gave approval for use of this data.

Annals of Surgery  Volume 273, Number 1, January 2021 www.annalsofsurgery.com | 163

Copyright © 2020 Wolters Kluwer Health, Inc. All rights reserved.


Attaar et al Annals of Surgery  Volume 273, Number 1, January 2021

Pulmonary resection was performed (n ¼ 2485; from Statistical Analysis


January 1, 2009 to June 30, 2014) at 8 hospital sites for malignant We compared baseline characteristics and outcomes using chi-
and benign lung tumors or nodules using International Classification squared test or Fisher exact test for categorical variables expressed as
of Diseases, Ninth Revision26 diagnosis codes (197.0, 212.3, 162.2, frequencies, and Student t test or Wilcoxon rank sum test for
162.3, 162.4, 162.5, 162.9, and 518.89), excluding pneumonectomy, continuous variables expressed as mean  1 SD or median/inter-
bilobectomy, and extended chest wall/diaphragm resections. Data- quartile range. We performed statistical analysis using Stata/SE 14.1
base variables included a wide range of perioperative data: patient (StataCorp, 2015; College Station, TX: StataCorp LP) and R version
demographics, preoperative evaluation, surgical procedures, cancer 3.2.3 (R Foundation for Statistical Computing, Vienna, Austria).
staging, and postoperative events. We excluded 59 patients who had Unless otherwise indicated, P values were 2-tailed; statistical signif-
undergone sleeve lobectomy to minimize confounding parenchymal icance was defined by P < 0.05.
and anastomotic air leaks; 31 patients from 2 hospital sites who did
not submit patients to the database over the entire study time frame; RESULTS
and 5 patients who died by day 5 or before (PAL definition). Multiple
lung resections were noted in 111 patients; data for the most recent Baseline Characteristics and Propensity Adjustment
surgery date were used and previous encounter(s) excluded. A total of 2280 patients underwent pulmonary resection for
lung cancer/lung tumors between January 2009 and June 2014. The
Outcomes distribution of the baseline characteristics and outcomes stratified by
PAL was defined as an air leak that persisted for >5 days PAL status are shown in Table 1 and Table 2, respectively. Incidence
postoperatively. Outcomes of interest included pulmonary, cardiac, of PAL was 8.3% (189/2280). The majority of operations in our study
infective, readmission, and mortality adverse events that occurred were video-assisted thoracoscopic (VATS) lobectomy/segmentec-
within 30 days of surgery or during the same admission regardless of tomy for primary lung cancer. A total of 2133 patients who had
length of stay. complete data on baseline characteristics were included in the
propensity score analysis. Before propensity score weighting, 19
Estimating Propensity Scores of the 31 variables were more than minimally imbalanced (stan-
Propensity scores for the probability of PAL conditional on dardized mean difference >0.1), and 7 variables more than moder-
observed characteristics were generated using generalized boosted ately imbalanced (standardized mean difference >0.2) with an
regression. Compared with traditional logistic regression models, average mean difference of 0.18  0.18. Procedure type (45%),
generalized boosted regression accounts for nonlinearity, better surgical approach (40%), COPD (38%), %FEV1 (36%), BMI
tolerates collinearity, remains insensitive to outliers, and models (34%), smoking history (26%), and reoperation (25%) had more
interaction terms, which is important in a moderate-size dataset that than moderate imbalance (Figure 1). The percent relative influence
has many potential confounders.25,27 We used the R package twang (a model diagnostic) of the variables (variables >1% shown) in the
using model refined specifications of 15,000 trees, maximum inter- creation of the propensity scores is shown in Figure 2. Figure 3 shows
action depth of 3, shrinkage of 0.0005, and 50% bag fraction28. We an unweighted and weighted histogram of the distribution of pro-
selected the 31 baseline variables shown in Table 1. The only pensity scores. It depicts that, after adjustment, the characteristics of
significantly missing data were for %FEV1 (percent forced expira- patients with and without PAL are more comparable with respect to
tory volume in 1 s); for our main model, we balanced on the observed their probability of PAL (ie, their propensity score). After weighting,
values of each covariable as well as the missing data pattern for the average mean standardized difference improved to 0.08  0.06
%FEV1.29,30 (Figure 1). To allow doubly-robust estimation of effect size, we
included age, BMI, surgical approach, and procedure type in addition
Propensity Weighting and Outcome Analysis to PAL in a multivariable weighted logistic regression; these cova-
We explored different balancing techniques (Supplemental riables are known important risk factors for PAL that had residual
Table 1, http://links.lww.com/SLA/B558)27 and chose IPTW imbalance after weighting.
(inverse probability treatment weighting) given good reduction in
standardized bias, advantage of retaining all patients in the outcome Risk of Cardiac, Pulmonary, and Infective
evaluation, and ability to estimate the average treatment effect Postoperative Adverse Events
(ATE).32 We assessed balance using absolute standardized mean After adjustment, PAL patients had 8.5 times the odds of
difference; differences 20% to 40% were considered as moderate empyema, 7.5 times the odds of pneumothorax, 2 times the odds of
imbalance, and between 10% and 20% residual imbalance, and blood transfusion with packed red blood cells, 3 times the odds of
<10% minimal imbalance.31 We further adjusted for those covari- other recorded pulmonary event, and 53 times the odds of being
ables that had postweighting standardized difference >10%, and are discharged with a chest tube, compared with patients without PAL
known significant risk factors for PAL from previous studies used a (Table 3). There was a trend toward higher odds of atelectasis
generalized logistic model (R packages survey and glm) to achieve requiring bronchoscopy. All other cardiac, pulmonary, and infective
doubly-robust estimation.31 postoperative events did not significantly differ between those with
and without PAL after adjustment. Initial ventilatory support >48 h,
Sensitivity Analyses reintubation, other pulmonary event, and atrial arrhythmia, which
We calculated the treatment effect estimates using 3 alternate had shown strong associations with PAL (P < 0.001) before propen-
models by performing the analysis described above except for the sity score adjustment, were no longer significant (Table 3). There was
following difference: (1) included patients if in the overlap in propensity a trend toward higher odds of pneumonia that was not significant
scores of PAL and non-PAL cases (ie, region of common support) 33; (2) after propensity adjustment. PAL patients had significantly higher
excluded patients missing on %FEV1 (ie, complete case analysis); (3) odds of 1 cardiac complication or 1 pulmonary complication
performed and compared results from 2:1 nearest neighbor propensity before adjustment. After adjustment, PAL patients had 3.5 times the
matching (R nonrandom package)32 (further information on the spec- odds of overall pulmonary complications, but no significantly higher
ifications of 2:1 matching available in appendix). risk of overall cardiac complications.

164 | www.annalsofsurgery.com ß 2019 Wolters Kluwer Health, Inc. All rights reserved.

Copyright © 2020 Wolters Kluwer Health, Inc. All rights reserved.


Annals of Surgery  Volume 273, Number 1, January 2021 Prolonged Air Leak Worsens Outcomes

TABLE 1. Baseline Characteristics of Patients With or Without Prolonged Air Leak Used to Generate Propensity Scores
Prolonged Air Leak
Variables Total n ¼ 2280 No n ¼ 2091 Yes n ¼ 189 P
Demographics
Age (mean  SD) 65  12 65  12 67  11 0.015
2
BMI, kg/m (mean  SD) 28  7 29  7 26  5 <0.001
Sex
Female 1277 (56) 1178 (56) 99 (52) 0.294
Male 1003 (44) 913 (44) 90 (48)
Race/ethnicity
White 2133 (94) 1950 (93) 183 (97) 0.056
Other race/ethnicity 147 (6) 141 (7) 6 (3)
Treatment variables
Surgery year
2009 431 (19) 396 (19) 35 (19) 0.984
2010 405 (18) 368 (18) 37 (20)
2011 445 (20) 411 (20) 34 (18)
2012 431 (19) 396 (19) 35 (19)
2013 392 (17) 359 (17) 33 (18)
2014 (end 6/14) 176 (8) 161 (8) 15 (8)
Surgeon <0.001
A 414 (18) 367 (18) 47 (25)
B 351 (15) 317 (15) 34 (18)
C 342 (15) 303 (15) 39 (21)
D 202 (9) 188 (9) 14 (7)
E 189 (8) 175 (8) 14 (7)
F 173 (8) 167 (8) 6 (3)
G 148 (7) 131 (6) 17 (9)
H 132 (6) 127 (6) 5 (3)
I 100 (4) 96 (5) 4 (2)
Surgeons <100 casesy 229 (10) 220 (11) 9 (5)
Disease category
Stage I 1012 (46) 916 (45) 96 (52) 0.002
Stage II 223 (10) 192 (9) 31 (17)
Stage III/IV 113 (5) 103 (5) 10 (5)
Benign tumor 80 (4) 74 (4) 6 (3)
Metastatic tumor 506 (23) 478 (24) 28 (15)
Benign nodule 284 (13) 269 (13) 15 (8)
Procedure type
Lobe/Segment 1500 (66) 1339 (64) 161 (85) <0.001
Wedge resection 780 (34) 752 (36) 28 (15)
Comorbidities
Smoking history
Never smoker 678 (30) 649 (31) 29 (15) <0.001
Past smoker 1155 (51) 1042 (50) 113 (60)
Current smoker 447 (20) 400 (19) 47 (25)
Zubrod score
0 162 (7) 152 (7) 10 (5) 0.032
1 1728 (76) 1594 (77) 134 (71)
2–5 382 (17) 338 (16) 44 (23)
ASA classification
I/II 276 (12) 262 (13) 14 (7) 0.077
III 1749 (77) 1596 (76) 153 (81)
IV 255 (11) 233 (11) 22 (12)
Hypertension 1269 (56) 1171 (56) 98 (52) 0.271
Coronary artery disease 479 (21) 434 (21) 45 (24) 0.329
Peripheral vascular disease 221 (10) 196 (9) 25 (13) 0.087
Interstitial fibrosis 27 (1) 23 (1) 4 (2) 0.217§
Diabetes 419 (18) 392 (19) 27 (14) 0.077
COPD 754 (33) 660 (32) 94 (50) <0.001
Cerebrovascular disease 138 (6) 125 (6) 13 (7) 0.622
Preoperative chemotherapy 464 (20) 430 (21) 34 (18) 0.395
Preoperative radiation Therapy 308 (14) 279 (13) 29 (15) 0.446
Steroids 124 (5) 112 (5) 12 (6) 0.565
Prior cardiothoracic surgery 503 (22) 450 (22) 53 (28) 0.040
z
Reoperation 283 (13) 245 (12) 38 (20) 0.001
Laboratory
FEV1, % predicted (13.0% missing; mean  SD) 83  22 84  21 76  24 <0.001
Last hemoglobin (mean  SD) 12.1  1.7 12.2  1.7 12.0  1.8 0.238
Last creatinine, median [IQR] 0.88 [0.7–1] 0.90 [0.7–1] 0.80 [0.7–1] 0.582ô
Operative details
Status Elective 2178 (96) 1993 (96) 185 (98) 0.181
Urgent/Emergent 89 (4) 85 (4) 4 (2)

Preoperative hospitalization <1 d 2122 (93) 1939 (93) 183 (97) 0.037
1 d 156 (7) 150 (7) 6 (3)
Laterality Left 912 (40) 851 (41) 61 (32) 0.017

ß 2019 Wolters Kluwer Health, Inc. All rights reserved. www.annalsofsurgery.com | 165

Copyright © 2020 Wolters Kluwer Health, Inc. All rights reserved.


Attaar et al Annals of Surgery  Volume 273, Number 1, January 2021

TABLE 1. (Continued)
Prolonged Air Leak
Variables Total n ¼ 2280 No n ¼ 2091 Yes n ¼ 189 P
Right 1343 (60) 1215 (59) 128 (68)
Missing 25 (1.1) 25 (1.2) 0 (0)
Robot-assisted surgery 82 (4) 79 (4) 3 (2) 0.121
Surgical approach Thoracotomy 566 (25) 489 (23) 77 (41) <0.001
VATS 1714 (75) 1602 (77) 112 (59)
Values n (%), and <1% missing unless indicated otherwise.

Entered in model as continuous variable.
ySurgeons anonymously lettered. Nine surgeons with <100 operated cases in study period were categorized together.
zCardiac or thoracic reoperation that affects operative field.
§Fisher exact test.
ôWilcoxon rank-sum test.
BMI indicates body mass index, %FEV1, percentage of predicted value of forced expiratory volume in 1 s; DLCO, diffusing capacity of carbon monoxide; ASA, American
Association of Anesthesiology; COPD, chronic obstructive pulmonary disease; VATS, video-assisted thoracic surgery SD, standard deviation; OR, operating room; IQR, interquartile
range.

TABLE 2. Postoperative Major Adverse Events, Mortality, and Length of Stay in Patients With or Without Prolonged Air Leak
Prolonged Air Leak
Complications Total n ¼ 2280 No n ¼ 2091 Yes n ¼ 189 P
Pulmonary
Atelectasis req. bronchoscopy 162 (7.1) 135 (6.5) 27 (14.3) <0.001
Pneumonia 145 (6.4) 124 (6.0) 21 (11.1) 0.005
ARDS 27 (1.2) 23 (1.1) 4 (2.1) 0.276
Bronchopleural fistula 6 (0.3) 4 (0.2) 2 (1.1) 0.082
Pulmonary embolus 23 (1.0) 20 (1.0) 3 (1.6) 0.432
Pneumothorax req. chest tube reinsertion 104 (5) 65 (3.1) 39 (20.6) <0.001
Initial ventilatory support >48 h 47 (2.1) 37 (1.8) 10 (5.3) 0.004
Reintubation 93 (4.1) 76 (3.6) 17 (9.0) <0.001
Tracheostomy 53 (2.3) 46 (2.2) 7 (3.7) 0.201
Discharged with chest tube 110 (4.8) 28 (1.3) 82 (43.6) <0.001
Other pulmonary event 80 (3.5) 63 (3.0) 17 (9.0) <0.000
>1 pulmonary eventy 392 (17.2) 313 (15.0) 79 (41.8) <0.000
Cardiovascular
Atrial arrhythmia req. treatment 183 (8.0) 157 (7.5) 26 (13.8) 0.002
Ventricular arrhythmia req. treatment 7 (0.3) 5 (0.2) 2 (1.1) 0.109
Myocardial infarct 7 (0.3) 7 (0.3) 0 (0.0) 1.000
DVT req. treatment 20 (0.9) 19 (0.9) 1 (0.5) 1.000
Other cardiovascular eventz 23 (1.0) 18 (0.9) 5 (2.6) 0.036
>1 cardiovascular event§ 219 (9.6) 189 (9.0) 30 (15.9) 0.002
Infective/Other
Empyema 15 (0.7) 6.0 (0.3) 9 (4.8) <0.001
Another infection req. antibiotics 38 (1.7) 33 (1.6) 5 (2.6) 0.239
Sepsis 26 (1.1) 20 (1.0) 6 (3.2) 0.017
Other events req. OR return with anesthesia 56 (2.5) 43 (2.1) 13 (6.9) <0.001
Chylothorax 11 (0.5) 8 (0.4) 3 (1.6) 0.056
Renal failure 26 (1.1) 23 (1.1) 3 (1.6) 0.471
Packed red blood cells 160 (7.0) 130 (6.2) 30 (15.9) <0.001
Readmission/Mortality
Unexpected return to ORô 77 (3.4) 53 (2.6) 24 (13) <0.001
Unexpected admission ICU 101 (4.5) 83 (4.0) 18 (9.5) <0.001
Mortality at discharge 28 (1.2) 23 (1.1) 5 (2.6) 0.076
30-d readmissionjj 171 (7.6) 143 (6.9) 28 (15.2) <0.001
30-d mortality 33 (1.4) 26 (1.2) 7 (3.7) 0.007
Length of stay, days; median [IQR] 4 [3–7] 4 [3–6] 10 [8–14] <0.001
Values n (%), and <1% missing unless indicated otherwise.

Other pulmonary event—another pulmonary event in the postoperative period.
Fisher exact test.
y1 pulmonary complication—any of the following: atelectasis requiring bronchoscopy, pneumonia, ARDS, bronchopleural fistula, pulmonary embolus, pneumothorax requiring
chest tube reinsertion, initial ventilator support >48 h, reintubation, tracheostomy, other pulmonary event.
z1 cardiac complication—any of the following: atrial arrhythmia req. treatment, ventricular arrhythmia requiring treatment, myocardial infarct, DVT requiring treatment, other
cardiovascular event.
§Other cardiovascular event—any other CV event including distal arterial embolism in the postoperative period.
ô1.7% missing (39/2280).
jj1.3% missing (30/2280).
Req. indicates requiring; ARDS, acute respiratory distress syndrome; DVT, deep vein thrombosis; OR, operating room; ICU, intensive care unit; IQR, interquartile range.

166 | www.annalsofsurgery.com ß 2019 Wolters Kluwer Health, Inc. All rights reserved.

Copyright © 2020 Wolters Kluwer Health, Inc. All rights reserved.


Annals of Surgery  Volume 273, Number 1, January 2021 Prolonged Air Leak Worsens Outcomes

%FEV1 Unmeasured
Procedure Type
Surgical Approach
COPD
%FEV1
BMI
Smoking History
Reoperaon
Age
Laterality
Surgeon
Prior CT Surgery
Race
Diabetes
Disease Category
Status
PVD
Zubrod Score
Robot -Assisted Surgery
Preop Hospitalizaon
Hemoglobin
ASA Class
Hypertension
Preop Radiaon
Therapy
Sex
Creanine Unweighted
Preop Chemotherapy
Intersal Fibrosis
Propensity Weighted
CAD
Cerebrovascular Disease
Surgery Year
Steroids
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Standardized % Bias
FIGURE 1. Absolute standard difference between patients with and without prolonged air leak for each of the 31 potential
confounders: unweighted and weighted samples.

Risk of Unexpected Return to Operating Room, and nearest neighbor 2:1 matching. All outcomes remained
Intensive Care Unit, and 30-day Readmission unchanged with regard to their significance except for blood trans-
After adjustment, PAL patients had 4 times the odds of fusion, which was no longer significant in complete case analysis and
unexpected return to the operating room and 2 times the odds of nearest neighbor matching, and pulmonary embolus, which became
30-day readmission compared with patients without PAL. There were significant in complete case analysis (Table 4).
significantly higher odds of unexpected admission to ICU before
adjustment, but after adjustment, PAL patients demonstrate no DISCUSSION
significantly higher risk. Previous works1– 6, including a clinical prediction model for
PAL published at our institution,7 have shown that patients who had
Mortality and Length of Stay an occurrence of PAL generally have poorer lung function, more
The overall rate of 30-day mortality was significantly higher comorbid illnesses, and had required more invasive surgical treat-
in patients with PAL (Table 2). Before adjustment, PAL patients had ment. To investigate the postoperative morbidity and mortality
2.5 times the odds of 30-day mortality, and a trend toward higher directly attributable to PAL requires controlling for differences in
odds of mortality at discharge compared with patients without PAL baseline characteristics. After forming 2 well-balanced groups using
(Table 3). After adjustment, there were no significantly higher odds propensity weighting under a generalized boosted regression frame-
of either 30-day or discharge mortality. PAL patients had a median work (overall mean bias was reduced from 18% to 8%), we found
length of stay of 10 days compared with 4 days in patients without that, compared with patients without PAL, PAL patients have
PAL. After propensity adjustment in terms of relative effect attribut- increased rates of empyema, blood transfusion, overall pulmonary
able to PAL, PAL patients had a statistically significant 5-day longer complications, unexpected return to operating room, 30-day read-
length of stay compared with 6 days estimated before adjustment mission, and requiring chest tube for pneumothorax. There was a
(Table 3). trend toward increased rate of atelectasis requiring bronchoscopy.
Equally significant, we found PAL does not lead to increased cardiac
Sensitivity Analyses complications (eg, atrial arrhythmia), unexpected admission to the
We calculated treatment effect estimates using 3 alternate ICU, or 30-day mortality. There was a 5-day increase in length of
propensity score models: IPTW using only region of common hospital stay that was attributable to PAL. We conclude that PAL
support; IPTW excluding the patients with missing %FEV1 data; increases the likelihood of postoperative morbidity and delays

ß 2019 Wolters Kluwer Health, Inc. All rights reserved. www.annalsofsurgery.com | 167

Copyright © 2020 Wolters Kluwer Health, Inc. All rights reserved.


Attaar et al Annals of Surgery  Volume 273, Number 1, January 2021

Surgeon

BMI

%FEV1

Hemoglobin

Age

Disease Category

Surgical Approach

Smoking History

Reoperaon

Creanine

Laterality

Surgery Year

Procedure Type

Zubrod Score

Preop Hospitalizaon

0% 5% 10% 15% 20% 25%


Relave Influence
FIGURE 2. Percent relative influence of each variable (variables >1% shown) contributing to creation of the propensity score.

hospital discharge; accordingly, the development and implementa- empyema,9,10,14,19 pneumonia,17,19 mortality,5,20 readmission to the
tion of clinical care pathways for PAL patients require consideration, ICU,5,17 and 30-day readmission.4 Before propensity score adjust-
anticipation, and proactive management of these complications. ment, we found that higher rates of pneumonia, atelectasis, atrial
Previous observational studies have reported PAL being asso- arrhythmia, initial ventilator support >48 hours, sepsis, unexpected
ciated with lengthened hospital stay,4,5,7–18 increased incidence of admission to the ICU, and 30-day mortality were all strongly

Histogram Propensity Scores Histogram Weighted Propensity Scores

FIGURE 3. Histogram of the distribution of unweighted and weighted propensity score for patient with and without PAL illustrative
of the efficacy of inverse probability of treatment weighting in creating a ‘‘pseudo-population’’ that is more similar in their
probability of having PAL. PAL patients who were less expected to get PAL conditional on observed baseline characteristics (ie, lower
propensity scores) are weighted up to account for the many patients like them who did not get PAL. PAL Patients who were expected
to get PAL (ie, higher propensity scores) are weighted down since they are overrepresented in the data.

168 | www.annalsofsurgery.com ß 2019 Wolters Kluwer Health, Inc. All rights reserved.

Copyright © 2020 Wolters Kluwer Health, Inc. All rights reserved.


Annals of Surgery  Volume 273, Number 1, January 2021 Prolonged Air Leak Worsens Outcomes

TABLE 3. Unadjusted and Adjusted Odds Ratio Estimates of Major Adverse Events, Mortality, and Length of Stay in Patients
With Prolonged Air Leak Compared With Those Without
Before Propensity Weighting After Propensity Weighting
Outcome OR 95% CI P OR 95% CI P
Pulmonary
Atelectasis req. bronchoscopy 2.48 1.59–3.88 <0.001 1.67 0.97–2.89 0.066
Pneumonia 2.03 1.23–3.36 0.055 1.15 0.57–2.33 0.695
ARDS 2.15 0.73–6.38 0.352 1.18 0.38–3.67 0.775
Pulmonary embolus 1.79 0.52–6.13 0.432 2.69 0.79–9.22 0.115
Pneumothorax req. chest tube reinsertion 7.59 4.89–11.78 <0.001 9.22 4.32–19.66 <0.001
Initial ventilatory support >48 h 3.10 1.46–6.60 0.003 1.72 0.75–3.97 0.203
Reintubation 2.83 1.63–4.94 <0.001 1.78 0.87–3.65 0.115
Tracheostomy 1.76 0.78–3.98 0.172 0.89 0.38–2.10 0.795
Discharged with chest tube 55.15 34.15–89.04 <0.001 52.81 29.72–93.86 <0.001
Other pulmonary eventy 3.28 1.87–5.76 <0.001 3.17 1.47–6.87 0.003
>1 pulmonary eventz 3.96 2.88–5.45 <0.001 3.67 2.22–6.06 <0.001
Cardiovascular
Atrial arrhythmia req. treatment 2.03 1.30–3.18 0.002 1.19 0.72–1.99 0.494
DVT req. treatment 0.59 0.08–4.44 0.608 0.24 0.03–1.84 0.169
Other cardiovascular event§ 3.40 1.23–9.38 0.018 2.62 0.77–8.87 0.122
>1 cardiovascular eventô 1.96 1.29–2.99 0.002 1.19 0.73–1.93 0.493
Infective/Other
Empyema 20.12 6.67–60.70 <0.001 8.39 2.56–27.45 <0.001
Another infection req. antibiotics 1.38 0.48–3.96 0.546 1.37 0.34–5.62 0.658
Sepsis 3.85 1.50–9.90 <0.005 1.71 0.62–4.65 0.297
Renal failure 1.53 0.45–5.18 0.495 1.40 0.31–6.23 0.660
Packed red blood cells 2.39 1.55–3.71 <0.001 2.16 1.05–4.41 0.036
Other events req. OR return with anesthesia 3.47 1.83–6.60 <0.001 1.85 0.93–3.70 0.073
Readmission/Mortality
Unexpected return to OR 4.92 2.92–8.27 <0.001 3.96 2.15–7.26 <0.001
Unexpected admission ICU 2.33 1.35–4.01 0.002 1.65 0.83–3.30 0.156
Mortality at discharge 2.58 0.96–6.93 0.060 1.20 0.43–3.39 0.693
30-d readmission 2.45 <0.001 1.57–3.83 2.12 1.20–3.73 0.009
30-d mortality 2.72 1.10–6.74 0.031 1.94 0.60–6.26 0.270
Length of stay, days [relative effect] 6.26 5.08–7.44 <0.001 5.04 3.77–6.30 <0.001
Unable to calculate adjusted logistic odds ratio estimates for bronchopleural fistula, ventricular arrhythmia, myocardial infarct, and chylothorax given small number of events.

Indicate type of propensity weighting.
yOther pulmonary event—another pulmonary event in the postoperative period.
z>1 pulmonary complication—any of the following: atelectasis requiring bronchoscopy, pneumonia, ARDS, bronchopleural fistula, pulmonary embolus, pneumothorax requiring
chest tube reinsertion, initial ventilator support >48 h, reintubation, tracheostomy, other pulmonary event.
§Other cardiovascular event—any other CV event including distal arterial embolism in the postoperative period.
ô>1 cardiac complication—any of the following: atrial arrhythmia req. treatment, ventricular arrhythmia requiring treatment, myocardial infarct, DVT requiring treatment, other
cardiovascular event.
Req. indicates requiring; ARDS, acute respiratory distress syndrome; DVT, deep vein thrombosis; OR, operating room; ICU, intensive care unit; CI, confidence interval.

associated with PAL. After adjustment, these complications were no biological plausibility due to residual space defect. The increased
longer significantly associated with PAL. rates of unexpected admission to the OR and 30-day readmission
Only 2 prior studies have studied the consequences of air leak could, in part, be explained by postoperative intervention for unre-
using propensity score methods.13,16 Okereke et al’s case matched solved air leaks. The increased rate of blood transfusion is an
lobectomy patients (n ¼ 82) with any air leak beginning in the outcome that has not been previously studied, and likely multifacto-
immediate postoperative period to those without air leak, and showed rial from prolonged chest tube drainage or length of stay. The finding
an increase in overall respiratory and overall cardiac complica- of no increased rate of 30-day mortality, unexpected admission to the
tions.16 Brunelli et al (n ¼ 85) in a case-matched analysis of PAL ICU, and overall cardiac complications corresponds with PAL rarely
(defined air leak at >7d) after lobectomy reported increased risk of being a life-threatening complication; rather the major consequence
empyema but not of overall cardiac and pulmonary morbidity.14 Both is prevention of patient discharge and return to mobility. Seely et al
studies found increase length of hospital stay and no increase in classify PAL in their severity of thoracic morbidity and mortality
hospital mortality. The specific events represented by our overall (TM&M) classification system as primarily a grade 2 minor com-
cardiac and pulmonary composite variables were similar to the 2 plication, which is one that requires pharmacologic treatment or
prior studies but with some differences in included outcomes and minor intervention; in 13% of cases in their population of lung
definitions. We had a broader inclusion criteria for the study popu- resection patients, however, PAL was a grade 3 or 4 major compli-
lation (inclusion of wedge resections, segmentectomy, and thoraco- cation, requiring more aggressive therapies or surgical interven-
scopic approach), larger sample size, and controlled on a larger set of tion.34 The study reaffirms the surgeon’s awareness that air leaks
potential confounders using a more rigorous statistical analysis. are frequent and manageable, but can quickly escalate the risk of
Our finding that PAL directly contributes to pulmonary com- morbidity and mortality if not properly and promptly addressed.
plications like pneumothorax, pulmonary space infection such as Three principal factors contribute to delayed hospital dis-
empyema, and the trend to toward increased rate of atelectasis has charge after lung resection: air leaks, pain control, and other

ß 2019 Wolters Kluwer Health, Inc. All rights reserved. www.annalsofsurgery.com | 169

Copyright © 2020 Wolters Kluwer Health, Inc. All rights reserved.


Attaar et al

TABLE 4. Sensitivity Analyses of Adjusted Odds Ratio Estimates of Reported Outcomes Shown in Table 3 to Varying Propensity Model Specifications
Model Specification IPTW of Table 3 IPTW Using Common Support IPTW Complete Case Analysis on %FEV1 Nearest Neighbor Matching 2:1
Sample Size (No PAL/PAL) 86/180
Outcome 1950/183 1742/169 Odds Ratio (95% CI) 356/179
Pulmonary
Atelectasis req. bronchoscopy 1.67 (0.97–2.89) 1.67 (0.95–2.91) 1.67 (0.96–2.91) 1.59 (0.91–2.76)

170 | www.annalsofsurgery.com
Pneumonia 1.15 (0.57–2.33) 1.08 (0.53–2.23) 1.21 (0.59–2.46) 0.94 (0.53–1.67)
ARDS 1.18 (0.38–3.67) 1.21 (0.39–3.72) 1.20 (0.38–3.82) 1.01 (0.30–3.42)
Pulmonary embolus 2.69 (0.79–9.22) 2.94 (0.83–10.42) 3.50 (1.01–12.12) 1.19 (0.27–5.28)
Pneumothorax req. chest tube reinsertion 9.22 (4.32–19.66) 8.60 (4.14–17.84) 6.28 (3.47–11.39) 6.16 (3.26–11.64)
Initial ventilatory support >48 h 1.72 (0.75–3.97) 1.87 (0.81–4.28) 2.18 (0.86–5.52) 1.76 (0.68–4.57)
Reintubation 1.78 (0.87–3.65) 1.74 (0.84–3.60) 1.88 (0.91–3.90) 1.49 (0.73–3.07)
Tracheostomy 0.89 (0.38–2.10) 0.92 (0.39–2.16) 0.83 (0.38–2.16) 0.98 (0.37–2.62)
Other pulmonary event 3.17 (1.47–6.87) 3.38 (1.57–7.25) 3.18 (1.63–7.95) 2.16 (1.12–4.18)
>1 pulmonary eventy 3.67 (2.22–6.06) 3.62 (2.22–5.91) 3.07 (2.02–4.66) 2.35 (1.55–3.55)
Cardiovascular
Atrial arrhythmia req. treatment 1.19 (0.72–1.99) 1.18 (0.69–2.00) 1.27 (0.76–2.13) 1.44 (0.82–2.55)
DVT req. treatment 0.24 (0.03–1.84) 0.24 (0.03–1.84) 0.24 (0.03–1.88) 0.39 (0.04–3.50)
Other cardiovascular eventz 2.62 (0.77–8.87) 2.48 (0.78–7.90) 3.08 (0.83–11.41) 2.37 (0.62–9.04)
>1 cardiovascular event§ 1.19 (0.73–1.93) 1.16 (0.70–1.91) 1.29 (0.79–2.12) 1.32 (0.77–2.26)
Infective/Other
Empyema 8.39 (2.56–27.45) 7.83 (2.10–29.28) 7.74 (2.40–25.02) 11.34 (2.25–57.15)
Another infection req. antibiotics 1.37 (0.34–5.62) 1.46 (0.36–5.88) 1.46 (0.35–6.08) 0.81 (0.24–2.70)
Sepsis 1.71 (0.62–4.65) 1.69 (0.61–4.67) 1.84 (0.66–5.08) 1.53 (0.50–4.62)
Renal failure 1.40 (0.31–6.23) 1.29 (0.31–5.28) 1.67 (0.33–8.35) 0.87 (0.22–3.41)
Packed red blood cells 2.16 (1.05–4.41) 2.19 (1.11–4.31) 1.52 (0.93–2.49) 1.25 (0.74–2.11)

ß
Other events req. OR return with anesthesia 1.85 (0.93–3.70) 1.87 (0.92–3.80) 1.91 (0.95–3.86) 2.10 (0.95–4.67)
Readmission/Mortality
Unexpected return to OR 3.96 (2.15–7.26) 3.84 (2.11–6.99) 4.36 (2.30–8.27) 3.51 (1.84–6.69)
Unexpected admission ICU 1.65 (0.83–3.30) 1.66 (0.80–3.42) 1.92 (0.94–3.90) 1.58 (0.82–3.08)
Discharged with chest tube 52.81 (29.72–93.86) 48.53 (27.39–86.01) 57.60 (32.85–100.98) 32.61 (15.07–70.54)
Mortality at discharge 1.20 (0.43–3.39) 1.42 (0.49–4.05) 1.34 (0.45–3.96) 1.72 (0.50–5.92)
30-d readmission 2.12 (1.20–3.73) 2.10 (1.20–3.70) 2.52 (1.44–4.42) 1.90 (1.07–3.36)
30-d mortality 1.94 (0.60–6.26) 2.17 (0.70–6.70) 2.58 (0.68–9.82) 2.01 (0.62–6.50)
Length of stay, days [relative effect] 5.04 (3.77–6.30) 4.99 (3.68–6.29) 4.68 (3.28–6.09) 4.46 (2.71–6.21)
Unable to calculate adjusted logistic odds ratio estimates for bronchopleural fistula, ventricular arrhythmia, myocardial infarct, and chylothorax given small number of events.

Other pulmonary event—another pulmonary event in the postoperative period.
y>1 pulmonary complication—any of the following: atelectasis requiring bronchoscopy, pneumonia, ARDS, bronchopleural fistula, pulmonary embolus, pneumothorax requiring chest tube reinsertion, initial ventilator support
>48 h, reintubation, tracheostomy, other pulmonary event.
p
zOther cardiovascular event—any other CV event including distal arterial embolism in the postoperative eriod.
§>1 cardiac complication—any of the following: atrial arrhythmia req. treatment, ventricular arrhythmia requiring treatment, myocardial infarct, DVT requiring treatment, other cardiovascular event.

Copyright © 2020 Wolters Kluwer Health, Inc. All rights reserved.


Req. indicates requiring; ARDS, acute respiratory distress syndrome; DVT, deep vein thrombosis; OR, operating room; ICU, intensive care unit; CI, confidence interval.
Annals of Surgery  Volume 273, Number 1, January 2021

2019 Wolters Kluwer Health, Inc. All rights reserved.


Annals of Surgery  Volume 273, Number 1, January 2021 Prolonged Air Leak Worsens Outcomes

postoperative complications.35 Fast tracking or enhanced recovery REFERENCES


pathways (ERPs) are multimodal clinical care pathways to reduce 1. Lee L, Hanley SC, Robineau C, et al. Estimating the risk of prolonged air leak
hospital costs and increase patient satisfaction through early dis- after pulmonary resection using a simple scoring system. J Am Coll Surg.
charge.36 A 2012 cost analysis looking at any air leak after pulmo- 2011;212:1027–1032.
nary resection using Medicare data found that an air leak diagnosis 2. Rivera C, Bernard A, Falcoz PE, et al. Characterization and prediction of
increases hospital costs by 15%.37 For PAL, the cost is greater; prolonged air leak after pulmonary resection: a nationwide study setting up the
index of prolonged air leak. Ann Thorac Surg. 2011;92:1062–1068.
another US study using a database of 27,336 lung resections found a
3. Brunelli A, Varela G, Refai M, et al. A scoring system to predict the risk of
$15,636 increase (35% increase) in health care expenditures due to prolonged air leak after lobectomy. Ann Thorac Surg. 2010;90:204–209.
PAL.38 Fast tracking studies in thoracic surgery have proposed 4. Liang S, Ivanovic J, Gilbert S, et al. Quantifying the incidence and impact of
modifying care elements such as chest tube protocols (eg, early postoperative prolonged alveolar air leak after pulmonary resection. J Thorac
water seal, discharge with Heimlich valve system, digital chest Cardiovasc Surg. 2013;145:948–954.
drainage systems),39 optimal pain control with adjunctive use of 5. Elsayed H, McShane J, Shackcloth M. Air leaks following pulmonary resec-
epidural and regional anesthesia,40 and implementation of aggressive tion for lung cancer: is it a patient or surgeon related problem? Ann R Coll Surg
Engl. 2012;94:422–427.
postoperative rehabilitation.41 A few small prospective studies have
6. Gilbert S, Maghera S, Seely AJ, et al. Identifying patients at higher risk of
investigated the feasibility of early discharge of PAL patients with prolonged air leak after lung resection. Ann Thorac Surg. 2016;102:1674–1679.
chest tube or Heimlich valves, and have demonstrated reductions in 7. Attaar A, Winger DG, Luketich JD, et al. A clinical prediction model for
hospital stay without increased patient morbidity.42–44 Our findings prolonged air leak after pulmonary resection. J Thorac Cardiovasc Surg.
support the possibility of early discharge for the majority of PAL 2017;153. 690-699.e2.
patients. Future risk modeling is needed to address the prevention and 8. Irshad K, Feldman LS, Chu VF, et al. Causes of increased length of
monitoring of the pulmonary complications due to PAL, to identify hospitalization on a general thoracic surgery service: a prospective observa-
tional study. Can J Surg. 2002;45:264–268.
the subset of PAL patients that will require hospital readmission and
9. Bhatnagar NK, Berndt S. A solution to prolonged air leak after video-assisted
surgical intervention, and to recognize the perioperative character- thoracoscopic lobectomy. Ann Thorac Surg. 1995;59:260–261.
istics that do directly lead to the increased mortality and cardiac 10. Varela G, Jimenez MF, Novoa N, et al. Estimating hospital costs attributable to
complications. Randomized controlled trials and multicenter inves- prolonged air leak in pulmonary lobectomy. Eur J Cardiothorac Surg.
tigation in this area would support the creation and adoption of early 2005;27:329–333.
discharge pathways that yield immense cost savings and stream- 11. Abolhoda A, Liu D, Brooks A, et al. Prolonged air leak following radical upper
lined, personalized care without negatively impacting patient safety. lobectomy: an analysis of incidence and possible risk factors. Chest.
1998;113:1507–1510.
12. Rice TW, Kirby TJ. Prolonged air leak. Chest Surg Clin North Am.
Strengths and Limitations 1992;2:803–811.
We developed our model from a large database using rigorous 13. Bardell T, Petsikas D. What keeps postpulmonary resection patients in
statistical analysis based on published guidelines.25,27,28,32 The hospital? Can Respir J. 2003;10:86–89.
appropriate implementation of propensity score analysis has been 14. Brunelli A, Xiume F, Al Refai M, et al. Air leaks after lobectomy increase the
advocated in thoracic surgery45; generalized boosted regression with risk of empyema but not of cardiopulmonary complications: a case-matched
IPTW represents a novel and underutilized technique that can more analysis. Chest. 2006;130:1150–1156.
accurately model the complexities of datasets, allowing for less 15. Stolz AJ, Schutzner J, Lischke R, et al. Predictors of prolonged air leak
following pulmonary lobectomy. Eur J Cardiothorac Surg. 2005;27:334–336.
biased and more precise estimates of exposure effect.
16. Okereke I, Murthy SC, Alster JM, et al. Characterization and importance of air
Propensity score methods assume that all confounding factors leak after lobectomy. Ann Thorac Surg. 2005;79:1167–1173.
that could influence PAL and/or the outcome be accounted for; 17. DeCamp MM, Blackstone EH, Naunheim KS, et al. Patient and surgical
however, the existence of unmeasured confounders that were not factors influencing air leak after lung volume reduction surgery: lessons
captured by STSGTSD was unavoidable. On the measured con- learned from the National Emphysema Treatment Trial. Ann Thorac Surg.
founders, we produced a well-balanced propensity score model with 2006;82:197–206.
minimal residual bias. Our additional sensitivity analyses confirmed 18. Stephan F, Boucheseiche S, Hollande J, et al. Pulmonary complications
following lung resection: a comprehensive analysis of incidence and possible
robustness of results to varying model specifications. The purpose of risk factors. Chest. 2000;118:1263–1270.
our study was to determine those complications directly attributable 19. Liberman M, Muzikansky A, Wright CD, et al. Incidence and risk factors of
to PAL, so we do not address a significant and separate research persistent air leak after major pulmonary resection and use of chemical
question of which baseline characteristics that differ between pleurodesis. Ann Thorac Surg. 2010;89:891–898.
patients with and without PAL most contribute to important and 20. Sanchez PG, Vendrame GS, Madke GR, et al. Lobectomy for treating
frequent complications like pneumonia, atrial arrhythmia, and unex- bronchial carcinoma: analysis of comorbidities and their impact on postoper-
ative morbidity and mortality. J Bras Pneumol. 2006;32:495–504.
pected admission to the ICU.
21. Pompili C, Falcoz PE, Salati M, et al. A risk score to predict the incidence of
prolonged air leak after video-assisted thoracoscopic lobectomy: an analysis
CONCLUSION from the European Society of Thoracic Surgeons database. J Thorac Car-
diovasc Surg. 2017;153:957–965.
PAL is associated with increased rates of empyema, blood
22. Brunelli A, Monteverde M, Borri A, et al. Predictors of prolonged air leak after
transfusion, overall pulmonary complications, and 30-day readmis- pulmonary lobectomy. Ann Thorac Surg. 2004;77:1205–1210.
sion, but not with cardiac complications, unexpected admission to the 23. Rubin DB. The use of matched sampling and regression adjustment to remove
ICU, or 30-day mortality after propensity score adjustment. Early bias in observational studies. Biometrics. 1973;29:185–203.
hospital discharge may be feasible and safe in appropriately selected 24. Rosenbaum PR, Rubin DB. Constructing a control group using multivariate
PAL patients. matched sampling methods that incorporate the propensity score. Am Stat.
1985;39:33–38.
ACKNOWLEDGMENTS 25. Lee BK, Lessler J, Stuart EA. Improving propensity score weighting using
machine learning. Stat Med. 2010;29:337–346.
The authors thank Connie Timko and Karan Moore for
26. Health UDo, Services H. ICD 9 CM. The International Classification of
assistance with data acquisition, Li Wang and Nalyn Siripong for Diseases. 9. Rev: Clinical Modification; Vol. 1: Diseases: Tabular List; Vol. 2:
their statistical expertise, and Aubri Drake for assistance with Diseases: Alphabetic Index; Vol. 3: Procedures: Tabular List and Alphabetic
manuscript submission. Index: US Government Printing Office; 1980.

ß 2019 Wolters Kluwer Health, Inc. All rights reserved. www.annalsofsurgery.com | 171

Copyright © 2020 Wolters Kluwer Health, Inc. All rights reserved.


Attaar et al Annals of Surgery  Volume 273, Number 1, January 2021

27. Harder VS, Stuart EA, Anthony JC. Propensity score techniques and the 37. Wood DE, Lauer LM, Layton A, et al. Prolonged length of stay associated with
assessment of measured covariate balance to test causal associations in air leak following pulmonary resection has a negative impact on hospital
psychological research. Psychol Methods. 2010;15:234. margin. Clinicoecon Outcomes Res. 2016;8:187–195.
28. Ridgeway G, McCaffrey D, Morral A, et al. Toolkit for Weighting and 38. Swanson S, Miller D, McKenna R, et al. Economic burden of prolonged air
Analysis of Nonequivalent Groups: A Tutorial for the R TWANG Package. leak after lung resection: open versus video-assisted thoracoscopic surgery
RAND Corporation. 2014. https://www.rand.org/pubs/tools/TL136z1.html. (VATS). Value Health. 2014;17:A80.
29. Cham H, West SG. Propensity score analysis with missing data. Psychol 39. Bryant AS, Cerfolio RJ. The influence of preoperative risk stratification on
Methods. 2016;21:427–445. fast-tracking patients after pulmonary resection. Thorac Surg Clin.
30. D’Agostino R Jr, Lang W, Walkup M, et al. Examining the impact of missing 2008;18:113–118.
data on propensity score estimation in determining the effectiveness of self- 40. Sokouti M, Aghdam BA, Golzari SE, et al. A comparative study of
monitoring of blood glucose (SMBG). Health Serv Outcomes Res Methodol. postoperative pulmonary complications using fast track regimen and con-
2001;2:291–315. servative analgesic treatment: a randomized clinical trial. Tanaffos.
31. Austin PC. An introduction to propensity score methods for reducing the 2011;10:12– 19.
effects of confounding in observational studies. Multivariate Behav Res. 41. Ueda K, Sudoh M, Jinbo M, et al. Physiological rehabilitation after video-
2011;46:399–424. assisted lung lobectomy for cancer: a prospective study of measuring daily
32. Austin PC, Stuart EA. Moving towards best practice when using inverse exercise and oxygenation capacity. Eur J Cardiothorac Surg. 2006;30:
probability of treatment weighting (IPTW) using the propensity score to estimate 533–537.
causal treatment effects in observational studies. Stat Med. 2015;34:3661–3679. 42. McKenna RJ Jr, Mahtabifard A, Pickens A, et al. Fast-tracking after video-
33. Stürmer T, Rothman KJ, Avorn J, et al. Treatment effects in the presence of assisted thoracoscopic surgery lobectomy, segmentectomy, and pneumonec-
unmeasured confounding: dealing with observations in the tails of the propensity tomy. Ann Thorac Surg. 2007;84:1663–1667.
score distribution—a simulation study. Am J Epidemiol. 2010;172:843–854. 43. Drahush N, Miller AD, Smith JS, et al. Standardized approach to prolonged
34. Seely AJ, Ivanovic J, Threader J, et al. Systematic classification of morbidity air leak reduction after pulmonary resection. Ann Thorac Surg. 2016;101:
and mortality after thoracic surgery. Ann Thorac Surg. 2010;90:936–942. 2097–2101.
35. Campos JH. Fast track in thoracic anesthesia and surgery. Curr Opin Anaes- 44. Cerfolio RJ, Bass CS, Pask AH, et al. Predictors and treatment of persistent air
thesiol. 2009;22:1–3. leaks. Ann Thorac Surg. 2002;73:1727–1730.
36. Fiore JF Jr, Bejjani J, Conrad K, et al. Systematic review of the influence of 45. Winger DG, Nason KS. Propensity-score analysis in thoracic surgery: when,
enhanced recovery pathways in elective lung resection. J Thorac Cardiovasc why, and an introduction to how. J Thorac Cardiovasc Surg. 2016;151:
Surg. 2016;151. 708-15 e1-6. 1484–1487.

172 | www.annalsofsurgery.com ß 2019 Wolters Kluwer Health, Inc. All rights reserved.

Copyright © 2020 Wolters Kluwer Health, Inc. All rights reserved.

You might also like