SEARCH

SEARCH BY CITATION

Keywords:

  • breast neoplasms;
  • breast neoplasm prevention and control;
  • breast neoplasm stage at diagnosis;
  • health services accessibility;
  • early diagnosis

Abstract

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. REFERENCES

BACKGROUND

The current study investigated the individual and community determinants of breast carcinoma stage at diagnosis (BCSAD) using multiple data sources merged with cancer registry data. The literature review yielded 5 studies that analyzed cancer registry data merged with community-level variables (1995–2004).

METHODS

Community variables constructed for the current study reflected social and economic risk factors, physician supply, and health maintenance organization penetration. Multivariate logistic regression was used to identify the significant predictors of increasingly progressive BCSAD.

RESULTS

Disparities remained for black and Hispanic females in California, who were least likely to be diagnosed early compared with their white counterparts. Younger (< 40 years) and middle-aged (40–64 years) females were less likely to be diagnosed with early BCSAD, compared with older females (≥ 65 years). Utilizing services at hospitals serving a lower volume of patients with breast carcinoma was associated with later BCSAD. After controlling for individual-level factors, community-level variables constructed at the census block group and county level were tested. If a woman resided in a neighborhood with greater percentages of female-headed households, persons living below the poverty level, less educated people, and more recent immigrants, then her chances of being diagnosed at an earlier stage were diminished. If, conversely, she resided in a neighborhood with greater percentages of females ≥ 65 years (a proxy for Medicare coverage), her access to medical care and the probability of earlier BCSAD increased. County-level insurance rates and residing in counties where greater percentages of women ever had a mammogram were associated with in situ and early-stage diagnosis. Similarly, the supply of primary care physicians and radiologists was associated positively with earlier BCSAD.

CONCLUSIONS

Results confirmed community-level predictors of socioeconomic and delivery system context matter, although the individual-level predictors showed a stronger effect. Nevertheless, analysis of community variables is promising for guiding and evaluating the effects of health policy and developing community and delivery system interventions for earlier detection and treatment of breast carcinoma. Cancer 2005. © 2005 American Cancer Society.

Breast carcinoma stage at diagnosis (BCSAD) is a powerful prognostic determinant of cancer outcome, which is linked directly to treatment effectiveness and survival.1–5 BCSAD provides a valuable means for recording patterns of disease, monitoring advances in diagnosis and therapy,6 guiding targeted interventions in high-risk subgroups, and assisting administrators' estimate resource needs.7 Although specific risk factors have been identified,8 breast carcinoma is not readily amenable to primary prevention. therefore, screening for the disease is crucial for early detection and cure.9

The predictors of BCSAD are similar to those found for medical care access. Lower educational attainment, racial/ethnic minority group, cultural values and beliefs, and lower household income are significant barriers to timely health care delivery. In addition, the unemployed, the uninsured and underinsured, and those having no usual source of care are significantly less likely than more affluent individuals to access medical care, delay seeking treatment, and may have worse health outcomes.2, 10–22

However, most researchers have failed to account for the effects of community determinants, which may explain to a greater extent than individual-level determinants differences in mammography screening rates and BCSAD. To assess the state of the research in this area, a literature review was conducted, which yielded five studies that investigated the community-level determinants of BCSAD. Four studies dichotomized BCSAD into early (Stages 0–I) and late-stage (Stages II–IV) carcinoma. All four included in situ disease (precancer stage) in the early-stage category.13, 23–25 The remaining study trichotomized BCSAD, applied multinomial regression analysis, and removed in situ cases.26 These studies analyzed cancer registry data from California, Florida, and Connecticut, merged with community variables constructed from other data sources (e.g., census data and/or area resource file [ARF] data).

Results from these studies show the major BCSAD predictors tested were percentage of community residents with low income and education,13, 23, 25 percentage of females in the labor force,13 neighborhood occupational class,26 urban/rural residence,13 and physician supply.13, 23, 24 The literature revealed our understanding of the effects of community predictors is limited, but the research is promising for guiding health policy and developing community and delivery system interventions for early detection and treatment of breast carcinoma, potentially leading to better cancer outcomes. The current study proposes to 1) develop a more comprehensive approach for investigating determinants of BCSAD using multiple data sources linked to cancer registry data, and to 2) empirically test a wider range of community predictors using California data.

MATERIALS AND METHODS

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. REFERENCES

Individual Variables

Breast carcinoma cases were obtained from the California Cancer Registry (CCR; 1994–1999), California's statewide population-based cancer surveillance system. For the outcome variable, BCSAD category levels reflect the American Joint Committee on Cancer staging system.27 More clinically oriented than the Surveillance, Epidemiology, and End Results summary staging system, the TNM method takes into account three dimensions in the etiology of all cancers: 1) local tumor growth; 2) spread to regional lymph nodes; and 3) metastasis. Other CCR patient-level data analyzed include age at diagnosis, race/ethnicity, marital status, and reporting hospitals categorized by volume of patients with breast carcinoma treated annually.

Community Variables

Community predictors were constructed from four data sources: the Census Bureau, ARF, the California Health Interview Survey (CHIS), and the California Statewide Health Maintenance Organization (HMO) and Special Programs Enrollment Study. For census data, the geographic unit of observation is the census block group (CBG), which contains 600–3000 people, with an optimum size of 1500 people.28 Five social risk factors were constructed: 1) percentage of the population living in a nonurban area, 2) percentage who recently immigrated to the United States, 3) percentage of older females (age ≥ 65 years), 4) percentage of households headed by females, and 5) percentage of racial/ethnic group. Five economic indicators were constructed: 1) percentage of the population that was unemployed, 2) percentage of females in the labor force, 3) percentage with less than a high school education, 4) median income, and 5) percentage reporting an annual income < 200% of the federal poverty level (FPL). Each observation in the CCR data contained a 1990 CBG identifier, which was used for merging the data. More current 2000 census tract data would produce erroneous results, because many census tracts have experienced boundary and name changes since the 1990 census.

Two CHIS variables were constructed at the county level: 1) ever had a mammogram and 2) insurance status in the past 12 months.29 Three ARF variables measuring physician supply were constructed at the county level: 1) total active physicians (Federal and non-Federal) per 10,000 population, 2) total radiologists per 100,000 population, and 3) total primary care physicians, which includes office-based physicians for general practice, general family practice, and general internal medicine.30 ARF values averaged for the 2 available years (1995 and 2000) were used to construct reliable county-level indicators. HMO penetration by county in 1997 was obtained from Cattaneo and Stroud's Statewide HMO and Special Programs Enrollment Study.31 Each observation in the CCR contains an Federal Information Processing Standards (FIPS) county code, which was used to merge these county-level variables.

Measurement Model

Figure 1 presents the community and individual-level determinants of the BCSAD measurement model. The model posits that community factors reflecting the social, economic, and delivery system environment of the patient with cancer will influence a woman's access to medical care and, subsequently, BCSAD. More proximal determinants are the individual characteristics of the patients with cancer recorded in the CCR. These community and individual predictors of the outcome variable will be evaluated at increasingly progressively stages of breast carcinoma.

thumbnail image

Figure 1. Community and individual determinants of breast carcinoma stage at diagnosis. HMO: Health Maintenance Organization.

Download figure to PowerPoint

Statistical Methods

Cases diagnosed as in situ (Stage 0), local (Stage I), regional (Stages II and III), and distant (Stage IV) were used in the multivariate analyses, whereas cases classified as unstaged were removed. Removed from this dataset were cases that did not specify the CBG, the racial/ethnic group, or the name of a reporting hospital. After the CCR file was merged, cases with nonmissing information on all individual and community variables were retained, resulting in a sample size of 112,471 women diagnosed with breast carcinoma, 1994–1999.

For the multivariate analysis, the predictors of BCSAD were tested using four models reflecting increasingly progressive disease. Four of the five studies reviewed in the current research included in situ and local as early stage at diagnosis. This BCSAD dependent variable is controversial because it includes the precancerous in situ cases. The fifth study used multinomial logistic regression to evaluate the three clinically significant stages of BCSAD: local, regional, and distant. Because multinomial logistic regression is more difficult to interpret, we evaluated the predictors using multiple logistic regression analysis applied to the four models reflecting increasingly progressive disease: Model I, in situ (Stage 0) versus local/regional/distant (Stages I–IV); Model 2, in situ/local (Stages 0–I) versus regional/distant (Stages II–IV); Model 3, local (Stage I) versus regional/distant (Stages II–IV); and Model 4, distant (Stage IV) versus local/regional (Stages I–III).

This statistical treatment of the dependent variable yields the same results as a multinomial approach, without the challenges of interpreting differences among three stages of BCSAD in the same equation. By categorizing the stages of the dependent variable, and using logistic regression to evaluate differences among stages, we elucidate the effect of predictors of increasingly progressive disease using a clear-cut and interpretable analytic technique.

Statistical software (Version 8; Statistical Analysis System, Cary, NC) was used to merge and edit the data sources and to compute descriptive statistics for the individual and community predictors. Hierarchical logistic regression was performed to determine which individual and community variables had a statistically significant effect on BCSAD. Stata 7.0 was used for the model selection procedures, because it allows dummy indicators representing the same variable to be treated as one unit (e.g., they are either entered or left out together in the stepwise selection process). The variables were entered using hierarchical logistic regression, whereby predictors were entered in three stages representing progressively larger units of observation: individual-level variables, CBG, and county-level variables.

Relative risks (RR) and odds ratio (OR) values are reported for the categorical individual predictors. ORs only are reported for the continuous community variables. The RR value is the incidence of an event among subjects exposed to a risk factor, divided by the incidence rate of an event among subjects not exposed to the risk. ORs represent the odds of exposure to a potential risk factor among subjects with a specific event, divided by the odds of exposure among those without the event. The Hosmer–Lemeshow goodness-of-fit test was used to compute a chi-square value from observed and expected frequencies. A nonsignificant P value indicates the logistic model is a good fit and the model explains the variance in the dependent variable to a significant degree. Also, the lower the computed chi-square statistic, the better the model fit.

RESULTS

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. REFERENCES

Table 1 shows the distribution of CCR cases by BCSAD for California females, 1994–1999. The highest percentage of cases was diagnosed at the local (41.2%) and regional (38.9%) stages, with smaller percentages reported for in situ or precancer (16.2%) and distant (3.7%) stages.

Table 1. Distribution of Patients by Breast Carcinoma Stage at Diagnosis, California Cancer Registry, 1994–1999 (n = 112,471)a
Stage at diagnosisNo. of patients (%)
  • a

    Patients with complete information (nonmissing values) for individual and community-level variables were used in the current analysis. Patients were removed for the following reasons: 1) if a census tract/block group identifier was missing, 2) if race was reported as other/unknown, and/or 3) if a reporting hospital was missing.

In situ18,238 (16.2)
Localized (Stage I)46,373 (41.2)
Regional (Stage II/III)43,721 (38.9)
Distant (Stage IV)4,139 (3.7)
Total112,471 (100.0)

Table 2 reports the number and percentage of breast carcinoma cases for relevant individual-level variables extracted from the CCR, 1994–1999. These include age at diagnosis, race/ethnicity, marital status, and reporting hospitals categorized by average volume of patients with breast carcinoma treated annually. Table 2 provides the individual-level predictors, the percentage and number of cases, and stratification by cancer stage.

Table 2. Percent and Number of Patients with Breast Carcinoma for Individual-Level Variables, California Cancer Registry, 1994–1999 (n = 112,471)
VariablesNo. of patients (%)In situ (%)Early (%)Regional (%)Distant (%)
Age (yrs)     
 < 406379 (5.7)12.425.957.14.7
 40–4920,612 (18.3)18.93245.83.4
 50–6436,626 (32.6)17.840.5383.7
 65–7426,594 (23.7)15.948.132.53.5
 ≥ 7522,260 (19.8)12.647.236.33.9
Race-ethnicity     
 White85,235 (75.8)16.243.636.83.4
 Black6379 (5.7)16.130.846.96.3
 Hispanic12,470 (11.1)14.633.347.74.4
 Asian-Pacific Islander8387 (7.5)18.937.2413
Marital status     
 Single13,130 (11.7)17.335.341.85.6
 Married64,693 (57.5)17.241.538.42.9
 Separated, divorced or widowed34,648 (30.8)13.94338.74.4
Hospital level     
 < 10 patients1711 (1.5)13.53546.55.1
 10–39 patients16,278 (14.5)13.739.542.34.6
 ≥ 40 patients94,482 (84)16.741.738.23.5

The mean and range of values for the candidate community-level variables are provided in Table 3, as are the community variables constructed at CBG level and at the county level. In California, approximately 7% of CBGs were nonurban territories (range, 0–100%). Within the CBGs, approximately one-third (32.3%) were inhabited by nonwhite residents (range, 0–100%). Percentages of other racial/ethnic groups were tested in preliminary multivariate analysis and found to be nonsignificant in the California results (data not shown). On average, approximately 8% of females residing in CBGs were ≥ 65 years (range, 0–69.5%). Less than 5% (4.3%) of residents in CBGs were recent immigrants (range, 0–59%). Almost one-fourth (22.8%) of those residing in CBGs reported income below the FPL (range, 0–100%). The median household income was approximately $44,000 (range, $4999–$150,001). On average, almost one-third of females (29.4%) in CBGs were employed (range, 0–72%). The general unemployment rate in the California CBGs was 3.6% (range, 0–48%). Approximately 19% of individuals reported less than a high school education (range, 0–100%). On average, approximately 5% of CBGs were female-headed households, range, 0–62.5% in some communities.

Table 3. Mean and Range of Values for Community-Level Variables, California, 1994–1999
VariablesMean (range)
  • HMO: health maintenance organization.

  • a

    Territory, population, and housing units not classified as urban.

  • b

    Five years or less in the United States.

  • c

    Incomes > $150,000 are categorized as $150,001.

  • d

    California Health Interview Survey data were collected in 2001.

  • e

    Females, ≥ 30 years.

  • f

    Both nonfederal and federal physicians.

  • g

    Only nonfederal primary care physicians.

  • h

    Only nonfederal radiologists.

Variables by census block group 
Community risk factors 
 Percent nonurban residencea7.0 (0.0–100)
 Percent nonwhite32.3 (0.0–100)
 Percent female, age ≥ 65 yrs8.2 (0.0–69.5)
 Percent recent immigrantb4.3 (0.0–58.7)
 Percent below federal poverty level (200%)22.8 (0.0–100)
 Median household income ($)c43,931 (4,999–150,001)
 Percent of females in labor force29.4 (0.0–71.7)
 Percent unemployed3.6 (0.0–48.5)
 Percent with less than high school education19.1 (0.0–100.0)
 Percent female-head households w/children5.3 (0.0–62.5)
Variables by county 
Community risk factors 
 Percent insured for past 12 mosd81.8 (74.6–91.5)
 Percent ever had a mammogramde72.6 (66.0–83.0)
Structure and market dynamics 
 Total active physicians per 10,000 populationf25.9 (3.0–66.9)
 Total active primary care physicians per 10,000 populationg6.3 (1.3–15.7)
 Total active radiologists per 100,000 populationh3.0 (0.0–9.5)
 HMO penetration (%)47.3 (2.0–73.7)

Table 3 shows that, on average, almost 82% of individuals residing in California counties reported having private or public health insurance coverage in the past 12 months (range, 74.6–91.5%). Among females ≥ 30 years old, 72.6% reported ever having a mammogram (range, 66.0–83.0%). Regarding physician supply, on average, approximately 25 active physicians practiced in California counties per 10,000 population (range, 3.0–67). Approximately 6 primary care physicians per 10,000 practiced in California counties (range, 1.3–15.7). Radiologists were fewer. On average, approximately 3 per 100,000 practiced in California (range, 0–9.5). In 1997, total HMO penetration (both commercial and public) in California counties as a percentage of the eligible population averaged 47.3% (range, 2–75%).

Table 4 shows the multivariate predictors of BCSAD. Individual and community variables were tested using four regression models to identify significant predictors of progressively later BCSAD. RR and OR values are reported for the categorical individual predictors and ORs are reported for the continuous community variables. Models 1 and 2 contain breast carcinoma cases diagnosed at the in situ stage. Although some experts do not consider this precancerous diagnosis clinically relevant, the in situ diagnosis still creates anxiety among women, who are subjected to further testing and treatment.32, 33

Table 4. Multivariate Predictors of Breast Carcinoma Stage at Diagnosis
Independent variablesDependent variables
Model 1: in situ vs. local/regional/distant (n = 112,471)Model 2: in situ/local vs. regional/distant (n = 112,471)Model 3: local vs. regional/distant (n = 94,233)aModel 4: distant vs. loca/regional (n = 94,233)a
RRORRRORRRORRROR
  • RR: relative risk; OR: odds ratio; PI: Pacific Islander: FPL: federal poverty level; HMO: Health Maintenance Organization.

  • a

    Patients with in situ disease were removed.

  • b

    Model fit statistic (Hosmer–Lemeshow) is reported from the hierarchical models to assess change in model fit as each successive block of variables is added. The RR and OR values are reported after all variables have been entered into the final hierarchical model.

  • c

    P < 0.05.

  • d

    P < 0.01.

Individual characteristics        
Age (yrs)        
 < 400.9160.902c0.6370.421d0.5470.374d1.1781.187c
 40–491.3811.485d0.8360.677d0.7190.555d0.9860.985
 50–741.2111.263d1.0670.907d1.0591.112d0.9800.979
 65–741.0001.0001.0001.0001.0001.0001.0001.000
 ≥ 75        
Race/ethnicity        
 Asian/PI1.1101.134d0.9710.934d0.9370.883d0.9200.916
 Black0.9970.9970.8240.663d0.7590.682d1.6231.669d
 Hispanic0.8920.874d0.8460.695d0.8130.611d1.2091.220d
 White1.0001.0001.0001.0001.0001.0001.0001.000
Marital status        
 Never married1.0231.0280.9430.873d0.9110.837d1.7901.853d
 Widowed/divorced/separated0.8670.844d0.9230.836d0.9220.859d1.4251.452d
 Married1.0001.0001.0001.0001.0001.0001.0001.000
Hospital level        
 < 10 patients0.8380.812d0.8560.715d0.8380.724d1.3081.327c
 10–39 patients0.8430.817d0.9220.831d0.9260.862d1.1991.210d
 ≥ 40 patients1.0001.0001.0001.0001.0001.0001.0001.000
Model fit: Hosmer–Lemeshowb12.53; P > 0.08418.41; P > 0.0187.98; P > 0.4363.10; P > 0.928
Census block group OR OR OR OR
Community risk factors        
 Percent female-headedhouseholds 0.994d   
 Percent female ≥ 65 yrs  1.005d 1.004d 0.990d
 Percent with < 200% FPL-poverty  0.996d 1.006d
 Median income 1.007d   
 Percent with education less than high school 0.996d 0.996d 0.994d 1.004c
 Immigration < 5 yrs   0.995d 
Model fit: Hosmer–Lemeshowb8.71; P > 0.3674.79; P > 0.7794.09; P > 0.8494.46; P > 0.814
Country level        
Community risk factors        
 Percent insured  1.011d 1.003 
 Percent with mammography use 1.022d 1.009d 1.007c 
Delivery system structure        
 Total physicians    
 Primary care physicians   1.014d 
 Radiologists 1.026d 1.010c  
 HMO penetration  0.998d  
Model fit: Hosmer–Lemeshowb8.30; P > 0.4057.10; P > 0.5254.99; P > 0.758  

Model 1 predicts the probability of an in situ BCSAD versus local/regional/distant stages (Table 4). Middle-aged females, 40–49 years (RR = 1.4) and 50–64 years (RR = 1.3), were most likely to be diagnosed at the in situ (Stage 0) compared to older females. Among racial/ethnic groups, the likelihood of being diagnosed at the in situ stage was not so different, with Asian/Pacific Islanders (PI) slightly more likely and Hispanics slightly less likely to receive a precancerous diagnosis. Less likely to be diagnosed at the in situ stage were widowed, divorced, or separated females (RR = 0.87) and patients from reporting hospitals delivering care to < 40 patients with breast carcinoma annually. Several community variables were significant predictors of an in situ diagnosis, including residing in neighborhoods with greater percentages of female-headed households (OR = 0.994, P ≤ 0.001) and greater percentages of uneducated persons (OR = 0.996, P ≤ 0.001). In contrast, women residing in neighborhoods with a higher median income (OR = 1.007, P ≤ 0.001), in counties with greater percentages of women reporting ever having a mammography (OR = 1.022, P ≤ 0.001), and in counties with more radiologists (OR = 1.026, P ≤ 0.001) had an increased probability of an in situ diagnosis. These results suggest that increasing the number of women screened will yield a higher probability for an in situ diagnosis. Also, increasing the number of specialty radiologists available at a hospital influences the intensity of screening by primary care physicians and, therefore, the number of in situ cases.

Model 2 predicts the probability of an early BCSAD (Table 4), which includes both in situ and local versus regional and distant stages. The results of this model show younger females were least likely to receive an early BCSAD, especially women < 40 years (RR = 0.64). Black (RR = 0.82) and Hispanic (RR = 0.85) females, as well as the never-married, divorced, widowed, and separated women, were less likely to be diagnosed early. Females utilizing services at hospitals delivering care to < 40 patients with breast carcinoma annually were diagnosed at a later stage. Residing in neighborhoods with more poverty or lower educational attainment was associated with lower odds of an early diagnosis. However, for women residing in a neighborhood with a greater percentages of older women &i.e., ≥ 65 years), the odds of an early BCSAD were increased. The number of women ≥ 65 years could be viewed as a proxy for Medicare coverage and greater access to mammography services. Residing in counties with more insured persons, more females reporting ever having a mammography, and more radiologists also were associated with better odds of an early BCSAD. In contrast, residing in a county with higher HMO penetration was associated with later BCSAD.

Models 3 and 4 remove in situ cases from the analysis, examining only clinically relevant cases (Table 4). Model 3 predicts the probability of early BCSAD versus regional/distant stages. Younger females (i.e., < 40 years) were approximately one-half as likely (RR = 0.55) to be diagnosed with early BCSAD, whereas middle-aged females (i.e., 40–49 years [RR = 0.72] and 50–64 years [RR = 0.89]) were less likely to be diagnosed with early BCSAD compared with older females (≥ 65 years). Younger and middle-aged women were not eligible for Medicare coverage. In descending order, Asian/PI (RR = 0.94), Hispanics (RR = 0.81), and black (RR = 0.76) females were less likely to present with early BCSAD compared with white females. Similar to the results for Models 1 and 2, females utilizing hospitals delivering care to < 40 patients with breast carcinoma annually were less likely to present with early BCSAD. Community variables generally had a smaller effect. Women residing in neighborhoods with greater percentages of older females had an increased probability of early BCSAD. Those residing in neighborhoods with greater percentages of lower educated persons or with greater numbers of recent immigrants had a decreased probability of early BCSAD. Those residing in counties with more insured persons or more females who reported a mammography had an increased probability of an early BCSAD. The number of primary care physicians practicing in a county (OR = 1.014, P < 0.01) was one of the strongest community predictors of earlier BCSAD. As the supply of primary care physicians increases, the odds of an early BCSAD increases.

Model 4 predicts the probability of distant BCSAD. Females < 40 years (RR = 1.178), Black (RR = 1.6), and to a lesser extent Hispanic (RR = 1.2) females had an increased likelihood of distant BCSAD. The never-married (RR = 1.8) and, to a lesser extent, divorced, widowed, and separated women (RR = 1.4) were more likely to present with a late BCSAD. Females visiting hospitals delivering care to < 40 patients with breast carcinoma annually were more likely to present with a late BCSAD. Among community predictors, the odds of a late BCSAD were slightly increased if a female resided in a neighborhood with high poverty or low educational attainment. The strongest community predictor shows that women residing in a neighborhood with a greater percentage of female residents ≥ 65 years (OR = 0.990) had a decreased probability of presenting with a late BCSAD. This proxy measure for Medicare coverage is a significant predictor in Models 2, 3, and 4. Table 4 shows the Hosmer–Lemeshow model fit statistic for each level of predictor, i.e., individual level, CBG, and county level. In general, the overall model fit improves after each level of variables is entered, as indicated by the lower computed chi-square statistic and the higher P values.

DISCUSSION

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. REFERENCES

The purpose of the current study was to develop a more comprehensive approach for investigating individual and community determinants of BCSAD using multiple data sources merged with cancer registry data. Most researchers have failed to account for the effects of community determinants, which were found to explain to a greater extent the differences in BCSAD among patients with breast carcinoma in California.

Regarding individual-level predictors of BCSAD, younger women (< 40 years) were least likely to be diagnosed at an early stage and were most likely to be diagnosed at the distant or metastic stage (Stage IV). Middle-aged women (40–64 years) were less likely to be diagnosed with an early BCSAD compared with older women (≥ 65 years), who have access to Medicare coverage. In 1997, the American Cancer Society updated its guidelines for breast carcinoma screening, appropriately recommending that women ≥ 40 years should begin annual screenings.34 However, these results indicate that women < 40 years were most likely to be diagnosed at a late BCSAD. Disparities remain for black and Hispanic females in California, who were less likely to be diagnosed early and more likely to be diagnosed at a later BCSAD. Most likely a reflection of insurance status, married women fared better than never-married, divorced, separated, and widowed women. Where a women received care matters. Those utilizing services at reporting hospitals delivering care to < 40 patients with breast carcinoma a year were at a higher risk of a later BCSAD, which has major implications for hospital costs, treatment effectiveness, and survival.35, 36

Community variables were constructed at the CBG and county level. If a woman resided in a neighborhood with more female-headed households, persons living below the FPL, less educated people, and more recent immigrants, then her chances of being diagnosed at an earlier stage were diminished. If, conversely, she resided in a neighborhood with more females ≥ 65 years (a proxy for Medicare coverage), her access to medical care and the probability of an earlier BCSAD increased. County-level insurance rates and residing in counties where greater percentages of women ever had a mammogram were associated with in situ and early-stage diagnosis. Similarly, the supply of primary care physicians and radiologists was associated positively with earlier BCSAD.

Other studies have examined the effects of community-level variables on BCSAD using cancer registry data from California,23, 26 Connecticut,25 and Florida.13, 24 Two studies showed that the percentage females in the labor force and the percentage of females employed in blue-collar occupations were significant predictors of BCSAD. However, in this California study, employment and labor force variables were not significant. A Florida study reported that the percentage of Hispanics was associated positively with late BCASD. In contrast, community-level race/ethnicity variables were not significant, after controlling for individual race/ethnicity of patients with cancer in California. However, our findings confirm that disparities between black and Hispanic females persist. Primary care physician supply significantly predicted earlier BCSAD in Florida.24 Similar to the Ferrante et al. study, we found that a greater supply of primary care physicians and radiologists was a significant predictor of earlier BCSAD.

The study results should be interpreted with caution because there are limitations. First, the CCR data are weak in measuring reliable social and demographic characteristics of patients with cancer. For this reason, researchers have turned to external data sources. These variables are crucial for targeting vulnerable subgroups, which may be uninsured, lack financial resources, and have limited medical care access. Second, before 2000, the CCR provided only the 1990 census tract identifiers. Future studies will use the 2000 census data to construct community predictors. Community variables constructed from the 1990 census data may account for the smaller effect sizes. Third, the quality resource systems, ARF was used to construct the physician supply variables. However, the California Medical Association (CMA) recognizes that current data for physician supply is inadequate, stating that the American Medical Association database substantially overstates the number of physicians in California and the Medical Board of California does not collect information on physician responsibilities (e.g., patient care, administration, research, teaching) or specialties. A partial solution would be CMA-sponsored legislation that would allow the Medical Board of California to collect information needed by policy makers, but currently it is impossible to correctly assess physician supply.37 Nevertheless, The ARF provides one of the greatest national sources of variables to be analyzed on a county level,38 even though these limitations exist in the California data. Fourth, multicollinearity concerns are always a challenge when testing community variables. We used preliminary multivariate analyses to select conceptually similar community variables showing a high correlation (≥ 0.80). Fifth, no policy or financing variables were tested in the multivariate models. In general, these types of measures are more difficult to operationalize in access studies.39

The CCR data (1994–1999) reflect a time period when major legislation had been enacted by federal and state governments to improve access to mammography screening and treatment services for the lowest income women in California. In the current study, our most predictive measure of screening mammography was a county-level indicator collected by the CHIS, which asked if a female (≥ 30 years) had ever had a mammography. Other CHIS variables measuring the “time interval for the most recent mammography” and “how many mammograms in the past 6 years” were not significant (data not shown). In future research, it would be helpful to construct a community-level variable that measures access to mammography screening in a smaller geographic area rather than on a county level (e.g., zip code). This would enable a finer-grained analysis and help to target subgroups of hard-to-reach women with lower screening rates and later BCSAD.

In summary, community variables explained, to a greater extent, the differences in BCSAD among patients with breast carcinoma in California. The CCR data reflect a time period parallel with the enactment of major changes in the policy, financing, organization, and delivery of breast carcinoma early detection programs in California related to the federally funded Breast and Cervical Cancer Control Program and the state-funded Breast Cancer Early Detection Program. Our future research will extend the current study to investigate the effects of health policy and financing on reducing inequities in California communities.

Acknowledgements

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. REFERENCES

The authors thank William Wright and researchers at the California Cancer Registry, who were enormously responsive to our requests for registry data and associated technical support.

REFERENCES

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. REFERENCES
  • 1
    Manjer J, Berglund G, Bondesson L, et al. Intra-urban differences in breast cancer mortality: a study from the city of Malmo in Sweden. J Epidemiol Community Health. 2000; 54: 279285.
  • 2
    Perkins CI, Allen M, Wright WE. Breast cancer in California: stage at diagnosis and medical status. Sacramento, CA: California Department of Health Services, Cancer Surveillance Section, March 2000.
  • 3
    Roetzheim RG, Gonzalez EC, Ferrante JM, Pal N, Van Durme DJ, Krischer JP. Effects of health insurance and race on breast carcinoma treatments and outcomes. Cancer. 2000; 89: 22022213.
  • 4
    Richards MA, Smith P, Ramirez AJ, Fentiman IS, Rubens RD. The influence on survival of delay in the presentation and treatment of symptomatic breast cancer. Br J Cancer. 1999; 79: 858864.
  • 5
    Sporn MB. The war on cancer. Lancet. 1996; 347: 13771381.
  • 6
    Gospodarowicz M, Benedet L, Hutter RV, Fleming I, Henson DE, Sobin LH. History and international developments in cancer staging. Cancer Prev Control. 1998; 2: 262268.
  • 7
    Evans WK, Crook J, Read D, Morriss J, Logan DM. Capturing tumour stage in a cancer information database. Cancer Prev Control. 1998; 2: 304309.
  • 8
    American Cancer Society. Breast cancer: early detection, importance of finding breast cancer early [monograph on the Internet]. Atlanta: American Cancer Society, 2005. Available from URL: http://www.cancer.org/docroot/CRI/content/CRI_2_6x_Breast_Cancer_Early_Detection.asp?sitearea [accessed January 2, 2005].
  • 9
    Menck HR, Mills PK. The influence of urbanization, age, ethnicity, and income on the early diagnosis of breast carcinoma: opportunity for screening improvement. Cancer. 2001; 92: 12992304.
  • 10
    Bibb SC. Access and late-stage diagnosis of breast cancer in the military health system. Milit Med. 2000; 165: 585590.
  • 11
    McCarthy EP, Burns RB, Coughlin SS, et al. Mammography use helps to explain differences in breast cancer stage at diagnosis between older black and white women. Ann Intern Med. 1998; 128: 729736.
  • 12
    Roetzheim RG, Pal N. The effects of race/ethnicity and health insurance on cancer stage at diagnosis [abstract]. Abstr Book Assoc Health Serv Res. 1998; 15: 49.
  • 13
    Amey CH, Miller MK, Albrecht SL. The role of race and residence in determining stage at diagnosis of breast cancer. J Rural Health. 1997; 13: 99108.
  • 14
    Jones BA, Kasi SV, Curnen MG, Owens PH, Dubrow R. Severe obesity as an explanatory factor for the black/white difference in stage at diagnosis of breast cancer. Am J Epidemiol. 1997; 146: 394404.
  • 15
    Snipes K. Stage at diagnosis of female breast cancer, 1988-1992. In: MorrisCR, WrightWE, editors. Breast cancer in California. Sacramento, CA: California Department of Health Services, Cancer Surveillance Section, 1996.
  • 16
    Bastani R, Kaplan CP, Maxwell AE, Nisenbaum R, Pearce J, Marcus AC. Initial and repeat mammography screening in a low income multi-ethnic population in Los Angeles. Cancer Epidemiol Biomarkers Prev. 1995; 4: 161167.
  • 17
    Mandelblatt J, Andrews H, Kao R, Wallace R, Kerner J. Impact of access and social context on breast cancer stage at diagnosis. J Health Care Poor Underserved. 1995; 6: 342351.
  • 18
    Bastani R, Marcus AC, Maxwell AE, Das IP, Yan KX. Evaluation of an intervention to increase mammography screening in Los Angeles. Prev Med. 1994; 23: 8390.
  • 19
    Bastani R, Maxwell AE, Carbonari J, Rozelle R, Baxter J, Vernon S. Breast cancer knowledge, attitudes, and behaviors: a comparison of rural health and non-health workers. Cancer Epidemiol Biomarkers Prev. 1994; 3: 7784.
  • 20
    Kasper A. Experiences of poor and low income U.S. women with breast cancer [abstract]. Abstr Book Assoc Health Serv Res. 1999; 16: 162.
  • 21
    Thorpe KE, Howard D. Health insurance and spending among cancer patients. Health Aff (Millwood). 2003; (Suppl. Web Exclusives) W3-189W3-198.
  • 22
    Bastani R, Marcus AC, Hollatz-Brown A. Screening mammography rates and barriers to use: a Los Angeles County survey. Prev Med. 1991; 20: 350363.
  • 23
    Perkins CI, Wright WE, Allen M, Samuels SJ, Romano PS. Breast cancer stage at diagnosis in relation to duration of medicaid enrollment. Med Care. 2001; 39: 12241233.
  • 24
    Ferrante JM, Gonzalez EC, Pal N, Roetzheim RG. Effects of physician supply on early detection of breast cancer. J Am Board Fam Pract. 2000; 13: 408414.
  • 25
    Gregorio DI, Walsh SJ, Tate JP. Diminished socioeconomic and racial disparity in the detection of early-stage breast cancer, Connecticut, 1986-1995. Ethnic Dis. 1999; 9: 396402.
  • 26
    Lee-Feldstein A, Feldstein PJ, Buchmueller T, Katterhagen G. The relationship of HMOs, health insurance, and delivery systems to breast cancer outcomes. Med Care. 2000; 38: 705718.
  • 27
    FlemingID, CooperJS, HensenDE, et al., editors. AJCC cancer staging manual, 5th ed. Philadelphia: Lippincott-Raven, 1998.
  • 28
    U.S. Census Bureau. Census block groups cartographic boundary files description and metadata. Washington, DC: U.S. Census Bureau, 2002.
  • 29
    University of California. California Health Interview Survey [database on the Internet]. CHIS survey methodology and sample design. Available from URL: http://www.chis.ucla.edu/methods_design.html [accessed October 1, 2003].
  • 30
    Quality Resource Systems Incorporated [on the Internet]. Area resource file (ARF) [updated 2003]. Available from URL: http://www.arfsys.com [accessed November 8, 2003].
  • 31
    Cattaneo & Stroud Incorporated [homepage on the Internet]. 2001 Statewide HMO and special programs enrollment study [updated July 14, 2001]. Consultants of the health care industry [about 2 screens]. Available from URL: http://www.cattaneostroud.com/2001HMO_Study/default.htm [accessed September 12, 2003].
  • 32
    Schwartz GF. The current treatment of ductal carcinoma in situ. Breast J. 2001; 7: 308310.
  • 33
    Swallow CJ, Van Zee KJ, Sacchini V, Borgen PI. Ductal carcinoma in situ of the breast: progress and controversy. Curr Probl Surg. 1996; 33: 553600.
  • 34
    Smith RA, Saslow D, Sawyer KA, et al. American Cancer Society guidelines for breast cancer screening: update 2003. CA Cancer J Clin. 2003; 53: 141169.
  • 35
    Kaija H, Matti H, Tapani H. Use of hospital services by breast cancer patients by stage of the disease: implications on the costs of cancer control. Breast Cancer Res Treat. 1996; 37: 237241.
  • 36
    Legorreta A, Brooks R, Leibowitz A, et al. Cost of breast cancer treatment. A 4-year longitudinal study. Arch Intern Med. 1996; 156: 21972201.
  • 37
    The California Medical Association [on the Internet]. 2001 Physician survey findings. [updated 2004]. Available from URL: www.cmanet.org/upload/Physician_Supply_(Acrobat).pdf [accessed October 1, 2004].
  • 38
    Best AE. Secondary data bases and their use in outcomes research: a review of the area resource file and the healthcare cost and utilizations project. J Med Syst. 1999; 23: 175181.
  • 39
    Davidson PL, Andersen RM, Wyn R, Brown ER. A framework for evaluating safety-net and other community-level factors on access for low-income populations. Inquiry. 2004; 41: 2138.