A Scoring System to Predict the Risk of Prolonged Air Leak After Lobectomy

SURGERY: The Annals of Thoracic Surgery CME Program is located online at http://cme.ctsnetjournals.org. To take the CME activity related to this article, you must have either an STS member or an individual non-member subscription to the journal. A Scoring System to Predict the Risk of Prolonged Air Leak After Lobectomy Alessandro Brunelli, MD, Gonzalo Varela, MD, PhD, Majed Refai, MD, Marcelo F. Jimenez, MD, Cecilia Pompili, MD, Armando Sabbatini, MD, and Jose Luis Aranda, MD Division of Thoracic Surgery, Ospedali Riuniti, Ancona, Italy; and Division of Thoracic Surgery, Salamanca University Hospital, Salamanca, Spain Background. Prolonged air leak (PAL) remains a frequent complication after lung resection. Perioperative preventative strategies have been tested, but their efficacy is often difficult to interpret due to heterogeneous inclusion criteria. The objective of this study was to develop and validate a practical score to stratify the risk of PAL after lobectomy. Methods. Six hundred fifty-eight consecutive patients were submitted to pulmonary lobectomy (2000 to 2008) in center A and were used to develop the risk-adjusted score predicting the incidence of PAL (> 5 days). Exclusion criteria were chest wall resection and postoperative assisted mechanical ventilation. No sealants, pleural tent, or buttressing material were used. To build the aggregate score numeric variables were categorized by receiver operating curve analysis. Variables were screened by univariate analysis and then used in stepwise logistic regression analysis (validated by bootstrap). The scoring system was developed by proportional weighing of the significant predictor estimates and was validated on patients operated on in a different center (center B). Results. The incidence of PAL in the derivation set was 13% (87 of 658 cases). Predictive variables and their scores were the following: age greater than 65 years (1 point); presence of pleural adhesions (1 point); forced expiratory volume in one second less than 80% (1.5 points); and body mass index less than 25.5 kg/m 2 (2 points). Patients were grouped into 4 risk classes according to their aggregate scores, which were significantly associated with incremental risk of PAL in the validation set of 233 patients. Conclusions. The developed scoring system reliably predicts incremental risk of PAL after pulmonary lobectomy. Its use may help in identifying those high-risk patients in whom to adopt intraoperative prophylactic strategies; in developing inclusion criteria for future randomized clinical trials on new technologies aimed at reducing or preventing air leak; and for patient counseling. (Ann Thorac Surg 2010;90:204 9) 2010 by The Society of Thoracic Surgeons Prolonged air leak (PAL) remains a frequent complication after pulmonary resection, occurring in approximately 10 to 15% of patients after pulmonary lobectomy [1, 2]. It may prolong hospital stay, impacting on costs [3 6], and increase the risk of other complications, including empyema [5 7]. For these reasons, several preventative strategies, including surgical techniques, sealants, or buttressing materials, have been tested in the clinical investigations. However, their reported cost efficacy has been often difficult to interpret owing to heterogeneous inclusion criteria. A commonly accepted system that may assist in stratifying the risk of PAL and standardize the selection of patients for these trials or for using prophylactic measures in the clinical practice is lacking. Accepted for publication Feb 3, 2010. Presented at the Forty-sixth Annual Meeting of The Society of Thoracic Surgeons, Fort Lauderdale, FL, Jan 25 27, 2010. Address correspondence to Dr Brunelli, Division of Thoracic Surgery, Ospedali Riuniti, Via Conca 1, Ancona, IT60020, Italy; e-mail: brunellialex@gmail.com. The objective of this study was to develop a practical and user-friendly aggregate risk score to stratify the risk of PAL after pulmonary lobectomy. Patients and Methods This is an observational multicenter analysis performed on prospective electronic databases. The study was approved by the local Institutional Review Boards and all patients gave their consent to use their data in the dataset. All consecutive pulmonary lobectomies operated on since January 2000 through April 2008 in center A for lung cancer were used as a derivation set to develop the scoring system predicting the risk of PAL. The risk score was then validated on a sample of patients operated on in another center (center B) from 2006 to 2008. In both centers, major lung resections were contraindicated in those patients with a predicted postoperative forced expiratory volume in one second and a predicted postoperative carbon monoxide lung diffusion capacity less than 30% of predicted, in addition to an insufficient exercise tolerance (height reached at stair climbing test 2010 by The Society of Thoracic Surgeons 0003-4975/$36.00 Published by Elsevier Inc doi:10.1016/j.athoracsur.2010.02.054

Ann Thorac Surg BRUNELLI ET AL 2010;90:204 9 SCORING SYSTEM FOR RISK OF PAL AFTER LOBECTOMY 205 less than 12 m or maximum oxygen consumption less than 10 ml/kg/min) [8]. Patients undergoing lung resections including chest wall or diaphragm resection, or those needing postoperative mechanical ventilation at any time after the operation were not included in this series. Mortality rate in both centers was 1% and these cases were not used for the analysis. All patients in both centers were operated on by qualified thoracic surgeons through a musclesparing anterolateral thoracotomy. Mechanical staplers were used to develop incomplete fissures in 80% of patients and to close the bronchus in all cases. Twenty percent of patients had completely developed or filmy fissures that did not require the use of staplers. No pleural tents, sealants, buttressing material, or pneumoperitoneum were applied in any of these patients. Systematic lymphadenectomy was performed in all cases after the pulmonary resection [9]. At completion of the operation the presence of an air leak was tested by submerging the lung parenchyma in sterile saline and reinflating the lung up to a pressure of 25 to 30 cm H 2 O. If any significant air leak was detected, any attempt was made to reduce it by applying sutures. One (center A after July 2007 and center B in all cases) or two chest tubes (center A before July 2007) were positioned at the end of the operation. Chest tubes were left on suction ( 20 cm H 2 O) until the morning of the first postoperative day and then either alternate suction (suction during night and no suction during day 10) or no suction was applied according to institutional policies. Patients were extubated in the operating room and admitted to a specialized dedicated thoracic ward. Perioperative treatment was standardized and focused on the control of postthoracotomy chest pain, chest physiotherapy, and early as possible mobilization, antibiotic and antithrombotic prophylaxis. Postoperative chest pain was assessed at least twice a day during morning and evening rounds. Treatment was titrated to achieve a pain score below 4 (range 0 to 10) during the first 72 postoperative hours by means of epidurals or continuous intravenous infusion of nonopioid analgesics. Physical rehabilitation and chest physiotherapy were performed in all patients starting from postoperative day 1 by qualified and dedicated physiotherapists according to standardized protocols. No positive pressure ventilation was used. Air leak was assessed twice daily (during morning and evening rounds). Patients were instructed to perform standardized repeated forced expiratory maneuvers (coughing and blowing). Chest tubes were removed if no air leak was detectable in the chest drain unit and the pleural effusion was less than 400 ml in the last 24 hours, after a chest X-ray was obtained to show satisfactory lung expansion. Statistical Analysis The derivation set consisted of 658 consecutive patients who underwent pulmonary lobectomy in center A. This sample was used to develop the risk-adjusted score predicting the incidence of PAL. For the purpose of this study, PAL was defined as an air leak lasting longer than 5 days. Initially the following series of perioperative factors were screened by univariate analysis for possible association with PAL: age, gender, serum albumin level, hemoglobin level, forced expiratory volume in one second (FEV 1 %), ratio of forced expiratory volume in one second to total lung capacity, ratio of residual volume to total lung capacity, body mass index, diffusing capacity of lung for carbon monoxide %, smoking pack-years, diabetes, preoperative systemic steroids, induction chemotherapy, side and site of lobectomy, presence of pleural adhesions (defined as dense adhesions occupying an entire lobe or at least 30% of the lung surface), and length of stapled parenchyma. All data were at least 95% complete. Sporadic missing data were imputed by averaging the nonmissing values (numeric variables) or taking the most frequent category (categoric variables). To avoid multicollinearity, only one variable in a set of variables with a correlation coefficient greater than 0.5 was selected (by bootstrap procedure), and used in the regression model. For the purpose of this analysis, significant (p 0.05) numeric variables were tested for a threshold effect and dichotomized by using receiver operating curve analysis (for identifying the best cutoff). Significant variables at univariate analysis were then used as independent predictors in a stepwise logistic regression analysis (dependent variable: presence of PAL 5 days). The reliability of the predictors was finally assessed by using a bootstrap resampling technique with 1,000 samples [10 12]. In the bootstrap procedure, repeated samples of the same number of observations as the original database were selected with replacement from the original set observations. For each sample, stepwise logistic regression was performed. The stability of the final stepwise model can be assessed by identifying the variables that enter most frequently in the repeated bootstrap models and comparing those variables with the variables in the final stepwise model. If the final stepwise model variables occur in a majority ( 50%) of the bootstrap models, the original final stepwise regression model can be judged to be stable. Only reliable (bootstrap frequency 50% in 1,000 simulated samples) predictors were used to construct the final aggregate score. The scoring system was developed by proportional weighing of the significant predictors estimates, assigning a value of 1 to the smallest coefficient. An aggregate risk score was generated for each patient by summing each estimate [13]. Finally, patients were grouped in classes of incremental risk according to their total score. The risk score was then validated on the patients operated on in center B (external validation set of 233 patients) and the risk of PAL was verified in each class in this external population. Moreover, to further validate the score, the population of center B was bootstrapped to obtain 1,000 simulated external samples. The proportion of patients with PAL was then verified for each class in each of these bootstrapped samples. Statistical analysis was performed on the Stata 9.0 statistical software (StataCorp, College Station, TX). All tests were 2-tailed with a statistical significance of 0.05.

206 BRUNELLI ET AL Ann Thorac Surg SCORING SYSTEM FOR RISK OF PAL AFTER LOBECTOMY 2010;90:204 9 Table 1. Characteristics of Patients in the Derivation and Validation Sets Variables Results Derivation Set (n 658) Validation Set (n 233) Age (years) 66.9 (10.2) 63.2 (12) Male gender (n,%) 514 (78%) 176 (79%) BMI m/kg 2 26.2 (4) 26.2 (4.4) FEV 1 % 87.3% (18.5) 87.3 (24) FEV 1 /FVC ratio 0.7 (0.1) 0.74 (0.1) D lco % 79.5% (19.5) 81.9% (18.7) RV/TLC 0.43 (0.08) 0.41 (0.1) Albumin level (g/dl) 4.2 (2.2) 4.1 (2.5) Pack-years 42.8 (32.4) 45.5 (35) Diabetes (n,%) 70 (11%) 34 (15%) Induction chemotherapy (n,%) 63 (9.6%) 25 (11%) Systemic steroids (n,%) 26 (3.9%) 7 (3.1%) Side of resection (right, n,%) 368 (56%) 132 (56%) Site of resection (upper, n,%) 409 (62%) 139 (60%) Pleural adhesions (n,%) 176 (27%) 96 (41%) Results are expressed as means standard deviations unless otherwise specified. BMI body mass index; D lco diffusing capacity of lung for carbon monoxide; FEV 1 forced expiratory volume in one second; FEV 1 / FVC forced expiratory volume to forced volume capacity ratio; RV/ TLC ratio of residual volume to total lung capacity. The characteristics of patients in the derivation and validation sets are illustrated in Table 1. The incidence of PAL in the derivation and validation sets were 13% (87 of 658 cases) and 14% (32 of 233 cases), respectively. After receiver operating curve analysis was used to categorize the numeric variables, stepwise logistic regression identified the following significant and reliable predictors of PAL: age greater than 65 years (p 0.04, coefficient 0.558, standard error [SE] 0.27, bootstrap frequency 57%), presence of pleural adhesions (p 0.01, coefficient 0.616, SE 0.25, bootstrap frequency 68%), FEV 1 less than 80% (p 0.001, coefficient 0.795, SE 0.24, bootstrap frequency 88%), body mass index (BMI) less than 25.5 kg/m 2 (p 0.0001, coefficient 1.03, SE 0.25, bootstrap frequency 98%). Based on their coefficients, the individual factor scores were the following: age greater than 65 years, 1 point; presence of adhesions, 1 point; FEV 1 less than 80%, 1.5 points; and BMI less than 25.5 kg/m 2, 2 points (Table 2). To obtain a cumulative score the individual points were summed in each patient to obtain a range from 0 to 5.5. For example, a 70-year-old patient, with a FEV 1 of 60% predicted, a BMI of 23 kg/m 2, and with pleural adhesions would have a score of 5.5 points. Patients were then grouped into four risk classes according to their aggregate scores, which were significantly associated with incremental risk of PAL in the validation set of 233 patients (Table 3). Patients in class A (score 0) had no PAL, whereas patients in class D (score 3) had 26% risk of developing PAL (p 0.003). When the risk classes were assessed in 1,000 bootstrapped samples from center B, we found that in class A 98% of samples had a PAL risk less than 5%. Class B had a risk less than 10% in 99% of samples. On the other hand, class C had a risk greater than 10% in 91% of samples (although in no cases greater than 20%) and class D had a PAL risk greater than 20% in 99% of samples. Comment The objective of this study was to develop and validate an aggregate score to stratify the risk of prolonged air leak after pulmonary lobectomy. For the purpose of this investigation, an air leak was defined as prolonged if it lasted more than 5 days in line with recent recommendations [14] and with the current definition in both European Society of Thoracic Surgeons and Society of Thoracic Surgeons thoracic databases. Prolonged air leak remains a frequent and bothering complication after lung resection. It has been reported to occur in approximately 10 to 15% of patients after lobectomy [1, 2]. Besides obliging the patients to live with a chest tube in place causing distress, anxiety, and pain, it may increase the risk of other cardiopulmonary complications [5, 7] and empyema [6]. It has been shown to be one of the most important determinants of a prolonged hospital stay and increased hospital costs [3 6]. Several papers have tried to identify risk factors for prolonged air leak after lung resection [14], but to our knowledge there is no user-friendly risk model or score available in the literature that could be used in the clinical setting to rapidly stratify the risk of PAL. Although previous studies have shown that certain factors are associated with increased risk of PAL, there is the need to integrate the most reliable factors in a single scoring system. This would greatly simplify the selection of patients for future efficacy studies on preventative interventions (like use of sealants or buttressed staple lines), making the interpretation of results across different investigations more consistent and meaningful. We chose to limit the analysis to lobectomies only, for the sake of homogeneity and because lobectomies carry a greater risk of PAL compared with minor resections as reported in the European Society of Thoracic Surgeons Database 2009 Annual Report (10% vs 4%). We found that significant and reliable predictors of PAL were older age ( 65 years), reduced pulmonary function (FEV 1 80%), presence of pleural adhesions, and low BMI ( 25.5 kg/ Table 2. Points Assigned at Individual Variables and to be Summed to Derive the Aggregate PAL Risk Score Variables Points Assigned at Individual Variables Age 65 1 BMI 25.5 m/kg 2 2 FEV 1 80% 1.5 Presence of pleural adhesions at 1 operation BMI body mass index; FEV 1 forced expiratory volume in one second; PAL prolonged air leak.

Ann Thorac Surg BRUNELLI ET AL 2010;90:204 9 SCORING SYSTEM FOR RISK OF PAL AFTER LOBECTOMY 207 Table 3. Distribution of Patients and Incidence of PAL in the Derivation and Validation Sets by Class of Risk Derivation Set (658 Patients) Validation Set (233 Patients) PAL Risk Class No. Patients (%) PAL Incidence No. Patients (%) PAL Incidence Class A (score 0) 68 (10%) 1.4% 23 (10%) 0 Class B (score 1) 136 (21%) 5% 30 (13%) 6.7% Class C (score 1.5 3) 313 (48%) 12.5% 110 (47%) 10.9% Class D (score 3) 141 (21%) 29% 70 (30%) 25.7% 2, p value 0.0001 0.003 C-index 0.71 (95% CL 0.65 and 0.77) 0.69 (95% CL 0.59 and 0.77) C-index receiver operating curve; CL confidence limits; PAL prolonged air leak. m 2 ). Although older age has never been reported to be a significant risk factor for PAL [14], elderly patients may have a more fragile lung parenchyma with a reduced healing capacity, which may predispose to the occurrence of this complication. A reduced pulmonary function has been reported as one of the most consistent risk factors for PAL [2, 3, 15 17]. Presence of important pleural adhesions has been previously found to be associated with this complication [2]. Tears in the lung parenchyma may ensue during mobilization of the lung and taking down of the adhesions. Finally, a low BMI may be a marker of a poor nutritional status, which in turn may negatively influence the healing of the surgically damaged tissue. We chose to construct an aggregate risk score to provide a practical tool for stratifying the risk for clinical and scientific purposes following previously reported methods [13]. In this way, four risk classes were identified. It readily appears that patients in class A had a minimal risk of PAL. Certainly these patients are not candidates for any preventative interventions or for efficacy trials testing new prophylactic devices or materials. Conversely, patients in class D were the ones to have the highest risk of PAL. These are the patients who could benefit most from the application and use of materials such as sealants or buttressed staples or preventative techniques such as pleural tents or pneumoperitoneum. Trials testing the cost efficacy of these measures or treatment should be ideally performed on these high-risk patients in order to obtain the most meaningful inferences. If they will show that such treatments or techniques will be able to reduce the risk of PAL in this set of patients then their systematic use could be warranted in the clinical practice. A risk stratification system such as this may therefore have a positive clinical and economic impact, avoiding unnecessary costly treatments in patients who most likely would not need them. This is particularly true for those devices, such as sealants or buttressed staple lines, which may significantly impact on surgical costs, or for those surgical techniques, such as pleural tent, which may increase the risk of other complications (bleeding). The risk score may further assist during preoperative counseling identifying the patients more at risk to develop PAL and permitting to instruct them in advance on the possibility to be discharged home with a portable chest drainage unit. The scoring system was validated in an external population showing a good discrimination across different populations. It was further validated in 1,000 external bootstrapped samples drawn from the validations set. This new statistical approach allowed us to assess the stability of the score across multiple external populations warranting its generalization. Bootstrap has been shown to be superior to the traditional training and test splitting of the population to assess reliability of predictors and risk models [12]. Although the model has been developed in a sizeable, homogeneous population and further validated in another external sample of patients operated on in another center, this study may have potential limitations. The retrospective and multicenter nature of the study may have implied some problems of definition and recording of variables and outcomes. Patients had partly different chest tube managements at the two centers (although most of the time tubes were left without suction in both centers) that could have potentially influenced duration of air leak [18, 19 22]. Nevertheless, the score appears to stratify the risk of PAL in both centers irrespective of the chest tube management. Air leak was assessed by several and different operators in two centers (staff surgeons on duty). Interobserver variability has been reported to occur in air leak assessment [23], but we think this is a general phenomenon which reflects the real life situation in the clinical ward and minimally affected, if any, the results of this study. In conclusion, we were able to develop and externally validate an easy-to-use risk score for prediction of PAL after pulmonary lobectomy. The primary application of this prediction score is for the targeted use of prophylactic therapy and interventions and for appropriate selection criteria in efficacy trials, thereby minimizing the expense and risk of such interventions or trials in those patients unlikely to derive benefit. References 1. Cerfolio RJ, Bass CS, Pask AH, Katholi CR. Predictors and treatment of persistent air leaks. Ann Thorac Surg 2002;73: 1727 31.

208 BRUNELLI ET AL Ann Thorac Surg SCORING SYSTEM FOR RISK OF PAL AFTER LOBECTOMY 2010;90:204 9 2. Brunelli A, Monteverde M, Borri A, Salati M, Marasco R, Fianchini A. Predictors of prolonged air leak after pulmonary lobectomy. Ann Thorac Surg 2004;77:1205 10. 3. Bardell T, Petsikas D. What keeps postpulmonary resection patients in hospital? Can Respir J 2003;10:86 9. 4. Irshad K, Feldman LS, Chu VF, et al. Causes of increased length of hospitalization on a general thoracic surgery service: a prospective observational study. Can J Surg 2002;45:264 8. 5. Varela G, Jiménez MF, Novoa N, Aranda JL. Estimating hospital costs attributable to prolonged air leak in pulmonary lobectomy. Eur J Cardiothorac Surg 2005;27:329 33. 6. Brunelli A, Xiumé F, Al Refai M, et al. Air leaks after lobectomy increase the risk of empyema but not of cardiopulmonary complications: a case-matched analysis. Chest 2006;130:1150 6. 7. 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. 8. Brunelli A, Charloux A, Bolliger CT. ERS/ESTS clinical guidelines on fitness for radical therapy in lung cancer patients (surgery and chemo-radiotherapy). Eur Respir J 2009;34:17 41. 9. Lardinois D, De Leyn P, Van Schil P, et al. ESTS guidelines for intraoperative lymph node staging in non-small cell lung cancer. Eur J Cardiothorac Surg 2006;30:787 92. 10. Blackstone EH. Breaking down barriers: helpful breakthrough statistical methods you need to understand better. J Thorac Cardiovasc Surg 2001;122:430 9. 11. Grunkemeier GL, Wu YX. Bootstrap resampling method: something for nothing? Ann Thorac Surg 2004;1142 4. 12. Brunelli A, Rocco G. Internal validation of risk models in lung resection surgery: bootstrap versus training and test sampling. J Thorac Cardiovasc Surg 2006;131:1243 7. 13. Passman RS, Gingold DS, Amar D, et al. Prediction rule for atrial fibrillation after major noncardiac thoracic surgery. Ann Thorac Surg 2005;79:1698 703. 14. Singhal S, Ferraris VA, Bridges CR, et al. Management of alveolar air leaks after pulmonary resection. Ann Thorac Surg 2010;89:1327 35. 15. Linden PA, Bueno R, Colson YL, et al. Lung resection in patients with preoperative FEV1 35% predicted. Chest 2005;127:1984 90. 16. Stolz AJ, Schützner J, Lischke R, Simonek J, Pafko P. Predictors of prolonged air leak following pulmonary lobectomy. Eur J Cardiothorac Surg 2005;27:334 6. 17. Abolhoda A, Liu D, Brooks A, Burt M. Prolonged air leak following radical upper lobectomy: an analysis of incidence and possible risk factors. Chest 1998;113:1507 10. 18. Brunelli A, Sabbatini A, Xiumé F, Refai MA, Salati M, Marasco R. Alternate suction reduces prolonged air leak after pulmonary lobectomy: a randomized comparison versus water seal. Ann Thorac Surg 2005;80:1052 5. 19. Cerfolio RJ, Bass C, Katholi CR. Prospective randomized trial compares suction versus water seal for air leaks. Ann Thorac Surg 2001;71:1613 7. 20. Marshall MB, Deeb ME, Bleier JI, et al. Suction vs water seal after pulmonary resection: a randomized prospective study. Chest 2002;121:831 5. 21. Alphonso N, Tan C, Utley M, et al. A prospective randomized controlled trial of suction versus non-suction to the under-water seal drains following lung resection. Eur J Cardiothorac Surg 2005;27:391 4. 22. Brunelli A, Monteverde M, Borri A, et al. Comparison of water seal and suction after pulmonary lobectomy: a prospective, randomized trial. Ann Thorac Surg 2004;77:1932 7. 23. Varela G, Jiménez MF, Novoa NM, Aranda JL. Postoperative chest tube management: measuring air leak using an electronic device decreases variability in the clinical practice. Eur J Cardiothorac Surg 2009;35:28 31. DISCUSSION DR DAVID HARPOLE, JR (Durham, North Carolina): I noticed that as a prerequisite to the trial you used sort of no provocative measures for spaces or leaks; you didn t do pleural tents or pneumoperitoneum and so forth. DR BRUNELLI: No. Pleural tents or pneumoperitoneum were not used in this series. DR HARPOLE: So in your center and in this trial, for a bilobectomy there was clearly a space afterwards. What was your usual mechanism for taking care of those? DR BRUNELLI: Although we previously published on the pleural tent we now use it sporadically only in some patients where a large residual pleural space is anticipated, but not routinely anymore. In this regard the cohort of patients used for the analysis was homogeneous and we really had the chance to test the influence of other factors on the duration of air leak eliminating the confounder of prophylactic measures. Since the objective was indeed to find a classification system to help selecting those patients who could benefit from their use. DR BRYAN F. MEYERS (St. Louis, MO): I just rise to congratulate you on your paper. I think that usually in papers like this, at meetings like this at the AATS (American Association for Thoracic Surgery), people will analyze their data, create a multivariable model, give us the result, and then expect us to accept that those data will extrapolate to other datasets. But what you have done here and what you generally do in work that you have done in this line is you always give us the derivation set and then have another set of data to validate it all within the same paper, and I think that really adds to the reliability of your findings, so I commend you on that. The one thing I see is that the overall numbers that you used to build the models were a bit smaller than you have used in the past for other similar models and I just wondered if you ran into any challenges in doing this work with the smaller numbers of patients that you had to deal with. DR BRUNELLI: We feel quite confident in using bootstrap. Especially when you have rare events or small numbers, we really feel that bootstrap adds to the analysis. We strongly recommend the use of bootstrap to assess model and factors reliability after logistic regression even in the case of small numbers, and particularly if you have rare events. DR TODD L. DEMMY (Buffalo, NY): I m curious regarding your model as far as the pathophysiology, particularly the low body mass index having a greater risk of air leak. One could look at these risk factors and wonder if it s a compliance problem, a relative disproportion of chest space and remaining lung to fill it. What do you think is the primary mechanism and is there maybe a more precise way to measure it, like lung compliance or maximum negative intrapleural pressure, at the end of the case? DR BRUNELLI: It is even more surprising when you think that BMI was the strongest, the most influential factor associated with PAL, and this was the first time we found this, and I m not aware in the literature of any study reporting on this association. We think a low BMI may be associated with perhaps a low

Ann Thorac Surg BRUNELLI ET AL 2010;90:204 9 SCORING SYSTEM FOR RISK OF PAL AFTER LOBECTOMY 209 nutritional status, hindering tissue healing and regeneration. This is our interpretation, but we cannot exclude that a low BMI is a surrogate measure reflecting or influencing chest mechanics. But I agree this should be analyzed further and using other technologies. With the data we have in our hands now, we cannot speculate too much on this. DR SCOTT J. SWANSON (Boston, Massachusetts): It s kind of a rare finding in the U.S. to have somebody with a low body mass index. What percentage of your patients were below 25? DR BRUNELLI: We had a rather high percentage, around 40%. DR SWANSON: They were below 25? DR BRUNELLI: Yes. DR JOHN A. ODELL (Jacksonville, FL): I think my planned question has briefly been answered. Italians are built differently from Americans and I just wonder whether your data would apply to people of American build, where obesity is so common. DR BRUNELLI: I don t know. That s a difficult question to answer. Although the model appeared quite generalizable by using external validation and bootstrap, it would be interesting to cross-validate the model in the North American population, perhaps using the STS [Society of Thoracic Surgeons] database. That could be an interesting analysis.