Radiation Therapy is Independently Associated with Worse Survival After R0-Resection for Stage I–II Non-small Cell Lung Cancer: An Analysis of the National Cancer Data Base

Abstract

Background.

The 1998 post-operative radiotherapy meta-analysis for lung cancer showed a survival detriment associated with radiation for stage I–II resected non-small cell lung cancer (NSCLC), but has been criticized for including antiquated radiation techniques. We analyzed the National Cancer Database (NCDB) to determine the impact of radiation after margin-negative (R0) resection for stage I–II NSCLC on survival.

Methods.

Adult patients from 2004 to 2014 were analyzed from the NCDB with respect to receiving radiation as part of their first course of treatment for resected stage I–II NSCLC; the primary outcome measure was overall survival.

Results.

A total of 197,969 patients underwent R0 resection for stage I–II NSCLC, and 4613 received radiation. Median radiation dose was 55 Gy with a 50–60 Gy interquartile range. On adjusted analysis, treatment at a community cancer program, sublobectomy, tumor size (3–7 cm), and pN1/Nx were associated with receiving radiation (odds ratio > 1, p < 0.05). The irradiated group had shorter median survival (45.8 vs. 77.5 months, p < 0.001), and radiation was independently associated with worse overall survival (hazard ratio (HR) 1.339, 95% confidence interval (CI) 1.282–1.399). After propensity score matching, radiation remained associated with worse overall survival (HR 1.313, 95% CI 1.237–1.394, p < 0.001).

Conclusions.

Radiotherapy was independently associated with worse survival after R0 resection of stage I–II NSCLC in the NCDB and was more likely to be delivered in community cancer programs.

Lung cancer is the leading cause of cancer mortality in the United States, with over 200,000 diagnoses expected in 2016.¹ For medically operable early-stage non-small cell lung cancer (NSCLC), surgical resection with mediastinal nodal staging is the standard of care.^2,3 For American Joint Committee on Cancer (AJCC), 7th edition, stage I–II, pT1–T2 N0–N1 disease, adjuvant treatment with radiotherapy (RT) is considered for positive surgical margins. However, for R0 resection without mediastinal involvement, RT is not recommended in the management of early-stage NSCLC.

Surgical resections achieve negative margins for the vast majority (95%) of stage I–III NSCLC.^4–6 With a lobectomy and mediastinal nodal dissection for early-stage NSCLC with negative surgical margins, local control rates are high without adjuvant RT. In an early trial comparing segmentectomy with lobectomy by Ginsberg et al. local failure for lobectomy group was only 6%.⁷ Contemporary lobectomy and sublobectomy series with mediastinal nodal dissection also have high rates of locoregional control.^8,9 Thus, the absolute local control benefit potentially provided by adjuvant RT for early-stage NSCLC would be small, carrying with it the expected toxicities and morbidities of thoracic RT.

In the Post-operative Radiotherapy (PORT) in Lung Cancer Meta-analysis, adjuvant RT was associated with a statistically significant overall survival (OS) detriment for AJCC stage I–II, N0–N1 NSCLC.^10,11 These findings led to the Cochrane Review recommending against the use of adjuvant RT for early-stage NSCLC.¹² In spite of these recommendations, there remain lingering doubts about the relevance of the PORT meta-analysis because of inclusion of antiquated radiation techniques such as cobalt teletherapy, lack of computed tomography-based planning, and no radiation quality assurance.

The National Cancer Data Base (NCDB) is a unique resource that has utility in understanding practice patterns such as RT for margin-negative early-stage NSCLC. The NCDB captures 70% of all newly diagnosed cancers in the United States and 82% of all lung cancer diagnoses.¹³ The NCDB has previously been used to understand the impact of adjuvant RT for surgically resected AJCC stage II–IIIA disease from 1998 to 2006, where PORT was associated with an OS benefit only for N2 nodal disease.¹⁴ A study by Mikell et al. also demonstrated a survival benefit associated with N2 disease for surgically resected NSCLC in the NCDB from 2004 to 2006.¹⁵ Thus, there is precedent to use the NCDB as a tool to evaluate the role of adjuvant RT for early-stage NSCLC with modern radiation techniques that have evolved and improved since the 1998 PORT meta-analysis.

It is important to understand the prevalence and predictors of adjuvant RT for early-stage NSCLC in the United States. To discern these practice patterns, we analyzed the NCDB to assess prevalence of RT after R0 surgical resection for AJCC stage I–II NSCLC from 2004 to 2014, comparing the characteristics of patients treated with radiation and those who did not. We have also evaluated the impact of RT on OS.

METHODS

Adult (≥18 years) patients with NSCLC diagnosed from 2004 to 2014 were retrieved from the NCDB. The International Classification of Diseases for Oncology, 3rd edition (ICD-O-3), codes for histologic types of NSCLC were grouped into squamous cell (8052, 8070–8073, 8076, 8078, 8083, 8084 and 8094), large cell (8012, 8014, 8020, and 8021), adenocarcinoma (8050, 8051, 8140–8147, 8230, 8250–8263, 8290, 8310, 8323, 8333, 8470–8490 and 8550), adenosquamous (8560), sarcomatoid (8022, 8030–8033, and 8575) and other NSCLC (8046, 8074, 8075, 8320, 8570, and 8572), totaling 1,089,880 patients. Histologies such as carcinoid tumors, other neuroendocrine tumors, and metastatic tumors to the lung were specifically excluded.

Selection criteria included (AJCC 6th or AJCC 7th edition) T1–T2, N0–N1, M0 tumors with R0 resection measuring less than 7 cm on pathologic examination. Patients were excluded if they received either no surgical procedure of the primary site, or had resection documented with non-negative or unknown margins. Patients who received RT either before, or during surgery were also excluded. Additionally, patients who received RT directed to a location other than the lung were excluded from analysis. Surgical procedure codes were grouped into excisions less than lobectomy, lobectomy or bi-lobectomy, pneumonectomy, and unspecified or unknown type of resection. Negative margins were selected by excluding patients classified in the NCDB as having “residual tumor, NOS,” “microscopic residual tumor,” and “macroscopic visible tumor.”

Pearson Chi square and Student’s t tests were used for descriptive analysis. For cases diagnosed from 2004 to 2013, survival curves were generated using the Kaplan–Meier method and OS was compared by the log-rank test. Cases diagnosed in 2014 were excluded from survival analysis due to inadequate follow-up data. Median follow-up was estimated using the reverse Kaplan–Meier method. The endpoint of interest was OS and time to event was indexed to the time of initial diagnosis, and measured as the time, in months, to last contact or death. Multivariable analysis was conducted using Cox proportional hazard and binary logistic regression models. Clinical variables of interest were first tested on univariate analysis and included in the multivariable models only if significant. A correlation matrix was generated to evaluate for multicollinearity and no variables in the final model were allowed to exhibiting significant correlation coefficients greater than 0.5. Additionally a variance important factor was calculated for each variable interaction, and a threshold of 5 or greater was considered unacceptable. Any covariates demonstrating significant multicollinearity were excluded from the final multivariable model. To minimize bias in predicting correlates of OS, 1:1 nearest neighbor propensity score matching was performed, without replacement to determine the propensity of patients to be selected for post-operative RT. Propensity scores were calculated using a multivariable binary logistic regression model that included only variables that were significant in the multivariable Cox proportional hazard model. Balance was assessed using mean standardized differences, with a value of <0.10 considered adequate.¹⁶ All tests were two tailed and statistical significance was considered for p values of <0.05. Data analysis was performed by SPSS 24 software (IBM SPSS, Chicago, IL).

RESULTS

There were 197,969 patients identified with margin negative pathologic AJCC stage I–II T1–T2, N0–N1, M0 NSCLC identified in the NCDB from 2004 to 2014. Of these patients, 193,356 (97.7%) were observed after surgery of the primary site and 4613 (2.3%) received RT directed to the lung as part of initial treatment. From 2004 to 2014, there was a trend toward decreased use of RT for these patients. For patients treated with radiation, the median dose was 55 Gy with a 50–60 Gy interquartile range (Fig. 1).

FIG. 1.

Trends for observation and radiation use in surgically resected margin negative (R0) AJCC stage I–II NSCLC from 2004 to 2014 in NCDB. Distribution of radiation doses (Gy) indicated for irradiated group. AJCC American Joint Committee on Cancer, NCDB National Cancer Data Base, NSCLC non–small cell lung cancer

The demographic and clinicopathologic characteristics of non-irradiated and irradiated patients are shown in Table 1. Patients who received RT were more likely to be treated in a community cancer program (p < 0.001) and live closer to a facility (22.6 vs. 30 miles, mean) than patients who underwent observation (p < 0.001). The type of surgery was significantly different between the radiated and non-radiated groups (p < 0.001), with sublobectomy in 39.8% of the radiation group and 20.9% of the non-irradiated group. When looking at the pathologic characteristics of the tumors, 38.4% of the adjuvant RT group had N1 or Nx nodal staging compared to 14.0% of the observed group, as well as less lymph nodes examined (p < 0.001). After surgery, 50.9% of patients treated with RT also received chemotherapy compared to 13.9% in the group observed (p < 0.001).

TABLE 1.

Clinicopathologic and demographic characteristics of National Cancer Data Base patients with and without radiotherapy for margin-negative (R0) early-stage non-small cell lung cancer from 2004 to 2014

Characteristic	R0 resection alone (%) (n = 193,356)	Radiotherapy (%) (n = 4613)	p value
Facility type
Community cancer program	7.9	14.0	<0.001
Comprehensive community cancer program	48.4	51.1
Academic/research program	36.6	28.6
Integrated network cancer program	7.0	6.0
Other specified types of cancer programs	0.18	0.22
Facility location
East Coast	45.3	44	<0.001
East Central	26.6	32.1
West Central	14.4	12.7
Mountain	3.7	2.7
Pacific	9.9	8.4
Gender
Male	47.1	52.3	<0.001
Female	52.9	47.7
Age (years), mean	68.2	67.3	0.056
Race
White	89.7	89.3	0.007
Black	7.8	8.7
Other	2.6	2.0
Hispanic identification
Non-hispanic	97.5	97.9	0.148
Hispanic	2.5	2.1
Insurance status
Not insured	1.5	1.7	0.001
Private insurance/managed care	29.5	30.9
Medicaid	3.8	4.6
Medicare	63.0	60.7
Other government	0.9	1.1
Insurance status unknown	1.4	1.0
Community size
Metro (20,000 to >1 million)	82.5	80.6	<0.001
Urban (2500–20,000)	15.4	17.3
Rural (<2500)	2.1	2.1
Patient distance from facility, miles	30	22.6	<0.001
Comorbidity score = 0 (%)	51.3	52.9	0.032
Mean tumor size (mm)	25.8	29.2	<0.001
Nodal disease
N0	86.0	61.7	<0.001
N1	11.2	34.0
Nx	2.8	4.3
Type of surgery
Less than lobectomy	20.9	39.8	<0.001
Lobectomy/bilobectomy	76.0	55.1
Pneumonectomy	3.0	4.5
Surgery, NOS	0.1	0.5
Nodes examined
0	10.2	21.6	<0.001
1–9	52.9	50.4
10+	29.6	21.2
FNA	0.1	0.4
Sampling, NOS	0.6	0.5
Dissection, NOS	3.8	2.7
Unknown	2.7	3.1
Chemotherapy
Not received	86.1	49.1	<0.001
Received	13.9	50.9

FNA fine needle aspiration, NOS not otherwise specified

Factors associated with receiving RT were then analyzed. Univariate analysis showed multiple demographic and clinicopathologic factors were significantly associated with receiving RT (Supplemental Appendix) and further explored using multivariable analysis (Table 2). Facility location in the northeast central region (odds ratio [OR] 1.155, 95% confidence interval [CI] 1.069–1.248, p < 0.001), and living in smaller urban populations of 2500 to 19,999 people (OR 1.316, 95% CI 1.111–1.557, p = 0.001) were associated with a higher likelihood of receiving RT on adjusted analysis. Non-community cancer programs (p < 0.001), female gender (OR 0.864, 95% CI 0.812–0.92, p < 0.001), residing more than 20 miles from treating facility (OR 0.762, 95% CI 0.698–0.832, p < 0.001), and metropolitan areas of more than one million people (OR 0.817, 95% CI 0.706–0.945, p = 0.007) were associated with a lower likelihood of receiving RT. Clinicopathologic features were also identified that were associated with a higher likelihood to receive RT including tumor size 3–7 cm (OR 1.336, 95% CI 1.245–1.434, p < 0.001), N1 or Nx nodal status (p < 0.001), and receiving chemotherapy (OR 5.229, 95% CI 4.855, 5.632, p < 0.001). More extensive surgery such as lobectomy, bilobectomy, or pneumonectomy (OR < 1, p < 0.001), and mediastinal nodal sampling of at least one node (OR < 1, p < 0.001) were associated with a significantly decreased likelihood of receiving RT on multivariable analysis.

TABLE 2.

Multivariable analysis of factors associated with receiving post-operative radiotherapy after R0 resection in National Cancer Data Base from 2004 to 2014 for early-stage non-small cell lung cancer

Characteristic	p value	OR	95% CI
			Upper	Lower
Facility type
Community cancer program (RL)
Comprehensive community cancer program	<0.001	0.707	0.641	0.779
Academic/research program	<0.001	0.566	0.508	0.630
Integrated network cancer program	<0.001	0.609	0.523	0.710
Other specified types of cancer programs	0.399	0.755	0.392	1.452
Facility location
East North Central	<0.001	1.155	1.069	1.248
West North Central	<0.001	0.663	0.580	0.759
Mountain	0.002	0.730	0.601	0.887
Gender
Male (RL)
Female	<0.001	0.864	0.812	0.920
Age (years)
≥65	0.540	1.021	0.955	1.092
Race
White (RL)
Black	0.019	1.141	1.022	1.274
Other	0.141	0.847	0.680	1.057
Urban/rural
Counties in metro areas of 1 million population or more	0.007	0.817	0.706	0.945
Counties in metro areas of 250,000 to 1 million population	0.239	0.914	0.786	1.062
Counties in metro areas of fewer than 250,000 population	0.393	1.071	0.915	1.255
Urban population of 20,000 or more, adjacent to a metro area	0.479	1.068	0.890	1.281
Urban population of 2500–19,999, adjacent to metro area	0.001	1.316	1.111	1.557
Patient distance from facility
>20 miles	<0.001	0.762	0.698	0.832
Comorbidity score
Score = 0	0.017	0.927	0.870	0.986
Tumor size
<3 cm (RL)
3–7 cm	<0.001	1.336	1.245	1.434
Nodal disease
N0 (RL)
N1	<0.001	2.779	2.562	3.014
Nx	0.006	1.249	1.066	1.464
Type of surgery
Less than lobectomy (RL)
Lobectomy/bilobectomy	<0.001	0.296	0.272	0.321
Pneumonectomy	<0.001	0.276	0.232	0.329
Surgery, NOS	0.243	1.307	0.834	2.049
Nodes examined
0 (RL)
1–9	<0.001	0.596	0.542	0.655
10+	<0.001	0.390	0.347	0.439
FNA	0.073	1.566	0.959	2.557
Sampling, NOS	0.001	0.466	0.300	0.724
Dissection, NOS	<0.001	0.453	0.368	0.557
Unknown	<0.001	0.674	0.554	0.819
Chemotherapy
Not received (RL)
Received	<0.001	5.229	4.855	5.632

CI confidence interval, FNA fine needle aspiration, NOS not otherwise specified, OR odds ratio, RL reference level

To better understand the impact of RT on survival for these patients, we compared the survival of irradiated and non-irradiated patients by Kaplan–Meier log-rank analysis (Supplemental Appendix). The median OS of the entire cohort was 76.6 months (95% CI 76.0–77.2), with a median follow-up of 5.1 years. The irradiated group had significantly worse OS (p < 0.001), with a median survival of 45.8 months compared to 77.5 months for the non-irradiated group. Univariate analysis was performed to identify factors associated with OS (Supplemental Appendix), followed by adjusted multivariable analysis for OS (Table 3). On multivariable analysis, RT remained associated with worse survival (hazard ratio (HR) 1.339, 95% CI 1.282, 1.399).

TABLE 3.

Multivariable analysis of factors associated with overall survival of patients in National Cancer Data Base treated with and without radiotherapy for margin-negative (R0) early-stage non-small cell lung cancer from 2004 to 2013

Characteristic	p value	HR	95% CI
			Upper	Lower
Facility type
Community cancer program (RL)
Comprehensive community cancer program	<0.001	0.924	0.897	0.951
Academic/research program	<0.001	0.871	0.844	0.898
Integrated network cancer program	<0.001	0.915	0.879	0.953
Other specified types of cancer programs	0.006	0.694	0.535	0.902
Facility location
New England (RL)
Middle Atlantic	0.626	0.990	0.952	1.030
South Atlantic	0.171	1.026	0.989	1.065
East North Central	<0.001	1.089	1.048	1.131
East South Central	<0.001	1.154	1.106	1.204
West North Central	0.572	1.013	0.969	1.059
West South Central	<0.001	1.087	1.039	1.137
Mountain	0.509	1.018	0.965	1.074
Pacific	0.998	1.000	0.959	1.043
Gender
Male (RL)
Female	<0.001	0.715	0.704	0.727
Age (years)
≥65	<0.001	1.397	1.364	1.432
Ethnicity
White (RL)
Black	0.089	1.027	0.996	1.059
Other	<0.001	0.794	0.747	0.843
Hispanic identification
Non-hispanic (RL)
Hispanic	<0.001	0.860	0.811	0.911
Insurance status
Not insured (RL)
Private insurance/managed care	<0.001	0.830	0.771	0.893
Medicaid	<0.001	1.213	1.116	1.317
Medicare	0.207	1.049	0.974	1.131
Other government	0.233	0.931	0.829	1.047
Insurance status unknown	0.069	0.910	0.821	1.007
Urban/rural
Counties in metro areas of 1 million population or more (RL)
Counties in metro areas of 250,000 to 1 million population	<0.001	1.084	1.062	1.107
Counties in metro areas of fewer than 250,000 population	<0.001	1.059	1.030	1.089
Urban population of 20,000 or more, adjacent to a metro area	<0.001	1.108	1.066	1.152
Urban population of 20,000 or more, not adjacent to a metro area	0.849	0.993	0.928	1.064
Urban population of 2500 to 19,999, adjacent to a metro area	<0.001	1.075	1.036	1.117
Urban population of 2500 to 19,999, not adjacent to a metro area	<0.001	1.115	1.063	1.170
Completely rural or less than 2500 urban population, adjacent to a metro area	0.021	1.098	1.014	1.189
Completely rural or less than 2500 urban population, not adjacent to a metro area	0.108	1.067	0.986	1.154
Patient distance from facility
>20 miles	0.575	0.994	0.973	1.016
Comorbidity score
Score = 0	<0.001	1.249	1.229	1.269
Tumor size
<3 cm (RL)
3–7 cm	<0.001	1.384	1.359	1.409
Nodal disease
N0 (RL)
N1	<0.001	1.902	1.855	1.949
Nx	<0.001	1.236	1.189	1.286
Type of surgery
Less than lobectomy (RL)
Lobectomy/bilobectomy	<0.001	0.780	0.763	0.798
Pneumonectomy	0.958	0.999	0.953	1.046
Surgery, NOS	0.681	0.962	0.799	1.158
Nodes examined
0 (RL)
1–9	<0.001	0.778	0.756	0.800
10+	<0.001	0.679	0.657	0.701
FNA	0.248	0.883	0.716	1.090
Sampling, NOS	<0.001	0.703	0.630	0.784
Dissection, NOS	<0.001	0.714	0.679	0.750
Unknown	<0.001	0.804	0.764	0.847
Chemotherapy
Not received (RL)
Received	<0.001	0.751	0.732	0.770
Post-operative radiotherapy
No (RL)
Yes	<0.001	1.339	1.282	1.399

CI confidence interval, FNA fine needle aspiration, HR hazard ratio, NOS not otherwise specified, RL reference level

Propensity score matching was then done to ameliorate the effect of confounding factors on OS. After censoring data for which there was no follow-up, 4147 patients who received post-operative RT were available for matching against 165,783 cases of patients with R0 surgery alone. The vast majority of cases (99.5%) matched successfully with 1:1 nearest neighbor method, and all matched variables were adequately balanced with the largest absolute mean standardized difference equal to 8.7% (Supplemental Appendix). After matching, RT remained associated with significantly worse OS (HR 1.313, 95% CI 1.237–1.394, p < 0.001), with a median survival of 46.1 months compared to 62.7 months in the non-irradiated group (Fig. 2).

FIG. 2.

Kaplan–Meier analysis of overall survival after propensity score matching comparing survival of non-irradiated and irradiated groups. Median survival was 62.7 months in non-irradiated group compared to 46.1 months in irradiated group (p < 0.001)

DISCUSSION

Despite the findings of the 1998 PORT Meta-analysis for lung cancer and subsequent guidelines recommending against the use of post-operative RT for resected stage I–II NSCLC, we identified a subset of patients treated with radiation in the NCDB from 2004 to 2014.¹⁰ While the PORT meta-analysis included 2128 patients, our analysis has included substantially more patients of whom over 4000 received RT.¹⁰ While the relative numbers of patients (2%) who received RT were small, the costs associated with the absolute number of patients who received RT are potentially high in terms of toxicity, financial burden, and quality of life. It is concerning that many patients received doses in excess of 60 Gy in absence of gross residual disease. Similarly, it is notable that a substantial number of patients received less than 40 Gy which would be considered a sub-therapeutic dose. Regardless of the criticisms of the 1998 PORT meta-analysis, our data shows no evidence that would refute its findings for early-stage NSCLC. These findings support current recommendations for post-operative radiation for early-stage NSCLC because our study suggests not only lack of benefit for radiation, but potential harm.

A number of factors were associated with delivery of post-operative RT for these patients. Patients who received RT were more likely to be treated at a community center, albeit this was only a small percentage of patients. In turn, we found that worse survival was associated with being treated at a community cancer program in comparison to other kinds of academic or integrated programs. This finding is in line with previous studies of NSCLC showing that center experience and volume of patients is highly correlated with OS outcomes in both surgically resected patients and for those who undergo definitive chemoradiation for locally-advanced disease.^17–22 There was a statistical trend for patients to be younger in the irradiated group, yet they still had worse OS. While our analysis has focused on delivery of radiation delivery in early-stage margin negative NSCLC, our findings suggest the need to better understand all practice patterns that could contribute to differential outcomes in community cancer programs.

An inherent limitation of our study being a retrospective population-based dataset is that the irradiated group may have had worse tumor or patient characteristics not captured in the NCDB. As an admission of the potential biases in patient selection, we used propensity score matching in our analysis as an attempt to address these confounding factors. Indeed, there were notable differences in surgeries and clinicopathologic features of patients who received post-operative radiation and those who did not. Patients who received radiation were more likely to have undergone sublobectomy, have fewer nodes dissected, and receive chemotherapy, suggesting that providers may have attempted to use RT to make up for suboptimal surgery or worse tumor features. Interestingly, N1 disease was also significantly associated with post-operative RT although it is not typically recommended for N1 disease. Though one potentially confounding variable is that the irradiated group was more likely to have had a sublobar resection, it is not captured in the NCDB whether these were wedge resections or formal segmentectomies, making it challenging to interpret the impact of surgical technique. Irrespective of potentially worse tumors in the irradiated group, it is notable that RT did not appear to compensate for possible uncaptured adverse features even after propensity score matching.

Although a retrospective study from the University of Pennsylvania did not show increased deaths related to modern post-operative radiation, it included stage III patients who are much more likely to die of their cancer, whereas we only evaluated stage I–II disease where majority of patients are cured.²³ The relatively late divergence of the survival curves at approximately 3 years in our study might be consistent with late radiation toxicity. However, it is not possible to explore causality as the NCDB does not provide the antecedent cause of death, provide toxicity data, or information on locoregional control. Our results are also consistent with a Surveillance, Epidemiology, and End Results analysis where patients with N0 or N1 disease were not shown to have a benefit from radiation.²⁴ Recently Wang et al. showed a similar finding in an analysis of the NCDB from 2011 to 2015.²⁵ It should be noted that our study has gone further than these previous analyses with robust use of propensity score matching as well showing factors associated with receiving radiation (such as being treated in a community center).

We have performed a large study of the NCDB showing persistent use of RT for resected margin negative early-stage NSCLC in approximately 500 people annually. The use of post-operative RT was independently associated with worse OS, consistent with the 1998 PORT meta-analysis.¹⁰ Our findings support current recommendations against post-operative RT for resected early-stage NSCLC and aim to provide further impetus to stop this practice in the United States.