The current study characterizes the overall survival (OS) and cause-specific survival (CSS) of patients with stage I nonsmall cell lung cancer (NSCLC) who were treated with radiotherapy alone, and analyzes the variables potentially affecting survival outcomes.
A total of 8524 patients with stage I NSCLC (according to the sixth edition of the American Joint Committee on Cancer staging manual) who were diagnosed between 1988 and 2008 were retrospectively analyzed using the population-based Surveillance, Epidemiology, and End Results database. Cox regression analysis was used to calculate hazard ratios (HR) from multivariate analyses.
The 1-year, 2-year, and 5-year OS rates were 62%, 37%, and 11%, respectively; the corresponding lung cancer CSS survival rates were 68%, 45%, and 20%, respectively. Approximately 77% of deaths were from lung cancer (5292 of 6891 total deaths). Cardiac (n = 477 deaths) and pulmonary (other than lung cancer deaths; n = 475 deaths) deaths accounted for 14% of deaths. From Cox proportional hazards analyses, male sex (HR, 1.2) and squamous cell carcinoma histology (HR, > 1.1) were found to be significantly (P < .0001) adverse prognostic factors for both OS and lung cancer CSS. A more recent calendar year of diagnosis was associated with significantly (P < .0001) improved OS (HR, 0.84 per decade) and lung cancer CSS. This trend was also significant (P < 0.0001) when restricting analyses to those patients with tumors measuring ≤ 5 cm (n = 5402 patients). T1 classification (vs T2 or T unknown) and smaller tumor size were found to be significantly (P < .0001) favorable factors.
Surgical resection remains the standard treatment of patients with stage I nonsmall cell lung cancer (NSCLC). For patients who are medically inoperable or decline surgical resection, definitive radiotherapy is an effective alternative. In these patients, the lack of surgical staging and death from comorbid conditions (more common in medically inoperable patients) have historically resulted in survival outcomes inferior to those of patients who undergo surgical resection. In a landmark pooled analysis from Sibley, published in 1998 in Cancer, approximately 15% of medically inoperable patients with stage I NSCLC were long-term survivors after radiotherapy; approximately 30% of deaths were from intercurrent illness and approximately 70% of deaths were from lung cancer, one-half of which were the result of distant metastases and one-half of which were the result of local disease recurrence.1 Higher prescribed doses were found to be correlated with better outcomes, whereas prophylactic lymph node irradiation was not.1
Since Sibley's article was published in 1998, radiotherapy planning and delivery have become more sophisticated, evolving from 2-dimensional (2D) anteroposterior/posteroanterior fields to 3D conformal radiotherapy, and more recently to novel modalities such as stereotactic body radiotherapy (SBRT)/stereotactic ablative body radiotherapy (SABR), image-guided radiotherapy, and 4D planning and delivery. These developments, particularly SBRT/SABR, have resulted in improved survival compared with historical outcomes, which is attributable to more accurate tumor targeting and the delivery of higher biologically effective doses.2 Furthermore, lung cancer staging with positron emission tomography/computed tomography (PET/CT) has become an accepted standard, reducing the risk of understaging patients with occult distant and/or lymph node metastases3 and aiding in radiotherapy planning. Although not diagnostic, PET/CT does help guide tissue biopsy, which has also recently become more sophisticated. Although minimally invasive bronchoscopy and CT-guided biopsies have been used for many years, recent advancements in endobronchial ultrasound-guided biopsy4 and navigational bronchoscopic biopsy5 can improve diagnostic yield.
In the current study, patients with stage I NSCLC based on the sixth edition of the American Joint Committee on Cancer (AJCC) staging manual6 who were treated with radiotherapy alone and registered in the population-based Surveillance, Epidemiology, and End Results (SEER) database were retrospectively analyzed to better understand the risk factors affecting their overall survival (OS) and cause-specific survival (CSS). There have been many SEER studies analyzing survival outcomes after resection of stage I NSCLC,7-12 although to our knowledge few SEER studies have focused on outcomes after radiotherapy alone. Two recent SEER studies demonstrated a significant benefit of radiotherapy (vs no radiotherapy) among patients with unresected, early stage NSCLC.13, 14 To our knowledge, this is the first analysis from SEER data focusing on prognostic factors in patients treated with radiotherapy alone for stage I NSCLC. We assessed whether survival outcomes improved over time with better lung cancer staging and radiotherapy treatment technology in recent years, and analyzed the clinicopathologic factors affecting survival after definitive radiotherapy.
MATERIALS AND METHODS
Patients were selected from the SEER 17 registries using the SEER*Stat case listing session with the following criteria: first or only primary cancer; “lung and bronchus” site; localized stage I disease, based on SEER historical and AJCC stage (third and sixth editions of the AJCC staging manual, abbreviated herein as AJCC-third and AJCC-sixth) database fields; treatment with beam radiotherapy, without cancer-directed surgery; and having the following histologies: nonsmall cell carcinoma (according to the third edition of International Classification of Diseases for Oncology [ICD-O-3] 8046/3; used from 1997-present, although infrequently before 2000), squamous cell carcinoma (ICD-O-3 8050-8084/3), adenocarcinoma (ICD-O-3 8140/3, 8255/3, 8260/3, and 8310/3), bronchioloalveolar adenocarcinoma (ICD-O-3 8250-8254/3), adenosquamous carcinoma (ICD-O-3 8560/3), and large cell carcinoma (ICD-O-3 8012/3). Patients with neuroendocrine, carcinoid, small cell carcinoma, oat cell carcinoma, sarcoma, and mesothelioma histologies were excluded. Patients with unspecified lung neoplasms (those that could not be distinguished from excluded histologies) were also excluded.
Beginning in 1988, the SEER registry recorded lung cancer stage and primary tumor size; those patients registered between 1973 and 1987 did not have their NSCLC stage recorded and therefore were excluded. For the current study, T1N0 NSCLC (stage IA) tumors measure ≤ 3 cm and are confined to the lung, whereas T2N0 NSCLC (stage IB) tumors measure > 3 cm and/or involve the mainstem bronchus ≥ 2 cm from the carina and/or with atelectasis that does not involve the entirety of the lung (based on AJCC-sixth). For patients diagnosed between 1988 and 2003, stage grouping (based on AJCC-third) and clinical tumor extent were recorded for the majority of patients by the SEER registry. For most patients diagnosed after 2003, the T classification was explicitly recorded. For patients for whom the T classification or stage grouping was not explicitly recorded by the SEER registries, primary lung cancers measuring ≤ 3 cm and confined to the lung were classified as T1, and lesions measuring > 3 cm in size and/or with T2-defining clinical extension (as per above) were classified as T2. Patients were assigned an “unknown” T classification (T1 or T2) if 1) the size of the tumor was ≤ 3 cm or not recorded and the extent of disease was not recorded (1988-2003); or 2) the T classification was not recorded, size was ≤ 3 cm or not recorded, and the extent of disease was not recorded (after 2003). It is important that lung cancer staging from the seventh edition of the AJCC staging manual (AJCC-seventh)15 differs from that of the AJCC-sixth in that T2 no longer includes tumors of any size measuring > 3 cm (without regional extension). Instead, lymph node-negative NSCLC with a primary tumor site measuring > 5 cm to 7 cm (without regional extension) is now staged as T2bN0, Stage IIA disease and a primary tumor site measuring > 7 cm is now staged as T3N0, Stage IIB disease.15 We opted not to exclude patients based on the size of their primary tumor, because those patients classified as having an “unknown” T classification (T1 or T2) and those patients with an unknown tumor size, classified as T2 based on the extent of disease, may include patients now classified as having T2b to T3 disease by AJCC-seventh. Instead, we performed separate analyses accounting for tumor size, as well as separate analyses excluding those patients with a tumor size > 5 cm, which represents an adverse prognostic group,16 prompting the change in AJCC staging.15
Stata version 9.2 statistical software (StataCorp, College Station, Tex) was used for data analysis. Actuarial overall survival (OS) and cause-specific survival (CSS) were calculated using the Kaplan-Meier method. Survival times were calculated from the date of diagnosis to the date of last recorded follow-up. For lung cancer CSS, death from lung cancer was considered an event, whereas those patients who died of other causes were censored at the time of death. For all other CSS, death from any cause other than lung cancer was considered an event, whereas those patients who died of lung cancer were censored at the time of death. Cardiac CSS reflects deaths from heart disease only, and pulmonary CSS reflects deaths only from pulmonary causes other than lung cancer (ie, chronic obstructive disease or pneumonia). For univariate analyses comparing survival outcomes between subgroups, the log-rank test was used for discrete variables, and Cox regression analysis was used for continuous variables. For multivariate analyses (MVA) assessing the significance and hazards ratios (HR) of prognostic variables, Cox proportional hazards models were used.
Patient and Tumor Characteristics
Only 1.4% of patients were diagnosed without histologic or cytologic confirmation. Table 1 summarizes the patient and tumor characteristics of the study patients, grouped by 4 eras: 1988 to 1995, 1996 to 2000, 2001 to 2004, and 2005 to 2008. The median age of the patients at the time of diagnosis was 74 years. Although age at diagnosis significantly increased with a more recent calendar year of NSCLC diagnosis, this change was modest. Other significant changes observed with a more recent year of diagnosis include relatively more female patients, fewer squamous cell carcinomas relative to adenocarcinomas, more T1 (vs T2) tumors, and smaller tumor size.
Table 1. Patient and Tumor Characteristics at Time of Lung Cancer Diagnosis
P values represent analysis of variance (ANOVA), analyzing the variable as a continuous variable. For age, although the ANOVA remained significant (P = .004) even when omitting patients diagnosed before 1996, the average ages are similar.
Chi-square test was used, omitting patients with an unknown variable.
The P value reflects differences in distribution between eras of squamous cell carcinomas and adenocarcinomas.
The P value reflects differences in distribution between eras of T1 and T2 classification (ie, omitting unknown T classification).
Among 3223 patients with a recorded tumor size ≤ 3 cm, 84% (n = 2722), 3% (n = 99), and 12% (n = 402) were classified as having T1, T2, and unknown disease, respectively.
The 6-month, 1-year, 2-year, and 5-year OS rates were 83%, 62%, 37%, and 11%, respectively. The 6-month, 1-year, 2-year, and 5-year lung cancer CSS rates were 86%, 68%, 45%, and 20%, respectively (Fig. 1). Significantly favorable prognostic factors for OS and lung cancer CSS included female sex, histology other than squamous cell carcinoma, more recent calendar year of diagnosis, T1 classification (vs others), and smaller tumor size (P < .00001 for all) (Table 2). For the eras between 1988 and 1995, 1996 and 2000, and 2001 and 2004, the 5-year OS rates were 7%, 11%, and 13%, respectively; for the years between 2005 and 2008, the 5-year OS was not reached, but was estimated (from 3-year and 4-year OS rates of 30% and 23%) to be between 17% and 18%.
Table 2. Univariate Analyses of Variables Potentially Affecting OS and CSS
Approximately 77% of deaths were from lung cancer (5292 of 6891 total deaths). Cardiac (n = 477) and pulmonary (other than lung cancer; n = 475) deaths accounted for 14% of deaths reported. The 1-year, 2-year, and 5-year CSS rates from any cause other than lung cancer were 91%, 82%, and 57%, respectively. Table 2 also shows the 5-year CSS rate from any cause other than lung cancer, pulmonary CSS, and cardiac CSS. Deaths from other causes were found to be significantly increased in older patients and male patients. Patients with larger tumors were more likely to die of lung cancer and less likely to die of other causes.
Figure 2 depicts the 2-year OS and lung cancer CSS for patients, relative to the calendar year of diagnosis. Amid fluctuations from year to year, there was a significant (P < .0001 using Cox regression analysis) increase in OS and lung cancer CSS with a more recent year of diagnosis. Also shown in Figure 2 is the 2-year OS and lung cancer CSS, including only those patients with a tumor measuring ≤ 5 cm (n = 5402, representing T1-T2aN0 NSCLC in AJCC-seventh). A similarly significant (P < .0001 with Cox regression analysis) improvement in OS and lung cancer CSS was noted with a more recent year of diagnosis.
All variables potentially impacting OS were analyzed using multivariate Cox proportional hazards analyses (Table 3). Male gender, T classification other than T1 (ie, T2 or unknown) and treatment in earlier decades, each resulted in a 1.2- to 1.3-fold significantly (P < .0001) increased risk of death. Older age at the time of diagnosis, squamous cell carcinoma histology, and tumor size (per cm) were also found to be significant factors. The MVAs shown Table 3 include only those patients whose tumor size was recorded. Omitting tumor size from the MVA allowed for analysis of all 8524 patients and did not appreciably change the HRs, with the exception of T classification (T1 vs others), for which the HR for OS was 0.67 (95% confidence interval, 0.64-0.71). For 2378 patients diagnosed between 2005 and 2008, older age, male sex, squamous cell histology, T classification other than T1, and increased tumor size were also found to be adverse prognostic factors for OS (Table 3). Table 3 also shows the Cox regression analyses, restricted to only those 5402 patients with a tumor size ≤ 5 cm. Aside from T classification being a less significant prognostic factor, with a slightly less pronounced HR, the outcomes are similar to those for all 8524 patients. The MVA models using the variable T1 versus T2 disease (n = 6166 patients, including all patients with obtainable T classification, and n = 5000 patients when including only those with tumors measuring ≤ 5 cm) compared with T1 disease versus all others did not appreciably change the HRs or P values (data not shown). The HRs and P values for lung cancer CSS were very similar with regard to magnitude and significance to the HRs and P values for OS (data not shown).
Table 3. Multivariate Analyses of Variables Potentially Affecting OSa
Major findings from the current study of patients treated with radiotherapy alone for T1-2N0 NSCLC (AJCC-sixth) include a significantly, albeit modestly, improved OS and lung cancer CSS associated with a more recent calendar year of diagnosis, female sex, non-squamous cell carcinoma histology, earlier T classification, and smaller tumor size. The 5-year OS rate of 11%, with the majority of deaths (77%) attributed to lung cancer, are consistent with the pooled analysis published by Sibley in 1998,1 although squamous cell histology was a favorable prognostic factor in that study, versus unfavorable in the current study.
Several recent SEER studies examining surgical resection for patients with stage I NSCLC have also demonstrated male sex,8, 10, 12 older age,8-10, 12 and larger tumor size9, 10, 12 to be significantly adverse risk factors for OS8-10, 12 and lung cancer CSS.8, 12 Extent of resection8, 12 and number of lymph nodes examined7, 9-12 have also been shown to be significant prognostic factors after surgical resection. It is interesting to note that, unlike our cohort of patients with stage I NSCLC (AJCC-sixth) who were treated with radiotherapy alone, some studies have indicated that black race is associated with significantly worse survival outcomes after surgical resection,12 and squamous cell carcinoma histology is not significant for lung cancer CSS12 or OS9, 10 after resection. In these studies,7-12 the calendar year of NSCLC diagnosis was not analyzed.
A population-based trend toward improved survival in patients with lung cancer with a more recent year of diagnosis has been described in the United States,17 although to our knowledge, this has not been previously reported among those patients treated with radiotherapy alone for localized NSCLC. A Dutch population-based study of 875 elderly patients (aged ≥ 75 years) treated with radiotherapy for stage I NSCLC from 1999 to 2007 demonstrated an increasing use of definitive radiotherapy after SBRT was introduced in 2003, and an improved survival that was correlated with SBRT use; younger age, earlier T classification, and more recent era (2005-2007) of diagnosis were found to be significantly favorable prognostic factors for OS.18
As described earlier, the survival improvement noted in the current study (with patients registered from 1988-2008) is likely multifactorial, and partly the result of advancements in lung cancer diagnostic methods, staging techniques, radiotherapy technology, and salvage therapy (radiotherapy or chemotherapy). PET imaging was unlikely to have been used in patients treated before the mid-2000s. Although the SEER database does not provide information regarding the radiotherapy technique used, patients treated in the late 1980s to the early 1990s were much less likely to have undergone 3D planning, and patients treated in the 1990s and early 2000s were much less likely to have been treated with more modern radiation planning and delivery techniques such as SBRT/SABR, 4D planning, or respiratory gating. In fact, Pan et al have shown that SBRT/SABR have become increasingly more commonly adopted in the United States since 2000, with the lung representing the most commonly treated site.19 The reported survival outcomes after SBRT/SABR for patients with stage I NSCLC2 are far superior to those reported by Sibley in 1998.1 As SBRT/SABR becomes used more frequently, it is anticipated that survival outcomes will continue to improve. It is interesting to note that technologic advances have been shown to improve the OS of patients with advanced NSCLC.20 It also will be intriguing to discover how the survival of patients with early stage NSCLC is impacted over the next years to decades with the anticipated increased rate of CT screening.21
Although a continued trend toward improved survival can be readily explained by medical advances, the year-to-year fluctuations (eg, the sharp decrease in OS and lung cancer CSS noted in 2004 [Fig. 2]), are more difficult to explain. In 2004, the SEER program underwent major coding changes (available at http://seer.cancer.gov/tools/codingmanuals/historical.html) and began to capture AJCC lung cancer T classification (based on AJCC-sixth, which had not changed from prior editions). The percentage of patients with an unknown T classification dropped from 21% in 2003 to 14% in 2004, the ratio of T1 to T2 cancers increased from 0.6 to 1.0 (a substantial change), and the median tumor size dropped from 3.4 cm to 3 cm (data not shown). Perhaps these coding changes impacted which patients were reported to the SEER program and thus the calculated survival, although it is not readily explained how this may have occurred, or why in 2005 survival increased again.
Patient and Tumor Characteristics
Among patients treated with radiotherapy alone for stage I NSCLC (AJCC-sixth), a more recent calendar year of diagnosis was found to be significantly associated with slightly older age, a reduction in male preponderance (nearly 50:50 among patients diagnosed between 2005-2008), a greater likelihood of adenocarcinoma histology, and a smaller reported tumor size. The data from the current study are consistent with the previously described increased incidence of adenocarcinoma (vs squamous cell carcinoma) NSCLC22 and an increase in the diagnosis of NSCLC in females.17 The increased age at the time of diagnosis likely reflects a more aging population. The slightly reduced tumor size noted over the past few decades may reflect earlier diagnosis, or the relatively more indolent disease progression associated with adenocarcinoma.
The strengths of the current study include the sizable number (> 8000) of patients identified from a large population-based setting. Substantial patient numbers allowed for analyses of outcomes by age, year of diagnosis, sex, NSCLC tumor size, and histology. Known limitations of SEER data include a lack of detailed information regarding radiotherapy (dose or dose fractionation, field, technique), and an absence of chemotherapy data (including salvage chemotherapy after disease recurrence) and tobacco history. Furthermore, SEER registries do not collect data concerning prognostic factors such as weight loss and performance status. Other weaknesses include the retrospective nature of the current study, a lack of pathologic or radiologic review to verify diagnosis and staging, and a lack of information regarding curative-intent versus palliative-intent therapy (likely relevant for the largest tumors). Presumably, the vast majority of patients received curative-intent therapy. Finally, the SEER program does not record dates of cancer progression or patterns of disease recurrence. With more sophisticated radiotherapy planning and delivery technology, it is expected that the predominant pattern of failure will be distant disease progression.23-25
From a hypothesis-generating, retrospective, population-based registry analysis of patients with lymph node-negative, early stage NSCLC (T1-T2N0 from AJCC-sixth), it was learned that a more recent calendar year of diagnosis, female sex, non-squamous cell carcinoma histology, earlier T classification, and smaller tumor size were associated with significantly more favorable (albeit modest) OS and lung cancer CSS rates. Improvements in survival in more recent years, even when correcting for other prognostic factors in Cox regression analyses, likely reflect technologic improvements in lung cancer diagnosis, staging, and radiotherapy delivery.