GEN

ERA

LT

HO

RA

CIC

Characterization and Prediction of Prolonged AirLeak After Pulmonary Resection: A NationwideStudy Setting Up the Index of Prolonged Air LeakCaroline Rivera, MD, Alain Bernard, PhD, Pierre-Emmanuel Falcoz, PhD,Pascal Thomas, PhD, Aurélie Schmidt, MS, Stève Bénard, PharmD, Eric Vicaut, PhD,and Marcel Dahan, PhDCHU Bordeaux, Haut Leveque Hospital, Pessac; CHU Dijon, Bocage Hospital, Dijon; CHU Strasbourg, Civil Hospital, Strasbourg;

CHU Marseille, North Hospital, Marseille; Stève consultants, Lyon; Fernand Widal Hospital, Paris; and CHU Toulouse, LarreyHospital, Toulouse, France

pt

Background. The objective of this study was to bettercharacterize prolonged air leak (PAL), defined as an airleak longer than 7 days, and to develop and validate apredictive model of this complication after pulmonaryresection.

Methods. All lung resections entered in Epithor, theFrench national thoracic database (French Society of Tho-racic and Cardiovascular Surgery), were analyzed. Datacollected between 2004 and 2008 (n � 24,113) were used tobuild the model using backward stepwise variable selec-tion, and the 2009 data (n � 6,813) were used for externalvalidation. The primary outcome was PAL. Results of thepredictive model were used to propose a score: the index ofPAL (IPAL).

Results. Prevalence of PAL after pulmonary resectionwas 6.9% (n � 1,655) in the development data set. In thefinal model, 9 variables were selected: gender, body massindex, dyspnea score, presence of pleural adhesions,

lobectomy or segmentectomy, bilobectomy, bulla resec-

CHU Bordeaux, Haut Leveque Hospital, Avenue de Magellan, 33604Pessac, France; e-mail: krorivera@yahoo.fr.

© 2011 by The Society of Thoracic SurgeonsPublished by Elsevier Inc

tion, pulmonary volume reduction, and location on up-per lobe. In the development data set, the C-index was0.71 (95% confidence interval [CI], 0.70 to 0.72). At exter-nal validation, the C-index was 0.69 (95% CI, 0.66 to 0.72)and the calibration slope (ie, the agreement betweenobserved outcomes and predictions) was 0.874 (<1). Ascore chart based on these analyses has been proposed.The formula to calculate the IPAL is the following:gender (F � 0; M � 4) - (body mass index-24) � 2 �dyspnea score � pleural adhesion (no � 0; yes � 4) �

ulmonary resection (wedge � 0; lobectomy or segmen-ectomy � 7; bilobectomy � 11; bulla resection � 2;

volume reduction � 14) � location (lower or middle lobe �0; upper � 4).

Conclusions. Surgeons can easily use the well-vali-dated model to determine intraoperative preventive mea-sures of PAL.

(Ann Thorac Surg 2011;92:1062–8)

© 2011 by The Society of Thoracic Surgeons

Prolonged air leak (PAL) is one of the most commoncomplications after pulmonary resection [1] but re-

mains incompletely characterized. In most of the cases,the air leak resolves rapidly when the visceral pleuraadheres to the chest wall but it usually leads to prolongedhospital stay [2–4], impacting on costs [5], and increasingthe risk of other complications such as empyema [4–6] .Prolonged air leak is often the only reason for prolongedlength of stay [3] and remains a frequent complicationdespite several preventive strategies, including surgicaltechniques [7], sealants [8], or buttressing materials [9].There is no consensus on the most effective method forPAL prevention [8]. There is some published informationregarding risk factors of postoperative PAL [6, 10–14], butthese previous studies are based on a very few number of

Accepted for publication April 6, 2011.

Presented at the Forty-seventh Annual Meeting of The Society of ThoracicSurgeons, San Diego, CA, Jan 31–Feb 2, 2011.

Address correspondence to Dr Rivera, Thoracic Surgery Department,

selected patients from only 1 institution and have in-cluded heterogeneous selected types of operation andindications.

The definition of the term “prolonged air leak” variesin multiple published series ranging from an air leaklasting beyond postoperative day 5 to day 10 [6, 13–16]. Inour study, we considered that a PAL is an air leak thatpersists when the patient could otherwise be discharged.In accordance with French practices [9], we defined aPAL as an air leak lasting 7 days or more. Despite this stilldisputed definition and limitations previously men-tioned, PAL has been reported to occur in approximately5% to 15% of patients after lobectomy [12, 16, 17]. As fasttrack strategies are being developed in thoracic surgery,this definition remains an unstandardized issue.

The management of PAL after pulmonary resectionrequires good surgical technique, appropriate chest tube

Aurélie Schmidt and Stève Bénard disclose that theyhave financial relationships with Nycomed.

0003-4975/$36.00doi:10.1016/j.athoracsur.2011.04.033

teiepmsstocp

1063Ann Thorac Surg RIVERA ET AL2011;92:1062–8 PROLONGED AIR LEAKS AFTER LUNG RESECTION

GEN

ERA

LT

HO

RA

CIC

management, and identification of risk factors. A predic-tive risk model of PAL may assist the surgeon in thetreatment strategy. Identifying preoperatively patients atincreased risk of PAL could allow the practitioner toanticipate the usefulness of additional operative preven-tive procedures. The purpose of this study was to bettercharacterize PAL and to develop and validate a predic-tive model of this complication after pulmonaryresection.

Patients and Methods

Data CollectionEpithor is a database created in 2002 by the FrenchSociety of Thoracic and Cardiovascular Surgery. It is acomputerized database collecting prospectively the datarelated to thoracic surgical procedures as a voluntaryinitiative of French general thoracic surgeons. Presently,98 private and public institutions daily contribute to thisdatabase, representing 82% of the centers throughout thecountry and including more than 140,000 procedures.The Epithor working process has been previously de-scribed [18–20]. Surgeons implement their local Epithordatabase with the help of a confidential password. Thesedata are automatically anonymized when sent to thenational database in order to guarantee patients’ andsurgeons’ confidentiality. Collected variables include in-formation about patient’s personal characteristics, med-ical history, surgical procedures, and outcomes.

The Institutional Review Board of the French Society ofThoracic and Cardiovascular Surgery approved thisstudy (Approval Number 2010-10-31-15-52-1-RiCa). Pa-tients’ consent was obtained for entry into the databaseand patients were aware that these data would be usedfor research purposes.

PatientsAn analysis was performed on all 30,926 pulmonaryresections (lobectomy, bilobectomy, segmentectomy,wedge resection, volume reduction, and bulla resection)performed from January 2004 to December 2009. Weexcluded pneumonectomies and explorative thoracoto-mies from our study because of the absence of paren-chyma dissection. The development sample contained24,113 procedures and the validation sample contained6,813 resections. During the study period, 90 centers werecontributing to Epithor. Baseline demographics includedage, gender, body mass index (BMI), American Society ofAnesthesiologists (ASA) score, forced expiratory volume(FEV), dyspnea score according to the Medical ResearchCouncil [21], and number of comorbid diseases. Numberof comorbid diseases per patient, considered as a cate-goric variable, was used because recent consistent databased on Epithor have suggested the superiority of thisvariable on the types of individual comorbidities in apredictive model for in-hospital mortality [19]. Data werealso collected on the following variables: pathology (ma-lignant, benign), type of procedure (limited resection,

segmentectomy, lobectomy, bilobectomy, bulla resection,

and lung volume reduction), surgical approach (openthoracotomy, video-assisted thoracic surgery [VATS]),presence of pleural adhesions, location (upper lobe, mid-dle lobe, lower lobe, and without precision), and side ofresection (left, right, and bilateral).

Statistical Analyses: Missing DataThe proportion of missing data was 1.3% for ASA score,less than 1% for location, number of comorbidities perpatient, and side of resection. For dyspnea score and FEV,the missing data were 19.5% (n � 4,733) and 37.7% (n �8,490), respectively. The variable FEV was excluded be-cause there were too many missing data. We assumedthat the missing data were missing at random (ie, the factthat the data were missing was not related to the true(unobserved) values of the missing data) [22, 23]. Missingprognostic factor data as ASA score, dyspnea score,location, and side were replaced by multiple imputationtechnique [23].

Developing the Predictive ModelThe comparison of patients with or without PAL wasperformed by the �2 test for categoric variables and the test for continuous variables. Variables with clinical rel-vance included in multivariate analysis were the follow-ng: age, gender, BMI, ASA score, dyspnea score, pres-nce of pleural adhesions, location, side of resection,athology, type of procedure, and surgical approach. Theodel was constructed using backward stepwise variable

election. All predictors were included in this process. Atep-down variable selection using AIC (Akaike informa-ion criteria) was used as a stopping rule (AIC � 2 degreef freedom). Continuous variables were converted intoategoric variables if the procedures led to a bettererformance of the model estimated by the AIC.

Measurement of Performances and External ValidationWe used the data set of the most recent patients whounderwent pulmonary resection between January 2009and December 2009 to perform the external validation.The R2 value and Brier score were used to measure theperformance of the model; ie, the difference betweenthe predicted outcome and the observed outcome [24,25]. For discriminative ability (ie, the accurate predic-tion between those with and those without PAL), theC-index was used [25]. Calibration by plotting pre-dicted against observed probability estimated inter-cepts and slopes of curves to quantify overfitting [24].The goodness-of-fit was tested by the Hosmer-Lemeshow test [26].

Predictive ScoreThe final predictive model was used to create a scorechart. The first step was to round updated shrunk regres-sion coefficients. For this score, the C-index was esti-mated and, for each level of score, the estimated proba-bility was calculated. Calculations were performed withStata 11 statistical software (StataCorp LP, College Sta-tion, TX) and R statistical software for which we used

Harrell’s Design library (http://www.r-project.org).

asqoB6fi

C0

fsrb1rtI

sgprrfr

2soc

1064 RIVERA ET AL Ann Thorac SurgPROLONGED AIR LEAKS AFTER LUNG RESECTION 2011;92:1062–8

GEN

ERA

LT

HO

RA

CIC

Results

Patient Characteristics and Description of theProceduresBetween January 2004 and December 2008, within the24,113 patients undergoing pulmonary resections in-cluded in the development dataset, 1,655 patients had aPAL (6.9%). This frequency decreased from 7.7% (n �310) in 2004 to 6.3% (n � 389) in 2008 (p � 0.003) (Fig 1).

The baseline patients’ characteristics and PAL for thedifferent variables are reported in Table 1. All predictorsexcept for the variable age influenced significantly theoccurrence of PAL. The main type of pulmonary resec-tion was lobectomy or segmentectomy (n � 14,056) with

PAL frequency of 8.3%. In patients with pleural adhe-ions requiring intraoperative adhesiolysis, the PAL fre-uency was 10.4%. The location of pulmonary resectionn the upper lobe increased the risk of PAL (9.1%).etween January 2009 and December 2009, within the,813 patients of the external sample used to validate thenal model, 391 patients had a PAL (5.7%).

Developing the Predictive ModelThe analysis of the development sample allowed us toidentify 9 independent predictors: gender, BMI, dyspneascore, presence of pleural adhesions, lobectomy or seg-mentectomy, bilobectomy, bulla resection, pulmonaryvolume reduction, and upper lobe location (Table 2).Body mass index and dyspnea score variables wereincluded in the model as continuous variables. Wechecked that results obtained after missing data replace-ment by multiple imputation technique were similar toresults obtained in complete cases.

Performance MeasuresThe C-index of the development data set was 0.71 (95%confidence interval [CI], 0.70 to 0.72). At external valida-tion, the C-index was 0.69 (95% CI, 0.66 to 0.72), close tothe C-index of the model development. At externalvalidation, the calibration slope (ie, the agreement be-

Fig 1. Evolution of the prolonged air leak frequency between 2004and 2008.

tween observed outcomes and predictions) was 0.874 (�

1), which reflects an overfitting of the development dataset. It indicates how much the effects of predictors needto be reduced on average to make the model wellcalibrated for new patients from the underlying popula-tion. The calibration slope has been used as a shrinkagefactor to adjust a model for future predictive score. TheHosmer-Lemeshow goodness-of-fit was nonsignificantfor model development and external validation (Table 3).

Predictive Score: Index of PAL (IPAL)First, we used a shrinkage technique with the value of thecalibration slope (0.874) at external validation to updatethe regression coefficients at the development model(Table 2). At external validation, the calibration slope ofthe final score chart was 0.99 close to 1 that indicates thatthe model is well calibrated for new patients. The Hos-mer-Lemeshow goodness-of-fit was nonsignificant forexternal validation (�2 � 11.4, p � 0.25). The estimated

-index of the predictive score was 0.71 (95% CI, 0.70 to.72).The formula to calculate the IPAL (score chart) is the

ollowing: gender (F � 0; M � 4) � (BMI-24) � 2 � dyspneacore � pleural adhesion (no � 0; yes � 4) � pulmonaryesection (wedge � 0; lobectomy or segmentectomy � 7;ilobectomy � 11; bulla resection � 2; volume reduction �4) � location (lower or middle lobe � 0; upper � 4). Theisk probability can be calculated thanks to the IPAL andhe following formula: 1/(1 � exp (�(�4.213 � 0.1167 �PAL))).

Distributions of estimated probability by predictivecore and actual probability according to patients’ sub-roup scores were reported (Table 4). A patient whoresents an IPAL � 5 has a risk of PAL after pulmonaryesection 5% or less (low risk), when 5 � IPAL � 10 theisk of PAL is between 5% and 10% (moderate risk), andor IPAL � 10 the risk of PAL is greater than 10% (highisk).

Comment

Prolonged air leak remains a frequent complication thatmay cause more severe morbidity and prolonged hospi-talization that can have negative economic effects anddelay adjuvant treatment. There are several techniquesand sealants, both old and recently developed, whichmay help in the prevention or treatment of air leak [27,8]. Their efficacy on duration of chest tubes and hospitaltay are, however, inconsistent. Accordingly, routine usef surgical sealants or tissue reinforcement material inlinical practice cannot be recommended to date [8]. As

PAL does not occur with the same frequency in allpatients, a predictive risk model of PAL may help toidentify the best candidates for additional preventiveprocedures and, in turn, to support a rationalized alloca-tion of resources. The purpose of our study was thus todevelop a logistic regression equation predicting the riskof PAL in patients undergoing pulmonary resection.

The reliability and exhaustivity of any data collectionremains the key issue when the information is used for

outcome analysis. Data must therefore be complete and

AA

D

N

P

S

P

P

L

S

ex;

1065Ann Thorac Surg RIVERA ET AL2011;92:1062–8 PROLONGED AIR LEAKS AFTER LUNG RESECTION

GEN

ERA

LT

HO

RA

CIC

accurate. The Epithor database is considered as repre-sentative; it currently represents more than 80% of allcenters performing thoracic surgery in France. The qual-ity of the data is constantly improved by various means.Multiple coherence tests are automatically carried out towarn the participating center of any inconsistency. Pa-tient data are recorded exclusively by means of pulldownmenus to avoid disparities of data entry. In addition,surgeons can check the quality of their way to enter thedata by comparing themselves to national data through aquality score. Moreover, the completeness and accuracy

Table 1. Patient Characteristics for Predictive Model in Devel

Characteristic Categories

Gender FemaleMale

BMI (kg/m2)ge (years)SA score 1

2345

yspnea score 012345

umber of comorbidities per patient None123

4 or 5ulmonary pathology Benign

Malignanturgical approach VATS

Thoracotomyleural adhesions No

Yesulmonary resection Wedge resection

LobectomySegmentectomyBilobectomyBulla resectionPulmonary volum

reductionocation Lower lobe

Middle lobeUpper lobeWithout precision

ide RightLeftBilateral

ASA � American Society of Anesthesiologists; BMI � body mass ind

of each center’s data are guaranteed by an on-site quality

audits. These audits are performed by thoracic surgeonsand take place in the departments involved in Epithor inorder to verify the coherence of the data on selected filesand to estimate the level of the missing data by variable.The main objective of these audits is pedagogic, advisingon the best methods to collect and enter the data. Theaudits are concluded by a report to help the practitionersinvolved to improve their data quality.

In our study, the frequency of PAL was 6.9% over a5-year period. This result is in the lower bound of theestimated figures in published literature. Indeed, the

ent Data Set (n � 24,113)

Patients (n) Prolonged Air Leak (%) p

7,396 378 (5.1%) 0.000116,717 1,277 (7.6%)24,113 22.9�4.4 0.000124,113 59.2�13 0.75,668 360 (6.4%) 0.0001

11,960 793 (6.6%)5,790 450 (7.8%)

368 37 (10.1%)311 2 (0.6%)

10,527 604 (5.7%) 0.00015,602 404 (7.2%)2,374 202 (8.5%)

522 59 (11.3%)148 12 (8.1%)227 26 (11.5%)

3,619 176 (4.9%) 0.00015,506 339 (6.2%)5,794 377 (6.5%)6,829 521 (7.6%)2,361 242 (10.2%)

10,017 535 (5.3%) 0.000114,096 1,120 (8.0%)4,803 149 (3.1%) 0.0001

19,310 1,506 (7.8%)18,730 1,095 (5.9%) 0.00015,383 560 (10.4%)7,653 252 (3.3%) 0.0001

13,100 1,087 (8.3%)955 76 (8%)931 103 (11.1%)

1,354 100 (7.4%)119 29 (24.4%)

7,039 400 (5.7%) 0.00011,063 30 (2.8%)

11,877 1,079 (9.1%)3,925 132 (3.4%)

14,055 112 (0.8%) 0.00019,839 534 (5.4%)

72 10 (13.9%)

VATS � video-assisted thoracic surgery.

opm

e

frequency of air leak lasting more than 5 days after

olqtitiapvtt1fon

T

DC

P

B

1066 RIVERA ET AL Ann Thorac SurgPROLONGED AIR LEAKS AFTER LUNG RESECTION 2011;92:1062–8

GEN

ERA

LT

HO

RA

CIC

pulmonary resection were previously estimated accord-ing to the studies between 5% and 13% [5, 16, 17], andbetween 9.7% and 17.6% when lasting more than 7 days[4, 12, 13]. These data were assessed on a single institu-tion base and it is difficult to compare the results from astudy with another because of the heterogeneity of meth-ods (retrospective or prospective series), of definition ofPAL and of analyzed procedures. Nonetheless, from ourknowledge this is the first study assessing the frequencyof PAL at a national level.

Moreover, we observed that the frequency of PALdecreased over time from 7.7% in 2004 to 6.3% in 2008(Fig 1). We have no consistent proof to explain this result.We wonder if it could be related to the variety of surgicalsealants that have been developed in recent years andused in thoracic surgery to prevent air leaks since 2004.However, the effectiveness of such sealants has not yetbeen fully established. As economic pressures to reducethe length of hospital stay have increased in the lastdecade, another hypothesis could be a reduction of PALinduced by a better surveillance of this complication andsubsequently a more appropriate postoperative manage-ment. This observation could also be related to theimpact of the literature, in particular to the randomizedprospective study of Marshall and colleagues [29], which

able 3. Performance Measures for Definitive Model

Aspect Measures

Overall performance R2

Brier scoreiscrimination C-indexalibration Slope

Test for miscalibration

Table 2. Logistic Regression Coefficients in Final Model Deve

Variable Categories Coeffi

Gender Female RMale 0.

BMI (kg/m2) Linear �0.Dyspnea score Linear 0.Pleural adhesion No R

Yes 0.ulmonary resection Wedge resection R

Lobectomy or segmentectomy 0.Bilobectomy 1.Bulla resection 0.Pulmonary volume reduction 1.

Location Lower or middle lobe RUpper lobe 0.

Intercept �1.

a Shrinkage regression coefficients are calculated with a slope of calibratiovalue (24) and multiplied by �1. c Value of dyspnea score is multipli

MI � body mass index; Ref � reference value (� 0).

Goodness of fit test Hosmer-Lemeshow

showed that the increased use of early waterseal, evenwhen a small air leak is present, may be responsible forthe decreasing incidence of PAL over time.

Previous studies on predictors of postoperative airleaks identified that risk factors of PAL were: importantpleural adhesions [12], diffuse emphysema [30], low FEVand chronic obstructive lung disease [12, 13, 31], poorwound-healing characteristics [14], and upper lobectomy

r bilobectomy [10, 12, 15, 32]. These studies had manyimits. The characterization of some of these predictors isuite subjective; for example, the definition of “impor-

ant pleural adhesions.” Patients included were recruitedn only 1 surgical center, limiting the variability of pa-ients’ characteristics. Studies that took place in only 1nstitution cannot reflect the heterogeneity of PAL man-gement. They are not representative of the differentostoperative and intraoperative practices used to pre-ent or manage air leaks. These studies did not representhe heterogeneity of types of indications and resectionhat can lead to PAL, being restricted to lung cancer [14,6] or to lobectomies [6, 12]. In some studies the timerame is very long, including patients from 8-year peri-ds; that could be an important bias as surgical tech-iques and staplers, as well as available sealants and

Development Data Set(n � 24,113)

External Validation(n � 6,813)

0.095 0.0730.06 0.053

0.71 (0.70–0.72) 0.69 (0.66–0.715)1 0.874 (0.74–1)p � 1 p � 0.07

2 2

for 2004 to 2008 Data Seta

s Standard Error p Coefficients Shrunk Scores

0.0001 00.06 0.39 40.007 0.0001 �0.109 �1b

0.024 0.0001 0.187 2c

0.0001 00.056 0.366 4

00.075 0.0001 0.717 70.126 0.0001 1.06 110.13 0.07 0.2 20.23 0.0001 1.43 14

00.06 0.0001 0.425 40.17 �1.3406

t at external validation (0.8738). b Variable BMI is centered by median2.

loped

cient

ef4467125214ef419ef822123635ef48189

n ploed by

� 8.4; p � 0.4 � 13; p � 0.24

1067Ann Thorac Surg RIVERA ET AL2011;92:1062–8 PROLONGED AIR LEAKS AFTER LUNG RESECTION

GEN

ERA

LT

HO

RA

CIC

reinforcement materials, underwent dramatic improve-

ments during the study period [12, 16].This study presents some shortcomings that should be

taken into account when interpreting the results. In ourdatabase we have no detailed information on surgicaltechnique performed or sealants used. The lack of im-portant covariates, such as precise duration of chest tubedrainage or detailed description on chest tube suctionmanagement, represents limitations that we cannot sur-pass in this study. Missing data on FEV1 (forced expira-tory volume in the first second of expiration) is also animportant shortcoming. Some authors reported postop-erative predicted FEV as a risk factor of PAL [12, 13, 31],whereas Okereke and colleagues [6] did not. Increasedairway resistance and pathologic parenchymal changesin chronic obstructive lung disease could lead to anelevated risk of PAL. The variable that brings informationon pulmonary function in our study is the dyspnea score.

We realized the external validation of the model due todata from the 2009 Epithor data. This temporal validationwith patients recently treated showed that the logisticregression equation obtained is applicable to presentclinical practice. The sample used for external validationrepresented almost 30% of the development data set.This is a largely acceptable volume.

The predictive value of the model we developed wasmeasurable due to the C-index. The estimated C-index ofour predictive score of PAL was 0.71 (with 95% CI 0.70 to0.72). This information reflects that some predictors ofPAL after pulmonary resection may not be included inthe model. This was a limit in our study. Variables suchas FEV1 may influence the occurrence of PAL and inducea decrease of our C-index. However, we noticed that the95% confidence interval of our C-index was very tight,meaning the robustness of our model was important. Ourmodel can be considered as an accurate tool to identifypatients in whom the balance of risks and benefits isconsistent with an intraoperative intervention in order tominimize the risk of persistent postoperative air leak. Italso can be used as a screening tool to determine whichpatients will need a digitalized drainage device to facili-tate the quantification of air leak.

Based on this first multicenter nationwide study, wecan conclude that PAL is a frequent complication afterpulmonary resection, affecting around 7% of the patients.A comprehensive strategy for PAL must include bothprevention and effective management. Identifying preop-

Table 4. Comparison of Predicted Probability and Actual ProbPredicted Risk Category

Risk Group Score SubgroupO

� 3% low �14 to 13 to 4.9% 2 to 55 to 6.9% moderate 6 to 77 to 10% 8 to 10�10% high 11 to 23

erative and intraoperative predictors of PAL may be

clinically useful for taking additional measures to pre-vent this complication and to improve healthcare costs. Arobust and relevant predictive risk model for postopera-tive PAL has been developed. It should be tested indifferent settings, using another database such as that ofthe Society of Thoracic Surgeons. This model has alsonow to be validated prospectively.

The authors would like to thank all the French thoracic surgeonswho participated in Epithor and subsequently in this study inorder to improve thoracic surgery quality. We are grateful toNycomed for the financial support that allowed statistical con-sulting by st[è]ve consultants and to the French National Insti-tute of Cancer (INCa) for the grant that allowed the on-sitequality audits of Epithor.

References

1. Rice TW, Okereke IC, Blackstone EH. Persistent air-leakfollowing pulmonary resection. Chest Surg Clin N Am2002;12:529–39.

2. Irshad K, Feldman LS, Chu VF, Dorval JF, Baslaim G, MorinJE. Causes of increased length of hospitalization on a generalthoracic surgery service: a prospective observational study.Can J Surg 2002;45:264–8.

3. Bardell T, Petsikas D. What keeps postpulmonary resectionpatients in hospital? Can Respir J 2003;10:86–9.

4. Brunelli A, Xiume F, Al Refai M, Salati M, Marasco R,Sabbatini A. Air leaks after lobectomy increase the risk ofempyema but not of cardiopulmonary complications: a case-matched analysis. Chest 2006;130:1150–6.

5. Varela G, Jiménez MF, Novoa N, Aranda JL. Estimatinghospital costs attributable to prolonged air leak in pulmo-nary lobectomy. Eur J Cardiothorac Surg 2005;27:329–33.

6. Okereke I, Murthy SC, Alster JM, Blackstone EH, Rice TW.Characterization and importance of air leak after lobectomy.Ann Thorac Surg 2005;79:1167–73.

7. Toloza EM, Harpole DH Jr. Intraoperative techniques toprevent air leaks. Chest Surg Clin N Am 2002;12:489–505.

8. Belda-Sanchis J, Serra-Mitjans M, Iglesias Sentis M, Rami R.Surgical sealant for preventing air leaks after pulmonaryresections in patients with lung cancer. Cochrane DatabaseSyst Rev:CD003051

9. Periquet Y, Poncelet AJ. Persistent air leak (PAL): conserva-tive vs. invasive approach? [Article in French] Rev MalRespir 2005;22(1 Pt 1):103–12.

10. Abolhoda A, Liu D, Brooks A, Burt M. Prolonged air leakfollowing radical upper lobectomy: an analysis of incidenceand possible risk factors. Chest 1998;113:1507–10.

11. Loran DB, Woodside KJ, Cerfolio RJ, Zwischenberger JB. Predic-tors of alveolar air leaks. Chest Surg Clin N Am 2002;12:477–88.

12. Brunelli A, Monteverde M, Borri A, Salati M, Marasco RD,Fianchini A. Predictors of prolonged air leak after pulmo-

ity of Prolonged Air Leak (PAL) in Selected Patients in

rved Frequencyof PAL (%)

Predicted Frequency ofPAL by the Model (%) No.

2 1.8 4,8244 4 6,3055.9 6.1 3,8308.8 8.6 5,010

15.4 15.5 4,144

abil

bse

nary lobectomy. Ann Thorac Surg 2004;77:1205–10.

1068 RIVERA ET AL Ann Thorac SurgPROLONGED AIR LEAKS AFTER LUNG RESECTION 2011;92:1062–8

GEN

ERA

LT

HO

RA

CIC

13. Stolz AJ, Schützner J, Lischke R, Simonek J, Pafko P. Predic-tors of prolonged air leak following pulmonary lobectomy.Eur J Cardiothorac Surg 2005;27:334–6.

14. Isowa N, Hasegawa S, Bando T, Wada H. Preoperative risk factorsfor prolonged air leak following lobectomy or segmentectomy forprimary lung cancer. Eur J Cardiothorac Surg 2002;21:951.

15. Cerfolio RJ, Bass C, Katholi CR. Prospective randomized trialcompares suction versus water seal for air leaks. Ann ThoracSurg 2001;71:1613–7.

16. Brunelli A, Varela G, Refai M, et al. A scoring system topredict the risk of prolonged air leak after lobectomy. AnnThorac Surg 2010;90:204–9.

17. Cerfolio RJ. Chest tube management after pulmonary resec-tion. Chest Surg Clin N Am 2002;12:507–27.

18. Falcoz PE, Conti M, Brouchet L, et al. The Thoracic SurgeryScoring System (Thoracoscore): risk model for in-hospitaldeath in 15,183 patients requiring thoracic surgery. J ThoracCardiovasc Surg 2007;133:325–32.

19. Bernard A, Rivera C, Pages PB, Falcoz PE, Vicaut E, DahanM. Risk model of in-hospital mortality after pulmonaryresection for cancer: A national database of the FrenchSociety of Thoracic and Cardiovascular Surgery (Epithor).J Thorac Cardiovasc Surg 2011;141:449–58.

20. Rivera C, Dahan M, Bernard A, Falcoz PE, Thomas P. Surgicaltreatment of lung cancer in the octogenarians: results of a nation-wide audit. Eur J Cardiothorac Surg 2011;39:981–6.

21. C. S. Dyspnea score The MRC breathlessness scale. (Lond).COM, editor, 2008: 226�7.

22. Little RJA, Rubin DB. Statistical analysis with missing data.New York: John Wiley; 1987.

mostly wedge resections for metastasectomy.

23. Schafer JL. Multiple imputation: a primer. Stat MethodsMed Res 1999;8:3–15.

24. Hilbe JM. Logistic regression models: London: Chapman &Hall: 2009.

25. Harrell PE. Regression modeling strategies with applicationsto linear models, logistic regression and survival analysis.New York: Springer; 2001.

26. Hosmer DW, Lemeshow SA. Goodness-of-fit tests for themultiple regression model. Communications in Statistics1980:1043–69.

27. Singhal S, Ferraris VA, Bridges CR, et al. Management ofalveolar air leaks after pulmonary resection. Ann ThoracSurg 2010;89:1327–35.

28. Venuta F, Rendina EA, De Giacomo T, et al. Technique toreduce air leaks after pulmonary lobectomy. Eur J Cardio-thorac Surg 1998;13:361–4.

29. Marshall MB, Deeb ME, Bleier JI, et al. Suction vs water sealafter pulmonary resection: a randomized prospective study.Chest 2002;121:831–5.

30. DeCamp MM, Blackstone EH, Naunheim KS, et al. Patientand surgical factors influencing air leak after lung volumereduction surgery: lessons learned from the National Em-physema Treatment Trial. Ann Thorac Surg 2006;82:197–206.

31. Linden PA, Bueno R, Colson YL, et al. Lung resection inpatients with preoperative FEV1 � 35% predicted. Chest2005;127:1984–90.

32. Brunelli A, Fianchini A. Prolonged air leak following upperlobectomy: in search of the key. Chest 1999;116:848.

DISCUSSION

DR MARCELO C. DASILVA (Boston, MA): Just one question.Do you use chest wall suction in your chest tube management?Do you have that information available or do you place the chesttubes to water seal right out of the OR [operating room]?

DR RIVERA: We do not have this information in our database.Our study includes, as I said, 86 centers, and there is animportant variability in postoperative chest tube management.

DR BRYAN FITCH MEYERS (St. Louis, MO): That was a verynicely presented paper. I wondered if you had any hypothesisabout the reasons for the decline in the air leaks over time. Is itassociated with increasing use of minimally invasive techniques,or do you have any other suggestions as to why the rate isdropping?

DR RIVERA: First of all, we don’t have any consistent proof toexplain this result, but we have some hypothesis. We think thatit could be related to the recent sealant that may be used inthoracic surgery. It also could be related to a modification ofpractices according to the literature on chest tube management,and it could also simply be related to a better surveillance of thiscomplication in the last decade because of the recent economicpressures on prolonged hospital stay.

Concerning video-assisted thoracic surgery [VATS], it is im-portant to note that in French practices and in this study, theprocedures performed by video-assisted thoracic surgery were

We do not currently practice VATS lobectomies in France yet,so maybe these results are not applicable to American practices.

DR DAVID PARK MASON (Cleveland, OH): It’s a very bigdatabase, but one question I would have is, did you look atsurgeon as the variable? I mean, honestly, if I were to considerall variables, the likelihood of leaving the operating room withan air leak or having a prolonged air leak, that to me would beone of the biggest factors to consider; the skill level or theexperience, or the whatever, with the surgeon.

Can you look at that without obviously naming the particularpeople? But I would assume that there is a large variability in airleak in terms of the surgeon as well.

DR RIVERA: We cannot look at this variable because surgeonsand departments are anonymous, we don’t know if it’s a privateor public institution or who is the surgeon.

DR MASON: You can’t look and see how you compare orsomebody compared? I mean, it doesn’t have to be a specificperson, but just to know that there is some variability in surgeon.

DR RIVERA: Once the data is in the national database, we don’tknow where it comes from.

DR MASON: Right.