Treatment Disparities for Disabled Medicare Beneficiaries With Stage I Non-Small Cell Lung Cancer

Data Sources 

SEER data include 11 population-based tumor registries, representing 14% of the U.S. population.14 SEER gathers information on all incident cases of cancer diagnosed within geographically defined areas including 5 states (Connecticut, Hawaii, Iowa, Utah, New Mexico) and 6 metropolitan areas (Atlanta, Detroit, San Francisco/Oakland, Seattle/Puget Sound, Los Angeles county, San Jose/Monterey). The last 2 registries joined SEER in 1992. The Centers for Medicare & Medicaid Services (CMS) links data from SEER-11 registries with Medicare enrollment and utilization information for Medicare beneficiaries diagnosed with cancer.14, 15

SEER registries identify cases primarily by reviewing hospital pathology reports and discharge diagnoses, achieving 98% case ascertainment. SEER collects information on each patient’s demographic and tumor characteristics at diagnosis, including primary tumor site, stage, size, histology, and grade. SEER records initial treatment (within 4 months of diagnosis from 1973 to 1998, within 12 months of diagnosis after 1998) and generally captures all surgery and radiation therapy.16, 17 For persons without surgery, SEER categorizes reasons for no surgery (eg, surgery not recommended, contraindications due to other conditions, patient refused). Registries collect chemotherapy information but do not release it because of concerns about incomplete data. SEER tracks vital status annually, obtaining underlying cause of death from death certificates.

CMS uses different algorithms to match SEER with Medicare data for beneficiaries under and over age 65 years. For persons over the age of 65 years, 94% of cases match.15 For persons under age 65 years, match rates are unknown but are probably considerably lower because a more stringent algorithm accepts only exact matches on Social Security number. With this strict requirement, we feel confident of the accuracy of matches for those persons under age 65 years who did match, but some unknown number of persons who did have SSDI/Medicare may not have been identified.

We analyzed SEER Public Use18 and Patient Entitlement and Diagnosis Summary files from the SEER-Medicare database, which contains demographic and annual enrollment information through 2001 from Medicare denominator files for program beneficiaries. We used Medicare enrollment dates extracted from the Medicare Enrollment Database to identify persons under age 65 years who qualified for SSDI, were on Medicare, and thus were disabled when diagnosed with non-small cell lung cancer (ie, SSDI/Medicare). SSA aggregate data indicate that neoplasms caused 9.8% of new SSDI disability determinations in 2002.19 As described elsewhere, we constructed an algorithm to identify and eliminate people who likely applied for SSDI and thus received Medicare because of cancer.20 We drew our nondisabled cases (ie, persons without SSDI/Medicare) from the SEER public use file.

SSDI is available to all working-age adults, regardless of their financial resources, who qualify as “insured” based on quarters of employee contributions to Social Security’s trust fund (in contrast, Supplemental Security Income [SSI] is the means-tested, income support program for disabled persons who are poor). To determine whether these people meet disability criteria, SSA requires specific medical evidence and diagnostic testing, physical examination, or other clinical data documenting disability expected to last 12 or more months. When people qualify for SSDI, SSA records their primary impairment or the reason that persons were “medically determined” disabled. SSA does not release impairment codes to nongovernmental investigators, although federal investigators have productively used these codes in research studies.21, 22 To explore whether specific disabilities had different associations with treatment and survival, officials at SSA, CMS, and the National Cancer Institute asked their data processing contractor to merge SSA impairment codes and diagnosis groups with our final analytic file, which we provided to them. To protect SEER registry confidentiality, we did not have access to these merged files but instead worked through the government contractor to analyze the SSA information.

Study Sample 

This retrospective cohort study involved persons ages 21 to 64 years diagnosed with stage I first primary non-small cell lung cancer between January 1, 1988, and December 31, 1999, residing in SEER-11 coverage areas (N=9500). We included persons with tumors that were classified as stage I by the American Joint Committee on Cancer criteria (ie, no regional or distant metastases). We excluded 59 patients with unknown month of diagnosis or whose cancer was reported at autopsy, was reported on death certificate, or was diagnosed without pathologic tissue confirmation. Our final study sample included 9441 persons; 1016 (10.8%) had SSDI/Medicare. For analyses using SSA disability determination information, we excluded 38 persons with SSDI/Medicare who did not link with SSA data.

Disability Diagnosis Groups 

After merging SSA data with our analytic file, the government contractor produced frequency distributions for each impairment code (SSA’s most granular listing of conditions) and diagnosis group (SSA’s groupings of impairment codes). Very few cases fell into individual impairment codes; many diagnosis groups also had too few cases for separate analysis. Of the 978 merged cases with SSDI/Medicare, 222 (22.7%) had codes that provided limited clinical insight (ie, 59 people with relatively rare, specific, but clinically heterogeneous conditions that we grouped as miscellaneous and 163 with unknown conditions). To increase sample sizes for statistical analyses, we combined some diagnosis groups. Here we present data for the 5 broad conditions with sufficient numbers for analysis (table 1 footnotes).

Table 1. Demographic and Tumor Characteristics by Medicare/SSDI Status and for Selected Disability Categories

	Demographic and Tumor Characteristics†	Not SSDI‡	All SSDI§	Selected SSA Diagnosis Groups⁎
				Mental∥	Neurologic¶	Circulatory#	Respiratory⁎⁎	Musculoskeletal††
	No.ofcases	8425	1016	170	74	179	136	197
	Meanage±SD(y)	56.6±6.5	58.7±5.3	56.8±6.0	57.7±5.2	59.9±4.0	59.1±5.2	59.3±4.6
	Female(%)	45.9	33.0	37.1	29.7	19.0	37.5	33.5
	Raceandethnicity(%)
	Non-Hispanicwhite	76.9	73.9	74.7	81.1	74.9	72.8	75.6
	Non-Hispanicblack	13.2	19.1	18.8	12.2	20.1	21.3	17.3
	Hispanic	3.5	3.7	3.5	§§	2.8	§§	4.1
	Maritalstatus(%)
	Nevermarried	11.0	15.9	27.7	17.6	8.4	14.0	12.2
	Married	65.5	49.3	31.2	48.7	59.2	52.9	52.8
	Widowed	6.7	10.3	7.1	10.8	10.1	11.0	8.6
	Divorced,separated,other	16.8	24.4	34.1	23.0	22.3	22.1	26.4
	Tumorsizeincm(%)
	≤1	5.3	3.9	4.1	§§	3.4	4.4	5.1
	1−2	24.7	23.9	28.2	25.7	22.9	20.6	27.4
	2−5	45.1	47.1	44.7	43.2	46.9	46.3	49.2
	>5	13.2	9.2	10.6	12.2	8.4	9.6	7.6
	Tumorgrade(%)‡‡
	Well-differentiated	9.8	7.8	8.2	10.8	7.8	5.2	4.6
	Moderatelydifferentiated	28.4	29.1	31.2	23.0	25.1	25.0	35.0
	Poorly/undifferentiated	36.0	34.3	31.2	40.5	37.4	28.7	40.1

Abbreviation: SD, standard deviation.

⁎ SSA diagnosis groups classified as miscellaneous (n=59) or unknown (n=163) are not shown. † Figures may not add to 100% because categories within characteristics with very small numbers of cases are not shown. ‡ Persons under age 65 years who did not receive SSDI/Medicare. § All persons under age 65 years who received SSDI/Medicare, regardless of SSA disability category. ∥ Mental health disorders and mental retardation. ¶ Nervous system and sense organ disorders and accidents, including spinal cord and other injuries. # Circulatory system disorders. ⁎⁎ Respiratory system disorders. †† Musculoskeletal and connective tissue conditions. ‡‡ P values for comparisons between persons with and without SSDI and among the 5 disability groups are all <.001, except for tumor grade, which is P=.05. §§ Cell sizes with fewer than 5 cases.

Non-Small Cell Lung Cancer Treatment 

Depending on tumor size, histology, and location, surgery can provide definitive treatment for non-small cell lung cancer.2, 3 Radiotherapy can potentially cure persons with resectable tumors who have medical contraindications to surgery.23 We defined initial treatment using only SEER information because Medicare claims, which could have identified chemotherapy, were available only for persons with SSDI/Medicare. Adjuvant cisplatin-based combination chemotherapy may prolong survival for surgical patients with stage IB disease,24 although chemotherapy use remains controversial.

As have others,4 we identified surgical resection with curative intent as follows: radical or partial pneumonectomy, lobectomy, bilobectomy, sleeve resection, segmentectomy, wedge resection, and local resection. All these surgeries were explicitly for lung cancer. As noted above, for patients without surgery, SEER records indicate whether surgery was not recommended or contraindicated. We also identified whether patients received radiotherapy.

Survival 

We examined survival (all-cause and cancer-specific) after diagnosis. We measured survival time as number of days from diagnosis until death or December 31, 2001, whichever came first. For all-cause mortality, we censored observations of persons alive when follow-up ended (n=3965). We also studied cancer-specific deaths, censoring observations of persons alive when follow-up ended or who died from causes other than lung cancer (n=5758).

Analysis 

All statistical analyses used SAS.a We conducted the analyses that used the SEER-Medicare data only. For analyses using merged SEER-Medicare-SSA data, we supplied SAS code to the government contractor, who ran analyses for us. After internal quality assurance audits, the contractor returned aggregated results (ie, we did not receive information on individual cases). Because all analyses used deidentified data, our institutional review boards exempted this study from review.

Using bivariable analyses, we compared demographic and tumor characteristics at diagnosis by SSDI/Medicare (disability) status. We conducted separate multivariable logistic regression to examine adjusted associations between disability status and each treatment (surgery, radiotherapy) after adjusting for age at diagnosis (continuous), sex, race and ethnicity (non-Hispanic white, non-Hispanic black, Hispanic, Asian American/Pacific Islander, other), marital status at diagnosis (married, widowed, never married, other), SEER tumor registry, year of diagnosis, tumor size (continuous, in centimeters), and grade (well differentiated, moderately differentiated, poorly and/or undifferentiated). The standard demographic characteristics could affect treatment and longevity risks, preferences for treatment, and possibly other practice pattern effects (eg, racial and ethnic differences in treatments).4, 5 Adjustments for SEER tumor registry aim to account for possible geographic differences in practice styles; adjustment for year of diagnosis accounts for possible changes over time in care (although surgery has remained the definitive therapy over the years studied). Tumor characteristics affect longevity and certain treatment choices. In each model, we compared persons with and without SSDI/Medicare. We converted odds ratios to relative risks (RRs)25 with 95% confidence intervals (CIs) for each treatment outcome.

We conducted multivariable Cox proportional hazards regression to estimate adjusted relative hazard ratios (HRs) for each mortality outcome (all-cause, cancer-specific). We fit 4 separate proportional hazards models for each mortality outcome. Model 1 estimated the unadjusted relative HR comparing persons with and without SSDI/Medicare. Model 2 adjusted this relative HR for age at diagnosis (continuous), sex, race and ethnicity, marital status, tumor registry, year of diagnosis, tumor size (continuous), and tumor grade. Model 3 further adjusted the relative HR for receipt of surgery. Finally, model 4 further adjusted the relative HR for receipt of radiotherapy. We present adjusted relative HRs and 95% CIs: adjusted relative HRs greater or less than 1.00 indicate longer survival times for disabled compared with nondisabled persons; adjusted relative HRs greater or less than 1.00 indicate shorter survival times.

Treatment Differences 

Rates of surgery and radiotherapy differed between persons with versus without SSDI/Medicare (table 2). Although 82.2% of nondisabled persons had surgery, only 68.5% of persons with disabilities received operations. In contrast, 25.4% of persons with SSDI/Medicare had radiotherapy, compared with 17.6% of persons without disabilities. Among those who did not undergo surgery, surgery was not recommended for 51.6% of persons with disabilities and for 50.1% of nondisabled persons; 15.6% of persons with disabilities had contraindications due to other conditions, as did 12.3% of nondisabled persons; and 6.9% of those with disabilities refused surgery, as did 5.6% of those without disabilities. Combined, all these factors do not explain the 13.7% difference in surgery rates observed between the 2 groups. Adjusting for demographic and tumor characteristics also did not explain the significantly lower use of surgery and higher use of radiotherapy among persons with disabilities.

Table 2. Treatment by Medicare/SSDI Status and Selected Diagnosis Groups

	Disability Status	Surgery		Radiotherapy
		Surgery (%)	Adjusted RR⁎ (95% CI)	Radiotherapy (%)	Adjusted RR⁎ (95% CI)
	Not SSDI/Medicare	82.2	1.0	17.6	1.0
	All persons with SSDI/Medicare	68.5	0.84(0.79−0.89)	25.4	1.43(1.23–1.65)
	Selected SSA diagnosis groups†
	Mental disorders	70.0	0.92(0.86−0.99)	23.0	1.41(1.02–1.93)
	Nervous system disorders	64.9	0.86(0.76−0.98)	25.7	1.54(0.98–2.43)
	Circulatory disorders	69.3	0.94(0.88−1.01)	25.1	1.27(0.91–1.78)
	Respiratory disorders	48.5	0.76(0.67−0.85)	34.6	2.12(1.61–2.79)
	Musculoskeletal disorders	79.2	0.99(0.94−1.05)	19.8	1.26(0.92–1.74)

⁎ Adjusted for age at diagnosis (continuous), sex, race and ethnicity, marital status, year of diagnosis, tumor registry, tumor size (continuous), and tumor grade. † In models with SSA diagnosis groups, 81% of persons without SSDI/Medicare (ie, reference group) had surgery and 17.4% of persons without SSDI/Medicare had radiotherapy.

Surgical and radiotherapy rates varied substantially across disability groups, with persons with respiratory disorders having the lowest surgery rate (48.5%) and the highest radiotherapy rate (34.6%). Surgery was not recommended for 67.1% and contraindicated because of other conditions for 12.7% of those with respiratory disorders who did not receive surgery; this was the only group that differed substantially in surgical recommendations from those without disabilities. In contrast, persons with musculoskeletal disorders had the highest surgical rate (79.2%) and the lowest radiotherapy rate (19.8%). After adjustment, persons with nervous system disorders (adjusted RR=.86; 95% CI, .76–.98), mental health/mental retardation disorders (adjusted RR=.92; 95% CI, .86–.99), and respiratory disorders (adjusted RR=.79; 95% CI, .67–.85) continued to have significantly lower surgical rates than persons without disabilities. Persons with the latter 2 conditions also had significantly higher use of radiotherapy.

Survival Differences 

Differences in rates of surgery appear to explain the significantly higher cancer-specific mortality observed for persons with versus without disabilities (table 3). Persons with disabilities had significantly higher cancer-specific mortality rates (adjusted relative HR=1.23; 95% CI, 1.10–1.39) after adjusting for demographic and tumor characteristics. However, adjustment for receipt of surgery eliminated statistically significant differences (adjusted relative HR=1.08; 95% CI, 0.96–1.21). Further adjustment for receipt of radiotherapy did not change the relative hazard ratio appreciably. A similar pattern occurred within all disability categories: adjusting for use of surgery eliminated statistically significant differences observed in cancer-specific mortality, even for persons with respiratory disabilities.

Table 3. Cancer-Specific and All-Cause Mortality by Medicare/SSDI Status and for Selected Diagnosis Groups

	Demographic and Tumor Characteristics	Not SSDI	All SSDI	Selected SSA Diagnosis Groups⁎
				Mental	Neurologic	Circulatory	Respiratory	Musculoskeletal
			HR (95% CI)†
	Cancer-specific mortality
	Unadjusted	1.0	1.37(1.24−1.51)	1.23(0.97−1.56)	1.67(1.22−2.28)	1.23(0.98−1.55)	1.88(1.47−2.34)	0.95(0.75−1.20)
	Adjusted for demographic and tumor characteristics‡	1.0	1.23(1.10−1.39)	1.20(0.92−1.56)	1.64(1.15−2.34)	1.04(0.79−1.37)	1.75(1.34−2.30)	0.85(0.65−1.11)
	Adjusted further for surgery	1.0	1.08(0.96−1.21)	1.12(0.86−1.46)	1.14(0.80−1.64)	0.95(0.73−1.25)	1.15(0.88−1.51)	0.84(0.64−1.10)
	Adjusted further for radiotherapy	1.0	1.07(0.95−1.20)	1.13(0.86−1.47)	1.12(0.78−1.60)	0.97(0.74−1.27)	1.14(0.87−1.50)	0.81(0.62−1.06)
	All-cause mortality
	Unadjusted	1.0	1.66(1.54−1.79)	1.50(1.25−1.79)	1.72(1.32−2.24)	1.65(1.39−1.95)	2.29(1.91−2.75)	1.24(1.04−1.47)
	Adjusted for demographic and tumor characteristics	1.0	1.50(1.37−1.64)	1.43(1.17−1.75)	1.62(1.20−2.18)	1.43(1.18−1.74)	2.15(1.74−2.65)	1.11(0.92−1.35)
	Adjusted further for surgery	1.0	1.34(1.22−1.46)	1.34(1.10−1.64)	1.17(0.87−1.57)	1.34(1.11−1.63)	1.46(1.19−1.81)	1.10(0.91−1.33)
	Adjusted further for radiotherapy	1.0	1.33(1.22−1.46)	1.35(1.10−1.65)	1.14(0.85−1.53)	1.36(1.12−1.66)	1.45(1.17−1.79)	1.07(0.88−1.30)

NOTE. See table 1 for footnote details.

⁎ All footnotes pertaining to Selected SSA Diagnosis Groups in table 1 also apply to table 3. † HR<1.00 indicates longer survival time among patients with SSDI/Medicare compared with patients without and HR>1.00 indicates shorter survival time. ‡ Adjusted for age at diagnosis (continuous), sex, race and ethnicity, marital status at diagnosis, SEER tumor registry, and year of diagnosis.

All-cause mortality rates, however, remained significantly higher for persons with versus without disabilities, even after adjusting for demographic and tumor characteristics, surgery, and radiotherapy (see table 3). All-cause mortality rates also remained significantly higher for persons within 3 of the 5 disability categories: mental health/mental retardation disorders; circulatory system disorders; and respiratory system disorders.

Study Limitations 

This study has other important limitations relating to the database. The SEER public use file does not contain information on adjuvant chemotherapy, although this treatment remains controversial. Although SEER data indicate whether tumors had pathologic tissue confirmation, they do not report whether the staging itself was pathologically confirmed. One study using Medicare procedure codes found that black lung cancer patients age 65 years and older were less likely to receive invasive staging procedures than whites.5 However, another study found high agreement between clinical and pathologic stage for persons with stage IA-IB disease.37 Although the SSA data provided useful insight into underlying disabling conditions, small numbers of cases and clinically imprecise categories complicated this effort. SEER does not provide information on patients’ treatment preferences, their physicians, or the health care system where they receive their care.

Finally, given the high population prevalence of disability in the United States, even among persons under age 65 years,10, 11 many persons in the non-SSDI population likely also had various disabilities. Our findings may not generalize to persons with disabilities who do not apply or qualify for SSDI and Medicare, for whatever reason. In particular, the data do not identify persons receiving only SSI, the income support program for persons with disabilities who are poor or have not paid sufficient payroll taxes. Impoverished SSI recipients and low income or uninsured persons with disabilities who have not yet applied for disability benefits face financial barriers to accessing health care. Medical indigence may heighten risks of poorer general health and reduce survival.38, 39

Despite these limitations, our study raises questions about treatment of disabled Medicare beneficiaries under age 65 years who develop stage I non-small cell lung cancer. Studies using similar methods and data for older Medicare beneficiaries found that adjusting for treatment differences explained disparate mortality rates by race: our study is the first to highlight similar conclusions for a large population of persons with disabilities. Thus, our study offers an early view of potential treatment and outcome disparities for people with disabilities—calling for future research to understand these findings just as has occurred over the last decade in investigations of racial and ethnic disparities.