Mammography Use Helps To Explain Differences in Breast Cancer Stage at Diagnosis between Older Black and White Women

Breast cancer is the second leading cause of cancer-related death among women in the United States. Women 65 years of age and older bear the greatest burden of disease, accounting for more than 43% of newly diagnosed cases of breast cancer [1]. Advanced-stage disease is also more commonly diagnosed in older women [1-4], and the breast cancer mortality rate in older women in seven times greater than that in women younger than 65 years of age [5]. Older black women are at higher risk for diagnosis of late-stage breast cancer than older white women [2, 6]. Data from the Surveillance, Epidemiology, and End Results (SEER) program obtained from 1981 through 1986 [6] demonstrate that in black women 65 years of age and older, late-stage breast cancer was diagnosed more frequently (47% compared with 40% of white women in the same age group) and localized breast cancer was diagnosed less frequently (39% compared with 49%). One possible explanation for the higher rate of late-stage disease observed among older black women is this group's lower rate of mammography use. Several recent studies of Medicare enrollees have found that rates of mammography use were appreciably lower for black women than for white women [7-10]. We sought to determine the relation between previous mammography use and cancer stage at diagnosis for older black and white women and to investigate the extent to which previous mammography use explains the observed black-white difference in cancer stage at diagnosis among older women. Methods Data Source We conducted a retrospective cohort study using the Linked Medicare-Tumor Registry Database [11]. The linked database was jointly created by the National Cancer Institute and the Health Care Financing Administration (HCFA) to enable researchers to conduct cancer-related health services research. The linked database contains cancer information on patients 65 years of age and older from the National Cancer Institute's SEER program that is linked with Medicare enrollment and utilization information from HCFA's Medicare Statistical System for the years 1985 to 1989. Two Medicare utilization files are available in the linked database. Medical Provider Analysis and Review (MEDPAR) is a 100% utilization file that contains one record for every inpatient hospitalization or stay in a skilled-nursing facility covered under Medicare Part A. The Physicians' Claims file is a 100% utilization file that contains one record for every physician and outpatient claim covered under Medicare Part B. Before 1991, the Physicians' Claims file was available for only 10 states. Data from the SEER and Medicare programs overlap in tumor registries for three areas: Connecticut; metropolitan Atlanta, Georgia and Seattle-Puget Sound, Washington. Specific information on the linkage between SEER and Medicare has been published elsewhere [11]. The match rates for Connecticut, Atlanta, and Seattle are 93.3%, 94.1%, and 91.5%, respectively. Study Sample Women were eligible for the study sample (n = 4438) if a first case of primary breast cancer was diagnosed between 1 January 1987 and 31 December 1989 and if they were 67 years of age or older; were black or white; and resided in Connecticut, Atlanta, or Seattle-Puget Sound. Although we selected these areas because physicians' claims were available for all cases, they also represent a geographically diverse population of older women with breast cancer. Women who were enrolled in a health maintenance organization and those with less than 2 full years of Medicare Part B coverage were not eligible for this study because physician claims data (which are required for identifying mammography use) are not available for these women. We limited our final study sample to women who were 67 years of age or older to ensure that all women had at least 2 years of Medicare utilization (claims) information before breast cancer was diagnosed. Women whose mammography use could not be categorized (21 black women and 271 white women) or whose disease was unstaged (8 black women and 133 white women) were excluded from the study. Measures We ascertained the following sociodemographic variables from the SEER file: age at diagnosis, marital status at diagnosis, and SEER area. Age at diagnosis (range, 67 to 100 years) was categorized as 67 to 74 years, 75 to 84 years, or 85 years and older for descriptive purposes but was modeled as a continuous variable. Marital status was defined as married or not married at diagnosis. The SEER area was classified according to the tumor registry of diagnosis: Connecticut, Atlanta, or Seattle. We used 1990 U.S. Census data as an ecologic measure of socioeconomic status. Women were assigned to the median household income of their ZIP code of residence; incomes were grouped as less than $15 000 or at least $15 000. We obtained information on race from the Medicare beneficiary enrollment file. A comparison of race between Medicare and SEER files demonstrated agreement for 99% of women. Enrollees are classified in Medicare files as black, white, Asian, Native American, Hispanic, or unspecified. Because this analysis focused on black and white women, women of other racial groups were not eligible (n = 125). We computed a modified Charlson Comorbidity Index using Deyo's method of classifying ICD-9-CM (International Classification of Diseases, 9th revision, Clinical Modification) diagnosis codes from inpatient claims [12]. For each woman, we identified all inpatient hospitalizations that began 2 years before diagnosis and ended 1 month after diagnosis. A priori, we extended the observation period to 1 month past the diagnosis because we expected that during the study years 1987 to 1989, most women would have had at least one hospitalization around the time of breast cancer diagnosis. We classified women as 1) nonhospitalized [that is, comorbidity could not be assessed], 2) having no comorbid conditions [Charlson Index, 0], or 3) having one or more comorbid conditions (Charlson Index 1). We measured mammography use by using Medicare physicians' claims. We identified women who had one or more bilateral mammographies (CPT [Physicians' Current Procedural Terminology] procedure code 76091) within 2 years before breast cancer diagnosis. We classified women as 1) nonusers (n = 833) if they did not have mammography during the 2 years before diagnosis, 2) regular users (n = 733) if they had at least two mammographies within the 2 years before diagnosis that were 10 or more months apart, and 3) peri-diagnosis users (n = 2267) if they had their only mammography or mammographies within 3 months before diagnosis. Women who did not fit into any of these categories were classified as uncertain (n = 292) and were excluded from the study. The peri-diagnosis users were a heterogeneous group that included women who underwent screening mammography and were given a diagnosis of breast cancer and women who underwent diagnostic mammography. Therefore, analyses relating previous mammography use to cancer stage at diagnosis considered only nonusers and regular users because these were two distinct groups. We measured cancer stage at diagnosis using the SEER historical staging system (in situ, localized, regional, distant, or unstaged) because it was available for all women. Stage of disease was dichotomized as early (in situ or localized) or late (regional or distant). Statistical Analysis All statistical analyses were done by using SAS statistical software, version 6.11 [13]. Black and white women were compared with respect to sociodemographic factors, comorbidity, stage at diagnosis, and previous mammography use. Chi-square statistics and Student t-tests were performed to identify characteristics that differed significantly between black and white women. Multivariable logistic regression was used to estimate the adjusted odds of late-stage disease for black women compared with white women [14]. To investigate the extent to which previous mammography use explains the observed black-white difference in stage at diagnosis, we compared simple models with more complex ones and examined changes in the estimated odds ratio for the race-stage association [15]. First, we compared a model that included only race with a model that included race and previous mammography use to determine how much of the excess late-stage disease among black women is explained by differences in previous mammography use. Next, we compared a model that included race, sociodemographic characteristics, and comorbidity information with a model that included these factors and previous mammography use to determine the additional amount of excess late-stage disease among black women that is explained by previous mammography use after adjustment for sociodemographic and comorbidity information. The odds ratio for race and the corresponding 95% CIs were estimated from the coefficient and SE from the logistic models [14]. We used the following formula to compute the percentage change in the estimated odds ratio to compare our results with those from a previous study [16]: Percentage change in OR = (Equation 1) Results Characteristics of the study sample (n = 4005) are given in Table 1. Overall, 172 (4%) of the women were black. Twenty-two percent of women had not had mammography within the 2 years before breast cancer diagnosis (nonusers), 19% of women had at least two mammographies within the 2 years preceding diagnosis that were at least done 10 months apart (regular users), and 59% had their only mammography within the 3 months before diagnosis (peri-diagnosis users). Late-stage disease was diagnosed in one third (32%) of the women. Table 1. Characteristics of the Study Sample by Race* Table 1 also shows the characteristics of the study sample by race. Race was confounded with the SEER area of residence. For example, although only 17% of women in our study resided in Atlanta, almost two thirds (66%) of the black women were from Atlanta. Black women