Data source
The Surveillance, Epidemiology and End Results (SEER)-Medicare linked database was utilized for this analysis. This database is a linkage of records from SEER, a large cancer registry, and Medicare, the national healthcare system for older persons. The SEER registry is designed to track primary cancer incidence and survival in the United States. Although not all US states report cancer cases to the SEER program, it is designed to be representative of data from the whole country. SEER participation has expanded greatly since its inception. While initially capturing approximately 10% of the US population in 1973, the SEER program has grown to comprise 26% of the US population currently [8]. The population covered by SEER is comparable to the general US population with regard to measures of poverty and education. However, the SEER registry population represents a somewhat more urban population and has a higher proportion of foreign-born persons than are found in the general US population. The compilers of the SEER registry continually monitor and evaluate their data to ensure high quality. The program's standard for registry completeness for the incidence of cancer cases is 98% [9]. The SEER registry collects information on gender, race, and dates of birth and death, cause of death, as well as tumor-specific information such as diagnosis date, tumor stage at diagnosis, tumor grade, ER and progesterone receptor status, and surgery and/or radiation treatments within four months after cancer diagnosis.
Medicare is a US system of federally funded health insurance for persons ages 65 years and older. Medicare part A covers hospital, skilled-nursing facility, hospice and some home health care services. Medicare part B covers physician and outpatient services. Virtually all beneficiaries receive part A, and 96% of beneficiaries choose to pay a monthly premium for part B. Medicare patients can be enrolled in either a fee-for-service (FFS) or health maintenance organization (HMO) plan. Claims files are not available for the 17% of Medicare recipients who are in an HMO plan [9]. Medicare data include demographic information and surgical and radiation treatment information that are already in the SEER database, as well as information on comorbid conditions and chemotherapy treatment. The SEER database has a unique code that does not include identifying information; therefore, the research that we conducted was exempt from institutional review board approval by the Colorado Multiple Institutional Review Board.
Study population
Women diagnosed with primary cancers of the breast at age 66 years or older and entered into the SEER data set from 1 January 1992 through 31 December 2000 were eligible for inclusion into our study. Women ages 65 years old were not included to allow for a one-year time period following Medicare enrollment during which comorbidities could be recorded in claims files. Women were excluded if they (1) lacked full insurance coverage under both Medicare part A and part B, (2) were enrolled in a Medicare HMO, (3) had an unknown month of cancer diagnosis, (4) had the same month of diagnosis and death, (5) had an unknown cause of death, and/or (6) had records considered to have unreliable Medicare coding for a comorbid condition (most commonly bills that were not encoded by a clinician for a specific reason for the visit). These exclusion criteria ensured that claims files were available for accurate detection of comorbidities and calculation of survival times from diagnosis to death.
A total of 96,954 women ages 66 years and older diagnosed with malignant breast cancer between 1992 and 2000 were potentially eligible for inclusion in this study. Just over one-third (33,388, or 34.4%) were excluded for the following reasons: being in a Medicare HMO (22.8%), unreliable diagnosis coding (4.0%), not having both Medicare parts A and B insurance coverage (5.5%), month and year of death the same as the diagnosis (2.6%), unknown month of diagnosis (0.7%) and/or unknown cause of death (0.5%).
Measurement of treatment and comorbidities
Medicare files were searched for diagnostic and procedural codes related to the comorbidities and treatments of interest [10]. Codes for diagnoses and procedures were derived from the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) [11], the Health Care Financing Administration Common Procedure Coding System and revenue center codes [12]. Medicare files for claims from inpatient, outpatient and physician visits were utilized.
In addition to SEER fields of surgery and radiation treatment within four months after breast cancer diagnosis, Medicare files were searched for chemotherapy and radiation codes. Patients with at least one chemotherapy code within six months after cancer diagnosis and patients with at least one radiation therapy code within nine months after cancer diagnosis were categorized as receiving those respective therapies. Data on hormonal therapy are not available in the SEER-Medicare database and therefore were not included in our analysis. Specific codes used to identify chemotherapy and radiation treatment are listed in Additional file 1.
Comorbidities included in this study are those 19 conditions that were found to affect survival by Charlson et al. [13, 14] and adapted by Deyo et al. [15] and Klabunde et al. [16] to be used with administrative claims data. Rather than using the combined Charlson Comorbidity Index, the current study assesses the individual effects of each of the comorbidities included in the Charlson Index. The Medicare files were searched for comorbidity diagnoses appearing during the time periods one year before and 30 days after the cancer diagnosis. This longer period before cancer diagnosis and the short period after cancer diagnosis allows for substantial time to identify diagnoses representing comorbidities without capturing conditions that may result from the cancer treatment. Diagnostic coding for comorbidities was made only if the comorbidity condition was included in Medicare inpatient files or if the condition had been coded at least twice at least 30 days apart in either the outpatient or physician files. The comorbid condition of previous cancer was determined from the SEER database.
For this analysis, we combined Charlson categories that were separated by severity (for example, diabetes and diabetes with complications) and further grouped conditions of myocardial infarction, congestive heart failure, peripheral vascular disease and cerebrovascular disease into one category of cardiovascular disease (CVD). Women without any of the 19 conditions comprised the group "No comorbidities." Only the four most common comorbid conditions within the Charlson index, previous cancer, CVD, chronic obstructive pulmonary disease (COPD) and diabetes, are included in the current study.
Underlying cause of death
The SEER registry uses US state death certificates to obtain information on the underlying cause of death encoded with the ICD-9 diagnostic codes. The underlying cause of death according to the ICD-9 diagnostic codes is defined as "the disease or injury that initiated the train of morbid events leading directly to death or the circumstances of the accident or violence that produced the injury" [17]. In the current study, deaths were attributed to either breast cancer or other causes. Deaths due to cancers other than breast cancer are included in the other-cause mortality group.
Statistical analysis
The study population was categorized into three study groups: (1) women who were alive at the end of the study period, (2) women who died as a result of breast cancer-specific reasons during the study period and (3) women who died as a result of other causes. For analyses involving breast cancer-specific mortality, deaths from other causes were censored at time of death and vice versa. Since the SEER registry does not include the specific diagnosis date or the specific death date, the 15th of the month was arbitrarily assigned to the reported month and year. The study population was followed through the end of 2005.
Descriptive statistics of the three study groups are presented. Cox proportional hazards models were used to calculate relative hazards of breast cancer mortality and other-cause mortality for population characteristics adjusted for age and fully adjusted for age and other characteristics. Causes of death were grouped into the three leading causes: breast cancer, other cancer and CVD. All other causes of death were grouped into the category "Other." Proportional distributions of cumulative causes of death were explored according to age group (66 to 74 years, 75 to 84 years and 85 years and older) and cancer stage (stage I, stage II and stages III and IV), as well as by duration of follow-up. Missing or unknown data were grouped into their own categories and were included in Cox proportional hazards models. Two approaches were used to assess whether missing predictors had an influence on the main findings: (1) records with missing or unknown values for any variables of interest were excluded from the analyses, and (2) the R package NestedCohort [18] was used to weight records with known values regarding age, race/ethnicity and death status to account for records with missing data. Accounting for missing or unknown data did not influence the results. SAS version 9.1 software (SAS Institute, Cary, NC, USA) was used for analysis.
Comments
View archived comments (1)