The Society of Thoracic Surgeons General Thoracic Surgery Database: Establishing Generalizability to National Lung Cancer Resection Outcomes

The Society of Thoracic Surgeons General Thoracic Surgery Database: Establishing Generalizability to National Lung Cancer Resection Outcomes Damien J. LaPar, MD, MS, Castigliano M. Bhamidipati, DO, MS, Christine L. Lau, MD, David R. Jones, MD, and Benjamin D. Kozower, MD, MPH Division of Thoracic and Cardiovascular Surgery, University of Virginia, Charlottesville, Virginia Background. The Society of Thoracic Surgeons General Thoracic Surgery Database (GTDB) has demonstrated outstanding results for lung cancer resection. However, whether the GTDB results are generalizable nationwide is unknown. The purpose of this study was to establish the generalizability of the GTDB by comparing lung cancer resection results with those of the Nationwide Inpatient Sample (NIS), the largest all-payer inpatient database in the United States. Methods. From 2002 to 2008, primary lung cancer resection outcomes were compared between the GTDB (n 19,903) and the NIS (n 246,469). Primary outcomes were the proportion of procedures performed nationally that were captured in the GTDB and differences in mortality rates and hospital length of stay. Observed differences in patient characteristics, operative procedures, and postoperative events were also analyzed. Results. Annual GTDB lung cancer resection volume has increased over time but only captures an estimated 8% of resections performed nationally. The GTDB and NIS databases had similar median patient age (67 vs 68 years) and female sex (50% vs 49%), lobectomy was the most common procedure (64.7% vs 79.7%; p < 0.001), and pneumonectomies were uncommon (6.3% vs 7.2%; p < 0.001). Compared with NIS, the GTDB had significantly lower unadjusted discharge mortality rates (1.8% vs 3.0%), median length of stay (5.0 vs 7.0 days; p < 0.001), and postoperative pulmonary complication rates (18.5% vs 23.6%, p < 0.001). Conclusions. The GTDB represents a small percentage of the lung cancer resections performed nationally and reports significantly lower mortality rates and shorter hospital length of stay than national results. The GTDB is not broadly generalizable. These results establish a benchmark for future GTDB comparisons and highlight the importance of increasing participation in the database. (Ann Thorac Surg 2012;94:216 21) 2012 by The Society of Thoracic Surgeons Accepted for publication March 12, 2012. Presented at the Forty-eighth Annual Meeting of The Society of Thoracic Surgeons, Fort Lauderdale, FL, Jan 28 Feb 1, 2012. Address correspondence to Dr Kozower, University of Virginia School of Medicine, PO Box 800679, Charlottesville, VA 22908; e-mail: bdk8g@virginia.edu. Lung cancer remains the leading cause of cancerrelated deaths in the United States, and surgical resection is the primary treatment for early-stage disease. Significant improvements in morbidity and survival have been achieved after lung cancer resections during the past several decades, with operative mortality and morbidity rates now approaching 2% and 8%, respectively [1, 2]. With advances in oncologic therapy and surgical technique, patient outcomes are expected to improve, and clinical documentation of patient-related details, cancer-specific data, operative features, and postoperative events will become increasingly emphasized to help guide treatment decisions and clinical research. During the past 2 decades, significant progress has been achieved in documentation of patient characteristics and outcomes within administrative and clinical databases. Although the nature and intent of documentation within each database may differ, substantial benefits have been derived by examining database content to advance patient care and facilitate clinical investigation. Among available databases, the Nationwide Inpatient Sample (NIS) exists as the largest, most representative, and publicly available administrative database [3]. The NIS uses inpatient discharge records to report patientlevel and hospital-level details as well as disposition and mortality data. Using a coding system based on the International Classification of Diseases-Ninth Revision, Clinical Modifications (ICD-9-CM), the NIS provides an estimate of nationwide trends for inpatient diagnoses, procedure-related details, and postoperative outcomes. Among clinical databases, The Society of Thoracic Surgeons (STS) General Thoracic Surgery Database (GTDB) represents the largest database specifically designed to capture and report general thoracic surgical outcomes. The GTDB was initiated in 2002 and has demonstrated outstanding outcomes for lung and esophageal cancer resection [2, 4]. We initiated this study to compare the current status of reported lung cancer resection outcomes between these administrative and clinical databases. Because the STS 2012 by The Society of Thoracic Surgeons 0003-4975/$36.00 Published by Elsevier Inc http://dx.doi.org/10.1016/j.athoracsur.2012.03.054

Ann Thorac Surg LAPAR ET AL 2012;94:216 21 GTDB VS NIS FOR LUNG CANCER RESECTIONS 217 GTDB has demonstrated outstanding results for lung cancer resections, the purpose of the present study was to establish the generalizability of the GTDB by comparing its lung cancer resection results with those of the NIS. We hypothesized that the GTDB would only capture a minority of procedures performed nationally and that it would report superior outcomes. Material and Methods Records for patients undergoing lung resections with a diagnosis of lung cancer were extracted from the NIS data sets and the STS GTDB for the years 2002 to 2008. The University of Virginia Institutional Review Board exempted this study from review because it was not human subjects research owing to the lack of discrete patient identifiers and because the analyzed data were not exclusively collected for research purposes. Patient Selection Within the NIS, appropriate discharge records were identified by querying the first five diagnosis and procedure categories using the ICD-9-CM diagnosis code 162.X for primary lung cancer and procedure codes 32.9, 32.3X, 32.4X, and 32.5X for lung resection. The STS GTDB was queried for all patient records for those undergoing primary lung cancer resections during the study period. Within both databases, lung cancer resections were characterized by the extent of anatomic resection to include pneumonectomy, lobectomy, and sublobar resections. Data were excluded for patients with benign pulmonary disease. During data extraction from the NIS data sets, less than 5% of patient records with missing data for patient age, sex, resection type, and mortality underwent case-wise deletion to obtain a complete data set for analysis. Variable and Outcome Definitions All variable definitions conformed to those used within the NIS and GTDB [3, 5]. All measured outcomes were established a priori before data analysis. The primary outcomes of this study included differences in reported frequency of lung cancer resections within both the NIS and GTDB as well as observed differences in discharge mortality rates and hospital length of stay (LOS). Reported mortality rates represent those deaths occurring during hospitalization for the NIS and within 30 days of lung cancer resection for the GTDB. The incidence of postoperative death and morbidity within the NIS data sets was determined using previously described methodology [6]. Statistical Analysis All statistical methodology was designed to test the null hypothesis that outcomes after performance of lung cancer resection wound not be significantly different as reported within the GTDB and NIS. Standard statistical significance was set to an 0.05. Descriptive statistics for all variable comparisons were calculated using appropriate univariate hypothesis tests. Categoric variables are expressed as within-group percentages and were compared for independent samples using either the Pearson Table 1. Patient Demographics and Risk Factors for Primary Lung Cancer Resections as Reported in The Society of Thoracic Surgeons General Thoracic Surgery Database and the Nationwide Inpatient Sample Factor a (n 19,903) (n 246,469) GTDB NIS 2 or Fisher exact test. Continuous variables are expressed as mean standard deviation (SD) or median (interquartile range), depending on overall variable distribution. Independent sample, single-factor analysis of variance was used for parametric data comparisons, and the Mann Whitney U test was used for all nonparametric data comparisons. Calculated test statistics were used to derive all reported two-tailed p values. All statistical analyses were performed using PASW 18 software (IBM Corp, Somers, NY). Results p Value Age 67 (60, 74) 68 (60, 74) 0.24 Sex 0.001 Male 9,754 (49.0) 123,686 (50.2) Female 10,149 (51.0) 122,620 (49.8) Year of operation 0.0001 2002 513 (2.6) 34,050 (13.8) 2003 1,308 (6.6) 32,675 (13.3) 2004 2,014 (10.1) 32,590 (13.2) 2005 2,635 (13.2) 37,682 (15.3) 2006 4,418 (22.2) 34,450 (14.0) 2007 5,786 (29.1) 37,422 (15.2) 2008 6,458 (32.4) 37,600 (15.3) Comorbid disease Hypertension 10,724 (53.9) 119,629 (48.5) 0.0001 Heart Failure 715 (3.6) 11,290 (4.6) 0.0001 Peripheral vascular 1,794 (9.0) 15,065 (6.1) 0.0001 disease Renal Failure 415 (2.1) 6,262 (2.5) 0.0001 Operation Sublobar resection 5,764 (29.0) 28,390 (11.5) 0.0001 Lobectomy 12,886 (64.7) 196,469 (79.7) 0.0001 Pneumonectomy 1,253 (6.3) 17,694 (7.2) 0.0001 a Continuous variables are reported as median (interquartile range) and categoric variables as number (%). GTDB general thoracic database; NIS Nationwide Inpatient Sample. A total of 19,903 patients were identified within the GTDB, and 246,469 patient records were extracted from the NIS (Table 1). Median patient age was similar in both databases (GTDB, 67 years; NIS, 68 years). Each database included a nearly equal number of men and women undergoing lung cancer resection. Among major comorbid disease states, nearly half of all patients presented with a preoperative history of hypertension, with a slightly higher prevalence among patients represented in the GTDB (53.9% vs 48.5%, p 0.0001). GTDB patients also had a higher prevalence of preoperative peripheral

218 LAPAR ET AL Ann Thorac Surg GTDB VS NIS FOR LUNG CANCER RESECTIONS 2012;94:216 21 Fig 1. Annual lung cancer resection volume recorded in the General Thoracic Surgery Database (GTDB, diamonds) and National Inpatient Sample (NIS, squares) database. vascular disease (9.0% vs 6.1%, p 0.0001), with a lower prevalence of congestive heart failure (3.6% vs 4.6%, p 0.0001) and renal failure (2.1% vs 2.5%, p 0.0001). The type of lung cancer resections performed differed between the GTDB and NIS. Lobectomy was the most common operation performed in both databases. However, the performance of lobectomy was higher in the NIS compared with the GTDB (79.7% vs 64.7%, p 0.0001). The GTDB contained a higher proportion of sublobar resections, including segmentectomies (29.0% vs 11.5%, p 0.0001). In both databases, performance of pneumonectomy occurred in a minority of patients. The annual volume of lung cancer resections reported within the NIS remained relatively constant during the examined study period, with resection volumes ranging from 32,590 to 37,682 operations/year. Lung cancer resection frequency consistently increased during the study period within the GTDB but only represented a small proportion (8%) of the estimated resections performed annually (Fig 1). Among postoperative events and primary outcomes, the GTDB reported lower mortality, morbidity, and hospital LOS (Table 2). The incidence of postoperative pulmonary complications was 18.5% in the GTDB compared with 23.6% in the NIS (p 0.0001). Discharge mortality in the GTDB was 1.8% after primary lung cancer resection vs 3.0% within the NIS (p 0.0001). The trend in operative mortality was similarly lower as reported in the GTDB for each operative year over the entire study period (Fig 2). Median total hospital LOS was significantly decreased by 2 days among patients represented in the GTDB compared with those within the NIS (5 vs 7 days, p 0.0001). Comment The present study represents an important comparison of primary lung cancer resection rates and outcomes as reported within the STS GTSDB and the NIS. Our results demonstrate significant differences in patient-level data, procedure-level data, and outcomes between samples of the United States general thoracic surgical patient population between two leading clinical and administrative databases. Although the trend in the number of lung cancer resections captured within the NIS remained relatively constant during the study period, the number of resections reported annually within the GTDB consistently increased over time. Despite this increasing frequency, however, the number of lung cancer resections reported within the GTDB only represented a minority of cases compared with the estimated nationwide performance of lung cancer resections reported within the NIS. Most important, postoperative mortality, morbidity, and hospital LOS were significantly better in the GTDB. These results have significant implications in the current and future status of the GTDB and the NIS with respect to their clinical utility and use as instruments for clinical investigation. The GTDB has become a leading clinical and academic resource within thoracic surgery. Since its introduction in Table 2. Postoperative Morbidity, Mortality, and Resource Utilization After Primary Lung Cancer Resections as Reported in The Society of Thoracic Surgeons General Thoracic Surgery Database and the Nationwide Inpatient Sample Outcome a (n 19,903) (n 246,469) GTDB NIS Pulmonary complication 3,683 (18.5) 58,065 (23.6) b Discharge mortality 367 (1.8) 7,456 (3.0) b Length of stay, days 5 (4, 7) 7 (5, 10) b a Categoric variables are reported as number (%) and continuous variables as the median (interquartile range). b p 0.0001 for all outcomes comparisons. GTDB general thoracic database; NIS Nationwide Inpatient Sample.

Ann Thorac Surg LAPAR ET AL 2012;94:216 21 GTDB VS NIS FOR LUNG CANCER RESECTIONS 219 Fig 2. Annual mortality rate after lung cancer resections as recorded in the General Thoracic Surgery Database (GTDB, diamonds) and National Inpatient Sample (NIS, squares) database. 2002, the GTDB has provided an increasingly useful foundation from which to perform comparative effectiveness studies and surgical quality assessment. Several reports have used the GTDB to analyze and track operative outcomes after lung and esophageal resections [2, 4, 7 10]. Within the reported lung cancer resection literature, the GTDB has been used to identify important predictors of morbidity and mortality and to compare hospital performance variation [2]. Wright and colleagues [10] used the GTDB to investigate prolonged LOS (PLOS) after lobectomy in a cohort of 4,979 patients. Their analysis found the incidence of PLOS was 7% after lobectomy and was significantly associated with increased mortality (PLOS, 10.8%; no PLOS, 0.7%; p 0.001). As a result, PLOS has been adopted as an important surgical quality metric for hospitals throughout the United States. The differences in reported annual frequency of lung cancer resections within the GTDB and the NIS were a principle finding in the present study. The use of annual surgical volumes captured within the NIS provided a valid nationwide estimate of lung cancer resection rates from which to evaluate the GTDB. To ensure a valid comparison between the data sets, comparisons were made for only those years with available data in both the GTDB and NIS. The reported frequency of annual lung cancer resection rates within the NIS in the present study is similar to that reported in other published series [11 13]. A recent series from Ellis and colleagues [11] reported intraoperative staging and outcomes among 222,233 patients undergoing lung cancer resections within the NIS from 1998 to 2007. In light of these estimates, the important finding that the GTDB only represents 8% of estimated lung cancer resections performed annually within the United States provides an important context from which to view the reported results. The low mortality and morbidity and the reported hospital LOS in this series are similar to other published reports [2, 7, 12, 13]. In the present series, discharge mortality rates ranged from 1.8% in the GTDB to 3.0% in the NIS, whereas pulmonary complications occurred in 18% to 25% of patients. These results compare favorably with the recently reported 2.2% perioperative mortality rates for lung cancer resections published by the STS in 2010 [2]. Other series have documented similar mortality rates for primary lung cancer resection ranging from 2.9% to 3.2%, with a mean hospital LOS of 8 days [12, 13]. The striking finding in the present analysis was the significantly lower mortality rate and reduced hospital LOS reported within the GTDB. The implication of these better outcomes and the small percentage of cases captured nationally is that the lung cancer resection results of the GTDB are not generalizable to all patients throughout the United States. A likely explanation for these differences is that the GTDB is composed of dedicated and experienced thoracic surgeons primarily practicing at tertiary care centers. Alternatively, the performance of a higher proportion of sublobar resections, as reported within the GTDB, may have influenced the lower mortality rates and decreased hospital LOS reported here. Moreover, regarding the observed differences in hospital LOS, the clinical implications of the slightly longer hospital LOS reported for patients within the NIS may reflect little impact on overall resource utilization because most costs associated with hospitalizations occur in the first few postoperative days. The present findings have significant clinical implications and provide empiric evidence to support increased participation in the GTDB by U.S. hospitals and surgeons. Since the introduction of the GTDB, there has been a consistent increase in the number of participating centers with an exciting exponential increase in participation within the past 3 to 5 years, reflecting an increase

220 LAPAR ET AL Ann Thorac Surg GTDB VS NIS FOR LUNG CANCER RESECTIONS 2012;94:216 21 Fig 3. Annual number of hospitals participating in the General Thoracic Surgery Database (GTDB). in the percentage of participating surgeons from 1.5% in 2002 to 17% in 2008. Figure 3 illustrates this dramatic trend in GTDB participation from 2003 through 2011. In light of the results of the present study, future comparisons of GTDB outcomes with other nationwide outcomes are warranted to establish the generalizability of its content with respect to evolving national trends. More important, these data imply that the superior outcomes represented within the GTDB should not be extrapolated to surgeons and centers not participating in the GTDB. This investigation has several limitations. The retrospective study design is subject to inherent selection bias. However, all lung cancer resections performed during the study period were included. A second limitation is that different definitions for patient and outcome variables exist between the GTDB and the NIS. Mortality is defined as in-hospital death by the NIS and as 30-day mortality by the GTDB. Importantly, the 30-day mortality rate is likely to be higher than in-hospital mortality, and this comparison may underestimate the outstanding results of the GTDB [14]. Further differences in data content limited examination of specific predictors of surgical outcomes, including pulmonary function, because these data are not captured or reported within the NIS data sets. Furthermore, differences in patient surgical risk profiles appear to exist between data sets because those patients reported within the NIS demonstrate higher operative risk due the higher prevalence of men with higher percentages of heart failure, renal failure, and performance of pneumonectomy. These patient characteristics remain wellestablished risk factors for death after lung cancer resection and are included in the STS GTDB mortality risk models. The potential for unrecognized miscoding and data entry errors should be considered in any secondary analysis of a data registry. The potential influence of a time bias in the reported shorter hospitalizations and lower mortality for the GTDB should also be considered because mortality rates reported in the GTDB appear to be approaching those reported within the NIS and because a larger percentage of patients in the NIS are from more recent times and hospital LOS has declined over time. For the included study period, external auditing of the GTDB had not been performed. However, auditing of the GTDB began in 2010, and the results of these and future audits will be essential for validating the accuracy of the database. Finally, a lack of long-term follow-up in these analyses limits the ability to provide a complete commentary on the current status of nationwide lung cancer resection results. Nevertheless, these results provide a valuable extension to previously reported series on U.S. nationwide lung cancer resection outcomes and provide a benchmark for future GTDB comparisons. In conclusion, the reported results suggest that the STS GTDB represents a small percentage of the lung cancer resections performed nationally and reports significantly lower mortality rates and resource utilization than national results. Thus, the GTDB is not broadly generalizable, and its reported outcomes should not be extrapolated to nonparticipant hospitals and surgeons. Continued auditing of GTDB content will be fundamental to ensuring its validity and long-term success. These results establish a benchmark for future GTDB comparisons and highlight the importance of increasing participation in the database. We thank Donna McDonald at the STS and Amelia Wallace at the Duke Clinical Research Institute for their assistance in data collection and analysis. References 1. Ginsberg RJ. Lung cancer surgery: acceptable morbidity and mortality, expected results and quality control. Surg Oncol 2002;11:263 6. 2. Kozower BD, Sheng S, O Brien SM, et al. STS database risk models: predictors of mortality and major morbidity for lung cancer resection. Ann Thorac Surg 2010;90:875 81; discussion 881 3. 3. Healthcare Cost and Utilization Project Nationwide Inpatient Sample (NIS). www.hcup-us.ahrq.gov/nisoverview.jsp. Accessed: Feb 1, 2012. 4. Wright CD, Kucharczuk JC, O Brien SM, Grab JD, Allen MS. Predictors of major morbidity and mortality after esophagectomy for esophageal cancer: a Society of Thoracic Surgeons General Thoracic Surgery Database risk adjustment

Ann Thorac Surg LAPAR ET AL 2012;94:216 21 GTDB VS NIS FOR LUNG CANCER RESECTIONS 221 model. J Thorac Cardiovasc Surg 2009;137:587 95; discussion 596. 5. Society of Thoracic Surgeons. General Thoracic Surgery Database. http://www.sts.org/quality-research-patientsafety/national-database/database-managers/generalthoracic-surgery-databa-1. Accessed: Feb 1, 2012. 6. LaPar DJ, Bhamidipati CM, Mery CM, et al. Primary payer status affects mortality for major surgical operations. Ann Surg 2010;252:544 50; discussion 550 1. 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;135:247 54. 8. Onaitis M, D Amico T, Zhao Y, O Brien S, Harpole D. Risk factors for atrial fibrillation after lung cancer surgery: analysis of the Society of Thoracic Surgeons general thoracic surgery database. Ann Thorac Surg 2010;90:368 74. 9. Shapiro M, Swanson SJ, Wright CD, et al. Predictors of major morbidity and mortality after pneumonectomy utilizing the Society for Thoracic Surgeons General Thoracic Surgery DISCUSSION DR DOUGLAS E. WOOD (Seattle, WA): Dr LePar, very nice presentation. And just to make a comment on the auditing of the General Thoracic Database, I think what is important knowledge is that that auditing shows 95% concordance of audited numbers with submitted numbers, so it actually suggests that the General Thoracic Database data is highly valid. I would like to push you one step further from where you went. You said that The Society of Thoracic Surgeons (STS) General Thoracic Database is not generalizable. But tell us why not, and why and how we might take that disparity and lift the whole boat of patients that are getting pulmonary resections nationwide. DR LAPAR: Thank you very much for that. I think that it is very encouraging in the audits that we have seen such a great concordance with what is being captured and what actually exists. I think with respect to our primary conclusion that in its current form the General Thoracic Database is not generalizable, what we are saying is that the outcomes as captured by participants of the General Thoracic Database are very, very encouraging but that there appears to be a disparity between these outcomes and those captured within a nationwide administrative database. Because the outcomes represent those of only 8% of lung cancer resections performed nationwide at the end of our study period, we think it is very important to make the case that the improved mortality and length of stay reported by the STS database should not be extrapolated to non-sts database participants. DR WOOD: So I am going to urge you to not be shy in recognizing the differences between the General Thoracic Database and the National Inpatient Sample; that, for example, those in the General Thoracic Database are board-certified thoracic surgeons, operating significantly in major academic centers as well as in larger practices. So those are some factors that might result in those differences. Database. Ann Thorac Surg 2010;90:927 34; discussion 934 5. 10. Wright CD, Gaissert HA, Grab JD, O Brien SM, Peterson ED, Allen MS. Predictors of prolonged length of stay after lobectomy for lung cancer: a Society of Thoracic Surgeons General Thoracic Surgery Database risk-adjustment model. Ann Thorac Surg 2008;85:1857 65; discussion 1865. 11. Ellis MC, Diggs BS, Vetto JT, Schipper PH. Intraoperative oncologic staging and outcomes for lung cancer resection vary by surgeon specialty. Ann Thorac Surg 2011;92:1958 63; discussion 1963 4. 12. LaPar DJ, Bhamidipati CM, Harris DA, et al. Gender, race, and socioeconomic status affects outcomes after lung cancer resections in the United States. Ann Thorac Surg 2011;92: 434 9. 13. LaPar DJ, Nagji AS, Bhamidipati CM, et al. Seasonal variation influences outcomes following lung cancer resections. Eur J Cardiothorac Surg 2011;40:83 90. 14. Bryant AS, Rudemiller K, Cerfolio RJ. The 30- versus 90-day operative mortality after pulmonary resection. Ann Thorac Surg 2010;89:1717 22; discussion 1722 3. DR LAPAR: Certainly these are important factors, and we believe that the participants contributing to the General Thoracic Database represent experienced thoracic surgeons, which translates into the improved outcomes that we are seeing in the GTDB. DR TODD DEMMY (Buffalo, NY): I would like to ask you about the concept of video-assisted thoracoscopic surgical (VATS) predominance. We just heard a concept for VATS reliability; for instance, once you are over 80% of your resections by VATS, it provides a normative framework for interpreting the data. A similar analysis was just presented downstairs by Dr Boffa. So, can you look at the percentage that an institution does either an open or a VATS as an indicator of quality or outcomes and then use that to normalize the data between these large data sets and the General Thoracic Database? DR LAPAR: Thanks for that question. I think that that is an important metric to look at in the future. In this study, we didn t look at a breakdown of VATS vs open techniques. But certainly, a higher performance of VATS lobectomies at a given institution may be a proxy of surgeon experience and may also serve as a quality indicator that we can look at in the future. DR AKIF TURNA (Istanbul, Turkey): I know it is different and difficult, another test, but did you happen to look at the Surveillance, Epidemiology and End Results (SEER) database and compare some of the main parameters with the General Thoracic Surgery Database? DR LAPAR: No, we did not look at outcomes in the SEER database as they compare to the General Thoracic Surgery Database in this study. However, this would be a great next step to take in future database comparisons.