Improved Outcomes Associated with Higher Surgery Rates for Older Patients with Early-Stage Non-Small Cell Lung Cancer

Abstract

Background

Although surgery offers the greatest chance of cure for patients with early stage non-small cell lung cancer (NSCLC), older and sicker patients often fail to undergo resection. The benefits of surgery in older patients and patients with multiple co-morbidities are uncertain.

Methods

We identified a national cohort of 17,638 Medicare beneficiaries, aged ≥66 years living in Surveillance, Epidemiology, and End Results (SEER) areas who were diagnosed with stage I or II NSCLC during 2001–2005. We compared areas with high and low rates of curative surgery for early-stage lung cancer to estimate the effectiveness of surgery in older and sicker patients. We used logistic regression models to assess mortality by quintile of area-level surgery rates, adjusting for potential confounders.

Findings

Fewer than 63% of patients underwent surgery in low-surgery areas while >79% underwent surgery in high-surgery areas. High-surgery areas operated on more patients with advanced age and COPD than low-surgery areas. Adjusted all-cause one year mortality was 18.0% in high-surgery areas vs. 22.8% in low-surgery areas (adjusted odds ratio (OR)=0.89 (95% confidence interval [CI] 0.86–0.93) for each 10% increase in surgery rates). One year lung-cancer-specific mortality was similarly lower in high-versus low-surgery areas (12.0% versus 16.9%), adjusted OR=0.86 (95% CI 0.82–0.91) for each 10% increase in surgery rates.

Interpretations

Higher rates of surgery for stage I/II NSCLC are associated with improved survival, even when older patients and sicker patients undergo resection. More work is needed to identify and reduce barriers to surgery for early-stage NSCLC.

Introduction

Surgical resection of non-small cell lung cancers (NSCLC) offers the greatest likelihood of cure for the 16% of patients who are diagnosed with stage I or II cancers.¹ Although data suggest that definitive radiation therapy can be used with curative intent in patients who are medically inoperable, strong retrospective evidence suggests that surgical resection is the optimal treatment for fit patients.^2–5 Nevertheless, many patients, particularly older patients and those with smoking-related illnesses, such as chronic obstructive lung disease (COPD), do not undergo curative surgery.^6–13

Recent research suggests that surgeons have substantial uncertainty about the benefits versus risks of surgery for patients with increasing severity of COPD and advanced age.¹⁴ Few data are available to help guide these decisions, as older patients and those with significant comorbid illness are not typically enrolled in randomized trials. Moreover, observational studies assessing outcomes following surgery are limited because treated versus untreated patients may differ by unobserved characteristics that may also influence outcomes.

Exploiting geographic variation in treatment rates is one strategy for understanding the association between treatments and outcomes when randomization is not feasible that may avoid bias due to unobserved confounders.^15–18 Previous research has demonstrated substantial geographic variation in practice patterns in cancer and other disease, but little variation in underlying health across areas.¹⁹ Thus area-level surgery rates can be thought of as a pseudo-randomization to treatment as patients living in areas with higher rates of surgery will be more likely to undergo surgery; but geographic location itself should not influence patient outcomes. In this study, we used a population approach to compare areas with high and low rates of curative surgery for early-stage lung cancer to estimate the effectiveness of surgery in older and sicker patients, for whom surgery is often omitted, likely due to uncertainty about the balance of risks versus benefits in such patients. If areas with higher rates of surgery for stage I/II NSCLC also have lower mortality, it will suggest that there is benefit to curative surgery even among individuals where there may be uncertainty about the likelihood of benefit. This population approach is similar to an analysis that uses area-level surgery rates as an instrumental variable for curative surgery.^15–18

Methods

Data

The SEER program of the National Cancer Institute collects uniformly reported data from 15 population-based cancer registries covering approximately 26% of the United States population.²⁰ SEER registrars collect information on patient demographics, tumor characteristics, and primary surgery and radiation for each incident cancer.

Since 1991, the SEER data have been merged with Medicare administrative data by a matching algorithm that has successfully linked files for more than 94% of SEER patients aged 65 or older.²¹ The Medicare claims data used in this study included the Medicare Provider Analysis and Review (MEDPAR) file (inpatient claims), the 100% Physician/Supplier file (claims for physicians' services and other medical services), and the Hospital Outpatient Standard Analytic file (claims for outpatient facility services). The study was approved by the Harvard Medical School Committee on Human Subjects.

Cohort

We identified all men and women diagnosed with stage I or II NSCLC in 2001–2005. We restricted the cohort to individuals aged 66 or older at diagnosis who were enrolled in parts A and B of fee-for-service Medicare as of 1 year before diagnosis, and we excluded a small number of patients with no claims during the period from 45 days before diagnosis through 180 days after diagnosis (because we were concerned about incomplete data). From this cohort of 19,363 patients we further excluded 1,218 patients who disenrolled from parts A and B of fee-for-service Medicare within 6 months of diagnosis, leaving 18,145 potentially eligible patients. We did not exclude patients who died within 6 months of diagnosis because we wanted to be certain to include patients who had negative outcomes related to their surgeries.

Approach

We use a population approach and compare areas with high and low rates of curative surgery for early-stage lung cancer to estimate the effectiveness of surgery in older and sicker patients. This approach is very similar to an instrumental variable analysis.^15–18 Both approaches rely on a similar set of assumptions: (1) area-level surgery use is related to treatment choice and (2) area-level surgery use is unrelated to the outcome of interest except through their effect on treatment. If these assumptions hold, then area-level surgery rates can be thought of as a pseudo-randomization to treatment as patients living in areas with higher rates of surgery will be more likely to undergo surgery; but geographic location itself should not influence patient outcomes.

Variables

Curative surgery

We assessed receipt of pneumonectomy, lobectomy, wedge or segmental resection within 180 days of diagnosis with stage I or II lung cancer based on International Classification of Disease, 9^th edition (ICD-9) and Common Procedure Technology (CPT) codes.²²

Primary Dependent Variable

Because there is substantial area-level variation in use of surgery for early-stage lung cancer,^{10, 23} we used the area-level rate of surgery as our primary independent variable to mimic randomization to curative surgery. Specifically, we calculated the proportion of patients with stage I/II NSCLCA in each year who underwent curative resection in each Health Service Area (HSA), and assigned to each patient the rate in their HSA of residence in the year of their diagnosis. HSAs were originally defined by the National Center for Health Statistics to be a single county or cluster of contiguous counties which are relatively self-contained with respect to hospital care. We excluded patients living in HSAs with <20 patients who were eligible for surgery because we could not be certain that the surgical rates were accurate (507 patients who lived in 59 of 156 HSAs were excluded for a final cohort of 17,638 patients). We categorized HSAs into quintiles of increasing rates of surgery for older individuals with stage I/II lung cancer based on the rate of surgery in the year when the patient was diagnosed. In a set of sensitivity analyses, we categorized HSAs into quintiles after averaging across all years; results were similar and are not presented.

Patient characteristics

We used registry data to characterize patients based on age, sex, race/ethnicity, marital status, year of diagnosis, and history of a prior cancer. We identified comorbid illness based on the Klabunde modification of the Charlson Index²⁴ using inpatient and outpatient claims in the year before cancer diagnosis. We removed COPD from the Charlson index so that we could assess it separately, since severe COPD is particularly likely to influence a patient's ability to tolerate lung resection. We obtained information on the proportion of individuals with a college degree in the census tract of residence from 2000 U.S. Census data. Variables are categorized as in Table 1.

Table 1.

Characteristics of patients by quintile of surgery rates (N=17,638)

	Quintile of Proportion of Patients Undergoing Surgery					P-value*
	Quintile 1 (lowest)	Quintile 2	Quintile 3	Quintile 4	Quintile 5 (highest)
Proportion of stage I/II patients undergoing surgery in area	≤62.6%	62.7–69.6%	69.7–74.2%	74.3–79.1%	≥79.2%
Age (%)
66–69	20.3	18.2	19.7	17.6	18.6	0.06
70–74	29.2	30.0	28.7	26.1	28.6	0.04
75–79	27.2	28.3	28.2	29.6	28.8	0.07
80–84	16.7	16.1	16.4	18.8	17.7	0.02
85+	6.6	7.4	7.1	7.9	6.3	0.83
Race (%)
White	89.3	87.6	88.3	86.3	90.5	0.46
Black	7.4	7.7	7.7	6.2	3.9	<0.0001
Hispanic	0.9	1.1	0.9	1.5	1.1	0.23
Unknown/other	2.4	3.6	3.1	6.1	4.5	<0.0001
Marital status (%)
Unmarried	41.6	42.6	43.4	43.1	40.9	0.69
Married	56.0	55.3	54.0	54.6	57.5	0.37
Unknown	2.5	2.1	2.7	2.3	1.6	0.09
% w/ college degree in zip code of residence (%)
Quartile 1 (lowest)	29.5	22.1	18.3	14.1	14.1	<0.0001
Quartile 2	22.1	21.0	18.8	15.4	20.4	<0.0001
Quartile 3	20.3	18.1	23.1	17.4	18.9	0.09
Quartile 4 (highest)	14.8	19.1	19.3	22.3	21.7	<0.0001
Unknown	13.2	19.7	20.6	30.9	24.9	<0.0001
Prior cancer (%)
No	73.4	71.5	70.6	69.6	70.1
Yes	26.6	28.5	29.4	30.4	29.9	<0.0001
Stage (%)
Stage I	82.6	82.7	82.2	83.7	82.9
Stage II	17.4	17.3	17.8	16.3	17.1	0.42
COPD (%)
No	54.9	57.4	56.4	59.1	57.0
Yes	45.1	42.6	43.6	40.9	43.0	0.02
Charlson (without COPD) (%)
0	59.7	60.9	59.9	61.4	59.6	0.92
1	24.3	23.4	23.9	23.3	23.7	0.60
2	9.0	10.0	9.7	9.4	10.3	0.23
3+	7.0	5.8	6.6	5.9	6.4	0.47

^*Using the Armitage test for trend; COPD=chronic obstructive pulmonary disease

Analysis

To test the underlying assumptions of our approach, we examined patient characteristics by quintile of surgery rates in the HSA and used the Armitage test for trend to see if characteristics were related to increasing (or decreasing) area-level surgery rates. Differences in observed characteristics associated with area-level treatment rates suggest violate of the key assumption that area-level surgery rates can be thought of as a pseudo-randomization to treatment. Next, among patients who underwent resection, we assessed if areas with high surgery rates for stage I/II lung cancer are more often operating on individuals who are older and have higher rates of COPD and other comorbid illnesses.

We then assessed unadjusted mortality rates by quintile of area-level rates of surgery using the Armitage test for trend. Finally, we conducted logistic regression models to assess all cause and lung-cancer specific mortality at 1 year from diagnosis for each quintile of surgery rate, adjusting for patient age, sex, race/ethnicity, marital status, history of prior cancer, stage, year of diagnosis, presence of COPD, presence of other comorbid illness, and proportion of individuals in the census tract of residence with a college education. Because lung cancer specific mortality was available through December 31, 2005, the cohort of patients included in the lung cancer specific mortality models was restricted to patients diagnosed through December 31, 2004 (n=14,001). We used generalized estimating equations to account for correlation among observations within HSAs. We calculated adjusted rates by quintile of surgery rates using direct standardization. Finally, we re-fit the logistic regression model using area-level surgery rate as the independent variable to test for trends between area-level rates and mortality and we calculated the adjusted odds of mortality associated with a 10% increase in surgery rates from this model.

To assess if other aspects of care for early-stage NSCLC also differed by area-level rates of surgery, we also assessed rates of radiation for 4955 patients with unresected stage I/II NSCLC (after excluding 235 patients who died within 6 months of diagnosis) and rates of adjuvant doublet chemotherapy among 12,148 patients who underwent surgery for NSCLC (after excluding 300 patients who died within 6 months of diagnosis) (coding radiation and chemotherapy as previously described²²). We used the Armitage test for trend to assess unadjusted rates and logistic regression models with the same covariates as for the mortality model to assess the adjusted association of area-level surgery rates with radiation for unresected patients and adjuvant chemotherapy among resected patients.

Finally, because there may be particular uncertainty about the risks versus benefits of curative lung cancer surgery for patients with COPD, where surgical rates overall are lower, we repeated analyses among patients with COPD.

Role of the funding source

The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; and preparation, review, or approval of the manuscript.

Results

Characteristics of patients by quintile of surgery rates are displayed in Table 1. In the areas with the highest rates of surgery for stage I/II NSCLC, more than 79% of patients underwent curative surgery, whereas fewer than 63% of patients underwent surgery in the areas with the lowest surgery rates. Highest-surgery areas had significantly fewer black patients (3.9% in the highest surgery areas vs. 7.4% in the lowest surgery areas), patients living in areas with more highly educated patients (21.7% in the top education quartile vs. 14.8% in the highest vs. lowest surgery areas), more patients with a prior history of cancer (29.9% vs. 26.6% in the highest vs. lowest surgery areas), and fewer patients with COPD (43.0% vs. 45.1% in the highest vs. lowest surgery areas). There were no significant differences in marital status, cancer stage or patient comorbidity by quintile. The relative balance by quintile of surgery rate suggests that this is an effective “instrument” by which to categorize patients.

Among patients who underwent surgery, the rates of advanced age and COPD were higher in areas with high surgical rates than in areas with lower surgical rates (Table 2), suggesting that surgeons in high-surgery areas are operating on more high-risk patients. For example, in the highest surgery areas, 41.4% of surgically resected patients had COPD, whereas in the lowest quintile areas, 37.8% had COPD.

Table 2.

Characteristics of patients who undergo surgery by quintile of surgery rates (N=12,448)

	Quintile of Proportion of Patients Undergoing Surgery					P-value*
	Quintile 1 (lowest)	Quintile 2	Quintile 3	Quintile 4	Quintile 5 (highest)
Age >85 years (%)	1.7	2.7	2.6	3.8	2.9	0.005
COPD (%)	37.8	38.5	40.5	38.6	41.4	0.02
Comorbidity >2 (%)	5.2	4.8	5.7	5.3	5.9	0.17

^*Using the Armitage test for trend

COPD=chronic obstructive pulmonary disease

Unadjusted and adjusted rates for one year all-cause and lung cancer specific mortality were significantly lower in higher surgical areas compared with lower surgical areas (Table 3). After adjusting for potential confounders, all-cause mortality at one year was 18.0% in areas with the highest surgery rates vs. 22.8% in areas with the lowest rates; with an adjusted odds ratio (OR) of 0.89 (95% confidence interval [CI] 0.86–0.93) for each 10% increase in surgery rates. Lung cancer specific mortality at one year was similarly reduced; with an adjusted OR of 0.86 (95% CI 0.82–0.91) for each 10% increase in surgery rates.

Table 3.

Unadjusted and adjusted^* rates of 1 year mortality of patients by quintile of surgery rates (N=17,638)

	Quintile of Proportion of Patients Undergoing Surgery					Adjusted odds-ratio associated with 10% increase in surgery rates (95% CI)*	P value
	Quintile 1 (lowest)	Quintile 2	Quintile 3	Quintile 4	Quintile 5 (highest)
All-cause mortality at 1 year (%)
Unadjusted	23.6	19.9	19.3	17.1	17.4	-	<0.0001¶
Adjusted†	22.8	19.6	19.3	17.3	18.0	0.89 (0.86 to 0.93)	<0.0001*
Lung cancer specific mortality at 1 year (%) ‡
Unadjusted	17.9	13.7	13.6	11.9	11.5	-	<0.0001¶
Adjusted†	16.9	13.4	13.7	11.6	12.0	0.86 (0.82 to 0.91)	<0.0001*

^*Adjusted odds ratio based on logistic regression model with linear term for area-level surgery rate to assess the association of area-level surgery rates with mortality, controlling for all variables in Table 1 and year of diagnosis.

^†Adjusted for all variables in Table 1 and year of diagnosis using logistic regression model with quintiles of area-level surgery rates; adjusted rates per quintile calculated using direct standardization.

^‡Lung cancer specific mortality was available through December 31, 2005, so the cohort was restricted to 14,001 patients diagnosed through December 31, 2004.

^¶Using the Armitage test for trend

CI=confidence interval

To better understand the mortality benefit associated with higher surgery rates, we questioned if patients in high-surgery areas were also getting other treatments likely to be of benefit, including use of radiation therapy for stage I and II patients who did not undergo surgical resection and use of adjuvant chemotherapy for stage I and II patients who underwent surgical resection (Table 4). After adjusting for potential confounders, patients in areas with higher surgery rates did not differ from patients in other areas in receipt of radiation for unresected disease (OR 0.96, 95% CI 0.87 to 1.05 for each 10% increase in surgery rates) or receipt of adjuvant chemotherapy for surgically resected disease (OR 0.99, 95% CI 0.92 to 1.06).

Table 4.

Unadjusted & adjusted rates of treatment ty quintile of treatment rate

	Quintile of Proportion of Patients Undergoing Surgery					Adjusted odds-ratio associated with 10% increase in surgery rates (95% CI)*	P value*
	Quintile 1 (lowest)	Quintile 2	Quintile 3	Quintile 4	Quintile 5 (highest)
Radiation for unresected stage I/II NSC lung cancer ‡
Unadjusted	68.2	65.8	65.5	63.3	63.3	-	0.009¶
Adjusted†	67.6	66.1	65.8	64.4	62.1	0.96 (0.87 to 1.05)	0.35*
Doublet chemotherapy among patients who received surgery §
Unadjusted	12.1	9.8	10.8	10.6	13.4	-	0.02¶
Adjusted†	12.2	11.0	10.5	11.9	11.5	0.99 (0.92 to 1.06)	0.75*

^*Adjusted odds ratio based on logistic regression model with linear term for area-level surgery rate to assess the association of area-level surgery rates with mortality, controlling for all variables in Table 1 and year of diagnosis.

^†Adjusted for all variables in Table 1 and year of diagnosis using logistic regression model with quintiles of area-level surgery rates; adjusted rates per quintile calculated using direct standardization.

^‡Among 4955 patients with stage I or II NSCLC who did not undergo surgical resection.

^¶Using the Armitage test for trend.

^§Among 12,148 patients with stage I or II NSCLC who underwent surgical resection.

CI=confidence interval

In sensitivity analyses, we repeated analyses among the 8584 patients with COPD, for whom the uncertainty about risks versus benefits of surgery may be greatest. Mortality rates were higher in these populations, but higher surgical areas had similarly better outcomes. The adjusted all-cause mortality at one year was 20.5% in areas with the highest surgery rates vs. 27.2% in areas with the lowest rates; with an adjusted odds ratio (OR) of 0.89 (95% confidence interval [CI] 0.84–0.94) for each 10% increase in surgery rates. Lung cancer specific mortality at one year was also lower in higher surgery areas; with an adjusted OR of 0.87 (95% CI 0.81–0.93) for each 10% increase in surgery rates.

Discussion

In this large, population-based study of early-stage lung cancer patients, we used geographic variation in surgery rates to mimic randomization to surgery to better understand the outcomes associated with surgery, particularly for patients where there may be uncertainty regarding benefits versus risks. We found that geographic areas with high rates of surgical resection more often operated on patients who were older and had COPD compared with geographic areas with lower rates of surgical resection. Even so, all-cause and lung cancer specific mortality at one year were significantly lower in high surgical areas than areas with lower surgery rates.

Our results are compelling because they provide evidence on a population level that higher rates of surgery for early-stage lung cancer improve mortality and that this mortality benefit is apparent even when older and sicker patients undergo operations. Although our findings do not suggest that all patients should undergo surgery, they suggest that for patients where the most uncertainty exists (e.g., patients where physicians may be uncertain about the risk/benefit trade-off), on average, the benefits of surgery are likely to outweigh the risks. Our findings are consistent with a prior observational study that demonstrated a survival benefit of surgery for NSCLC in an older population,¹² with significantly higher two year survival among patients with local/regional NSCLC cancer who underwent surgery (66%) compared with patients who received no treatment (17%) or radiation alone (15%). Such findings from observational studies could be biased if treated patients differed from untreated patients in ways that could not be measured (such as performance status or smoking status); our analysis avoids this concern by sorting patients into groups based on the likelihood of getting surgery in the area where they live. This population approach is similar to an analysis that uses area-level surgery rates as an instrumental variable for curative surgery¹⁸ and has been used previously to estimate effectiveness of treatments for cardiac disease in elderly populations.¹⁶

Previous research suggests that surgeons have substantial uncertainty about the benefits versus risks of surgery for patients with increasing severity of COPD and advanced age.¹⁴ Historically, advanced age has been a relative contraindication for lung resection—in the 1970's age greater than 70 years was considered prohibitory for thoracotomy.^{25, 26} Newer operative and anesthetic techniques combined with improved knowledge about pre-operative risk stratification expanded use of surgery for older NSCLC patients in the 1980's and 1990's.²⁶ Despite these trends, older patients are still less likely to undergo surgical resection than their younger counterparts.^{6–11, 13} This may reflect surgeons' concerns about older patients' risks for operative complications and post-operative death. Some studies suggest that older patients are at higher risk for operative complications and post-operative death than younger patients,^{27, 28} although other studies have failed to find such associations.^{29, 30} Importantly, a recent review found that older lung cancer patients had similar rates of operative complications, post-operative mortality and quality of life as those found in younger patients.²⁹

If relatively older and sicker patients are likely to benefit from higher surgery rates, it will be important to understand patients' preferences related to surgery and identify potential patient-related barriers to surgical excision. A recent study reported that older patients were more likely to decline lung cancer surgery, as were patients who did not believe their diagnosis, who had negative perceptions about quality of life after surgery, and who had more negative perceptions of doctor-patient communication.¹³ Beliefs that surgery causes cancer to spread are also associated with lower rates of surgery.³¹ Future work is needed to understand why older patients are more likely to decline surgery for early-stage lung cancer and if patient education interventions can improve resection rates among eligible patients.

In light of the large survival advantage associated with higher rates of surgery, we sought to understand if areas that did more surgery also provided patients with higher rates of radiation or chemotherapy. We found that higher-surgical areas did not differ from other areas in rates of radiation therapy for patients who did not undergo surgery or rates of adjuvant chemotherapy use post-operatively. Thus, care for early-stage lung cancer in high-surgery areas is not necessarily more aggressive overall, suggesting that the higher surgery rates are driving the better outcomes via higher cure rates.

There are several limitations of our study. First, our primary dependent variable is the rate of surgery among patients with lung cancer in the HSA of patients' residence, but could reflect care delivered outside of the HSA if some patients traveled beyond their HSA for treatment. Second, the use of area-level rates to mimic randomization to therapy requires the assumption that potential confounders, such as performance status or severe comorbid illness, do not differ by geographic region and that the area-level surgery rates influence the survival only through its effect on rates of surgery (and other post-surgical treatments). We observed some differences in patient characteristics by geographic area (for example, patients in higher surgery areas had higher incomes), and there may be unobserved differences as well. Further work is needed to validate the use of area-level treatment rates to mimic random assignment to treatments.

Additionally, we studied care only for patients aged 65 and older living in SEER areas and enrolled in fee-for-service Medicare; whether our findings are generalizable to other patients and non-SEER areas is a question for further study. Finally, although we had rich data about patient and tumor characteristics, we lacked data on potential confounders such as performance status and individual patient education. Nevertheless, to the extent that patients in different areas have generally similar performance status and other characteristics, the lack of such information is not likely to influence our results.

Despite the fact that surgical resection offers the greatest chance of cure for patients with early-stage NSCLC, many patients never receive surgery. Our findings suggest that surgery offers a benefit to patients with stage I and II NSCLC who may be older and/or sicker and may not otherwise have undergone surgery. The mortality benefit for high surgery areas is evident even though a greater proportion of surgeries in these areas are done in older patients and patients who have COPD. Future work should be aimed at decreasing patient and physician-related barriers to surgical resection in early-stage lung cancer and increasing surgical rates for individuals who may be perceived as borderline surgical candidates because of age or COPD alone. Randomized trials that include older patients and those with comorbid illness would be useful to better identify patients who are most likely to benefit from surgery.

Uncertainty exists about the benefits versus risks of surgery for early-stage non-small cell lung cancer in older patients. Using a geographic analysis to address bias, we observed that higher rates of surgery for stage I/II NSCLC are associated with improved survival, even when older patients and sicker patients undergo resection.