ca

Estimating the Risk of Prolonged Air Leakafter Pulmonary Resection Using aSimple Scoring SystemLawrence Lee, MD, Stephen C Hanley, PhD, Catherine Robineau, RN, Christian Sirois, MD, FRCSC,David S Mulder, MD, FACS, Lorenzo E Ferri, MD, PhD, FACS

BACKGROUND: The high rate of prolonged air leak (PAL) after pulmonary resection has prompted interest in surgicaladjuncts designed to prevent this complication. However, these adjuncts are costly and might not bebeneficial if used routinely. Identification of patients at highest risk might allow for more effective useof these adjuncts. Therefore, we sought to develop a simple scoring system to predict PAL.

STUDY DESIGN: A derivation set of 580 patients was identified from a prospectively entered database of consec-utive pulmonary resections at a single institution from 2002 to 2007. Patient and operativecharacteristics were compared using Student’s t-test and chi-square tests. Significant variableson univariate analysis were entered into a stepwise logistic regression to establish a simplepredictive model to estimate the risk of PAL. This scoring system was then validated in aconsecutive set of 381 patients operated at the same institution from 2007 to 2009.

RESULTS: The rate of PAL was 14% in the derivation set and 18% in the validation set. Poor pulmonaryfunction (forced expiratory volume in 1 second and carbon monoxide diffusing capacity,percent predicted) and pleural adhesions were significantly associated with PAL in the deriva-tion set. A weighted scoring system was devised using pleural adhesions (�2 points), forcedexpiratory volume in 1 second (�1 per 10% below 100%), and carbon monoxide diffusingcapacity (�1 per 20% below 100%). Total number of points estimated the probability of PAL.Hosmer-Lemeshow goodness-of-fit test confirmed validity (p � 0.2) of this scoring system inthe validation set.

CONCLUSIONS: We have devised and validated a simple scoring system to predict the probability of PAL afterpulmonary resection. (J Am Coll Surg 2011;212:1027–1032. © 2011 by the American College

of Surgeons)

ttctnP

Prolonged air leak (PAL) is a common complication afterpulmonary resection, with a reported incidence of up to26% after surgery for lung cancer.1 PAL represents a majorclinical problem because it is associated with considerableeconomic costs and increased rates of other postoperativecomplications, length of hospitalization, and mortality.2,3

Intraoperative pneumostasis, postoperative maintenanceof lung expansion, and pleural apposition remain impera-tive in preventing this complication.4 Surgical adjuncts in-luding biologic sealants, autologous blood pleurodesis,nd staple-line buttressing have been developed in an effort

Disclosure Information: Nothing to disclose.

Received December 18, 2010; Revised March 4, 2011; Accepted March 7,2011.From the Department of Surgery, Division of Thoracic Surgery, McGill Uni-versity Health Center, Montreal, QC, Canada.Correspondence address: Lorenzo E Ferri, MD, PhD, FACS, Department of

Surgery, McGill University, Montreal General Hospital, L9-112, 1650 CedarAve, Montreal, Quebec H3G 1A4, Canada. email: lorenzo.ferri@mcgill.ca

1027© 2011 by the American College of SurgeonsPublished by Elsevier Inc.

o decrease postoperative air leak. However, routine use ofhese adjuncts are of unproven use, and are often veryostly. A Cochrane meta-analysis investigating the effec-iveness of biologic and synthetic sealants or glues couldot demonstrate a substantial reduction in the incidence ofAL or duration of hospitalization.5 Pleural-based ad-

juncts, such as pleural tenting or autologous blood patchpleurodesis, have been described, but they are technicallydifficult and associated with their own set of complications.

The true effectiveness of these adjuncts might be hiddenby the fact that these trials did not specifically target pa-tients at high risk for PAL. Multiple risk factors for PALhave been described previously, but predicting the patientsin whom PAL will develop remains an inexact science.Identifying high-risk patients can lead to more targeted andeffective use of these expensive surgical sealants and otheradjuncts. Therefore, the aim of this study was to identifyrisk factors for PAL and to derive a simple scoring system to

predict the risk of PAL.

ISSN 1072-7515/11/$36.00doi:10.1016/j.jamcollsurg.2011.03.010

02(

rtvGwori

tswtmcwrvriv

pa

i

1028 Lee et al Predicting Prolonged Air Leak J Am Coll Surg

METHODSA derivation set was identified from a prospectively entereddatabase of patients who underwent pulmonary resectionbetween August 2002 and March 2007 at a single univer-sity teaching hospital. Patients with potential confoundingoperative and postoperative factors were excluded (pneu-monectomy, immediate postoperative mechanical ventila-tion, trauma, blebectomy, major chest wall or diaphrag-matic resections, and trans-sternal resection). Patients werealso excluded if surgical sealants or other buttressing mate-rials were used. Real-time documentation of all postopera-tive outcomes, including air leak, was performed by a nurseclinician rounding daily with the ward service and enteredin the database. PAL was defined as an air leak lastinglonger than 7 days.

Variables were compared between patients with PAL andthose without. Variables considered were patient sex andage, open thoracotomy versus video-assisted thoracoscopicsurgery, anatomic versus wedge resection, presence of pleu-ral adhesions at the time of operation, surgeon, and pul-monary function testing. Results are presented as mean �SEM. Statistical significance between means was deter-mined using two-tailed Student’s t-test for continuous vari-ables, and chi-square analysis for categorical variables. Dif-ferences were considered significant at p � 0.05.

Given that patients having undergone wedge resectionswere at limited risk of PAL as compared with those havingundergone anatomic resection, additional analyses were re-stricted to the subset of patients having undergone ana-tomic resection. Stepwise logistic regression analysis wasperformed to establish a simple predictive model to esti-mate the risk of PAL. Coefficients were then rounded toprovide a simple weighted scoring system to allow a rapidestimation of risk of PAL.

This scoring system was then validated in a set of pa-tients who underwent pulmonary resection between April2007 and June 2009 at the same university teaching hos-pital identified here. Patients were assigned into bins usingthe developed scoring system and observed and expec-ted frequencies were compared by Hosmer-Lemeshowgoodness-of-fit test.

RESULTSFrom August 2002 to March 2007, five hundred andeighty patients underwent pulmonary resection for sus-pected or confirmed malignancy and fit the inclusioncriteria described here. Patient and operative variablesare presented in Table 1. Overall incidence of PAL was14.1%. Of note, the vast majority of resections were

performed by open thoracotomy and two-thirds of re- r

sections were anatomic (lobectomy or segmentectomy),with the remainder being wedge resections (Table 1).

To identify factors that predict the risk of PAL, weperformed univariate analysis of these variables. Of thevariables considered, procedures performed by openthoracotomy (98% PAL� versus 87% PAL�; p �0.082), anatomic resection (89% versus 64%; p �

.001), the presence of pleural adhesions (34% versus3%; p � 0.031), forced expiratory volume in 1 secondFEV1; 74 versus 87% predicted; p � 0.001) and carbon

monoxide diffusing capacity (DLCO; 71 versus 84%predicted; p � 0.001) were associated with an increasedisk of PAL. Of these variables, the risk of PAL appearedo correlate most with anatomic resection (wedge: 4.8%ersus lobectomy/segmentectomy: 18.6%, p � 0.001).iven the limited risk of PAL in patients undergoingedge resection and the fact that more than two-thirdsf patients underwent anatomic resection, we elected toestrict additional analysis to the subset of patients hav-ng undergone anatomic resection.

Patient and operative variables for the subset of pa-ients having undergone anatomic resection are pre-ented in Table 2. The incidence of PAL in this groupas 18.6%. Univariate analysis of this group indicated

hat the presence of pleural adhesions and impaired pul-onary function continued to be associated with in-

reased risk of PAL (Table 2), and open thoracotomyas no longer significantly correlated. This stands to

eason, as most wedge resections were performed byideo-assisted thoracoscopic surgery and most anatomicesections were performed by open thoracotomy, as ev-denced by the significant association between these twoariables (p � 0.001).

To establish a model to predict the risk of PAL, weerformed stepwise logistic regression using the vari-bles identified in Table 2. To account for the increased

Table 1. Derivation and Validation Set Variables (All Pa-tients)

Variable

Derivationset

(n � 580)

Validationset

(n � 381)

Sex, male (%) 51 45Age (y), mean � SD 63 � 1 68 � 1Open thoracotomy (%) 88 76Anatomic resection (%) 68 67Presence of pleural adhesions (%) 25 18FEV1 (% predicted), mean � SD 85 � 1 86 � 1DLCO (% predicted), mean � SD 82 � 1 77 � 1PAL (%) 14 18

DLCO, carbon monoxide diffusing capacity; FEV1, forced expiratory volumen 1 second; PAL, prolonged air leak.

isk of PAL associated with impaired pulmonary func-

en

carn

e in 1

1

i

DR

1029Vol. 212, No. 6, June 2011 Lee et al Predicting Prolonged Air Leak

tion, pulmonary function testing values were enteredinto the model as 100 � FEV1, and 100 � DLCO. Thecoefficients from the logistic regression model (Table 3)were then rounded to provide a simple weighted scoringsystem to allow a rapid estimation of risk of PAL (Table4).

In this model, baseline risk for a patient undergoinganatomic resection with normal pulmonary functiontests and no pleural adhesions is 8.2%. A patient withpleural adhesions and poor pulmonary function is mostat risk for PAL developing, with the presence of pleuraladhesions having the same prognostic value as a 20%decrease in FEV1, or a 40% decrease in DLCO. Forxample, a patient with mild chronic obstructive pulmo-ary disease (predicted FEV1 � 70% and DLCO �

80%) undergoing a pulmonary lobectomy with the pres-ence of intrapleural adhesions has an estimated proba-bility of PAL of 28.5%.

To validate this model, the scoring system was applied toa set of 381 consecutive patients who underwent pulmo-nary resection between April 2007 and June 2009 at thesame university teaching hospital identified above (Ta-ble 1). Scores were calculated for each patient, and ob-served and expected frequencies were compared betweenweighted groups of patients by Hosmer-Lemeshowgoodness-of-fit test (Table 5). Although the scores ap-peared less accurate at predicting low risk of PAL, themodel was found to be well-calibrated (p � 0.2).

Table 2. Derivation Set Variables Associated with Prolonge

Variable Overall

Sex, male (%) 50Age (y), mean � SD 64 � 1Open thoracotomy (%) 98Presence of pleural adhesions (%) 27FEV1 (% predicted), mean � SD 84 � 1DLCO (% predicted), mean � SD 82 � 1

DLCO, carbon monoxide diffusing capacity; FEV1, forced expiratory volum

Table 3. Logistic Regression of Variables Predicting Pro-longed Air Leak (Anatomic Resection)Variable � coefficient p Value

Constant 2.420 �0.001Presence of pleural adhesions �0.483 0.140100 � FEV1 (% predicted) �0.026 0.00300 � DLCO (% predicted) �0.013 0.091

DLCO, carbon monoxide diffusing capacity; FEV1, forced expiratory volumen 1 second. i

DISCUSSIONPAL after pulmonary resection is a common complication,occurring in up to 26% of cases.1 It has been reported toincrease complications, length of hospitalization, and hos-pital costs.2,3 Prevention of this common yet highly morbidomplication begins with meticulous surgical technique tovoid tears in the lung parenchyma and the closure of mac-oscopic air leaks.6 Lack of an intraoperative air leak doesot reliably indicate the absence of postoperative air leak.7

Considerable efforts have been made to develop sur-gical adjuncts to prevent PAL. Several surgical tech-niques have been described, including pleural tenting,pneumoperitoneum, mechanical pleurodesis, and stapleline buttressing, but all of these procedures have theirown limitations or are prohibitively expensive to be usedroutinely.8-10 New biologic and synthetic sealants havebecoming increasingly popular as a method to prevent

Leak (Anatomic Resection)Air leak

p Value<7 Days >7 Days

48 59 0.09263 � 1 66 � 1 0.053

97 99 0.55824 37 0.029

87 � 1 74 � 3 �0.00184 � 1 71 � 3 �0.001

second.

Table 4. Scoring System to Estimate the Risk of ProlongedAir LeakAnatomic resection Score

Wedge resection Risk � 4.8%Presence of pleural adhesions �2FEV1 (% predicted) �1 per 10% below 100%

LCO (% predicted) �1 per 20% below 100%isk (%)8.2 010.2 112.8 215.8 319.5 423.7 528.5 633.8 739.7 845.8 952.0 1058.2 11

d Air

DLCO, carbon monoxide diffusing capacity; FEV1, forced expiratory volumen 1 second.

pduts3

aatG

�

ncdrrlia

Eap

f

1030 Lee et al Predicting Prolonged Air Leak J Am Coll Surg

postoperative air leaks, but results have been equivocal.Single-center trials have reported reductions in chesttube duration and incidence of PAL, but a decreasedlength of hospitalization has not been consistentlydemonstrated.11-13 A Cochrane meta-analysis also re-ported that PAL was reduced only in two of the seventrials that reported outcomes on PAL, and length of staywas considerably reduced in only 3 of 14 trials.5 Basedon these results, authors of the Cochrane review couldnot recommend systematic use of surgical sealants.However, these adjuncts can be effective if used only inpatients who are at high-risk of PAL.

Therefore, we have derived a simple scoring system topredict the probability of PAL, and then validated it in aconsecutive series of 381 patients. The incidence of PALin the derivation and validation sets was 14% and 18%,respectively. PAL was strongly associated with anatomicresection, poor pulmonary function, and presence ofpleural adhesions on unvariate analysis. Because of thelimited incidence of PAL in nonanatomical resections inour derivation set (4.8%), we decided to restrict analysisto anatomic resections only. Even after excluding non-anatomical resections, poor pulmonary function andpleural adhesions remained significantly associated withPAL. Pleural adhesions, FEV1 (each 10% below 100%

redicted) and DLCO (each 10% below 100% pre-icted) were assigned a score based on the � coefficientsing stepwise logistic regression. The aggregate scorehen estimated the probability of PAL (Table 4). Thiscoring system was then validated in a consecutive set of81 patients treated at the same institution (Table 5).Age approached statistical significance on univariate

nalysis in the original derivation set of patients, as wells the subset of patients who underwent anatomic resec-ion (Table 2), but was not included in the final model.iven that FEV1 and DLCO values were expressed rel-

ative to predicted values corrected for age, sex, height,and weight, it was inherent that the effect of age wouldbe included indirectly in the final model. To this end,re-analysis including age did not improve the accuracyof the model. In addition, Pearson correlation analysisdemonstrated age to be correlated with FEV1 (r �

0.100, p � 0.077) and DLCO (r � �0.215, p �0.001).

Several studies have previously described the association

Table 5. Hosmer-Lemeshow Goodness-of-FitTest in Valida-tion SetScore <1 2�3 4 5 6�7 >8

Expected risk (%) 9.0 14.5 19.5 23.7 31.6 44.4Observed risk (%) 16.7 17.5 16.0 31.8 22.2 46.2

of poor preoperative pulmonary function with increased

pulmonary complications and particularly PAL. Patientswith pulmonary function tests consistent with chronic ob-structive pulmonary diesase are at especially high risk forPAL. Abolhoda and colleagues found an FEV1/FVC ratio�50% to be associated with a higher incidence of PAL intheir series of 100 consecutive right upper lobectomies.1

Similarly, Stolz and colleagues identified COPD (definedas a preoperative predicted FEV1 �70% and FEV1/FVC�70%) as predictive for PAL.14 Studies by Cerfolio,Brunelli and their colleagues have similarly demonstrateddecreased FEV1 to be associated with PAL.15,16 These ab-

ormal spirometric values represent changes in lung me-hanics of COPD, namely increased airway resistance,ecreased pulmonary compliance, and emphysematous pa-enchyma, which result in incomplete apposition of theaw pulmonary surfaces essential in closing alveolar aireaks.17 The diffusing capacity of the lung has been shownn several studies to be a strong predictor of complicationsfter pulmonary resection.18-20 Unlike flow and volumetric

studies, which measure airflow limitation and lung vol-umes, DLCO assesses gas exchange at the alveolar level.

ven in patients without COPD, DLCO can be abnormalnd has been reported by to be associated with increasedostoperative morbidity.21

Upper lobectomies, bi-lobectomies, and pleural adhe-sions have also been associated with PAL previously.15,16,22

This might be due, in part, to the large residual space afterlobectomy, preventing complete parieto-visceral pleuralapposition and tears in the lung parenchyma caused byblunt or sharp dissection of pleural adhesions. Other riskfactors include age, sex, and factors impairing woundhealing.9,16,23,24

There have been several attempts at creating a scoringsystem to predict PAL. Brunelli and colleagues first re-ported a prediction model for prolonged air leak.15 Thisirst model incorporated predicted postoperative FEV1

and the presence or absence of upper lobectomy andpleural adhesions. However, the regression equation wasdifficult to calculate and predicted postoperative FEV1

is not routinely available in all centers. The authors thendeveloped a second scoring system using a derivation setof 658 patients and validated with 233 patients fromdifferent center. This model used an aggregate score sim-ilar to the present study, including age older than 65years, body mass index �25.5, FEV1 �80% of pre-dicted, and the presence or absence of pleural adhesions.Although similar at first glance, several important dif-ferences can be noted. In our model, pulmonary func-tion is ranked according to severity of impairment. Itstands to reason that a 60-year-old with FEV1 70% pre-

dicted and a 60-year-old with FEV1 40% predicted

oaBtbnscam

simpootptwtdsvPsmab

1031Vol. 212, No. 6, June 2011 Lee et al Predicting Prolonged Air Leak

would not have the same degree of air leak. In the modelby Brunelli and colleagues, these 2 patients would bothbe assigned the same score, but the patient with thelower FEV1 would be appropriately given a higher risk

f PAL in the present study. There are also concernsbout the generalizability of the scoring system byrunelli and colleagues, given that different body habi-

us of Italians and North Americans. Additionally, a lowody mass index can be a confounder because pulmo-ary function testing is not only adjusted for age andex, but usually also height and weight as well. In thisase, body mass index could have been representative ofltered chest wall and pulmonary mechanics instead ofalnutrition.Limitations of the present study include the retro-

pective single-center study design and lack of standard-zed protocol for air leak assessment and chest tube re-

oval. Chest tube management was dictated by surgeonreference; however, the majority of patients were keptn suction postoperatively until the absence of air leakn forced expiration. The chest tube was then removed ifhe no substantial pneumothorax was seen on chest x-rayerformed 4 hours after suction was discontinued. Suc-ion was reapplied if a pneumothorax was seen. Air leakas assessed clinically. The scoring system was also in-

ernally validated, so the selection biases present in theerivation set might have been present in the validationet as well. Despite these limitations, this study remainsaluable as we have derived a scoring system to predictAL that is easy to use. Patients identified by this scoringystem to have a low or high probability of PAL can beanaged accordingly. Patients at high risk can receive

dditional surgical adjuncts, and these interventions cane avoided in patients who are unlikely to benefit.

CONCLUSIONSPAL remains a common and morbid complication afterpulmonary resection. Surgical adjuncts have been devel-oped to prevent PAL, but are of limited use and oftenprohibitively expensive. On multivariate analysis, poorpulmonary function and pleural adhesions were stronglyassociated with PAL. We have devised a simple, easy-to-use scoring system using these varibales to estimate theprobability of PAL after pulmonary resection. Identifi-cation of patients at high risk for PAL can lead to moreselective and effective use of surgical adjuncts.

Author Contributions

Study conception and design: Lee, Sirois, Mulder, FerriAcquisition of data: Lee, Hanley, Robineau

Analysis and interpretation of data: Lee, Hanley, Ferri

Drafting of manuscript: Lee, Hanley, FerriCritical revision: Robineau, Sirois, Mulder, Ferri

REFERENCES

1. Abolhoda A, Liu D, Brooks A, Burt M. Prolonged air leakfollowing radical upper lobectomy: an analysis of incidenceand possible risk factors. Chest 1998;113:1507–1510.

2. Varela G, Jimenez MF, Novoa N, Aranda JL. Estimating hospitalcosts attributable to prolonged air leak in pulmonary lobectomy.Eur J Cardiothorac Surg 2005;27:329–333.

3. Okereke I, Murthy SC, Alster JM, et al. Characterization andimportance of air leak after lobectomy. Ann Thorac Surg 2005;79:1167–1173.

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

5. Belda-Sanchis J, Serra-Mitjans M, Iglesias Sentis M, Rami R.Surgical sealant for preventing air leaks after pulmonary resec-tions in patients with lung cancer. Cochrane Database Syst Rev2010;(1):CD003051.

6. Hazelrigg SR, Nunchuck SK, LoCicero J 3rd. Video Assistedthoracic surgery study group data. Ann Thorac Surg 1993;56:1039–1043.

7. Wain JC, Kaiser LR, Johnstone DW, Y, et al. Trial of a novelsynthetic sealant in preventing air leaks after lung resection. AnnThorac Surg 2001;71:1623–1628.

8. Brunelli A, Al Refai M, Monteverde M, et al. Pleural tent afterupper lobectomy: a randomized study of efficacy and durationof effect. Ann Thorac Surg 2002;74:1958–1962.

9. Hazelrigg SR, Boley TM, Naunheim KS, et al. Effect of bovinepericardial strips on air leak after stapled pulmonary resection.Ann Thorac Surg 1997;63:1573–1575.

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

11. Belboul A, Dernevik L, Aljassim O, et al. The effect of autolo-gous fibrin sealant (Vivostat) on morbidity after pulmonary lo-bectomy: a prospective randomised, blinded study. Eur J Car-diothorac Surg 2004;26:1187–1191.

12. Gagarine A, Urschel JD, Miller JD, et al. Effect of fibrin glue onair leak and length of hospital stay after pulmonary lobectomy.J Cardiovasc Surg (Torino) 2003;44:771–773.

13. Fabian T, Federico JA, Ponn RB. Fibrin glue in pulmonary re-section: a prospective, randomized, blinded study. Ann ThoracSurg 2003;75:1587–1592.

14. Stolz AJ, Schutzner J, Lischke R, et al. Predictors of prolongedair leak following pulmonary lobectomy. Eur J CardiothoracSurg 2005;27:334–336.

15. Brunelli A, Monteverde M, Borri A, et al. Predictors of pro-longed air leak after pulmonary lobectomy. Ann Thorac Surg2004;77:1205–1210; discussion 1210.

16. Cerfolio RJ, Bass CS, Pask AH, Katholi CR. Predictors and treatmentof persistent air leaks. Ann Thorac Surg 2002;73:1727–1730.

17. Loran DB, Woodside KJ, Cerfolio RJ, Zwischenberger JB. Predictorsof alveolar air leaks. Chest Surg Clin N Am 2002;12:477–488.

18. Cerfolio RJ, Bryant AS. Different diffusing capacity of the lungfor carbon monoxide as predictors of respiratory morbidity. AnnThorac Surg 2009;88:405–410.

19. Brunelli A, Refai MA, Salati M, et al. Carbon monoxide lung

diffusion capacity improves risk stratification in patients

1032 Lee et al Predicting Prolonged Air Leak J Am Coll Surg

without airflow limitation: evidence for systematic mea-surement before lung resection. Eur J Cardiothorac Surg2006;29:567–570.

20. Ferguson MK, Little L, Rizzo L, et al. Diffusing capacity predictsmorbidity and mortality after pulmonary resection. J ThoracCardiovasc Surg 1988;96:894–900.

21. Ferguson MK, Vigneswaran WT. Diffusing capacity predictsmorbidity after lung resection in patients without obstructive

lung disease. Ann Thorac Surg 2008;85:1158–1164.

22. Cerfolio RJ, Bass C, Katholi CR. Prospective randomized trialcompares suction versus water seal for air leaks. Ann ThoracSurg 2001;71:1613–1617.

23. Rice TW, Okereke IC, Blackstone EH. Persistent air-leak fol-lowing pulmonary resection. Chest Surg Clin N Am 2002;12:529–539.

24. Cerfolio RJ, Tummala RP, Holman WL, et al. A prospectivealgorithm for the management of air leaks after pulmonary re-

section. Ann Thorac Surg 1998;66:1726–1731.