- Email: [email protected]
Prediction of Long-Term Survival After Lung Cancer Surgery for Elderly Patients in The Society of Thoracic Surgeons General Thoracic Surgery Database
J. MAXWELL CHAMBERLAIN MEMORIAL PAPER
Prediction of Long-Term Survival After Lung Cancer Surgery for Elderly Patients in The Society of Thoracic Surgeons General Thoracic Surgery Database Mark W. Onaitis, MD, Anthony P. Furnary, MD, Andrzej S. Kosinski, PhD, Sunghee Kim, PhD, Daniel Boffa, MD, Betty C. Tong, MD, Patricia Cowper, PhD, Jeffrey P. Jacobs, MD, Cameron D. Wright, MD, Joe B. Putnam, Jr, MD, and Felix G. Fernandez, MD, MSc Department of Surgery, University of California-San Diego, La Jolla, California; Department of Surgery, Starr-Wood Cardiac Group, Portland, Oregon; Duke Clinical Research Institute, Duke University, Durham, North Carolina; Department of Surgery, Yale University, New Haven, Connecticut; Department of Surgery, Johns Hopkins All Children’s Heart Institute, St. Petersburg, Florida; Department of Surgery, Massachusetts General Hospital, Boston, Massachusetts; Department of Surgery, Baptist MD Anderson Cancer Center, Jacksonville, Florida; and Department of Surgery, Emory University, Atlanta, Georgia
Background. Prior risk models using the STS General Thoracic Surgery database (STS-GTSD) have been limited to 30-day outcomes. We have now linked STS data to Medicare data and sought to create a risk prediction model for long-term mortality after lung cancer resection in patients older than 65 years. Methods. The STS-GTSD was linked to Medicare data for lung cancer resections from 2002 to 2013 as previously reported. Successful linkage was performed in 29,899 lung cancer resection patients. Cox proportional hazards modeling was used to create a long-term survival model. Variable selection was performed using statistically significant univariate factors and known clinical predictors of outcome. Calibration was assessed by dividing the cohort into deciles of predicted survival and discrimination assessed with a C-statistic corrected for optimism via 1,000 bootstrap replications. Results. Median age was 73 years (interquartile range, 68 to 78 years), and 48% of the patients were male. Of the 29,094 patients with nonmissing pathologic stage, 69% were stage I, 18% stage II, 11% stage III, and 2% stage IV. Procedure performed was lobectomy in 69%, bilobectomy in 3%, pneumonectomy in 3%, segmentectomy in 7%,
sleeve lobectomy in 1%, and wedge resection in 17%. Thoracoscopic approach was performed in 47% of resections. The final Cox model reveals that stage and age are the strongest predictors of long-term survival. Even after controlling for stage, wedge resection, segmentectomy, bilobectomy, and pneumonectomy are all associated with increased hazard of death in comparison with lobectomy. Thoracoscopic approach is associated with improved long-term survival in comparison with thoracotomy. Other modifiable predictive factors include smoking and low body mass index. Calibration of the model demonstrates excellent performance across all survival deciles and a C-statistic of 0.694. Conclusions. The STS-GTSD-Medicare long-term risk model includes several novel factors associated with mortality. Although medical factors predict long-term survival, age and stage are the strong predictors. Despite this, procedure choice and thoracoscopic/open approach are potentially modifiable predictors of longterm survival after lung cancer resection.
T
patient records in the STS-GTSD were recently linked to data from the Centers for Medicare and Medicaid Services (CMS) [1]. The resulting data set is a unique resource permitting analysis of oncologic outcomes.
he STS General Thoracic Surgery Database (GTSD) has identified predictors of short-term (30-day) outcomes. To analyze longer-term outcomes, lung cancer
(Ann Thorac Surg 2017;-:-–-) Ó 2017 by The Society of Thoracic Surgeons
Accepted for publication June 28, 2017. Presented at the Fifty-third Annual Meeting of The Society of Thoracic Surgeons, Houston, TX, Jan 21-25, 2017. Winner of the J. Maxwell Chamberlain Memorial Award for General Thoracic Surgery. Address correspondence to Dr Onaitis, Department of Surgery, University of California-San Diego, 9300 Campus Point Dr, Mailcode 7892, LaJolla, CA 92037; email: [email protected].
Ó 2017 by The Society of Thoracic Surgeons Published by Elsevier Inc.
The Supplemental Figures can be viewed in the online version of this article [http://dx.doi.org/10.1016/ j.athoracsur.2017.06.071] on http://www.annalsthoracic surgery.org.
0003-4975/$36.00 http://dx.doi.org/10.1016/j.athoracsur.2017.06.071
2
CHAMBERLAIN MEMORIAL PAPER ONAITIS ET AL LONG-TERM GTSD SURVIVAL MODEL
Improved methods of data sharing and statistical analysis have allowed refinement of the American Joint Committee on Cancer staging system. These efforts have confirmed the importance of stage as well as tumor size and location [2]. Our previous report also demonstrated that age and sex are significant predictors [1]. Others have used national databases to define factors that predict survival after lung cancer resection [3, 4]. Here, we sought to create a long-term survival model for lung cancer patients aged over 65 years who underwent resection in STS-GTSD hospitals.
Patients and Methods Medicare Database of CMS The data source for this study is the 100% Medicare inpatient claims file, which contains information on hospitalizations of patients enrolled in fee-for-service Medicare. The Standard Analytic Files were used. These include dates of service and diagnostic codes from the International Classification of Diseases, Ninth Revision, Clinical Modification. The database contains anonymous patient identifiers that enable follow up of beneficiaries over time. The 100% Medicare denominator file, which links to the inpatient file and contains information on beneficiary eligibility, demographic characteristics, and date of death, was also used.
Linkage The STS-GTSD was linked to Medicare data for lung cancer resections from 2002 to 2013 [1]. Briefly, the STS GTSD was linked to CMS claims files using combinations of nonunique indirect identifiers through a deterministic matching algorithm [5, 6]. Once the individual patients were linked, longitudinal records were created containing follow-up information, including subsequent death. Successful linkage was performed in 29,899 lung cancer resection patients.
Variables The following independent variables were considered in the multivariable analyses: age, sex, body mass index (BMI, in kg/m2), American Society of Anesthesiologists (ASA) class, Zubrod score, presence of coronary artery disease (CAD), presence of cardiovascular disease (CVD), congestive heart failure (CHF), hypertension, diabetes mellitus (DM), steroid use, peripheral vascular disease (PVD), renal insufficiency (creatinine > 2 or hemodialysis [HD]), FEV1% predicted, smoking status, prior thoracic surgery, approach, and procedure. Urgent/emergent procedures were excluded. Pathologic stage was used instead of clinical stage as there was a high percentage of missingness for the clinical stage variable (9.4%) as opposed to only 2.7% for the pathologic stage. DLCO also was not included because of missingness.
Statistical Analyses Unadjusted survival was plotted using the Kaplan-Meier method. Cox modeling was used to create a long-term survival model. Clustering within hospitals was accounted for by consideration of robust sandwich covariance matrix estimates [7]. Variable selection was performed
Ann Thorac Surg 2017;-:-–-
using statistically significant univariate factors and known clinical predictors of outcome. Violations of the proportional hazard assumption were assessed graphically. No violations were detected. No meaningful collinearity between covariates was present—all the condition indices were less than 16, with 88% of them less than 5; thus all the condition indices were substantially lower than the commonly accepted threshold of 30. Calibration was assessed by a calibration slope, as well as by plot at 5 years— that is, by dividing the cohort into deciles of predicted survival and comparing the decile specific observed survival at 5 years (Kaplan-Meier) with the average predicted 5-year survival within the corresponding deciles. Calibration slope and C-statistic were both corrected for optimism via 1,000 bootstrap replications. As an alternative optimism correction of the C-statistic, we also randomly split data into equally sized training and testing samples, applied model coefficients based on the training sample to the testing sample data, and subsequently computed a testing sample C-statistic; repetition of this process 1,000 times resulted in the reported average testing sample C-statistic.
Results The STS-GTSD linkage resulted in 29,899 patients. Table 1 presents the demographics. Median age was 73 years (range, 68 to 78 years], and 48% of the patients were male. At pathologic analysis, 69% were stage I, 18% stage II, 11% stage III, and 2% stage IV. Procedure performed was lobectomy in 69%, bilobectomy in 3%, pneumonectomy in 3%, segmentectomy in 7%, sleeve lobectomy in 1%, and wedge resection in 17%. Thoracoscopic approach was performed in 47% of resections. Figure 1 depicts overall survival. The 1-, 3-, and 5-year survival proportions of the entire study population are 86.7%, 67.5%, and 53.2%, respectively. The 90-day mortality is 4.5%, and the 180-day mortality is 7.3%. Cox proportional hazards survival analysis was then performed. Table 2 demonstrates the models. In the multivariable model, increasing age, increasing stage, male sex, BMI less than 18.5, increasing ASA and Zubrod score, CAD, CVD, CHF, DM, steroid use, PVD, acute kidney injury, decreasing FEV1% predicted, and smoking history predict mortality. The calibration and discrimination of the model were assessed. Calibration across deciles of predicted survival was excellent (calibration plot, Supplemental Fig 1) and the calibration slope is 0.987. The C-statistic of the linked model is 0.696 (95% confidence interval, 0.689 to 0.702), the optimism corrected via bootstrap C-statistic is 0.694, and the average testing sample C-statistic is 0.693 (range over 1,000 data splits, 0.685 to 0.704) (Supplemental Fig 2). Figure 2 demonstrates survival curves for three predictor variables using Kaplan-Meier and Cox model prediction. Figure 2A shows a stepwise decrement in long-term survival as FEV1% predicted decreases. Similarly, current smoking is associated with decreased longterm survival (Fig 2B). Procedure performed is predictive. Figure 2C demonstrates that lobectomy patients perform significantly better independent of stage than patients undergoing
Ann Thorac Surg 2017;-:-–-
Table 1. Continued
Table 1. Patient Demographics Characteristic Age (years) 65–69 70–74 75–79 80þ Sex Male Female Body mass index <18.5 18.5–25 25–30 30–35 >35 ASA class I–II III IV–V Zubrod score 0 1 2–5 Coronary artery disease Cardiovascular disease Congestive heart failure Diabetes mellitus Steroid use Peripheral vascular disease Acute kidney injury FEV1 % predicted >80 60–80 40–60 <40 Pathologic stage I II III IV Clinical stage I II III IV Cigarette use Never Past Current Reoperation Procedure Lobectomy Bilobectomy Pneumonectomy
3
CHAMBERLAIN MEMORIAL PAPER ONAITIS ET AL LONG-TERM GTSD SURVIVAL MODEL
n (%)
8,671 8,886 7,147 5,195
(29.0) (29.7) (23.9) (17.4)
14,412 (48.2) 15,474 (51.8) 794 9,568 12,323 4,961 2,253
(2.7) (32.0) (41.2) (16.6) (7.5)
3,885 (13.1) 22,281 (75.4) 3,391 (11.5) 12,443 15,451 1,700 7,959 3,000 1,152 5,677 1,036 3,599 697
(42.0) (52.2) (5.7) (26.6) (10.0) (3.9) (19.0) (3.5) (12.0) (2.3)
13,613 8,637 3,910 842
(50.4) (32.0) (14.5) (3.1)
20,038 5,155 3,311 590
(68.9) (17.7) (11.4) (2.0)
20,635 3,588 2,366 487
(76.2) (13.3) (8.7) (1.8)
3,819 20,645 5,435 1,624
(12.8) (69.0) (18.2) (5.4)
Characteristic Segmentectomy Sleeve lobectomy Wedge resection Video-assisted thoracoscopic surgery
n (%) 2,079 282 5,203 14,182
ASA ¼ American Society of Anesthesiologists; FEV1 ¼ forced expiratory volume, 1 second.
other procedures. At 5 years, survival after lobectomy is 56.5%, whereas pneumonectomy survival is 34.3% and segmentectomy survival is 52.5%. In addition to procedure, thoracoscopic approach is associated with improved long-term survival (Fig 3). Given that a previous analysis documented lower N1 upstaging in clinical N0 patients undergoing thoracoscopic as compared to open lobectomy [8], we assessed the predictors of survival in clinical stage I patients in the linked data set. Unadjusted upstaging to stage II was 13.9% in the open group and 8.9% in the video-assisted thoracoscopic surgery (VATS) group and to upstaging to stage III was 7.0% in the open group and 4.8% in the VATS group. In the Cox survival model, as Table 3 demonstrates, in the 20,635 patients who were clinical stage I patients, all of the same variables predict long-term survival. In order to more thoroughly examine the effect of individual variables on long-term survival, the effects of modifying key prognostic factors are examined in Table 4. For the purposes of illustration, 5-year survival of patients with the listed predictor variables was obtained from the Cox model in Table 2 by multiplying particular patient covariate values by estimated model coefficients. For simplicity, medical comorbidities were set to “0,” ASA/ Zubrod score was set to the lowest categories, and BMI was set to 20. These data demonstrate the importance of age and stage in comparison with procedural variables.
Comment We have created a long-term risk model for non-small cell lung cancer (NSCLC) resection patients aged over 65 years in the STS GTDB. The resulting model allows assessment of survival after the index hospitalization,
20,578 (68.8) 842 (2.8) 915 (3.1) (Continued)
(7.0) (0.9) (17.4) (47.4)
Fig 1. Overall survival of the entire cohort, Kaplan-Meier.
4
CHAMBERLAIN MEMORIAL PAPER ONAITIS ET AL LONG-TERM GTSD SURVIVAL MODEL
Ann Thorac Surg 2017;-:-–-
Table 2. Results of Cox Models for Long-Term Overall Survival Characteristic
Univariate HR (95% CI)
Age group (years) 65–69 70–74 75–79 80þ Sex Male Female Body mass index 18.5–25 <18.5 25–30 30–35 >35 ASA group I–II III IV–V Zubrod score 0 1 2–5 Coronary artery disease Cardiovascular disease Congestive heart failure Diabetes mellitus Steroid use Peripheral vascular disease Chronic kidney disease % FEV1 predicted >80 60–80 40–60 <40 Pathologic stage I II III IV Smoking Never Past Current Reoperation Procedure Lobectomy Bilobectomy Pneumonectomy Segmentectomy Sleeve Wedge Video-assisted thoracoscopic surgery approach ASA ¼ American Society of Anesthesiologists;
p
Multivariable HR (95% CI)
<0.001 1.00 1.17 (1.11, 1.24) 1.35 (1.28, 1.44) 1.72 (1.62, 1.82)
<0.001 1.00 1.19 (1.12, 1.27) 1.40 (1.31, 1.50) 1.90 (1.78, 2.03)
<0.001 1.00 0.66 (0.63, 0.68)
<0.001 1.00 0.73 (0.70, 0.76)
<0.001 1.58 0.93 0.87 0.88
1.00 (1.41, (0.87, (0.82, (0.82,
1.78) 1.00) 0.92) 0.96)
<0.001 1.58 0.87 0.82 0.87
1.00 (1.41, (0.82, (0.77, (0.80,
1.78) 0.93) 0.87) 0.95)
<0.001 1.00 1.53 (1.40, 1.68) 2.13 (1.86, 2.43)
<0.001 1.00 1.23 (1.14, 1.33) 1.37 (1.21, 1.54)
<0.001 1.41 2.33 1.40 1.35 1.75 1.23 1.47 1.51 1.93
1.00 (1.33, (2.12, (1.34, (1.28, (1.61, (1.17, (1.34, (1.43, (1.72,
1.49) 2.57) 1.47) 1.44) 1.92) 1.29) 1.63) 1.61) 2.17)
<0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001
1.00 1.32 (1.26, 1.39) 1.48 (1.40, 1.57) 1.60 (1.40, 1.83)
<0.001 1.26 1.69 1.09 1.18 1.34 1.12 1.33 1.21 1.57
1.00 (1.19, (1.50, (1.04, (1.10, (1.22, (1.07, (1.19, (1.13, (1.40,
1.33) 1.91) 1.15) 1.26) 1.48) 1.18) 1.49) 1.29) 1.76)
<0.001
<0.001 1.00 1.80 (1.69, 1.91) 2.62 (2.43, 2.81) 2.86 (2.45, 3.35)
<0.001
1.65 2.26 1.08 1.46 1.29 0.75
1.00 (1.47, (2.07, (0.95, (1.21, (1.21, (0.71,
CI ¼ confidence interval;
1.85) 2.48) 1.23) 1.76) 1.38) 0.80)
<0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001
1.00 1.17 (1.11, 1.24) 1.26 (1.18, 1.35) 1.42 (1.25, 1.61)
1.00 1.85 (1.75, 1.97) 2.58 (2.43, 2.73) 2.86 (2.49, 3.27) 1.00 1.58 (1.46, 1.71) 1.68 (1.53, 1.84) 1.38 (1.27, 1.50)
p
<0.001 <0.001
<0.001
<0.001 1.00 1.35 (1.23, 1.48) 1.54 (1.38, 1.71) 1.13 (1.03, 1.24)
1.30 1.58 1.10 0.93 1.22 0.86
1.00 (1.15, (1.40, (0.99, (0.75, (1.14, (0.82,
FEV1 ¼ forced expiratory volume, 1 second;
1.46) 1.80) 1.23) 1.14) 1.30) 0.92)
0.012 <0.001
<0.001
HR ¼ hazard ratio.
Ann Thorac Surg 2017;-:-–-
CHAMBERLAIN MEMORIAL PAPER ONAITIS ET AL LONG-TERM GTSD SURVIVAL MODEL
5
Fig 3. Univariate Kaplan-Meier and multivariable predicted survival based on approach. (Open ¼ open lobectomy; VATS ¼ video-assisted thoracoscopic surgery.)
Fig 2. Univariate Kaplan-Meier and multivariable predicted survival based on (A) FEV1 % predicted, (B) smoking status, and (C) procedure. FEV1 ¼ forced expiratory volume, 1 second.
which is important because studies with high 90-day mortality have recently been published [9]. The previous STS short-term lung cancer resection risk model identified several factors as predictive of 30-day survival: age, male sex, FEV1%, body mass index, cerebrovascular disease, steroids, coronary artery disease, peripheral vascular disease, renal dysfunction, Zubrod score, ASA rating, thoracotomy approach, induction therapy, reoperation, tumor stage, and greater extent of resection [10, 11]. Induction therapy was not included in the current model because of its association with stage. Interestingly, all of these factors were also predictive in the linked, long-term model.
The model sheds light on several variables that contribute to long-term survival. Age is obviously a significant predictor in any long-term survival analysis. The average life expectancy in the United States is 78.8 years. Other analyses have identified age as a strong predictor of long-term survival in lung cancer patients [3, 4]. In addition, previous studies have also demonstrated that female sex is protective in survival analyses [3, 12], and this is confirmed in our study. Pathologic stage strongly predicts survival. Increases in stage are associated with a larger hazard ratios for death than any medical factor analyzed. These results emphasize the need for thorough assessment of lymph nodes and highlight the need for improvements in adjuvant therapies. All of the medical factors are associated with survival in this series with a large n. Although CHF, steroid use, and creatinine greater than 2/HD are the strongest chronic medical predictors of long-term survival, none of these factors is as impactful as stage. Future analyses including patients undergoing stereotactic body radiation therapy and ablative therapy will be helpful in individualizing therapies. Increasing ASA/Zubrod scores were predictive of death. The majority of patients were ASA III and Zubrod I, and the hazard ratio for death increases sharply above these values. These scores are useful in predicting long-term survival of these patients and may be used in addition to oncologic factors in planning treatment strategies. One modifiable risk factor that significantly impacts long-term survival in this data set is smoking. A recent NIH-AARP cohort study in more than 160,000 patients over age 70 years demonstrated decreased hazard ratios for allcause mortality for former smokers as compared with current smokers [13]. A recent meta-analysis demonstrated worse survival in early-stage NSCLC continuing smokers [14]. Our study is much smaller and demonstrates survival benefit for never smokers and a smaller benefit for former smokers. Smoking cessation is clearly recommended to decrease perioperative mortality and lung cancer risk [15]. BMI is another modifiable risk factor. Short-term analysis has demonstrated increased risk of perioperative events at the extremes of BMI [16]. Underweight patients (BMI < 18.5) maintain significantly increased risk (hazard ratio [HR] 1.58) independently over time.
6
CHAMBERLAIN MEMORIAL PAPER ONAITIS ET AL LONG-TERM GTSD SURVIVAL MODEL
Ann Thorac Surg 2017;-:-–-
Table 3. Results of Cox Models for Long-Term Overall Survival in Clinical Stage I Patients Characteristic
Univariate HR (95% CI)
Age group 65–69 70–74 75–79 80þ Sex Male Female Body mass index group 18.5–25 <18.5 25–30 30–35 >35 ASA group I–II III IV–V Zubrod score 0 1 2–5 Coronary artery disease Cardiovascular disease Congestive heart failure Diabetes mellitus Steroid use Peripheral vascular disease Chronic kidney disease % FEV1 predicted >80 60–80 40–60 <40 Pathologic stage I II III IV Smoking Never Past Current Reoperation Procedure Lobectomy Bilobectomy Pneumonectomy Segmentectomy Sleeve Wedge Video-assisted thoracoscopic surgery approach ASA ¼ American Society of Anesthesiologists;
p
Multivariable HR (95% CI)
<0.001 1.00 1.23 (1.14, 1.32) 1.45 (1.35, 1.56) 1.93 (1.80, 2.07)
<0.001 1.00 1.23 (1.14, 1.33) 1.47 (1.35, 1.59) 2.07 (1.91, 2.25)
<0.001 1.00 0.66 (0.63, 0.69)
<0.001 1.00 0.72 (0.68, 0.76)
<0.001 1.67 0.91 0.89 0.91
1.00 (1.43, (0.84, (0.83, (0.83,
1.95) 0.98) 0.97) 1.01)
<0.001 1.65 0.83 0.84 0.88
1.00 (1.41, (0.77, (0.76, (0.79,
1.93) 0.89) 0.92) 0.99)
<0.001 1.00 1.61 (1.43, 1.83) 2.28 (1.93, 2.69)
<0.001 1.00 1.30 (1.17, 1.44) 1.45 (1.26, 1.67)
<0.001 1.41 2.28 1.47 1.40 1.80 1.24 1.46 1.65 2.23
1.00 (1.32, (2.04, (1.39, (1.31, (1.62, (1.16, (1.29, (1.53, (1.96,
1.50) 2.55) 1.56) 1.51) 2.01) 1.32) 1.66) 1.77) 2.53)
<0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001
1.00 1.33 (1.25, 1.42) 1.57 (1.46, 1.69) 1.63 (1.40, 1.91)
<0.001 1.29 1.70 1.07 1.16 1.36 1.13 1.33 1.26 1.62
1.00 (1.22, (1.50, (1.01, (1.07, (1.20, (1.06, (1.16, (1.16, (1.44,
1.37) 1.93) 1.14) 1.27) 1.53) 1.20) 1.53) 1.37) 1.84)
<0.001
<0.001 1.00 1.76 (1.61, 1.93) 2.47 (2.21, 2.77) 2.69 (2.01, 3.58)
<0.001
1.57 2.16 1.11 1.14 1.33 0.80
1.00 (1.33, (1.75, (0.95, (0.81, (1.23, (0.75,
CI ¼ confidence interval;
1.86) 2.67) 1.30) 1.62) 1.43) 0.85)
0.03 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001
1.00 1.17 (1.10, 1.25) 1.29 (1.20, 1.40) 1.36 (1.19, 1.55)
1.00 1.75 (1.61, 1.91) 2.35 (2.14, 2.59) 2.62 (2.07, 3.32) 1.00 1.67 (1.52, 1.85) 1.86 (1.67, 2.08) 1.44 (1.29, 1.61)
p
<0.001 <0.001
<0.001
<0.001 1.00 1.42 (1.28, 1.59) 1.73 (1.53, 1.95) 1.20 (1.05, 1.37)
1.29 1.71 1.13 0.79 1.22 0.88
1.00 (1.07, (1.33, (0.98, (0.56, (1.14, (0.83,
FEV1 ¼ forced expiratory volume, 1 second;
1.56) 2.18) 1.31) 1.12) 1.32) 0.94)
0.007 <0.001
<0.001
HR ¼ hazard ratio.
Ann Thorac Surg 2017;-:-–-
CHAMBERLAIN MEMORIAL PAPER ONAITIS ET AL LONG-TERM GTSD SURVIVAL MODEL
7
Table 4. Survival Prediction in Hypothetical Patients Age (years)
Sex
65 80 65 65 80 65 65 80 65 65 65 80
M M F M M M M M M M M M
FEV1 %
Pathologic Stage
90 90 90 90 90 90 90 90 90 50 50 50
I I I I I III I I I I I I
CI ¼ confidence interval; F ¼ female; assisted thoracoscopic surgery.
Smoking Status Past Past Past Past Past Past Past Past Never Past Past Past
Procedure Lobectomy Lobectomy Lobectomy Lobectomy Lobectomy Lobectomy Segmentectomy Segmentectomy Lobectomy Lobectomy Segmentectomy Segmentectomy
FEV1 ¼ forced expiratory volume, 1 second;
Procedure performed is predictive. Previous short-term STS database analysis in propensity matched patients has demonstrated that wedge resection is associated with lower morbidity and mortality than anatomic resection [17]. While we await results of the CALGB 140503 trial [18], this analysis identifies both wedge resection and segmentectomy as predictors of long-term mortality. Although such a finding may be related to confounding in this large retrospective data set with inability to capture sufficiently important factors such as frailty, identification of subgroups of patients appropriate for limited resection will be important [19, 20]. This project is funded as a sub-aim to analyze sublobar and lobar resection in propensity-matched patients. Also of interest is thoracoscopic approach. Using this approach is significant in the short-term model (HR 1.51 for thoracotomy), and it remains significant in the present analysis (HR 1.16 for thoracotomy). This persists when only clinical stage I patients are considered in the model (HR 1.14 for thoracotomy). Although the reasons for short-term benefit are well known, explanations for longterm survival advantage are less clear. As with procedure choice, approach choice is complex and involves subtle decision-making that may not be captured adequately in the STS database. Thus, unmeasured confounders may drive this finding. Perhaps collection of central/ peripheral location and subsequent inclusion in the model would render the difference nil. Also, perhaps thoracoscopic lobectomy results in lower prometastatic/ inflammatory cytokines that impact long-term cancerspecific mortality [21, 22]. Another explanation is that patients who have access to thoracoscopic resection techniques also have access to better comprehensive cancer care and health care in general. A more detailed analysis of VATS versus open approach is forthcoming. Five-year survival stratified by pathologic stage in the STS-CMS data set is slightly worse than that in the latest International Association for the Study of Lung Cancer (IASLC) Lung Cancer Staging Project and comparable to that of the National Cancer Database (NCDB). Five-year
Approach VATS VATS VATS Open Open VATS VATS VATS VATS VATS VATS VATS
M ¼ male;
Predicted 5-Year Survival (%) (95% CI) 79.3 64.4 84.5 76.5 60.1 54.5 77.5 61.5 84.2 74.7 72.6 54.3
(77.4, (61.0, (83.0, (74.3, (56.7, (50.7, (74.5, (57.2, (82.5, (72.2, (69.0, (49.2,
open ¼ open lobectomy;
81.3) 67.9) 86.0) 78.7) 63.6) 58.7) 80.5) 66.2) 86.0) 77.4) 76.3) 59.9)
VATS ¼ video-
survival based on pathologic stage in the IASLC project was as follows: IA, 85%; IB, 73%; IIA, 65%; IIB, 56%; IIIA, 41%; IIIB, 24%. In the NCDB, 5-year survival was as follows: IA, 70%; IB, 60%; IIA, 55%; IIB 47%; IIIA, 38%; IIIB 24%. Using STS-CMS data, pathologic stage-specific 5-year survival was observed to be as follows: I, 59.7%; II, 40.7%; III, 29.9%; and IV, 26.7. Reasons for the decreased survival relative to IASLC are unclear but may relate to possible overrepresentation of Asian patients with well-differentiated tumors in IASLC and exclusion of patients younger than 65 years in the present study. The current study is limited in several important ways. First, the data are only strictly applicable to patients over age 65 years. Younger patients may have slightly different risk factors for long-term survival. The outcomes are from selected surgeons who may have better outcomes than non-STS participants. Important prognostic factors—such as differentiation status, positron emission tomography standard uptake value, and genetic/genomic alterations— are not measured. Also, interaction terms are not considered in the analysis. Finally, unmeasured confounders are present in any retrospective analysis. In conclusion, the long-term risk model created by linkage of the STS GTSD and CMS data identifies risk factors for long-term mortality. Future analyses of these data will be useful in answering oncologic questions. This project was supported by grant number R01 HS022279 from the Agency for Healthcare Research and Quality. The content is solely the responsibility of the authors and does not necessarily represent the official views of the Agency for Healthcare Research and Quality.
Audio Discussion: Audio of the discussion that followed the presentation of this paper at the STS Annual Meeting can be accessed in the online version of this article [http://dx.doi.org/10.1016/j.athoracsur. 2017.06.071] on http://www.annalsthoracicsurgery.org.
CHAMBERLAIN MEMORIAL PAPER ONAITIS ET AL LONG-TERM GTSD SURVIVAL MODEL
References 1. Fernandez F, Furnary A, Kosinski A, et al. Longitudinal follow-up of lung cancer resection from the society of thoracic surgeons general thoracic surgery database in patients 65 years and older. Ann Thorac Surg 2016;101:2067–76. 2. Goldstraw P, Chansky K, Crowley J, et al. The IASLC Lung Cancer Staging Project: Proposals for revision of the TNM stage groupings in the forthcoming edition of the TNM classification for lung cancer. J Thorac Oncol 2016;11:39–51. 3. Khullar O, Gillespie T, Nickleach D, et al. Socioeconomic risk factors for long-term mortality after pulmonary resection for lung cancer: an analysis of more than 90,000 patients from the national cancer data base. J Am Coll Surg 2015;220: 156–68.e4. 4. 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 2003. Cancer 2007;110:1532–41. 5. Jacobs JP, Edwards F, Shahian D, et al. Successful linking of the Society of Thoracic Surgeons Adult Cardiac Surgery Database to Centers for Medicare and Medicaid Services Medicare data. Ann Thorac Surg 2010;90:1150–6; discussion 1156-7. 6. Dokholyan R, Muhlbaier L, Falletta J, et al. Regulatory and ethical considerations for linking clinical and administrative databases. Am Heart J 2009;157:971–82. 7. Lee EW, Wei LJ, Amato DA. Cox-type regression analysis for large numbers of small groups of correlated failure time observations. In: Klein JP, Goel PK (eds). Survival Analysis: State of the Art. Dordrecht, Netherlands: Kluwer Academic Publishers; 1992:237–47. 8. Boffa D, Kosinski A, Paul S, Mitchell J, Onaitis M. Lymph node evaluation by open or video-assisted approaches in 11, 500 anatomic lung cancer resections. Ann Thorac Surg 2012;94:347–53; discussion 353. 9. McMillan R, Berger A, Sima C, et al. Thirty-day mortality underestimates the risk of early death after major resections for thoracic malignancies. Ann Thorac Surg 2014;98:1769–74; discussion 1774-5. 10. Fernandez F, Kosinski A, Burfeind W, et al. The society of thoracic surgeons lung cancer resection risk model: higher quality data and superior outcomes. Ann Thorac Surg 2016;102:370–7.
Ann Thorac Surg 2017;-:-–-
11. Kozower B, O’Brien S, Kosinski A, et al. The Society of Thoracic Surgeons composite score for rating program performance for lobectomy for lung cancer. Ann Thorac Surg 2016;101:1379–86; discussion 1386-7. 12. Tong B, Kosinski A, Burfeind W Jr, et al. Sex differences in early outcomes after lung cancer resection: analysis of the Society of Thoracic Surgeons general thoracic database. J Thorac Cardiovasc Surg 2014;148:13–8. 13. Nash S, Liao L, Harris T, Freedman N. Cigarette smoking and mortality in adults aged 70 years and older: results from the NIH-AARP cohort. Am J Prev Med Nov 2016;52:276–83. 14. Parsons A, Daley A, Begh R, Aveyard P. Influence of smoking cessation after diagnosis of early stage lung cancer on prognosis: systematic review of observational studies with meta-analysis. BMJ 2010;340:b5569. 15. Tanner N, Kanodra N, Gebregziabher M, et al. The association between smoking abstinence and mortality in the national lung screening trial. Am J Respir Crit Care Med 2016;193:534–41. 16. Williams T, Gulack B, Kim S, Fernandez F, Ferguson M. Operative risk for major lung resection increases at extremes of body mass index. Ann Thorac Surg 2017;103:296–302. 17. Linden P, D’Amico T, Perry Y, Saha-Chaudhuri P, Sheng S, Kim S, Onaitis M. Quantifying the safety benefits of wedge resection: a Society of Thoracic Surgery database propensitymatched analysis. Ann Thorac Surg 2014;98:1705–11; discussion 1711-2. 18. Altorki N, Kohman L, Veit L, You YN, Boughey J. Limited resection as a cure for early lung cancer: time to challenge the gold standard? Bull Am Coll Surg 2015;100:57–8. 19. Altorki N, Kamel M, Narula N, et al. Anatomical segmentectomy and wedge resections are associated with comparable outcomes for patients with small ct1n0 non-small cell lung cancer. J Thorac Oncol 2016;11:1984–92. 20. Dai C, Shen J, Ren Y, et al. Choice of surgical procedure for patients with non-small-cell lung cancer aL’d’ 1 cm or > 1 to 2 cm among lobectomy, segmentectomy, and wedge resection: a population-based study. J Clin Oncol 2016;34:3175–82. 21. Ng C, Wan I, Yim A. Impact of video-assisted thoracoscopic major lung resection on immune function. Asian Cardiovasc Thorac Ann 2009;17:426–32. 22. Zhang LB, Wang B, Wang XY, Zhang L. Influence of videoassisted thoracoscopic lobectomy on immunological functions in non-small cell lung cancer patients. Med Oncol 2015;32:201. ˇ
8
Prediction of Long-Term Survival After Lung Cancer Surgery for Elderly Patients in The Society of Thoracic Surgeons General Thoracic Surgery Database Mark W. Onaitis, MD, Anthony P. Furnary, MD, Andrzej S. Kosinski, PhD, Sunghee Kim, PhD, Daniel Boffa, MD, Betty C. Tong, MD, Patricia Cowper, PhD, Jeffrey P. Jacobs, MD, Cameron D. Wright, MD, Joe B. Putnam, Jr, MD, and Felix G. Fernandez, MD, MSc Department of Surgery, University of California-San Diego, La Jolla, California; Department of Surgery, Starr-Wood Cardiac Group, Portland, Oregon; Duke Clinical Research Institute, Duke University, Durham, North Carolina; Department of Surgery, Yale University, New Haven, Connecticut; Department of Surgery, Johns Hopkins All Children’s Heart Institute, St. Petersburg, Florida; Department of Surgery, Massachusetts General Hospital, Boston, Massachusetts; Department of Surgery, Baptist MD Anderson Cancer Center, Jacksonville, Florida; and Department of Surgery, Emory University, Atlanta, Georgia
Background. Prior risk models using the STS General Thoracic Surgery database (STS-GTSD) have been limited to 30-day outcomes. We have now linked STS data to Medicare data and sought to create a risk prediction model for long-term mortality after lung cancer resection in patients older than 65 years. Methods. The STS-GTSD was linked to Medicare data for lung cancer resections from 2002 to 2013 as previously reported. Successful linkage was performed in 29,899 lung cancer resection patients. Cox proportional hazards modeling was used to create a long-term survival model. Variable selection was performed using statistically significant univariate factors and known clinical predictors of outcome. Calibration was assessed by dividing the cohort into deciles of predicted survival and discrimination assessed with a C-statistic corrected for optimism via 1,000 bootstrap replications. Results. Median age was 73 years (interquartile range, 68 to 78 years), and 48% of the patients were male. Of the 29,094 patients with nonmissing pathologic stage, 69% were stage I, 18% stage II, 11% stage III, and 2% stage IV. Procedure performed was lobectomy in 69%, bilobectomy in 3%, pneumonectomy in 3%, segmentectomy in 7%,
sleeve lobectomy in 1%, and wedge resection in 17%. Thoracoscopic approach was performed in 47% of resections. The final Cox model reveals that stage and age are the strongest predictors of long-term survival. Even after controlling for stage, wedge resection, segmentectomy, bilobectomy, and pneumonectomy are all associated with increased hazard of death in comparison with lobectomy. Thoracoscopic approach is associated with improved long-term survival in comparison with thoracotomy. Other modifiable predictive factors include smoking and low body mass index. Calibration of the model demonstrates excellent performance across all survival deciles and a C-statistic of 0.694. Conclusions. The STS-GTSD-Medicare long-term risk model includes several novel factors associated with mortality. Although medical factors predict long-term survival, age and stage are the strong predictors. Despite this, procedure choice and thoracoscopic/open approach are potentially modifiable predictors of longterm survival after lung cancer resection.
T
patient records in the STS-GTSD were recently linked to data from the Centers for Medicare and Medicaid Services (CMS) [1]. The resulting data set is a unique resource permitting analysis of oncologic outcomes.
he STS General Thoracic Surgery Database (GTSD) has identified predictors of short-term (30-day) outcomes. To analyze longer-term outcomes, lung cancer
(Ann Thorac Surg 2017;-:-–-) Ó 2017 by The Society of Thoracic Surgeons
Accepted for publication June 28, 2017. Presented at the Fifty-third Annual Meeting of The Society of Thoracic Surgeons, Houston, TX, Jan 21-25, 2017. Winner of the J. Maxwell Chamberlain Memorial Award for General Thoracic Surgery. Address correspondence to Dr Onaitis, Department of Surgery, University of California-San Diego, 9300 Campus Point Dr, Mailcode 7892, LaJolla, CA 92037; email: [email protected].
Ó 2017 by The Society of Thoracic Surgeons Published by Elsevier Inc.
The Supplemental Figures can be viewed in the online version of this article [http://dx.doi.org/10.1016/ j.athoracsur.2017.06.071] on http://www.annalsthoracic surgery.org.
0003-4975/$36.00 http://dx.doi.org/10.1016/j.athoracsur.2017.06.071
2
CHAMBERLAIN MEMORIAL PAPER ONAITIS ET AL LONG-TERM GTSD SURVIVAL MODEL
Improved methods of data sharing and statistical analysis have allowed refinement of the American Joint Committee on Cancer staging system. These efforts have confirmed the importance of stage as well as tumor size and location [2]. Our previous report also demonstrated that age and sex are significant predictors [1]. Others have used national databases to define factors that predict survival after lung cancer resection [3, 4]. Here, we sought to create a long-term survival model for lung cancer patients aged over 65 years who underwent resection in STS-GTSD hospitals.
Patients and Methods Medicare Database of CMS The data source for this study is the 100% Medicare inpatient claims file, which contains information on hospitalizations of patients enrolled in fee-for-service Medicare. The Standard Analytic Files were used. These include dates of service and diagnostic codes from the International Classification of Diseases, Ninth Revision, Clinical Modification. The database contains anonymous patient identifiers that enable follow up of beneficiaries over time. The 100% Medicare denominator file, which links to the inpatient file and contains information on beneficiary eligibility, demographic characteristics, and date of death, was also used.
Linkage The STS-GTSD was linked to Medicare data for lung cancer resections from 2002 to 2013 [1]. Briefly, the STS GTSD was linked to CMS claims files using combinations of nonunique indirect identifiers through a deterministic matching algorithm [5, 6]. Once the individual patients were linked, longitudinal records were created containing follow-up information, including subsequent death. Successful linkage was performed in 29,899 lung cancer resection patients.
Variables The following independent variables were considered in the multivariable analyses: age, sex, body mass index (BMI, in kg/m2), American Society of Anesthesiologists (ASA) class, Zubrod score, presence of coronary artery disease (CAD), presence of cardiovascular disease (CVD), congestive heart failure (CHF), hypertension, diabetes mellitus (DM), steroid use, peripheral vascular disease (PVD), renal insufficiency (creatinine > 2 or hemodialysis [HD]), FEV1% predicted, smoking status, prior thoracic surgery, approach, and procedure. Urgent/emergent procedures were excluded. Pathologic stage was used instead of clinical stage as there was a high percentage of missingness for the clinical stage variable (9.4%) as opposed to only 2.7% for the pathologic stage. DLCO also was not included because of missingness.
Statistical Analyses Unadjusted survival was plotted using the Kaplan-Meier method. Cox modeling was used to create a long-term survival model. Clustering within hospitals was accounted for by consideration of robust sandwich covariance matrix estimates [7]. Variable selection was performed
Ann Thorac Surg 2017;-:-–-
using statistically significant univariate factors and known clinical predictors of outcome. Violations of the proportional hazard assumption were assessed graphically. No violations were detected. No meaningful collinearity between covariates was present—all the condition indices were less than 16, with 88% of them less than 5; thus all the condition indices were substantially lower than the commonly accepted threshold of 30. Calibration was assessed by a calibration slope, as well as by plot at 5 years— that is, by dividing the cohort into deciles of predicted survival and comparing the decile specific observed survival at 5 years (Kaplan-Meier) with the average predicted 5-year survival within the corresponding deciles. Calibration slope and C-statistic were both corrected for optimism via 1,000 bootstrap replications. As an alternative optimism correction of the C-statistic, we also randomly split data into equally sized training and testing samples, applied model coefficients based on the training sample to the testing sample data, and subsequently computed a testing sample C-statistic; repetition of this process 1,000 times resulted in the reported average testing sample C-statistic.
Results The STS-GTSD linkage resulted in 29,899 patients. Table 1 presents the demographics. Median age was 73 years (range, 68 to 78 years], and 48% of the patients were male. At pathologic analysis, 69% were stage I, 18% stage II, 11% stage III, and 2% stage IV. Procedure performed was lobectomy in 69%, bilobectomy in 3%, pneumonectomy in 3%, segmentectomy in 7%, sleeve lobectomy in 1%, and wedge resection in 17%. Thoracoscopic approach was performed in 47% of resections. Figure 1 depicts overall survival. The 1-, 3-, and 5-year survival proportions of the entire study population are 86.7%, 67.5%, and 53.2%, respectively. The 90-day mortality is 4.5%, and the 180-day mortality is 7.3%. Cox proportional hazards survival analysis was then performed. Table 2 demonstrates the models. In the multivariable model, increasing age, increasing stage, male sex, BMI less than 18.5, increasing ASA and Zubrod score, CAD, CVD, CHF, DM, steroid use, PVD, acute kidney injury, decreasing FEV1% predicted, and smoking history predict mortality. The calibration and discrimination of the model were assessed. Calibration across deciles of predicted survival was excellent (calibration plot, Supplemental Fig 1) and the calibration slope is 0.987. The C-statistic of the linked model is 0.696 (95% confidence interval, 0.689 to 0.702), the optimism corrected via bootstrap C-statistic is 0.694, and the average testing sample C-statistic is 0.693 (range over 1,000 data splits, 0.685 to 0.704) (Supplemental Fig 2). Figure 2 demonstrates survival curves for three predictor variables using Kaplan-Meier and Cox model prediction. Figure 2A shows a stepwise decrement in long-term survival as FEV1% predicted decreases. Similarly, current smoking is associated with decreased longterm survival (Fig 2B). Procedure performed is predictive. Figure 2C demonstrates that lobectomy patients perform significantly better independent of stage than patients undergoing
Ann Thorac Surg 2017;-:-–-
Table 1. Continued
Table 1. Patient Demographics Characteristic Age (years) 65–69 70–74 75–79 80þ Sex Male Female Body mass index <18.5 18.5–25 25–30 30–35 >35 ASA class I–II III IV–V Zubrod score 0 1 2–5 Coronary artery disease Cardiovascular disease Congestive heart failure Diabetes mellitus Steroid use Peripheral vascular disease Acute kidney injury FEV1 % predicted >80 60–80 40–60 <40 Pathologic stage I II III IV Clinical stage I II III IV Cigarette use Never Past Current Reoperation Procedure Lobectomy Bilobectomy Pneumonectomy
3
CHAMBERLAIN MEMORIAL PAPER ONAITIS ET AL LONG-TERM GTSD SURVIVAL MODEL
n (%)
8,671 8,886 7,147 5,195
(29.0) (29.7) (23.9) (17.4)
14,412 (48.2) 15,474 (51.8) 794 9,568 12,323 4,961 2,253
(2.7) (32.0) (41.2) (16.6) (7.5)
3,885 (13.1) 22,281 (75.4) 3,391 (11.5) 12,443 15,451 1,700 7,959 3,000 1,152 5,677 1,036 3,599 697
(42.0) (52.2) (5.7) (26.6) (10.0) (3.9) (19.0) (3.5) (12.0) (2.3)
13,613 8,637 3,910 842
(50.4) (32.0) (14.5) (3.1)
20,038 5,155 3,311 590
(68.9) (17.7) (11.4) (2.0)
20,635 3,588 2,366 487
(76.2) (13.3) (8.7) (1.8)
3,819 20,645 5,435 1,624
(12.8) (69.0) (18.2) (5.4)
Characteristic Segmentectomy Sleeve lobectomy Wedge resection Video-assisted thoracoscopic surgery
n (%) 2,079 282 5,203 14,182
ASA ¼ American Society of Anesthesiologists; FEV1 ¼ forced expiratory volume, 1 second.
other procedures. At 5 years, survival after lobectomy is 56.5%, whereas pneumonectomy survival is 34.3% and segmentectomy survival is 52.5%. In addition to procedure, thoracoscopic approach is associated with improved long-term survival (Fig 3). Given that a previous analysis documented lower N1 upstaging in clinical N0 patients undergoing thoracoscopic as compared to open lobectomy [8], we assessed the predictors of survival in clinical stage I patients in the linked data set. Unadjusted upstaging to stage II was 13.9% in the open group and 8.9% in the video-assisted thoracoscopic surgery (VATS) group and to upstaging to stage III was 7.0% in the open group and 4.8% in the VATS group. In the Cox survival model, as Table 3 demonstrates, in the 20,635 patients who were clinical stage I patients, all of the same variables predict long-term survival. In order to more thoroughly examine the effect of individual variables on long-term survival, the effects of modifying key prognostic factors are examined in Table 4. For the purposes of illustration, 5-year survival of patients with the listed predictor variables was obtained from the Cox model in Table 2 by multiplying particular patient covariate values by estimated model coefficients. For simplicity, medical comorbidities were set to “0,” ASA/ Zubrod score was set to the lowest categories, and BMI was set to 20. These data demonstrate the importance of age and stage in comparison with procedural variables.
Comment We have created a long-term risk model for non-small cell lung cancer (NSCLC) resection patients aged over 65 years in the STS GTDB. The resulting model allows assessment of survival after the index hospitalization,
20,578 (68.8) 842 (2.8) 915 (3.1) (Continued)
(7.0) (0.9) (17.4) (47.4)
Fig 1. Overall survival of the entire cohort, Kaplan-Meier.
4
CHAMBERLAIN MEMORIAL PAPER ONAITIS ET AL LONG-TERM GTSD SURVIVAL MODEL
Ann Thorac Surg 2017;-:-–-
Table 2. Results of Cox Models for Long-Term Overall Survival Characteristic
Univariate HR (95% CI)
Age group (years) 65–69 70–74 75–79 80þ Sex Male Female Body mass index 18.5–25 <18.5 25–30 30–35 >35 ASA group I–II III IV–V Zubrod score 0 1 2–5 Coronary artery disease Cardiovascular disease Congestive heart failure Diabetes mellitus Steroid use Peripheral vascular disease Chronic kidney disease % FEV1 predicted >80 60–80 40–60 <40 Pathologic stage I II III IV Smoking Never Past Current Reoperation Procedure Lobectomy Bilobectomy Pneumonectomy Segmentectomy Sleeve Wedge Video-assisted thoracoscopic surgery approach ASA ¼ American Society of Anesthesiologists;
p
Multivariable HR (95% CI)
<0.001 1.00 1.17 (1.11, 1.24) 1.35 (1.28, 1.44) 1.72 (1.62, 1.82)
<0.001 1.00 1.19 (1.12, 1.27) 1.40 (1.31, 1.50) 1.90 (1.78, 2.03)
<0.001 1.00 0.66 (0.63, 0.68)
<0.001 1.00 0.73 (0.70, 0.76)
<0.001 1.58 0.93 0.87 0.88
1.00 (1.41, (0.87, (0.82, (0.82,
1.78) 1.00) 0.92) 0.96)
<0.001 1.58 0.87 0.82 0.87
1.00 (1.41, (0.82, (0.77, (0.80,
1.78) 0.93) 0.87) 0.95)
<0.001 1.00 1.53 (1.40, 1.68) 2.13 (1.86, 2.43)
<0.001 1.00 1.23 (1.14, 1.33) 1.37 (1.21, 1.54)
<0.001 1.41 2.33 1.40 1.35 1.75 1.23 1.47 1.51 1.93
1.00 (1.33, (2.12, (1.34, (1.28, (1.61, (1.17, (1.34, (1.43, (1.72,
1.49) 2.57) 1.47) 1.44) 1.92) 1.29) 1.63) 1.61) 2.17)
<0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001
1.00 1.32 (1.26, 1.39) 1.48 (1.40, 1.57) 1.60 (1.40, 1.83)
<0.001 1.26 1.69 1.09 1.18 1.34 1.12 1.33 1.21 1.57
1.00 (1.19, (1.50, (1.04, (1.10, (1.22, (1.07, (1.19, (1.13, (1.40,
1.33) 1.91) 1.15) 1.26) 1.48) 1.18) 1.49) 1.29) 1.76)
<0.001
<0.001 1.00 1.80 (1.69, 1.91) 2.62 (2.43, 2.81) 2.86 (2.45, 3.35)
<0.001
1.65 2.26 1.08 1.46 1.29 0.75
1.00 (1.47, (2.07, (0.95, (1.21, (1.21, (0.71,
CI ¼ confidence interval;
1.85) 2.48) 1.23) 1.76) 1.38) 0.80)
<0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001
1.00 1.17 (1.11, 1.24) 1.26 (1.18, 1.35) 1.42 (1.25, 1.61)
1.00 1.85 (1.75, 1.97) 2.58 (2.43, 2.73) 2.86 (2.49, 3.27) 1.00 1.58 (1.46, 1.71) 1.68 (1.53, 1.84) 1.38 (1.27, 1.50)
p
<0.001 <0.001
<0.001
<0.001 1.00 1.35 (1.23, 1.48) 1.54 (1.38, 1.71) 1.13 (1.03, 1.24)
1.30 1.58 1.10 0.93 1.22 0.86
1.00 (1.15, (1.40, (0.99, (0.75, (1.14, (0.82,
FEV1 ¼ forced expiratory volume, 1 second;
1.46) 1.80) 1.23) 1.14) 1.30) 0.92)
0.012 <0.001
<0.001
HR ¼ hazard ratio.
Ann Thorac Surg 2017;-:-–-
CHAMBERLAIN MEMORIAL PAPER ONAITIS ET AL LONG-TERM GTSD SURVIVAL MODEL
5
Fig 3. Univariate Kaplan-Meier and multivariable predicted survival based on approach. (Open ¼ open lobectomy; VATS ¼ video-assisted thoracoscopic surgery.)
Fig 2. Univariate Kaplan-Meier and multivariable predicted survival based on (A) FEV1 % predicted, (B) smoking status, and (C) procedure. FEV1 ¼ forced expiratory volume, 1 second.
which is important because studies with high 90-day mortality have recently been published [9]. The previous STS short-term lung cancer resection risk model identified several factors as predictive of 30-day survival: age, male sex, FEV1%, body mass index, cerebrovascular disease, steroids, coronary artery disease, peripheral vascular disease, renal dysfunction, Zubrod score, ASA rating, thoracotomy approach, induction therapy, reoperation, tumor stage, and greater extent of resection [10, 11]. Induction therapy was not included in the current model because of its association with stage. Interestingly, all of these factors were also predictive in the linked, long-term model.
The model sheds light on several variables that contribute to long-term survival. Age is obviously a significant predictor in any long-term survival analysis. The average life expectancy in the United States is 78.8 years. Other analyses have identified age as a strong predictor of long-term survival in lung cancer patients [3, 4]. In addition, previous studies have also demonstrated that female sex is protective in survival analyses [3, 12], and this is confirmed in our study. Pathologic stage strongly predicts survival. Increases in stage are associated with a larger hazard ratios for death than any medical factor analyzed. These results emphasize the need for thorough assessment of lymph nodes and highlight the need for improvements in adjuvant therapies. All of the medical factors are associated with survival in this series with a large n. Although CHF, steroid use, and creatinine greater than 2/HD are the strongest chronic medical predictors of long-term survival, none of these factors is as impactful as stage. Future analyses including patients undergoing stereotactic body radiation therapy and ablative therapy will be helpful in individualizing therapies. Increasing ASA/Zubrod scores were predictive of death. The majority of patients were ASA III and Zubrod I, and the hazard ratio for death increases sharply above these values. These scores are useful in predicting long-term survival of these patients and may be used in addition to oncologic factors in planning treatment strategies. One modifiable risk factor that significantly impacts long-term survival in this data set is smoking. A recent NIH-AARP cohort study in more than 160,000 patients over age 70 years demonstrated decreased hazard ratios for allcause mortality for former smokers as compared with current smokers [13]. A recent meta-analysis demonstrated worse survival in early-stage NSCLC continuing smokers [14]. Our study is much smaller and demonstrates survival benefit for never smokers and a smaller benefit for former smokers. Smoking cessation is clearly recommended to decrease perioperative mortality and lung cancer risk [15]. BMI is another modifiable risk factor. Short-term analysis has demonstrated increased risk of perioperative events at the extremes of BMI [16]. Underweight patients (BMI < 18.5) maintain significantly increased risk (hazard ratio [HR] 1.58) independently over time.
6
CHAMBERLAIN MEMORIAL PAPER ONAITIS ET AL LONG-TERM GTSD SURVIVAL MODEL
Ann Thorac Surg 2017;-:-–-
Table 3. Results of Cox Models for Long-Term Overall Survival in Clinical Stage I Patients Characteristic
Univariate HR (95% CI)
Age group 65–69 70–74 75–79 80þ Sex Male Female Body mass index group 18.5–25 <18.5 25–30 30–35 >35 ASA group I–II III IV–V Zubrod score 0 1 2–5 Coronary artery disease Cardiovascular disease Congestive heart failure Diabetes mellitus Steroid use Peripheral vascular disease Chronic kidney disease % FEV1 predicted >80 60–80 40–60 <40 Pathologic stage I II III IV Smoking Never Past Current Reoperation Procedure Lobectomy Bilobectomy Pneumonectomy Segmentectomy Sleeve Wedge Video-assisted thoracoscopic surgery approach ASA ¼ American Society of Anesthesiologists;
p
Multivariable HR (95% CI)
<0.001 1.00 1.23 (1.14, 1.32) 1.45 (1.35, 1.56) 1.93 (1.80, 2.07)
<0.001 1.00 1.23 (1.14, 1.33) 1.47 (1.35, 1.59) 2.07 (1.91, 2.25)
<0.001 1.00 0.66 (0.63, 0.69)
<0.001 1.00 0.72 (0.68, 0.76)
<0.001 1.67 0.91 0.89 0.91
1.00 (1.43, (0.84, (0.83, (0.83,
1.95) 0.98) 0.97) 1.01)
<0.001 1.65 0.83 0.84 0.88
1.00 (1.41, (0.77, (0.76, (0.79,
1.93) 0.89) 0.92) 0.99)
<0.001 1.00 1.61 (1.43, 1.83) 2.28 (1.93, 2.69)
<0.001 1.00 1.30 (1.17, 1.44) 1.45 (1.26, 1.67)
<0.001 1.41 2.28 1.47 1.40 1.80 1.24 1.46 1.65 2.23
1.00 (1.32, (2.04, (1.39, (1.31, (1.62, (1.16, (1.29, (1.53, (1.96,
1.50) 2.55) 1.56) 1.51) 2.01) 1.32) 1.66) 1.77) 2.53)
<0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001
1.00 1.33 (1.25, 1.42) 1.57 (1.46, 1.69) 1.63 (1.40, 1.91)
<0.001 1.29 1.70 1.07 1.16 1.36 1.13 1.33 1.26 1.62
1.00 (1.22, (1.50, (1.01, (1.07, (1.20, (1.06, (1.16, (1.16, (1.44,
1.37) 1.93) 1.14) 1.27) 1.53) 1.20) 1.53) 1.37) 1.84)
<0.001
<0.001 1.00 1.76 (1.61, 1.93) 2.47 (2.21, 2.77) 2.69 (2.01, 3.58)
<0.001
1.57 2.16 1.11 1.14 1.33 0.80
1.00 (1.33, (1.75, (0.95, (0.81, (1.23, (0.75,
CI ¼ confidence interval;
1.86) 2.67) 1.30) 1.62) 1.43) 0.85)
0.03 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001
1.00 1.17 (1.10, 1.25) 1.29 (1.20, 1.40) 1.36 (1.19, 1.55)
1.00 1.75 (1.61, 1.91) 2.35 (2.14, 2.59) 2.62 (2.07, 3.32) 1.00 1.67 (1.52, 1.85) 1.86 (1.67, 2.08) 1.44 (1.29, 1.61)
p
<0.001 <0.001
<0.001
<0.001 1.00 1.42 (1.28, 1.59) 1.73 (1.53, 1.95) 1.20 (1.05, 1.37)
1.29 1.71 1.13 0.79 1.22 0.88
1.00 (1.07, (1.33, (0.98, (0.56, (1.14, (0.83,
FEV1 ¼ forced expiratory volume, 1 second;
1.56) 2.18) 1.31) 1.12) 1.32) 0.94)
0.007 <0.001
<0.001
HR ¼ hazard ratio.
Ann Thorac Surg 2017;-:-–-
CHAMBERLAIN MEMORIAL PAPER ONAITIS ET AL LONG-TERM GTSD SURVIVAL MODEL
7
Table 4. Survival Prediction in Hypothetical Patients Age (years)
Sex
65 80 65 65 80 65 65 80 65 65 65 80
M M F M M M M M M M M M
FEV1 %
Pathologic Stage
90 90 90 90 90 90 90 90 90 50 50 50
I I I I I III I I I I I I
CI ¼ confidence interval; F ¼ female; assisted thoracoscopic surgery.
Smoking Status Past Past Past Past Past Past Past Past Never Past Past Past
Procedure Lobectomy Lobectomy Lobectomy Lobectomy Lobectomy Lobectomy Segmentectomy Segmentectomy Lobectomy Lobectomy Segmentectomy Segmentectomy
FEV1 ¼ forced expiratory volume, 1 second;
Procedure performed is predictive. Previous short-term STS database analysis in propensity matched patients has demonstrated that wedge resection is associated with lower morbidity and mortality than anatomic resection [17]. While we await results of the CALGB 140503 trial [18], this analysis identifies both wedge resection and segmentectomy as predictors of long-term mortality. Although such a finding may be related to confounding in this large retrospective data set with inability to capture sufficiently important factors such as frailty, identification of subgroups of patients appropriate for limited resection will be important [19, 20]. This project is funded as a sub-aim to analyze sublobar and lobar resection in propensity-matched patients. Also of interest is thoracoscopic approach. Using this approach is significant in the short-term model (HR 1.51 for thoracotomy), and it remains significant in the present analysis (HR 1.16 for thoracotomy). This persists when only clinical stage I patients are considered in the model (HR 1.14 for thoracotomy). Although the reasons for short-term benefit are well known, explanations for longterm survival advantage are less clear. As with procedure choice, approach choice is complex and involves subtle decision-making that may not be captured adequately in the STS database. Thus, unmeasured confounders may drive this finding. Perhaps collection of central/ peripheral location and subsequent inclusion in the model would render the difference nil. Also, perhaps thoracoscopic lobectomy results in lower prometastatic/ inflammatory cytokines that impact long-term cancerspecific mortality [21, 22]. Another explanation is that patients who have access to thoracoscopic resection techniques also have access to better comprehensive cancer care and health care in general. A more detailed analysis of VATS versus open approach is forthcoming. Five-year survival stratified by pathologic stage in the STS-CMS data set is slightly worse than that in the latest International Association for the Study of Lung Cancer (IASLC) Lung Cancer Staging Project and comparable to that of the National Cancer Database (NCDB). Five-year
Approach VATS VATS VATS Open Open VATS VATS VATS VATS VATS VATS VATS
M ¼ male;
Predicted 5-Year Survival (%) (95% CI) 79.3 64.4 84.5 76.5 60.1 54.5 77.5 61.5 84.2 74.7 72.6 54.3
(77.4, (61.0, (83.0, (74.3, (56.7, (50.7, (74.5, (57.2, (82.5, (72.2, (69.0, (49.2,
open ¼ open lobectomy;
81.3) 67.9) 86.0) 78.7) 63.6) 58.7) 80.5) 66.2) 86.0) 77.4) 76.3) 59.9)
VATS ¼ video-
survival based on pathologic stage in the IASLC project was as follows: IA, 85%; IB, 73%; IIA, 65%; IIB, 56%; IIIA, 41%; IIIB, 24%. In the NCDB, 5-year survival was as follows: IA, 70%; IB, 60%; IIA, 55%; IIB 47%; IIIA, 38%; IIIB 24%. Using STS-CMS data, pathologic stage-specific 5-year survival was observed to be as follows: I, 59.7%; II, 40.7%; III, 29.9%; and IV, 26.7. Reasons for the decreased survival relative to IASLC are unclear but may relate to possible overrepresentation of Asian patients with well-differentiated tumors in IASLC and exclusion of patients younger than 65 years in the present study. The current study is limited in several important ways. First, the data are only strictly applicable to patients over age 65 years. Younger patients may have slightly different risk factors for long-term survival. The outcomes are from selected surgeons who may have better outcomes than non-STS participants. Important prognostic factors—such as differentiation status, positron emission tomography standard uptake value, and genetic/genomic alterations— are not measured. Also, interaction terms are not considered in the analysis. Finally, unmeasured confounders are present in any retrospective analysis. In conclusion, the long-term risk model created by linkage of the STS GTSD and CMS data identifies risk factors for long-term mortality. Future analyses of these data will be useful in answering oncologic questions. This project was supported by grant number R01 HS022279 from the Agency for Healthcare Research and Quality. The content is solely the responsibility of the authors and does not necessarily represent the official views of the Agency for Healthcare Research and Quality.
Audio Discussion: Audio of the discussion that followed the presentation of this paper at the STS Annual Meeting can be accessed in the online version of this article [http://dx.doi.org/10.1016/j.athoracsur. 2017.06.071] on http://www.annalsthoracicsurgery.org.
CHAMBERLAIN MEMORIAL PAPER ONAITIS ET AL LONG-TERM GTSD SURVIVAL MODEL
References 1. Fernandez F, Furnary A, Kosinski A, et al. Longitudinal follow-up of lung cancer resection from the society of thoracic surgeons general thoracic surgery database in patients 65 years and older. Ann Thorac Surg 2016;101:2067–76. 2. Goldstraw P, Chansky K, Crowley J, et al. The IASLC Lung Cancer Staging Project: Proposals for revision of the TNM stage groupings in the forthcoming edition of the TNM classification for lung cancer. J Thorac Oncol 2016;11:39–51. 3. Khullar O, Gillespie T, Nickleach D, et al. Socioeconomic risk factors for long-term mortality after pulmonary resection for lung cancer: an analysis of more than 90,000 patients from the national cancer data base. J Am Coll Surg 2015;220: 156–68.e4. 4. 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 2003. Cancer 2007;110:1532–41. 5. Jacobs JP, Edwards F, Shahian D, et al. Successful linking of the Society of Thoracic Surgeons Adult Cardiac Surgery Database to Centers for Medicare and Medicaid Services Medicare data. Ann Thorac Surg 2010;90:1150–6; discussion 1156-7. 6. Dokholyan R, Muhlbaier L, Falletta J, et al. Regulatory and ethical considerations for linking clinical and administrative databases. Am Heart J 2009;157:971–82. 7. Lee EW, Wei LJ, Amato DA. Cox-type regression analysis for large numbers of small groups of correlated failure time observations. In: Klein JP, Goel PK (eds). Survival Analysis: State of the Art. Dordrecht, Netherlands: Kluwer Academic Publishers; 1992:237–47. 8. Boffa D, Kosinski A, Paul S, Mitchell J, Onaitis M. Lymph node evaluation by open or video-assisted approaches in 11, 500 anatomic lung cancer resections. Ann Thorac Surg 2012;94:347–53; discussion 353. 9. McMillan R, Berger A, Sima C, et al. Thirty-day mortality underestimates the risk of early death after major resections for thoracic malignancies. Ann Thorac Surg 2014;98:1769–74; discussion 1774-5. 10. Fernandez F, Kosinski A, Burfeind W, et al. The society of thoracic surgeons lung cancer resection risk model: higher quality data and superior outcomes. Ann Thorac Surg 2016;102:370–7.
Ann Thorac Surg 2017;-:-–-
11. Kozower B, O’Brien S, Kosinski A, et al. The Society of Thoracic Surgeons composite score for rating program performance for lobectomy for lung cancer. Ann Thorac Surg 2016;101:1379–86; discussion 1386-7. 12. Tong B, Kosinski A, Burfeind W Jr, et al. Sex differences in early outcomes after lung cancer resection: analysis of the Society of Thoracic Surgeons general thoracic database. J Thorac Cardiovasc Surg 2014;148:13–8. 13. Nash S, Liao L, Harris T, Freedman N. Cigarette smoking and mortality in adults aged 70 years and older: results from the NIH-AARP cohort. Am J Prev Med Nov 2016;52:276–83. 14. Parsons A, Daley A, Begh R, Aveyard P. Influence of smoking cessation after diagnosis of early stage lung cancer on prognosis: systematic review of observational studies with meta-analysis. BMJ 2010;340:b5569. 15. Tanner N, Kanodra N, Gebregziabher M, et al. The association between smoking abstinence and mortality in the national lung screening trial. Am J Respir Crit Care Med 2016;193:534–41. 16. Williams T, Gulack B, Kim S, Fernandez F, Ferguson M. Operative risk for major lung resection increases at extremes of body mass index. Ann Thorac Surg 2017;103:296–302. 17. Linden P, D’Amico T, Perry Y, Saha-Chaudhuri P, Sheng S, Kim S, Onaitis M. Quantifying the safety benefits of wedge resection: a Society of Thoracic Surgery database propensitymatched analysis. Ann Thorac Surg 2014;98:1705–11; discussion 1711-2. 18. Altorki N, Kohman L, Veit L, You YN, Boughey J. Limited resection as a cure for early lung cancer: time to challenge the gold standard? Bull Am Coll Surg 2015;100:57–8. 19. Altorki N, Kamel M, Narula N, et al. Anatomical segmentectomy and wedge resections are associated with comparable outcomes for patients with small ct1n0 non-small cell lung cancer. J Thorac Oncol 2016;11:1984–92. 20. Dai C, Shen J, Ren Y, et al. Choice of surgical procedure for patients with non-small-cell lung cancer aL’d’ 1 cm or > 1 to 2 cm among lobectomy, segmentectomy, and wedge resection: a population-based study. J Clin Oncol 2016;34:3175–82. 21. Ng C, Wan I, Yim A. Impact of video-assisted thoracoscopic major lung resection on immune function. Asian Cardiovasc Thorac Ann 2009;17:426–32. 22. Zhang LB, Wang B, Wang XY, Zhang L. Influence of videoassisted thoracoscopic lobectomy on immunological functions in non-small cell lung cancer patients. Med Oncol 2015;32:201. ˇ
8