Validation of a Model to Predict Perioperative Mortality from Lung Cancer Resection in the Elderly

Transcription

1 Validation of a Model to Predict Perioperative Mortality from Lung Cancer Resection in the Elderly Max Kates 1, Xavier Perez 2, Julie Gribetz 1, Scott J. Swanson 3, Thomas McGinn 2, and Juan P. Wisnivesky 2,4 1 Mount Sinai School of Medicine, New York, New York; 2 Division of General Internal Medicine, Mount Sinai School of Medicine, New York, New York; 3 Division of Thoracic Surgery, Brigham and Women s Hospital, Harvard Medical School, Boston, Massachusetts; and 4 Division of Pulmonary, Critical Care, and Sleep Medicine, Mount Sinai School of Medicine, New York, New York Rationale: Surgical resection is the mainstay therapy for localized non small cell lung cancer (NSCLC), yet elderly patients are less likely to be treated due to concerns about morbidity and mortality related to surgery. Objectives: To validate and refine a clinical model to predict 30-day perioperative mortality (POM) in elderly patients undergoing curative resection for lung cancer. Methods: We identified 14,297 patients aged 65 years and older with stage I, II, or IIIA NCSLC from the Surveillance, Epidemiology, and End-Results Registry linked to Medicare claims. We used logistic regression analysis to identify independent risk factors for POM and to validate and refine a previously derived prediction model. Measurements and Main Results: Overall, POM was 4.6% (95% confidence interval, %). Multiple regression analysis revealed that greater age, male sex, resections of multiple lobes, advanced stage, greater tumor size, and certain comorbidities were associated with increased risk for POM. These risk factors were similar to those observed in the prior model. When patients were stratified according to their predicted risk of POM, the observed mortality increased from 1.2 to more than 10%. Conclusions: Among elderly patients with lung cancer, a prediction rule can identify those patients at higher risk for fatal complications from surgery. Further studies should evaluate whether use of the model can lead to improvements in treatment decision making. Keywords: lung malignancy; lung resection; risk assessment; surgical outcomes With a median age at diagnosis of 69 years, rates of lung cancer are increasing in elderly patients relative to the younger population (1, 2). Lung resection is the mainstay therapy for non small cell lung cancer (NSCLC) localized to the chest (stages I IIIA), with 5-year survival rates of up to 70% for earlier stages (3). However, surgical management of the elderly can be challenging due to increased frailty and a higher prevalence of comorbid conditions, and, for this reason, many physicians are more reluctant to recommend surgery for these patients (4, 5). These factors may contribute to lower overall rates of resection among the elderly and an increased use of safer, yet potentially less effective, limited resections (6, 7). When making decisions regarding resection, physicians must balance the potential long-term benefits of surgery with the risk of surgery-related deaths, particularly among patients with multiple comorbid conditions or limited life expectancy. Some studies have shown that certain characteristics of the patient, the tumor, or the surgery are associated with increased 30-day perioperative (Received in original form August 26, 2008; accepted in final form November 20, 2008) Correspondence and requests for reprints should be addressed to Juan P. Wisnivesky, M.D., M.P.H., Department of Medicine, Mount Sinai School of Medicine, One Gustave L. Levy Place, Box 1087, New York, NY juan.wisnivesky@mssm.edu This article has an online supplement, which is accessible from this issue s table of contents at Am J Respir Crit Care Med Vol 179. pp , 2009 Originally Published in Press as DOI: /rccm OC on November 21, 2008 Internet address: AT A GLANCE COMMENTARY Scientific Knowledge on This Subject For elderly patients with lung cancer, the risk of perioperative mortality from resection is an important concern. A clinical rule can help physicians and patients make informed treatment decisions and anticipate potential complications from surgery. What This Study Adds to the Field Using a population-based cancer registry, we validated and refined a model to predict perioperative mortality in an elderly population undergoing lung cancer resection. mortality (POM) (8 10). Using these risk factors, Strand and colleagues (11) developed a clinical rule to predict the risk of POM in patients undergoing resection for lung cancer. This predictive rule has several potential clinical implications. First, it could be used to assist physicians and patients make informed decisions regarding surgery. Second, it could help select high-risk patients who should undergo further testing before surgery and receive closer follow-up care in the postoperative period. Ultimately, high-risk patients identified by the rule could be the target of interventions to decrease POM. Conversely, the rule can be used to identify low-risk patients who otherwise may not be operated on because of their advanced age. However, before a predictive model can be applied to clinical practice, it should be externally validated among different patient populations to assess the generalizability of the model (12, 13). The aim of this study was to use data from a populationbased cancer registry to validate and refine a clinical model to predict 30-day POM among elderly patients undergoing curative resection for lung cancer. METHODS Since 1973, the Surveillance, Epidemiology, and End Results (SEER) program has collected clinicopathologic data on all incident cancer cases in specified geographic areas of the United States (14). The SEER registry has been combined with Medicare claims to provide additional data on inpatient treatment, as well as outpatient physician services (15). From SEER Medicare, we selected all patients aged 65 years and older with histologically confirmed stages I IIIA NSCLC who underwent surgical resection between 1991 and Additional information regarding the identification of cases, the coding of variables, and the statistical analysis is available in the online supplement. Baseline sociodemographic information was obtained from SEER and Medicare databases. To obtain a summary measure of the burden of comorbidities within our sample population, we employed the Deyo adaptation of the Charlson comorbidity index (16 18). Additionally, we created several dichotomous variables indicating the presence or absence of individual comorbidities. Stage was classified according to the American Joint Committee on Cancer criteria (19). In terms of

2 Kates, Perez, Gribetz, et al.: Predicting Mortality from Lung Surgery 391 histology, cases were classified as adenocarcinoma, bronchoalveolar cell carcinoma, squamous, or large-cell carcinoma (20). Patients were grouped by their type of resection according to the SEER site-specific codes (segmentectomy, wedge resection, lobectomy, bilobectomy, pneumonectomy, and chest wall resection with lung resection [en bloc resection]). Chemotherapy and radiotherapy use was ascertained using previously reported codes (21, 22). The outcome of the study, POM, was defined as death within 30 days of surgery. The date of surgery was identified from inpatient Medicare files, and the date of death was obtained from Medicare, which receives this information from the U.S. Social Security Administration. Statistical Analysis Differences in distribution of sociodemographic and tumor characteristics, comorbidities, and type of surgery between patients who died within 30 days of surgery and those that did not were evaluated using the Chi-square test. To validate the model published by Strand and colleagues (11), we fitted a logistic regression model using the same covariates included in the previous study. To refine the model, we evaluated whether the addition of some of the variables that were associated with 30-day POM in the univariate analysis, or the inclusion of individual comorbidities (instead of the summary score), improved the model s predictive ability. These models were compared in terms of discrimination using the area under the curve or the c statistic. The Akaike information criteria (AIC) and the Hosmer and Lemeshow test were used to assess the goodness of fit of the different models (23, 24). Once the final model was fitted, we divided patients into quintiles of increasing predicted probability of POM. The predicted and observed POM was compared for each quintile. All analyses were performed using the SAS statistical software program (SAS Institute Inc., Cary, NC). RESULTS Of the 14,297 patients in the study cohort, 72% had stage I, 14% had stage II, and 14% had stage IIIA NSCLC. Lobectomy was the most common procedure, comprising 74% of the surgeries. Overall, 655 patients (4.6%; 95% confidence interval [CI], %) died within 30 days of surgery. POM rates were 4.1% for limited resection, 3.7% for lobectomy, and 13.1% for pneumonectomy (P, 0.001). The baseline patient characteristics are summarized in Table 1. Greater age (P, 0.001) and male gender (P, 0.001) were significantly associated with increased 30-day POM. Additionally, patients with squamous cell carcinomas had higher rates of 30-day mortality compared with those with adenocarcinoma (P, 0.001). POM also increased with tumor size (P, 0.001) and stage (P, 0.001). Having multiple comorbid conditions increased 30-day POM, with 14% of those who died having a Charlson comorbidity score greater than 4, as opposed to 8% of the patients who did not experience POM events (P, 0.001). When evaluated as individual conditions, a history of acute myocardial infarction (P, 0.001) was the comorbidity with the strongest association for risk of postoperative death, although many other conditions affected surgical outcomes as well. Validation of POM Prediction Model As shown in Table 2, we obtained comparable results to those reported by Strand and colleagues (11) when we fitted a model using the same covariates. Overall, there was considerable agreement in terms of the significance of predictor variables, as well as the strength of association between each factor and the risk of POM. Similar to Strand and colleagues (11), our analysis showed that the odds of 30-day POM increased with older age (ages years: odds ratio [OR], 1.48; 95% CI, ; age.80 years: OR, 2.18; 95% CI, ) and men (OR, 1.45; 95% CI, ). Bilobectomy (OR, 2.16; 95% CI, ) and pneumonectomy (OR, 3.61; 95% CI, ) also significantly increased the mortality risk compared with lobectomy. However, our analysis also found a tumor size greater than 5 cm (OR, 1.31; 95% CI, ) to be a significant predictor, whereas Strand and colleagues did not (11). As expected, the odds of POM increased significantly when comorbidity scores rose (scores 3 4: OR, 2.2; 95% CI, ; scores.5: OR, 2.92; 95% CI, ). Refinement of the Predictive Model Once we confirmed the validity of the model, we assessed whether inclusion of other key predictors or a different categorization of some variables could improve the model s predictive performance. As shown in Figure 1, the initial validation of the model yielded a c statistic of and an AIC of 5,006. We found the highest c statistic (0.724) and lowest AIC value (4,955) with a model that included the original variables as well as individual comorbidities, instead of the summary comorbidity score. To simplify the model, we limited the comorbid conditions to those that were significantly associated with POM. This final model had a c statistic of and an AIC of 4,960. The Hosmer and Lemeshow test showed that this model had an adequate fit (P ). We then divided our cohort into quintiles, with the first quintile representing those with the smallest predicted probability of 30-day death, and the fifth quintile representing those patients that the model predicted would have the highest likelihood of POM. Figure 2 shows that as a predictive risk score increased from the first to the fifth quintile, the observed mortality increased from 1.2 to more than 10%. DISCUSSION Although surgical resection is the only means to obtain good longterm outcomes for patients with localized NSCLC, physicians often have concerns about perioperative deaths, particularly among elderly patients. In this study, we validated and refined a model to predict POM on a large cohort of elderly patients with stage I IIIA NSCLC. Our results show that this model can be used to stratify patients into groups with considerably different risks of POM. Identification of patients with low risk of POM may increase use of surgery within the elderly, whereas identification of those at high risk can help inform decision making and select candidates for targeted interventions to reduce POM. The decision to operate on elderly patients with lung cancer with multiple comorbidities or limited life expectancy can be difficult. Patients, clinicians, and surgeons are required to weigh the potential long-term benefits of surgery against the risk of perioperative death (25). Strand and colleagues (11) derived a clinical prediction rule to aide this decision-making process, using data from 4,395 patients (1,844 with available comorbidity data) from the Cancer Registry of Norway. In that study, the investigators showed that the model was able to identify a subset of patients at increased risk of 30-day POM. However, the derivation of a model is only the first step in the development of a clinical prediction rule. Before a predictive model can be used clinically, it must also go through external validation, as well as an impact analysis (13). Validation is essential to ensure that the original model is applicable to other patient populations. During the derivation process, it may occur that some of the predictive variables included in the model are unique to the study population, and therefore cannot be generalized to other settings. Thus, the clinical rule must be validated in independent populations with varying spectrum of disease and different demographic characteristics to demonstrate the reproducibility of the model s predictive ability (26). Finally, impact analysis is necessary to determine whether the prediction rule is easy to

3 392 AMERICAN JOURNAL OF RESPIRATORY AND CRITICAL CARE MEDICINE VOL TABLE 1. BASELINE CHARACTERISTICS AND 30-DAY PERIOPERATIVE OUTCOMES OF ELDERLY PATIENTS WITH STAGE I, II, AND IIIA NON SMALL CELL LUNG CANCER Characteristics Death within 30 D of Resection (n 5 655) No. (%) No Death within 30 D of Resection (n 5 13,642) No. (%) P Value Age, yr (25) 4,631 (34) (35) 4,628 (34) (26) 3,075 (22) (14) 1,295 (10) Male 431 (65) 7113 (52) Race 0.10 White 587 (90) 12,123 (89) African American 41 (6) 753 (5) Hispanic 9 (1) 132 (1) Other 18 (3) 628 (5) Marital status 0.92 Married 393 (60) 8151 (60) Not married 262 (40) 5478 (40) Income quartile* 0.02 First quartile 192 (29) 3,280 (24) Second quartile 170 (26) 3,571 (26) Third quartile 162 (25) 3,683 (27) Fourth quartile 130 (20) 2,080 (23) Histology Adenocarcinoma 250 (38) 6,154 (45) Bronchoalveolar carcinoma 52 (8) 1,507 (11) Squamous cell carcinoma 286 (44) 4,547 (33) Large-cell carcinoma 35 (5) 804 (6) Other 32 (5) 617 (5) Tumor size, cm < (42) 7,368 (54) (31) 3,777 (28) (21) 1,915 (14) Not reported 36 (6) 569 (4) Stage I 425 (65) 9,892 (73) II 101 (15) 1,888 (14) IIIA 129 (20) 1,849 (13) Side of resection Left 243 (37) 5,961 (44) Right 412 (63) 7,668 (56) Surgical procedure Sublobar resection 86 (13) 2,035 (15) Upper lobectomy 242 (37) 6,054 (43) Middle lobectomy (,1) 497 (4) Lower lobectomy 131 (20) 3,396 (25) Other lobectomy 11 (2) 255 (2) Bilobectomy 38 (6) 362 (3) Pneumonectomy 118 (18) 779 (6) En bloc resection 24 (4) 264 (2) Presurgical radiation therapy (,1) 115 (,1) 0.85 Presurgical chemotherapy 23 (3) 518 (4) 0.69 Charlson Comorbidity Index score (20) 4,057 (29) (33) 5,427 (40) (33) 3,072 (23).4 98 (14) 1073 (8) Individual comorbidities Acute myocardial infarction 63 (10) 280 (2) Old myocardial infarction 34 (5) 858 (6) 0.25 Congestive heart failure 119 (18) 1,329 (10) Cerebrovascular disease 94 (14) 1118 (8) Peripheral vascular disease 77 (12) 1,156 (8) 0.01 COPD 394 (60) 7,508 (55) 0.01 Moderate/severe renal disease 23 (4) 222 (2) Diabetes 117 (18) 1,895 (14) 0.01 Diabetes with complications 36 (6) 381 (3) Rheumatologic disease 25 (4) 425 (3) 0.32 Paralysis 13 (2) 124 (1) 0.01 Dementia (,1) 78 (1) 0.53 Mild ulcer disease 19 (3) 327 (2) 0.41 Severe ulcer disease 16 (2) 118 (1) Mild liver disease (,1) 72 (1) 0.42 Moderate/severe liver disease (,1) 16 (,1) 0.19 Definition of abbreviation: COPD 5 chronic obstructive pulmonary disease. * First income quartile includes patients with the lowest estimated income. Exact number of patients is not reported to maintain patient confidentiality.

4 Kates, Perez, Gribetz, et al.: Predicting Mortality from Lung Surgery 393 TABLE 2. VALIDATION AND REFINEMENT OF MODEL TO PREDICT 30-DAY PERIOPERATIVE MORTALITY SEER Medicare Data (n 5 14,297) Strand and colleagues (n 5 1,844) Validated Model Refined Model Variable OR (95% CI) OR (95% CI) OR (95% CI) Sex Female Reference ( ) ( ) ( ) Male 1.84 ( ) 1.45 ( ) 1.45 ( ) Age,50 Reference ( ) ( ) ( ) ( ) ( ) ( ) ( ) 1.38 ( ) ( ) ( ) 1.59 ( ) ( ) 1.48 ( ) ( ) ( ) 2.18 ( ) 2.09 ( ) Side of resection Left Reference ( ) ( ) ( ) Right 1.91 ( ) 1.48 ( ) 1.48 ( ) Surgical procedure Sublobar resection 0.71 ( ) 1.01 ( ) 1.03 ( ) Upper lobectomy Reference ( ) ( ) ( ) Middle lobectomy 1.78 ( ) 0.27 ( ) 0.27 ( ) Lower lobectomy 3.05 ( ) 0.93 ( ) 0.93 ( ) Bilobectomy 4.68 ( ) 2.16 ( ) 2.18 ( ) Pneumonectomy 6.49 ( ) 3.61 ( ) 3.55 ( ) En bloc resection ( ) 1.83 ( ) 1.84 ( ) Histopathology type Adenocarcinoma Reference ( ) ( ) ( ) Squamous cell 1.34 ( ) 1.20 ( ) 1.23 ( ) Large cell ( ) ( ) 0.94 ( ) Bronchioloalveolar ( ) ( ) 0.89 ( ) Other 2.52 ( ) 1.04 ( ) 1.18 ( ) Stage I Reference ( ) ( ) ( ) II 1.36 ( ) 1.01 ( ) 1.00 ( ) III* 1.66 ( ) 1.33 ( ) 1.27 ( ) Tumor size, cm,3 Reference ( ) ( ) ( ) ( ) 1.21 ( ) 1.22 ( ) ( ) 1.31 ( ) 1.29 ( ) Unknown ( ) 1.44 ( ) 1.48 ( ) Comorbidity score 0 Reference ( ) ( ) ( ) ( ) 1.26 ( ) ( ) ( ) 2.20 ( ) ( ) ( ) 2.92 ( ) ( ) Surgical approach VATS Reference ( ) ( ) ( ) Open thoracotomy 0.77 ( ) ( ) ( ) Hospital volume ( ),20 Reference ( ) ( ) ( ) ( ) ( ) ( ) Individual comorbidities Congestive heart failure ( ) Cerebrovascular disease ( ) ( ) 1.75 ( ) Diabetes with complications ( ) ( ) 1.70 ( ) Acute myocardial infarction ( ) ( ) 4.03 ( ) Severe ulcer ( ) ( ) 2.9 ( ) Definition of abbreviations: CI5 confidence interval; OR 5 odds ratio. SEER 5 Surveillance, Epidemiology, and End-Results Registry; VATS 5 video-assisted thoracic surgery. * The reference group in SEER Medicare is years of age. The refined model divided age group into and years. For the refined model, data for age years is shown. Additionally, the analysis using SEER Medicare data was limited to patients with stage IIIA disease. Hospital volume is measured in terms of patients/year. The data from the model by Strand and colleagues (11) is limited to the 1,844 patients that have comorbidity data available (which were from years ). use in routine care and to assess if its use is beneficial to patient care (27). In this study, we showed that the model was able to predict the risk of POM among a large cohort of elderly patients with NSCLC a population at increased risk of surgical complications. Thus, our results strongly support the generalizability of the prediction model. The overall 30-day mortality of 4.6% in our study was similar to the 5.2% rate reported by the American College of Sur-

5 394 AMERICAN JOURNAL OF RESPIRATORY AND CRITICAL CARE MEDICINE VOL Figure 1. Refinement of the predictive model. Each logistic model was compared using the Akaiki information criteria (AIC) and the c statistic. A low AIC indicates better goodness of fit, whereas a higher c statistic denotes improved discrimination of a model. The validation model used the same risk factors as the initial model by Strand and colleagues. Histology and age were then redefined to include large-cell carcinoma and bronchoalveolar carcinoma as separate histologic groups, and ages and years as individual age groups. The subsequent model used all individual comorbidities instead of the composite comorbidity score. Only five comorbidities were statistically significant predictors, and, thus, were included in the final refined model. geons in a national hospital survey (28). Similarly, a study by Memtsoudis and colleagues (29), using data from the National Hospital Discharge Survey, found a POM of 4.8%. As previously reported, POM varied considerably according to the type of procedure 3.6% after lobectomy compared with 13.1% after pneumonectomy. When compared with studies including older patients, Strand and colleagues (11) reported POM of 4.1% for lobectomy/limited resection, and 13.7% for pneumonectomy among patients older than 70 years of age. A study using a smaller cohort of patients from a single center found a POM rate of 5.0% for lobectomy and 8.0% for pneunectomy among octogenarians (30). There are several potential uses of this model. First, the predictive model can help inform the discussion between elderly patients and clinicians about treatment options for potentially resectable lung cancers. Additionally, surgery is underused in the elderly, with concerns of morbidity and mortality trumping the potential benefits of curative resection (4, 6, 9, 31). A clinical prediction rule can help identify low-risk patients, and thus reassure physicians that surgery is appropriate. Thus, the implementation of this rule could lead to increases in surgery rates among the elderly by illuminating the low probability of a fatal complication. This model could also identify patients at the highest risk of POM. Because lung resection is the only definitively curative treatment, it may still be the best option, even among the highest-risk patients (with POM. 10%). However, the identification of these patients may be useful for ordering additional preoperative tests, providing close postoperative monitoring, or implementing targeted interventions to reduce their POM. Lastly, the predictive model can be used for risk adjustment to compare surgical outcomes among different institutions or for public reporting. This model is particularly well suited for such institutional comparisons, because it makes use of administrative data that should be easily available to hospital administrators. There are some strengths and limitations of this study that should be mentioned. As SEER Medicare is a population-based cancer registry, the generalizability of the model should be strong. Because our study was limited to patients older than 65 years of age, this clinical rule still must be validated in a younger population. However, the elderly are at higher risk for POM and, as a result, there is a greater need for a clinical rule for these patients. Additionally, both the initial study by Strand and colleagues and the SEER Medicare registry lacked data on pulmonary function tests and other pulmonary performance measures; thus, we are not able to asses whether their inclusion would improve the accuracy of the rule. Information regarding hospital volume and surgical technique (video-assisted thoracic surgery versus open thoracotomy) were not available in SEER Medicare, and, thus, are not included in our model. However, these variables were not significant predictors of 30-day POM in the model published by Strand and colleagues, and their exclusion should not have affected the model substantially. In conclusion, risk assessment is critical for physicians to make decisions about the best therapy for their lung cancer patients, and particularly for the elderly, who are at high risk for POM. In this study, we validated and refined a previously derived prediction rule, showing that it can be used to stratify these patients

6 Kates, Perez, Gribetz, et al.: Predicting Mortality from Lung Surgery 395 Figure 2. Predicted versus observed rates of perioperative mortality (POM). Using the final refined model, patients were divided into predicted risk quintiles, with quintile 1 representing the lowest-risk group. The predicted and observed rates of 30-day POM was assessed for each quintile. according to their risk of 30-day POM. An impact analysis is necessary to assess if routine use of the rule in clinical practice improves the management and outcomes of elderly patients with potentially resectable NSCLC. Conflict of Interest Statement: None of the authors has a financial relationship with a commercial entity that has an interest in the subject of this manuscript. Acknowledgment: The authors thank Wen Liu for outstanding data management, and acknowledge the efforts of the following: Applied Research Branch, Division of Cancer Prevention and Population Science, National Cancer Institute; the Office of Information Services, and the Office of Strategic Planning, Health Care Finance Administration; Information Management Services, Inc.; and the Surveillance, Epidemiology, and End Results (SEER) Program tumor registries in the creation of the SEER Medicare Database. The interpretation and reporting of these data are the sole responsibilities of the authors. References 1. Ries LAG, Melbert D, Krapcho M, Stinchcomb DG, Howlader N, Horner MJ, Mariotto A, Miller BA, Feuer EJ, Altekruse SF, et al., editors. SEER cancer statistics review, [Internet]. Bethesda, MD: National Cancer Institute; 2007 [accessed 2009 Jan 8]. Available from: 2. Wingo PA, Cardinez CJ, Landis SH, Greenlee RT, Ries LA, Anderson RN, Thun MJ. Long-term trends in cancer mortality in the United States, Cancer 2003;97: Strand TE, Rostad H, Moller B, Norstein J. Survival after resection for primary lung cancer: a population based study of 3211 resected patients. Thorax 2006;61: Mery CM, Pappas AN, Bueno R, Colson YL, Linden P, Sugarbaker DJ, Jaklitsch MT. Similar long-term survival of elderly patients with nonsmall cell lung cancer treated with lobectomy or wedge resection within the surveillance, epidemiology, and end results database. Chest 2005;128: Schneider T, Pfannschmidt J, Muley T, Reimer P, Eberhardt R, Herth FJ, Dienemann H, Hoffmann H. A retrospective analysis of short and long-term survival after curative pulmonary resection for lung cancer in elderly patients. Lung Cancer 2008;62: Janssen-Heijnen ML, Smulders S, Lemmens VE, Smeenk FW, van Geffen HJ, Coebergh JW. Effect of comorbidity on the treatment and prognosis of elderly patients with non small cell lung cancer. Thorax 2004;59: Beshay M, Dorn P, Ris HB, Schmid RA. Influence of comorbidity on outcome after pulmonary resection in the elderly. Asian Cardiovasc Thorac Ann 2007;15: Pennathur A, Abbas G, Christie N, Landreneau R, Luketich JD. Video assisted thoracoscopic surgery and lobectomy, sublobar resection, radiofrequency ablation, and stereotactic radiosurgery: advances and controversies in the management of early stage non small cell lung cancer. Curr Opin Pulm Med 2007;13: Heerdt PM, Park BJ. The emerging role of minimally invasive surgical techniques for the treatment of lung malignancy in the elderly. Anesthesiol Clin 2008;26: Matsuoka H, Okada M, Sakamoto T, Tsubota N. Complications and outcomes after pulmonary resection for cancer in patients 80 to 89 years of age. Eur J Cardiothorac Surg 2005;28: Strand TE, Rostad H, Damhuis RA, Norstein J. Risk factors for 30-day mortality after resection of lung cancer and prediction of their magnitude. Thorax 2007;62: Hall BL, Hirbe M, Waterman B, Boslaugh S, Dunagan WC. Comparison of mortality risk adjustment using a clinical data algorithm (American College of Surgeons National Surgical Quality Improvement Program) and an administrative data algorithm (Solucient) at the case level within a single institution. J Am Coll Surg 2007;205: McGinn TG, Guyatt GH, Wyer PC, Naylor CD, Stiell IG, Richardson WS. Users guides to the medical literature: XXII. How to use articles about clinical decision rules. Evidence-Based Medicine Working Group. JAMA 2000;284: Warren JL, Klabunde CN, Schrag D, Bach PB, Riley GF. Overview of the SEER Medicare data: content, research applications, and generalizability to the United States elderly population. Med Care 2002;40: IV-3 IV Cooper GS, Virnig B, Klabunde CN, Schussler N, Freeman J, Warren JL. Use of SEER Medicare data for measuring cancer surgery. Med Care 2002;40:IV-43 IV Deyo RA, Cherkin DC, Ciol MA. Adapting a clinical comorbidity index for use with ICD-9-CM administrative databases. J Clin Epidemiol 1992;45: Klabunde CN, Potosky AL, Legler JM, Warren JL. Development of a comorbidity index using physician claims data. J Clin Epidemiol 2000;53: Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis 1987;40: Mountain CF. Revisions in the International System for Staging Lung Cancer. Chest 1997;111: Ravdin PM, Davis G. Prognosis of patients with resected non small cell lung cancer: impact of clinical and pathologic variables. Lung Cancer 2006;52: Warren JL, Harlan LC, Fahey A, Virnig BA, Freeman JL, Klabunde CN, Cooper GS, Knopf KB. Utility of the SEER Medicare data to identify chemotherapy use. Med Care 2002;40:IV-55 IV Virnig BA, Warren JL, Cooper GS, Klabunde CN, Schussler N, Freeman J. Studying radiation therapy using SEER Medicare linked data. Med Care 2002;40:IV-49 IV Baldwin LM, Klabunde CN, Green P, Barlow W, Wright G. In search of the perfect comorbidity measure for use with administrative claims data: does it exist? Med Care 2006;44: Hosmer DW, Hosmer T, Le Cessie S, Lemeshow S. A comparison of goodness-of-fit tests for the logistic regression model. Stat Med 1997; 16: Little AG. Risk and benefit: the eternal yin and yang of thoracic surgery. Thorax 2007;62: Bleeker SE, Moll HA, Steyerberg EW, Donders AR, Derksen-Lubsen G, Grobbee DE, Moons KG. External validation is necessary in prediction research: a clinical example. J Clin Epidemiol 2003;56: Reilly BM, Evans AT. Translating clinical research into clinical practice: impact of using prediction rules to make decisions. Ann Intern Med 2006;144: Little AG, Rusch VW, Bonner JA, Gaspar LE, Green MR, Webb WR, Stewart AK. Patterns of surgical care of lung cancer patients. Ann Thorac Surg 2005;80: Memtsoudis SG, Besculides MC, Zellos L, Patil N, Rogers SO. Trends in lung surgery: United States 1988 to Chest 2006;130: Dominguez-Ventura A, Allen MS, Cassivi SD, Nichols FC III, Deschamps C, Pairolero PC. Lung cancer in octogenarians: factors affecting morbidity and mortality after pulmonary resection. Ann Thorac Surg 2006;82: Farjah F, Wood DE, Yanez D III, Symons RG, Krishnadasan B, Flum DR. Temporal trends in the management of potentially resectable lung cancer. Ann Thorac Surg 2008;85: (discussion 1856).