Bradley CJ, Given CW, Dahman B, Fitzgerald TL. Adjuvant Chemotherapy After Resection in Elderly Medicare and Medicaid Patients With Colon Cancer. Arch Intern Med. 2008;168(5):521–529. doi:10.1001/archinternmed.2007.82
This study investigated the influence of Medicaid enrollment on the receipt and completion of adjuvant chemotherapy and the likelihood of evaluation by an oncologist for those patients who do not initiate chemotherapy.
Medicaid and Medicare administrative data were merged with the Michigan Tumor Registry to extract a sample of patients who had resection for a first primary colon tumor diagnosed between January 1, 1997, and December 31, 2000 (n = 4765). We used unadjusted and adjusted logistic regression to assess the relationship between Medicaid enrollment and the outcomes of interest.
Relative to Medicare patients, Medicaid patients were less likely to initiate chemotherapy (odds ratio, 0.50; 95% confidence interval, 0.39-0.65) or complete chemotherapy (odds ratio, 0.52; 95% confidence interval, 0.31-0.85). When the sample was restricted to patients with TNM-staged disease, Medicaid patients were less likely to initiate chemotherapy. Older patients and patients with comorbidities were also less likely to initiate or, in some cases, to complete chemotherapy.
Medicaid enrollment is associated with disparate colon cancer treatment, which likely compromises the long-term survival of these patients.
The survival benefits of adjuvant chemotherapy for stage III colon cancer have been demonstrated,1- 3 even among elderly patients.4,5 Once patients initiate adjuvant chemotherapy, completion rates have been found to be as high as 78%.6 Modest survival benefits have been demonstrated for patients treated with chemotherapy for stage II colon cancer,7,8 but the use of chemotherapy for treating stage II cancer remains controversial and is recommended only in high-risk patients.9 However, chemotherapy administration across oncology practices suggests low levels of awareness of guidelines for chemotherapy, particularly among rural, community-based cancer care centers.10
In addition, patient characteristics such as female sex,6,7 comorbidity burden,11 African American race,6,12- 15 low income,6,14 and older age5,6,12,14 appear to be related to low rates of chemotherapy initiation. Furthermore, many patients are not referred to a medical oncologist for evaluation,16 indicating that these patients may not have had the opportunity to evaluate the benefits and risks of chemotherapy.
Medicaid-insured patients have poor colon cancer survival,17 but whether these patients receive less treatment than patients with other health insurance is unknown. We examined adjuvant chemotherapy initiation rates, completion rates, and evaluation by an oncologist in a sample of elderly patients who underwent resection for colon cancer. We were primarily interested in Medicaid enrollment and if it influenced chemotherapy initiation and completion and evaluation by an oncologist. Medicaid patients are of interest because they often embody the characteristics (eg, low income or racial or ethnic minority status) associated with disparities in the receipt of health care18 and poor cancer survivorship.
We used statewide Medicaid and Medicare data merged with the Michigan Tumor Registry to extract a study sample of patients with a first primary colon cancer diagnosis. The Michigan Cancer Surveillance Program, which maintains the Michigan Tumor Registry, ascertains more than 95% of all incident cancer cases based on external audit findings. This study was approved by institutional review boards at the Michigan Department of Community Health (Lansing), Michigan State University, and Virginia Commonwealth University.
Patients were matched to the Michigan state segment of the Medicare denominator file for the period from January 1, 1997, through December 31, 2000, using the patient's Social Security number. We extracted, from statewide Medicare files, all claims for inpatient, outpatient, and physician services during the study period for all patients who were correctly matched to the Michigan state segment of the Medicare denominator file (approximately 89% of patients) and were enrolled in Parts A and B. Michigan proved to be an ideal state for merging cancer registry and Medicare data because less than 3% of Michigan Medicare beneficiaries were enrolled in managed care. The process for linking the Michigan Tumor Registry, Medicare, and Medicaid data sets is described more fully elsewhere.19 We also used the unique physician identifying numbers listed in the claim files and linked with the Medicare Physician Identification and Eligibility Registry file to extract physician specialty information. To the linked Michigan Tumor Registry, Medicare, and Medicaid files, we added data from the 2000 Census Summary File using patient address information recorded in the Michigan Tumor Registry.
From this sample, we excluded patients who resided in a nursing home (Medicaid only, n = 228). Patients in private pay nursing homes remained in the Medicare sample because we could not identify them in the data set. Their presence increased the illness severity in the overall Medicare sample. We identified 8043 patients with colon cancer who were 66 years and older and who had undergone resection (International Classification of Diseases, Ninth Revision [ICD-9] codes 45.71-45.79, 45.8, 48.41-48.49, 48.50, and 48.61-48.69).20 We excluded all patients with less than 9 months of data after the month of diagnosis (n = 695) and those whose conditions were diagnosed in 1996 because they had less than 12 months of data before the month of diagnosis (n = 2291). We also excluded patients with invasive but unknown stage disease (n = 292); the remaining sample size was 4765, of which 458 were insured by Medicaid in addition to Medicare.
We used Medicare claims files to assess the outcomes of interest for the full sample and a sample restricted to patients with a TNM stage. We used this approach because the TNM stage was missing for 50% of the patients, but a Surveillance, Epidemiology, and End Results (SEER) summary stage was available for all patients.
Although TNM stage is more informative than the SEER summary stage, the National Program of Cancer Registries does not require state registries to report TNM stage. Michigan began requiring TNM stage in 1997 from facilities that had registries but excluded smaller facilities that did not have a registry (as may be the case in rural and underserved areas) from this requirement. Among patients for whom both SEER summary stage and TNM stage were known, there was a 90% or greater correspondence between TNM stage I and local stage, TNM stage III and regional stage, and TNM stage IV and distant stage. However, TNM stage II cancers were split between local (40%) and regional (57%) stages, making it difficult to infer TNM stage from a summary stage of regional or local. A TNM stage II tumor that has penetrated nonperitonealized pericolonic tissue is considered regional even in the absence of positive lymph nodes. We report findings from the sample with a TNM stage because it is more precise with regard to stage but also report findings from the entire sample so as not to exclude patients treated in small, rural, and/or underserved regions of the state.
Chemotherapy initiation was identified by at least 1 claim indicating the administration of chemotherapy within 6 months after diagnosis (Current Procedural Terminology codes 96400-96599; Health Care Common Procedural Codes Q0083-Q0085, J8510, J8520, J8521, J8530-J8999, J9000-J9999, and J0640; and ICD-9 codes E0781, E9331, and V58.1). Hershman et al21 found that 91% of elderly patients with colon cancer initiate chemotherapy within 3 months of diagnosis but that older patients and patients with comorbid conditions were more likely to start chemotherapy 3 or more months after diagnosis.
We defined a complete course of adjuvant chemotherapy as 5 consecutive months of chemotherapy with 1 claim day in a month. To avoid misclassifying chemotherapy for cancer recurrences, we counted claims that ended (1) with the claim date after which there were 3 months without any type of colon cancer treatment, (2) with a cancer recurrence, or (3) 9 months after diagnosis, whichever came first. Recurrent cancer was identified by the following codes: ICD-9 codes 50.20 to 50.22, 50.29, 50.3, or 50.4; ICD-9 code 197.7; or Current Procedural Terminology codes 36246, 36247, 47120, 47122, 47125, 47130, 47370, 47371, 47380, 47381, 47382, 76362, 76394, 76490, 36260, or 47100. Secondary malignancies were identified by the following ICD-9 codes: 197.0, 197.1, 197.2 197.3, 197.8, 198.3 to 198.5, 198.41, 198.45, 198.48, 198.51, 197.04, or 197.08. This method for counting complete chemotherapy administration has been validated by a prior nationwide study that examined practice patterns during our study period.6 In the assessment of chemotherapy completion, we excluded patients diagnosed as having distant stage disease or TNM stage IV disease, depending on the sample, because a complete course of chemotherapy was not defined for these patients.
We assessed whether patients who did not initiate chemotherapy were evaluated by an oncologist. Our process for identifying oncologists was as follows. First, we examined the physician specialty codes listed in the Medicare Physician Identification and Eligibility Registry file that coincided with the date of any chemotherapy administration, regardless of cancer site. Most chemotherapy claims (68%) were submitted by either a medical oncologist (specialty code 90) or hematologist/oncologist (specialty code 83). The remaining claims were coded as either internal medicine or clinic or group practice. Second, we designated all physicians' unique physician identifying numbers that were ever associated with an oncology specialty or the administration of chemotherapy as an “oncology specialist,” using all claims (including those from cancer sites other than colon) in the Michigan state segments of the carrier file for 1996 through 2000. Our method overestimates the number of oncologists, which reduces the chance of observing a statistically significant result. For example, among patients who did not initiate chemotherapy, 84% had been evaluated by an oncologist.
Data on patient age, race, and sex were obtained from the Michigan Tumor Registry. Age was grouped into the following categories: 66 to 69 years, 70 to 74 years, 75 to 79 years, and 80 years and older. Race was categorized as white or other, although most of these latter patients were African American (86%). In addition to these variables, we included information on patients' census tract median household income. The income categories were less than $25 000, $25 001 to $35 000, $35 001 to $45 000, and more than $45 000. We also included variables that indicated whether the patient lived in a metropolitan area, urban area adjacent to a metropolitan area, urban area not adjacent to a metropolitan area, rural area adjacent to a metropolitan area, or an isolated rural area. Address information was not available for 5% of the patients. Therefore, we used the monotone multiple imputations method with logistic regression to impute the missing categorical variables for both census median income and rural and urban residence.22 Analyses were conducted with and without imputed values, and the findings were nearly identical.
Subsequent hospitalization after resection was negatively associated with chemotherapy completion in other published studies.6 Therefore, we constructed a variable that reflected subsequent hospitalization 1 to 6 months after resection of the colon. We did not include hospitalizations that were for chemotherapy or radiation therapy. In addition, we included a variable that indicated whether the patient's resection occurred in a teaching hospital.
To estimate patient comorbidity burden, we used the adaptation of Deyo et al23 and Klablunde et al24 of the Charlson Comorbidity Index,25 which has been used to explain the probability and extent of cancer treatment.24,26 We counted comorbidities by using all inpatient, outpatient, and physician claims for services rendered to patients in the year before diagnosis. We classified comorbidity scores into 3 groups: 0, 1, and 2 or higher. We chose the modifications of Deyo and colleagues and Klablunde and colleagues of the Charlson Comorbidity Index because they are conducive to assessing comorbidity burden with administrative claims data. However, in elderly patients with cancer, the Charlson Comorbidity Index score does not adequately reflect functional ability or predict tolerance to treatment.27,28 Finally, we controlled for SEER summary stage or TNM stage, depending on the sample.
We described the characteristics of all patients with colon cancer by Medicare and Medicaid enrollment and used χ2 tests to test for statistical differences. We also used χ2 tests to test for unadjusted differences in the outcomes of interest and the independent variables. Adjusted logistic regression modeling was then used to measure the relationship between the independent variables and the initiation and completion of chemotherapy and evaluation by an oncology specialist. We report herein odd ratios (ORs), 95% confidence intervals (CIs), and P values. P values are derived from likelihood ratio tests and are 2-sided. All analyses were conducted using SAS statistical software, version 9.13 (SAS Institute Inc, Cary, NC).
In the full sample, Table 1 reports the descriptive characteristics for patients by Medicare and Medicaid status. Patients were similar in terms of age, urban and rural residency, and SEER summary stage but were different along every other dimension studied. Medicaid-insured patients were more likely to be African American or another minority race and female, to have more comorbid conditions, to be readmitted to the hospital after resection, and to live in low-income census tracts. When we restricted the sample to patients with TNM staging, these differences persisted with the exception of hospital readmissions and comorbidity burden, which were no longer statistically significantly different between Medicaid- and Medicare-insured patients. Among patients with TNM-staged disease, Medicaid patients were more likely to have later-stage disease relative to Medicare patients.
Table 2 reports the unadjusted relationship between the independent variables and the outcomes we studied for the full sample. Medicaid insurance, older age, high comorbidity, low census tract income, and early-stage disease were all statistically significant and negatively associated with chemotherapy initiation and completion (P < .05). Women were significantly less likely to initiate chemotherapy (P < .001) than men, but once started, women were equally as likely to complete chemotherapy as men. Minority race was marginally statistically significant (P = .10) and negatively associated with chemotherapy initiation. Living in urban areas was negatively associated with chemotherapy completion. Medicaid insurance, high comorbidity, hospital readmission, low census tract income, living in an isolated area, and having advanced-stage disease were all negatively associated with being evaluated by an oncologist (P < .05).
Table 3 reports similar results for the restricted sample with some important exceptions. Medicaid insurance was no longer a statistically significant predictor of chemotherapy completion or oncology evaluation. In the case of chemotherapy completion, the lack of statistical significance was most likely related to a diminished sample of Medicaid patients.
Table 4 reports the adjusted logistic regressions for all patients with colon cancer. Patients insured by Medicaid were considerably less likely than Medicare patients to initiate chemotherapy (OR, 0.50; 95% CI, 0.39-0.65). Older age, high comorbidity, resection in a teaching hospital, and early-stage disease were all negatively and statistically significantly associated with initiating chemotherapy. Medicaid patients were half as likely to complete chemotherapy as Medicare patients (OR, 0.52; 95% CI, 0.31-0.85), as were patients 80 years and older and patients with comorbid conditions relative to their younger and healthier counterparts. Regarding evaluation by an oncologist, Medicaid-insured patients were less likely to be evaluated by an oncologist (OR, 0.72; 95% CI, 0.53-0.97). Other patients at risk for not being evaluated by an oncologist included low-income patients, patients living outside metropolitan areas, and patients with a disease stage other than regional. When the sample was restricted (Table 5), the predictors of chemotherapy initiation and completion were remarkably similar, although with the loss of statistical significance with respect to chemotherapy completion and Medicaid; the ORs for Medicaid insurance were nearly identical.
We examined chemotherapy initiation, completion, and, for those patients who did not initiate chemotherapy, the likelihood of being evaluated by an oncologist. The analysis was conducted on 2 samples: one that consisted of the entire state of Michigan and the other that was confined to patients with a TNM stage. The latter sample contained a greater proportion of patients treated in a teaching hospital and patients who resided in a metropolitan area. Regardless of these differences, the findings were remarkably similar. This study did not examine the appropriateness of cancer treatment (for example, the proportion of patients with TNM stage III disease who received chemotherapy) but instead predicted treatment patterns for all patients with resected colon cancer, regardless of stage. The evidence suggests that chemotherapy is not confined to TNM stage III disease and that perhaps some patients are treated inappropriately.
The main finding of this study is that Medicaid insurance is associated with a low likelihood of chemotherapy initiation and completion, and there is conflicting evidence regarding Medicaid patients' likelihood of being evaluated by an oncologist. Several explanations for these observations are plausible. First, Medicaid-insured patients may have poorer health status than Medicare patients. However, Medicaid patients residing in nursing homes were removed from the sample, which should have eliminated cases of greater dependence. Second, Medicaid-insured patients may have fewer support services (eg, transportation or caregivers) to help them get to physicians' offices and to care for them after treatment. Finally, Medicaid-insured patients may have a lower acceptance rate of chemotherapy.
Another important finding of the study is that older patients are less likely to initiate chemotherapy. Several other published studies5,29 have found treatment disparities between older and younger adults. These studies conclude that cancer survival and time to recurrences could be improved for these patients with the administration of chemotherapy.5,29 We speculate that conservative practice patterns, poor physician-patient communication about the benefits and risks of chemotherapy, and/or differences in baseline functional status, which may not be captured by comorbidity measures, lead to low chemotherapy initiation rates. Nevertheless, evidence suggests that patients with colon cancer and chronic conditions such as heart failure, diabetes mellitus, and chronic obstructive pulmonary disease benefit from chemotherapy.30 Once chemotherapy is initiated, only the oldest patients discontinue chemotherapy prematurely relative to their younger counterparts.
The study has 3 main limitations. First, the study is specific to Michigan, so it may not be generalizable to other states or regions. However, the only way to identify Medicaid-insured patients at this time is at the state level. The state buy-in variable, which is in the Medicare denominator file, does not adequately identify Medicaid patients. Second, Medicare nursing home patients could not be identified. Had we been able to identify them, we may have observed greater differences between Medicaid and Medicare patients in the outcomes we studied. Finally, published estimates indicate that only half of elderly Medicare beneficiaries with incomes at or below the poverty level enroll in Medicaid.31 The inclusion of elderly patients who qualify but are not enrolled in Medicaid would diminish the relationship between Medicaid and the outcomes we studied.
This study found that Medicaid-insured patients, relative to Medicare-insured patients, are less likely to initiate and complete chemotherapy. These patients may have poorer health status, but they may also have inadequate financial and social support to help get and continue chemotherapy. Furthermore, physicians may need to spend extra time with these patients explaining the benefits and risks of treatment. Clearly, further research is needed to understand the underlying causes for poor chemotherapy initiation. Nevertheless, as Medicaid administrators across the nation move toward a managed care model of care delivery, they need to be cognizant of differential treatment of their patients for cancer and include initiatives that increase adherence to cancer treatment regimens. Progress toward increasing compliance also needs to be measured. As long as these substantially large groups of patients (Medicaid-insured and elderly patients) have disparate treatment uptake and compliance, the nation as a whole will have difficulty reaching its goals for reduced cancer mortality.
Correspondence: Cathy J. Bradley, PhD, Department of Health Administration and Massey Cancer Center, Virginia Commonwealth University, Grant House, 1008 Clay St, PO Box 980203, Richmond, VA 23298-0203 (email@example.com).
Accepted for Publication: September 26, 2007.
Author Contributions:Study concept and design: Bradley and Dahman. Acquisition of data: Bradley, Given, and Dahman. Analysis and interpretation of data: Bradley, Given, Dahman, and Fitzgerald. Drafting of the manuscript: Bradley, Given, and Fitzgerald. Critical revision of the manuscript for important intellectual content: Bradley and Dahman. Statistical analysis: Bradley and Dahman. Obtained funding: Bradley and Given. Administrative, technical, and material support: Bradley. Study supervision: Bradley.
Financial Disclosure: None reported.
Funding/Support: This research was supported by National Cancer Institute grant R01-CA101835-01, “In-Depth Examination of Disparities in Cancer Outcomes” (Dr Bradley).