Screening mammography use and chemotherapy among female stage II colon cancer patients: a retrospective cohort study

The 2006 SEER-Medicare linked data were used in this study. The details of SEER-Medicare data can be found at the National Cancer Institute SEER website [13]. Briefly, as of 2002, SEER included all cancer patients residing in 17 geographic areas, covering about 26% of total US populations. For cancer patients 65 years of age and older, 97% of them were linked to Medicare claims data [14]. We identified all incident female colon cancer patients from 1996 to 2002 using the first primary cancer site code (15-23, colon cancer) (n = 49,086) in the SEER Patient Entitlement and Diagnosis Summary File (PEDSF). We excluded patients who had a second malignancy, including colon cancer, within the first year, and included only those with stage II cancer (n = 13,609). Since identifying screening mammography use prior to the cancer diagnosis requires two years of Medicare claims, and mammography rate is significantly lower among people 80 years and older [15], we restricted patients to those with 67 to 79 years of age at the time of cancer diagnosis (n = 5,912). In addition, previous studies have also shown colon cancer patients 80 years and older also have significantly lower chemotherapy rates than younger patients [16]. Similar to other studies using SEER-Medicare data, we hierarchically excluded the cancer patients who did not have both Medicare Part A (hospital insurance) and Part B (medical insurance for office visit) enrollment (Both Part A and B were administered by the US Center for Medicare and Medicaid Service, i.e., CMS) (n = 428), were enrolled in managed care (commercial insurance and health care groups contracted with CMS to manage Medicare beneficiaries) (n = 1,810), or had end-stage renal disease (n = 46) during two years before and one year after the cancer diagnosis. We also excluded those who died (n = 247) or were admitted to hospice (n = 17) within six months around the cancer diagnosis because the treatment decisions for these patients were likely different from others. The date of death was based on Medicare entitlement information included in PEDSF. In-hospital surgical treatment for colon cancer was searched using the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) procedure codes 45.7x, 45.8, 48.4x, 48.5, and 48.6x, in the Medicare Provider Analysis and Review (MedPAR) claims within six months around the cancer diagnosis. Those who did not have a colon cancer related surgery or a surgeon's claim in the Medicare National Claim History (NCH) Carrier during that time window were also excluded (n = 199). Teaching hospital was defined as any hospital with a non-zero indirect cost for medical education in the MedPAR hospitalization claims for the colon surgery.

The study has been approved by the Institutional Review Board (IRB) at the University of Minnesota where the study was initially conducted. The data used in this study were obtained from National Cancer Institute through IMS under the SEER-Medicare data user agreement.

The use of adjuvant chemotherapy treatment was identified from Medicare NCH Carrier claims (The NCH file contain claims for services provided by physicians and stand-alone clinics), Durable Medical Device and Outpatient file claims were searched using Healthcare Common Procedure Coding System (HCPCS) level II codes for commonly used chemotherapy drugs [14]. 99.3% patients who had chemotherapy received 5-fluorouracil (5-FU, J9190) or capecitabine (J8520, J8521), both of which are used either alone, or with leucovorin (J0640), floxuridine (J9200), or oxaliplatin (J9263), irinotecan (J9206). The waiting period for the chemotherapy initiation was defined as the time window from the date of cancer surgery to the date of the first claim containing chemotherapy drugs [17]. Adjuvant chemotherapy is usually started within three months of surgery [16, 18]. In our study, 4.2% of patients started chemotherapy after six months. They were excluded (n = 134) because the purpose for the chemotherapy was uncertain [17].

The duration of chemotherapy was defined as the time window from the first to the last claim that contained chemotherapy drug information. Since 1995, typical chemotherapy usually lasts for about six months [3]. We excluded those who were on chemotherapy for more than one year (4.0%, n = 121) because these patients might have a recurrence or the intention of treatment might be a long treatment regimen [19].

Screening mammography within the two years before the cancer diagnosis was identified using HCPCS codes 76092, G0202, and G0203, and ICD-9-CM code V76.12, in the Medicare NCH Carrier and Outpatient files [15, 20].

Certain tumor characteristics have been related to poor prognosis and may affect chemotherapy decision among stage II colon cancer [5]. We distinguished tumors based on the detailed staging information using SEER extent of diseases codes (defined in SEER EOD-88 3rd edition (1998): 40: invasion through muscularis; 45: extension to adjacent tissue; 50: invasion of/through serosa), and identified bowel obstruction (ICD-9 diagnosis code: 560.89 and 560.90) and bowel perforation (ICD-9 diagnosis codes: 569.83) in the MedPAR claims within four months of the cancer surgery. Those with tumors that have invaded of/through serosa or having bowel obstruction or perforation were Stage IIb colon cancer. Tumor grade from SEER PEDSF was regrouped as well/moderately differentiated and poorly differentiated/other. The "other" group includes unknown grade or ungraded (3.3%, n = 95). They were combined with the poorly differentiated group based on the principle of conservative analysis. In addition, since the number of lymph node examined during the surgery has been related to the quality of surgery and cancer survival, it was grouped as less than 12, and 12 or more [21].

Chemotherapy is mainly administered by oncologists. In this study, we used all chemotherapy claims from 1996 to 2002 for breast cancer, colorectal cancer, and uterine cancer to identify all physicians who were likely to administer chemotherapy to Medicare cancer patients. Including all these cancer types allows us to have a larger and more complete Medicare claim history to search for possible oncologists and capture those who only saw a small number of elderly patients. An oncologist was defined as a physician who had 5 or more chemotherapy administration or drug claims per year and had seen 2 or more of cancer patients during the study period. Physicians with the HCFA specialty codes (90: medical oncology, 82/83: hematology/oncology, 91: surgical oncology and 70: multiple specialty, 11: internal medicine) accounted for more than 95% of total oncologists identified.

Patient socio-demographic variables included age group (67-69, 70-74, and 75-79), race/ethnicity (black, white, and other), state buy-in status (state buy-in means the state pays part or all of the patient's Medicare Part B premium or the person is in the Medicaid program), median household income at the census tract based on the 2000 census (If the value was missing, it was imputed from the corresponding zip code median household income), metropolitan residence status (yes versus no), and marital status at the time of cancer diagnosis (married/living with a partner versus living alone). They were obtained from SEER PEDSF.

Patient comorbidities were assessed using Charlson score [22] which was calculated from ICD-9 diagnosis codes in the Medicare NCH, MedPAR and Outpatient claims, in the two years before the month of cancer diagnosis using Deyo and Klabunde modified algorithm [23, 24]. The Charlson score was grouped as zero, 1, 2, and 3+. Cancer diagnosis was excluded in calculating the above score.

Descriptive statistics were calculated for patient socio-demographic variables, comorbidities, tumor characteristics, and oncologist visit, stratified by mammography use before the cancer diagnosis. Statistical significance was assessed using t-test for continuous variables and chi-square test for categorical variables. Multivariate adjusted chemotherapy rates were computed as 1/(1+exp(-bx)) from the logistic regression with chemotherapy use (yes/no) as the dependent variable. The time from the cancer surgery to the first chemotherapy use was presented using Kaplan Meier survival curves stratified by prior mammography use. Curves were inverted and also truncated at 160 days to be more interpretable. Odds ratios for the determinants of chemotherapy use were obtained from the above multivariate logistic regression in which all determinants were mutually adjusted. To take account of possible clustering effect of patients, we adopted the Generalized Estimate Equations (GEE) method in the multivariate logistic regression with the hospital where patient received the surgery as the cluster variable. All the statistical analyses were performed using SAS 9.1.4 (Proc GLM, GENMOD, and LIFETEST) (SAS Institute, Cary, NC).