Conditional Cancer-Specific Versus Cardiovascular-Specific Survival After Lobectomy for Stage I Non-Small Cell Lung Cancer

Transcription

1 Conditional Cancer-Specific Versus Cardiovascular-Specific Survival After Lobectomy for Stage I Non-Small Cell Lung Cancer Shawn S. Groth, MD, MS, Natasha M. Rueth, MD, James S. Hodges, PhD, Elizabeth B. Habermann, PhD, MPH, Rafael S. Andrade, MD, Jonathan D Cunha, MD, PhD, and Michael A. Maddaus, MD Divisions of General Thoracic and Foregut Surgery and Basic and Translational Research, Department of Surgery, University of Minnesota, and Division of Biostatistics, University of Minnesota School of Public Health, Minneapolis, Minnesota GENERAL THORACIC Background. Conditional cancer-specific survival rates account for changes in the risk of death from cancer over time. There may be a point during follow-up when patients who undergo lobectomy for stage I non-small cell lung cancer (NSCLC) are as likely to die of cardiovascular disease as of cancer. Methods. Using the Surveillance Epidemiology and End Results Database (1988 through 2005), we identified patients 50 years old and older who underwent lobectomy for stage I NSCLC. We used competing risks methods to calculate conditional survival rates and to ascertain if there is a point in follow-up where the risk of dying of cancer is equivalent to the risk of dying of cardiovascular disease. Results. In all, 22,518 patients met our inclusion criteria. The difference in the 5-year conditional probability of dying of cancer and the 5-year conditional probability of dying of cardiovascular disease decreased with time; in the whole cohort, these probabilities were equivalent if patients survived to 7 years after lobectomy (p 0.11). With increasing age, the probability of dying of cancer and the probability of dying of cardiovascular disease became equivalent at earlier time points. Furthermore, the 5-year probability of dying of cardiovascular disease was significantly greater than the 5-year probability of dying of cancer for patients aged 70 to 79 years who survived to 7 years and for patients aged 80 years and older who survived to 5 years after lobectomy. Conclusions. For patients undergoing lobectomy for stage I NSCLC, cardiovascular-specific mortality becomes increasingly important over the course of followup, especially among elderly patients. (Ann Thorac Surg 2010;90:375 82) 2010 by The Society of Thoracic Surgeons Patients who undergo lobectomy for stage I non-small cell lung cancer (NSCLC) have a 5-year cumulative survival rate of 55% to 90% [1-5]. Although conventional cumulative survival rates (such as these), which are calculated from the time of diagnosis or the initiation of treatment, are often used to provide prognostic information and to measure treatment effects, they fail to account for changes in the risk of death over the course of follow-up. Specifically, the risk of cancer-specific mortality is highest in the first years after cancer diagnosis and subsequently diminishes with time. To illustrate this point, as shown in a cancer-specific survival curve for patients with early stage (I or II) NSCLC based on data from a large population-based cancer dataset (Fig 1), the risk of death (the slope of the curve) is not constant it changes throughout the course Accepted for publication April 23, Presented at the Forty-sixth Annual Meeting of The Society of Thoracic Surgeons, Fort Lauderdale, FL, Jan 25 27, Winner of the Geriatric Patient Care Award. Address correspondence to Dr Maddaus, Department of Surgery, University of Minnesota, Division of General Thoracic and Foregut Surgery, MMC 207, 420 Delaware St SE, Minneapolis, MN 55455; madda001@umn.edu. of follow-up (Fig 1A). For example, the risk of dying of cancer in the subsequent 5 years is much greater for patients at the time of diagnosis than for patients who survive at least 3 years (Fig 1B). Conditional survival rates (survival predicated on living to a certain point in follow-up) account for this change in risk over time and, therefore, may be more informative than cumulative survival rates [6]. Other investigators have assessed conditional survival rates for a number of malignancies, including lung cancer [6, 7], colon cancer [8], breast cancer [9], brain cancer [10], melanoma [11], and prostate cancer [12]. Based on a number of sources in the literature, we know that most NSCLC patients are older and often have cardiovascular comorbidities. A recent study using The Society of Thoracic Surgeons (STS) database estimated that the median age of patients with primary lung cancer is 67 years. That study also found a relatively high prevalence of cardiovascular comorbidities among lung cancer patients: hypertension (66%), coronary artery disease (26%), peripheral vascular disease (11%), cerebrovascular disease (8%), and congestive heart failure (4%) [13]. Because the risk of cancer-specific mortality decreases over time, patients who undergo lobectomy for 2010 by The Society of Thoracic Surgeons /$36.00 Published by Elsevier Inc doi: /j.athoracsur

2 GENERAL THORACIC 376 GROTH ET AL Ann Thorac Surg CONDITIONAL SURVIVAL AFTER LOBECTOMY FOR NSCLC 2010;90: Data The Surveillance, Epidemiology, and End Results (SEER) database is a set of population-based cancer registries that was founded in 1973 by the National Cancer Institute (Bethesda, MD). Currently, 17 US cancer registries, selected to include a diverse sample (about 26%) of the national population, participate in the SEER program. We used the SEER database that was based on the November 2007 submission, which provides data through December 31, 2005 [14]. The SEER registries began collecting American Joint Committee on Cancer (AJCC) staging data in 1988 [15]; therefore, this study represents data collected by the SEER registries from 1988 through Because of the negative impact of Hurricane Katrina on data collection by the Louisiana SEER registry, we excluded information obtained by that registry in We collected information on patient characteristics, primary tumor characteristics, and treatment regimens. Fig 1. Cancer-specific survival curve (solid line) for patients with early stage (I and II) non-small cell lung cancer. (A) Representative (dotted) tangent lines demonstrate that the risk of death (slope of the curve) changes throughout the course of follow-up. (B) For example, the risk of dying of cancer in the subsequent 5 years is greater for patients at the time of cancer diagnosis ( 1 ) than for patients 3 years after diagnosis ( 2 ). Data from the Surveillance, Epidemiology, and End Results Database ( ). stage I NSCLC and who survive to a certain point in follow-up may be as likely to die from cardiovascular disease as from cancer. Using a large population-based cancer dataset, we sought to (1) compare conditional survival rates for patients who underwent lobectomy for stage I NSCLC with their survival rate at the time of diagnosis, and (2) compare conditional cancer-specific survival rates with conditional cardiovascular-specific survival rates using a competing risks model to ascertain if there is a point in follow-up when patients are as likely to die from cardiovascular disease as they are from cancer. Material and Methods The Human Subjects Committee of the University of Minnesota determined that this study was exempt from formal review by the Institutional Review Board. Inclusion Criteria We selected patients from the SEER database for inclusion in our study using the following International Classification Disease for Oncology, Third Edition (ICD-O-3), topography codes: C34.0 to C34.3, C34.8, and C34.9 [16]. We only included patients 50 years and older who underwent lobectomy (the standard of care) [2] for AJCC (sixth edition) stage I (T1N0M0 and T2N0M0) NSCLC [15]. We limited our study to patients with stage I disease to create a homogeneous study group and to eliminate the confounding influence of different extents of resection, different stages of disease, and use of chemotherapy (which until recently SEER did not collect) and radiation therapy on survival. The following carcinoma histologic subtypes were included in our analysis and were categorized as follows: adenocarcinomas, bronchoalveolar carcinomas, squamous cell carcinomas, and other NSCLC histologies (large cell, epithelial, undifferentiated, anaplastic, pleomorphic, giant cell, mixed cell, clear cell, mucoepidermoid, signet cell, and adenosquamous cell carcinomas; cystadenocarcinomas, and NSCLC not otherwise specified [see Appendix for ICD-O-3 morphology codes]). Exclusion Criteria We excluded patients with a prior or subsequent cancer diagnosis. Due to the potential for confounding, we also excluded those patients who were unlikely to have received aggressive cancer treatment based on the following reporting sources: a hospice/nursing home, autopsy, or death certificate. Statistical Analysis Data were extracted from the SEER database using SAS version 9.1 (SAS Institute, Cary, NC). The Kaplan-Meier method was used to calculate conditional cancer-specific survival rates. We then used the empirical transition matrix (etm) package, version 0.3-7, in the R system, version (R Foundation for Statistical Computing, Vienna, Austria) to perform competing risks analyses. At

3 Ann Thorac Surg GROTH ET AL 2010;90: CONDITIONAL SURVIVAL AFTER LOBECTOMY FOR NSCLC each point in follow-up, the probability of surviving, dying from cancer, dying from cardiovascular disease classified in SEER as follows: heart disease, hypertension, cerebrovascular disease, atherosclerosis, aortic aneurysms and dissection, and other arterial diseases), and dying from other causes was estimated using the Aalen- Johansen estimator with covariance estimated using Greenwood s method [17]. We used Wald s method to compute p values for the difference in the probability of cancer-specific mortality within 5 years and the probability of cardiovascular disease-specific mortality within 5 years conditioned on survival to 0 to 11 years after lobectomy. We had insufficient follow-up to perform a competing risk analyses beyond 11 years after lobectomy. We adjusted for multiple comparisons by using a more conservative two-sided significance level of 0.01, rather than the traditional level of Because no available statistical software performs a competing-risks regression analysis while comparing the two conditional probabilities of central interest, we qualitatively assessed the effects of patient characteristics (age, race, and sex) and tumor characteristics (tumor [T] stage and histology) by dividing the population into subgroups and then analyzing those groups separately. 377 GENERAL THORACIC Results Of the 496,458 patients in the SEER database with invasive lung cancer, 22,518 met our inclusion criteria (Fig 2). Most of the patients in our study were between the ages of 60 and 79 years (72.8%), and most were Caucasian (86.5%). Most had T1 tumors (62%); adenocarcinoma was the most common histology (44.6%; Table 1). The 30-day mortality rate was 1.5%. Our results confirmed our hypothesis that 5-year cancer-specific conditional survival rates improved with time (Fig 3). This effect was most profound in the first 5 years after lobectomy. Our competing risks analysis demonstrated that the difference in the conditional probability of dying of cancer and the conditional probability of dying of cardiovascular disease decreased with time (Fig 4). At 7 years after lobectomy, this difference was no longer statistically significant (p 0.11; Fig 4E). We also assessed the effect of several patient and tumor characteristics on our results (Fig 5). Age had the largest impact on our findings (Fig 5A). With increasing age, the conditional probability of dying of cancer and the conditional probability of dying of cardiovascular disease became equivalent at earlier time points after lobectomy. Furthermore, the conditional probability of dying of cardiovascular disease was significantly greater than the conditional probability of dying of cancer for patients aged 70 to 79 years old who survived 7 years and for patients aged 80 years and older who survived 5 years after lobectomy. The difference in the conditional probability of dying of cardiovascular disease and the conditional probability of dying of cancer became statistically equivalent at slightly earlier time points for non-caucasians than for Caucasians (Fig 5B). It occurred slightly earlier in men than women (Fig 5C). It also occurred earlier in patients with squamous cell carcinomas and other NSCLC histologies than in patients Fig 2. Patients included in and excluded from the study. (NSCLC non-small cell lung cancer; SEER Surveillance, Epidemiology, and End Results.) with adenocarcinomas or BAC (Fig 5D). Interestingly, even though patients with T2 tumors had a higher risk of cancerspecific mortality immediately after lobectomy, the difference in the conditional probability of dying of cardiovascular disease and the conditional probability of dying of cancer became equivalent at similar time points (Fig 5E). For patients with T1 tumors, the probability of dying of cardiovascular disease exceeded the probability of dying of cancer for patients who survived 10 years after lobectomy. Comment Using a large-population based cancer dataset, we demonstrated that the risk of cancer-specific mortality after lobectomy for stage I NSCLC is not constant; it improves throughout the course of follow-up. Unlike conventional cumulative survival rates, conditional survival rates account for this change in risk. Therefore, they provide more useful information when counseling patients preoperatively and postoperative about the risk of death from cancer as compared with other causes of death over the course of follow-up. In particular, we demonstrated that the improvement in conditional 5-year cancerspecific survival rates after lobectomy for stage I NSCLC is most profound in the first 5 to 7 years after lobectomy, which has important implications for posttreatment surveillance and NSCLC survivorship. Current National Comprehensive Cancer Network (NCCN) guidelines for postoperative surveillance for stage

4 GENERAL THORACIC 378 GROTH ET AL Ann Thorac Surg CONDITIONAL SURVIVAL AFTER LOBECTOMY FOR NSCLC 2010;90: Table 1. Patient Characteristics a n (%) Age, years 50 to 59 4,208 (18.7) 60 to 69 8,083 (35.9) 70 to 79 8,316 (36.9) 80 1,910 (8.5) Race White 19,446 (86.5) Black 1,626 (7.3) Other 1,398 (6.2) Sex Male 11,540 (51.2) Female 10,978 (48.8) pt stage T1 13,858 (62.0) T2 8,502 (38.0) Histology Adenocarcinoma 10,037 (44.6) BAC 3,195 (14.2) Squamous cell carcinomas 6,730 (29.9) Other NSCLC 2,556 (11.3) a Patient subgroups may not add up to 22,528 (the size of the total cohort) owing to missing data. BAC bronchoalveolar carcinoma; NSCLC non-small cell lung cancer; pt pathologic tumor stage. I NSCLC patients recommend chest computed tomography scans every 4 to 6 months for the first 2 years after lobectomy and annually thereafter [18]. Although the primary utility of posttreatment radiographic surveillance is to assess for locoregional recurrence, SEER registries do not collect information on recurrence, and, therefore, we cannot provide data to support (or refute) this recommendation. However, we did show that the conditional probability of cancer-specific mortality (which is temporally correlated with disease recurrence) was highest in the first 5 to 7 years after lobectomy, suggesting the need to focus the intensity of surveillance early in the course of follow-up [19]. Because the risk of cancer-specific mortality decreases over time, we hypothesized that cardiovascular disease would become an increasingly important cause of mortality over the course of follow-up. Indeed, although the risk of dying of cancer exceeds the risk of dying of cardiovascular disease immediately after lobectomy, the disparity between these risks diminishes with time. For patients who survive 7 years after lobectomy for stage I NSCLC, the risk of dying of cancer is not statistically different than the risk of dying of cardiovascular disease. These findings have important implications for NSCLC survivors. Because cardiovascular-specific mortality becomes an increasingly important cause of mortality over the course of follow-up, health promotion and management of medical comorbidities should be incorporated into the care of stage I NSCLC patients after lobectomy. We divided patients into subgroups to qualitatively assess the effect of potentially important prognostic factors on our findings. Age had the most profound impact on our results; with increasing age, the risk of dying of cancer becomes equivalent to the risk of dying of cardiovascular disease at earlier time points after lobectomy. Because the risk of both cancer-specific and cardiovascular disease-specific mortality increases with age, this finding is not surprising [5, 6]. Similarly, in other subgroups with a higher risk of death (non-caucasians, men, T2 tumors, squamous cell carcinomas and other NSCLC histologies [which includes large cell and undifferentiated NSCLC]), the risk of cancer-specific mortality and cardiovascular diseasespecific mortality converged earlier in the course of follow-up than in lower risk subgroups [5]. Although our subgroup analysis is largely qualitative, each of its findings is expected (ie, mortality rates are higher for older patients as compared with younger patients). Had there been large discrepancies in the observed and expected outcomes, our competing risks analysis would be brought into question. We recognize a number of limitations of our study, some of which are inherent limitations of using SEER data. In particular, the SEER cancer registries do not collect information on several important prognostic factors, such as performance status, comorbidities, time to local recurrence, and surgeon and hospital volume. Therefore, we could not determine the effects of these Fig 3. Five-year cancer-specific survival rates, with error bars indicating 95% confidence intervals (vertical axis) conditioned on surviving 0 to 10 years after lobectomy (horizontal axis).

5 Ann Thorac Surg GROTH ET AL 2010;90: CONDITIONAL SURVIVAL AFTER LOBECTOMY FOR NSCLC 379 GENERAL THORACIC Fig 4. Competing risks curves for the probability of the following outcomes: surviving (solid line), cancer-specific mortality (dotted line), cardiovascular disease-specific mortality (dashed line), and death from other causes (dotted and dashed line) at (A) the time of diagnosis and conditioned on living (B) 1 year, (C) 3 years, (D) 5 years, and (E) 7 years after lobectomy. The p values are for tests comparing the conditional probabilities of cancer-specific and cardiovascular disease-specific mortality within 5 years.

6 GENERAL THORACIC 380 GROTH ET AL Ann Thorac Surg CONDITIONAL SURVIVAL AFTER LOBECTOMY FOR NSCLC 2010;90: Fig 5. Analysis demonstrating the year at which the probability (P) of dying of cancer is not significantly different (p 0.01) from the probability of dying of cardiovascular (CV) disease (vertical axis) for each of the following subgroups: (A) age, (B) race, (C) sex, (D) pathologic tumor (T) stage, and (E) histology. (Adeno adenocarcinoma; BAC bronchoalveolar carcinoma; squam squamous cell carcinoma.) covariates on our findings. Our study is also prone to the inherent potential sources of error of observational studies. Of these, selection bias may be especially important. In particular, because we only included patients who underwent lobectomy, we excluded those patients who lacked sufficient cardiopulmonary reserve to tolerate a lobectomy. Consequently, though our data indicate that the risk of dying of cancer exceeds the risk of dying of cardiovascular disease (regardless of age) at the time of diagnosis, our competing risks analysis cannot be used as a widely generalizable indication for lobectomy. Each patient must be evaluated to ascertain whether or not he or she has sufficient cardiopulmonary and functional reserve to tolerate a lobectomy. In addition, we limited our analysis to patients with stage I disease who underwent standard of care treatment (eg, lobectomy) to create a homogenous group and eliminate the confounding influence of different extents of resection, different stages of disease, and use radiation and chemotherapy (which until recently SEER registries did not record). Consequently, our results may not be generalizable to NSCLC patients who undergo lesser resections or who have more advanced stages of disease. Because our data encompassed a relatively long period of time (17 years), it is possible that improvements in overall and cancer-specific survival over this time period could confound our results. However, although the life expectancy has increased [20] and the overall mortality rate for lung cancer has declined in the United States [21] over the period of our study, the mortality rate for stage I

7 Ann Thorac Surg GROTH ET AL 2010;90: CONDITIONAL SURVIVAL AFTER LOBECTOMY FOR NSCLC NSCLC has not changed [5]. Finally, our competing risks model does not allow us to perform a regression analysis to simultaneously adjust our results for a number of potentially confounding covariates. Nonetheless, we were able to evaluate the effect of several patient and tumor characteristics using another well-accepted method of assessing for confounding stratification. Despite these limitations, we also recognize a number of strengths of our study. We had a large sample size (22,518 patients) with relatively long period of follow-up (as long as 17 years), which allowed us to conduct a survival analysis that would not be possible with many other sources of data. Because we used population-based data, our results may be more generalizable to the US population than data from tertiary cancer centers. Furthermore, this study is the first in the literature to assess conditional survival rates specifically for patients who underwent lobectomy (the standard of care) for stage I NSCLC. Previous studies included patients with more advanced stages of disease [7, 22], included other histologies (ie, small cell carcinoma [6, 7, 22]), and did not adjust for standard of care treatment for the patients that were included [7, 12, 22]. Although Skuladottir and Olsen [6] examined conditional survival rates for stage I NSCLC patients, they did not adjust for extent of resection. This study is also the first that used competing risks analyses to compare the conditional probability of cancer-specific mortality with the conditional probability of cardiovascular disease-specific mortality (or other chronic diseases). In conclusion, using a large population-based cancer dataset, we found that the risk of dying of cancer exceeds the risk of dying of cardiovascular disease at the time of lobectomy for stage I NSCLC. However, the difference in the conditional probabilities of these outcomes changes over the course of follow-up. While the risk of dying of cancer decreases over time, the risk of dying of cardiovascular disease becomes an increasingly important cause of death, especially for elderly patients. These findings have potentially important implications for counseling NSCLC patients, posttreatment surveillance, and for risk factor modification for cancer survivors. References 1. Koike T, Yamato Y, Yoshiya K, Shimoyama T, Suzuki R. Intentional limited pulmonary resection for peripheral T1 N0 M0 small-sized lung cancer. J Thorac Cardiovasc Surg 2003;125: Ginsberg RJ, Rubinstein LV. Randomized trial of lobectomy versus limited resection for T1 N0 non-small cell lung cancer. Lung Cancer Study Group. Ann Thorac Surg 1995; 60: Whitson BA, Groth SS, Duval SJ, Swanson SJ, Maddaus MA. Surgery for early-stage non-small cell lung cancer: a systematic review of the video-assisted thoracoscopic surgery versus thoracotomy approaches to lobectomy. Ann Thorac Surg 2008;86: Mountain CF. Revisions in the International System for Staging Lung Cancer. Chest 1997;111: Ou SH, Zell JA, Ziogas A, Anton-Culver H. Prognostic factors for survival of stage I nonsmall cell lung cancer patients: a population-based analysis of 19,702 stage I patients in the California Cancer Registry from 1989 to Cancer 2007;110: Skuladottir H, Olsen JH. Conditional survival of patients with the four major histologic subgroups of lung cancer in Denmark. J Clin Oncol 2003;21: Merrill RM, Henson DE, Barnes M. Conditional survival among patients with carcinoma of the lung. Chest 1999;116: Merrill RM, Henson DE, Ries LA. Conditional survival estimates in 34,963 patients with invasive carcinoma of the colon. Dis Colon Rectum 1998;41: Henson DE, Ries LA, Carriaga MT. Conditional survival of 56,268 patients with breast cancer. Cancer 1995;76: Davis FG, McCarthy BJ, Freels S, Kupelian V, Bondy ML. The conditional probability of survival of patients with primary malignant brain tumors: Surveillance, Epidemiology, and End Results (SEER) data. Cancer 1999;85: Janssen-Heijnen ML, Houterman S, Lemmens VE, Brenner H, Steyerberg EW, Coebergh JW. Prognosis for long-term survivors of cancer. Ann Oncol 2007;18: Wingo PA, Ries LA, Parker SL, Heath CW. Long-term cancer patient survival in the United States. Cancer Epidemiol Biomark Prev 1998;7: Boffa DJ, Allen MS, Grab JD, Gaissert HA, Harpole DH, Wright CD. Data from The Society of Thoracic Surgeons general thoracic surgery database: the surgical management of primary lung tumors. J Thorac Cardiovasc Surg 2008;135: Graham EA, Singer JJ. Successful removal of the entire lung for carcinoma of the bronchus. JAMA 1933;101: Greene FL, Page DL, Fleming ID, Fritz AG. AJCC cancer staging manual. 6th ed. New York: Springer-Verlag, Fritz AG, Percy C, Jack A, Sobin LH. International Classification of Diseases for Oncology. 3rd ed. Geneva: World Health Organization, Andersen PK, Borgan O, Gill RD, Keiding N. Statistical models based on counting processes. Springer series in statistics. New York: Springer, The NCCN clinical practice guidelines in oncology, nonsmall cell lung cancer (version ), National Comprehensive Cancer Network, Inc. Available at: NCCN.org. Accessed November 22, Martini N, Rusch VW, Bains MS, et al. Factors influencing ten-year survival in resected stages I to IIIa non-small cell lung cancer. J Thorac Cardiovasc Surg 1999;117: Centers for Disease Control and Prevention. Health, United States, Updated trend tables as of November Available at: htm. Accessed January 19, Jemal A, Siegel R, Ward E, Hao Y, Xu J, Thun MJ. Cancer statistics, CA Cancer J Clin 2009;59: Kato I, Severson RK, Schwartz AG. Conditional median survival of patients with advanced carcinoma: surveillance, epidemiology, and end results data. Cancer 2001; 92: Appendix 381 The following carcinoma histologic subtypes were included in our analysis (International Classification Disease for Oncology, Third Edition, morphology codes in parenthesis) and were categorized as follows: adenocarcinomas (8140 to 8146, 8200, 8201, 8211, 8230, 8231, 8255, 8260, 8263, 8333, 8340, 8345, 8350, 8550); bronchoalveolar carcinomas (8250 to 8254); squamous cell carcinomas (8050 to 8052, 8070 to 8073, 8075, 8076, 8082 to 8084); and other NSCLC histologies (large cell [8003], epithelial [8010], undifferentiated [8020], anaplastic [8021], pleomorphic [8022], giant cell [8031]; NSCLC not otherwise specified [8046], mixed cell [8323], clear cell [8310], mucoepidermoid [8430], cystadenocarcinomas [8440, 8441, 8470, 8471, 8480, 8481], signet cell [8490], and adenosquamous cell carcinoma [8560, 8570]). GENERAL THORACIC

8 GENERAL THORACIC 382 GROTH ET AL Ann Thorac Surg CONDITIONAL SURVIVAL AFTER LOBECTOMY FOR NSCLC 2010;90: DISCUSSION DR DAVID H. HARPOLE (Durham, NC): This is very nice, and I want to emphasize how important these data are. A number of us who have been involved with clinical trials for years have been in this perpetual bout with the National Cancer Institute CTIP Committee on what should be the outcome for early stage lung cancer studies, and the standard for breast cancer and so forth has always been overall survival, but in young women with breast cancer, the vast majority of deaths are related to breast cancer and their other comorbid death rate is low, so, frankly, it s a reasonable endpoint; whereas in our elderly population with lung cancer, those of us have argued that with time, looking at patients for more than 5 years doesn t make a lot of sense because all of the curves go down because of their issues with comorbid disease that is unrelated to their cancer diagnosis, and these data actually demonstrate that very nicely. Why I think this is so important is that, although we can use randomization of patients for trials, it makes me wonder, and my question is, can we come up with some sort of algorithm that we should be using when we re looking at these randomized trial data to make sure that truly the arms are balanced. In other words, just because it s randomized doesn t mean there could be more males or more smaller tumors, and so then it would take into account, if you re using overall survival, that the difference that you re seeing, if it s close in a trial, might be attributable to the difference that you re actually predicting due to their noncancerrelated death. So I would like you to comment on that. DR GROTH: Thank you for your comments, Dr Harpole. To answer your first question, one of the goals of randomization in an adequately powered clinical trial is to construct treatment groups that have a similar distribution of both recognized and unrecognized confounding variables. You are correct. Due to either random error or systematic error, the treatment groups in a randomized trial may have significant differences. There are methods that can be used when designing a clinical trial to reduce this risk. In particular, stratification is a procedure that can be used as part of a randomization scheme to ensure that certain variables of particular prognostic interest, such as tumor size or sex, have a similar distribution for each of the treatment groups. Although such schemes are more difficult to implement, stratified randomization would eliminate the risk of unbalanced groups. To address your last comment, overall survival does not differentiate between cancer-related and noncancer-related causes of death. As we found, noncancer-related causes of death likely become an increasingly important cause of death in the course of follow-up. For this reason, cancer-specific survival can be a useful outcome to compare with overall survival to assess if other, noncancer-related causes of death are contributing to an observed survival difference in a clinical trial. DR MICHAEL P. POULLIS (Liverpool, United Kingdom): Did you look at the effect of aspirin? DR GROTH: Unfortunately, SEER does not collect information on medication use. DR POULLIS: We have looked at our data the same way as you. We think there are two reasons for dying. About 30% of people coming for lobectomy are on aspirin in our practice, and they have all the risk factors for dying previous heart attacks, strokes, CVAs, peripheral vascular disease and actually they do better. We presented that at the European meeting and it will be out in the next couple of months in the European Journal and will be a basis of a trial we re going to look at on aspirin ( Aspirin has COX-2 beneficial effects for the adenocarcinomas as well. DR MALCOLM M. DECAMP (Chicago, IL): I enjoyed that very much. I think what you are doing is to flesh out in numerical form what our clinical judgment tells us when we are evaluating the physiologic state of our patients and trying to equate that with all the very well-defined parameters that you can get from SEER, for example, about the cancer. I m wondering if this actually isn t almost an underestimate of the competing risks, because you have a lot of detailed information about the cancer, its therapy and stage, and yet the details for cardiovascular comorbidity aren t quite as elaborate. For example, from SEER, it would be very nice to be able to combine data and be able to match or propensity-match patients from the STS databases or some other very well-defined cardiovascular database to really start to drill down on what the competing (noncancer) risks are. But I think this is a fabulous first approximation at doing what we all typically do in terms of clinical judgment of operative risk and potential overall survival. Thank you. DR GROTH: Thank you for your comments, Dr DeCamp. Although SEER provides some data on cancer treatment, it doesn t collect any information on risk factors for developing comorbidities or on the treatment of comorbidities. Although we were able to limit our analysis to patients who underwent lobectomy, the standard of care treatment for stage I disease, we have no way of assessing whether or not these patients also received standard of care treatment for cardiovascular disease. Had we been able to stratify our analysis based on the presence or severity of cardiovascular disease risk factors or had we been able to restrict our analysis to patients who also received standard of care treatment for cardiovascular disease, our results may have differed slightly. Unfortunately, the limitations of the SEER dataset prevent us from performing such an analysis. However, we did use a large population-based dataset, and, therefore, our results are likely reflective of the treatment patterns of cardiovascular disease in the United States. To address your last comment, we are considered using other datasets, such as the STS database, for future studies. We appreciate your comments. DR SUDISH C. MURTHY (Cleveland, OH): Congratulations on a very nice paper. To get back to what Dr Harpole was alluding to, perhaps one of the primary endpoints in future studies of lung cancer should be lung cancer recurrence. We have looked at cancer recurrence in a fairly homogeneous population of stage I lung cancers. You get a good sense of what the risk factors are for cancer recurrence. You then you can then study the competing risk of death or death before recurrence, and at some level, that helps address Dr Harpole s question and might clarify some of these data. DR GROTH: That s a good point, Dr Murthy. Unfortunately, SEER does not collect information on cancer recurrence, and therefore, we couldn t assess the effect of cancer recurrence on our results.