Neighborhood-Level Socioeconomic Predictors of Nonadherence to Mammography Screening Guidelines

There is ample evidence showing that persons of higher socioeconomic status (SES) do better on most measures of health status (1). Traditionally, SES has been measured on an individual level, taking into account measures such as income, education level, and occupational ranking. Although individual-level variables are of particular importance in health disparities research, interest in neighborhood- or community-level SES has increased markedly (2, 3). Numerous studies have reported independent neighborhood effects on health outcomes, including all-cause mortality, chronic and infectious disease outcomes, and some health-damaging behaviors (4).

Despite the increasing volume in studies reporting area-level effects on health outcomes, a limited number of published studies have examined the influence of neighborhood effects on mammography screening use. In 2002, Coughlin et al. (5) found that women living in rural areas were less likely to have received a mammogram in the previous 2 years compared with women living in metropolitan areas. A Canadian study by Kothari and Birch (6) found that a regional-level variable (percent high school graduates in provinces' public health agency boundaries) had a significant effect on mammography use (measured as ever/never had a mammogram) independent of individual-level education. Conversely, in a study of African-American women enrolled in the Black Women's Health Study, Rosenberg et al. (7) reported that neighborhood-level socioeconomic characteristics were not associated with regular mammography use, measured at three 2-year follow-up periods. Most recently, Schootman et al. (8) found that area-level poverty was associated with never having a mammogram and Litaker and Tomolo (9) reported that the number of primary care physicians per capita and proportion of female-headed households were independently associated with breast cancer screening. In the health services literature, studies have shown that mammography screening has been associated with area-level health maintenance organization market share and other system-level factors (10-12). Although several novel studies have considered neighborhood-level predictors of mammography use, there is significant variability in the neighborhood-level indicators examined and the evidence remains inconclusive.

The aim of this study was to build on this relatively new area of research by determining if neighborhood-level SES was an independent predictor of nonadherence to mammography screening guidelines in a cohort of African-American and White women, ages 40 to 79, living in Connecticut who participated in the Race Differences in the Screening Mammography Process Study. The purpose of this prospective study was to better understand race differences in maintenance of mammography screening and related outcomes by examining a host of individual-level biomedical and psychosocial factors. This study was expanded to include neighborhood-level SES variables to potentially further explain racial/ethnic and socioeconomic disparities in adherence to mammography screening guidelines.

As previously reported (13-16), women who presented for a screening mammogram (hereafter referred to as the “index mammogram”) between October 1996 and January 1998, at five hospital-based mammography facilities in the Connecticut cities with the largest African-American populations, were recruited for enrollment. As African-Americans compose only 9.1% of the Connecticut population (17), we used 1990 U.S. Census data (18) and our own (1994) survey of mammography facilities in Connecticut (19) to identify where African-Americans lived as well as the mammography facilities that were most likely to provide their screening mammograms. As in most of the Northeast, the African-American population in Connecticut is largely urban (20). Thus, study subjects were recruited from hospital-based facilities in the five Connecticut cities with the largest African-American populations (ranged from 16% to 38% of total population); furthermore, four of these five cities were among the five most populous cities in Connecticut. The state-wide survey of all facilities in Connecticut showed that, with one exception, only hospital-based facilities met our enrollment criteria of reporting a high monthly volume of screening mammograms and ≥20% African-American patient population. The five hospital-based sites were located in New Haven, New London, Bridgeport, Hartford, and Waterbury. As expected from a community-based sample, the race-specific sociodemographic profiles in the final study population are similar to those of the general population in Connecticut (21-26).

All eligible African-American women who obtained index mammograms at these five facilities during the study period were invited to participate. White women were selected by a computer-generated random selection process and frequency matched to the African-American women on facility and date of mammogram. Asymptomatic women ages 40 to 79, who self-identified as African-American or White, with no previous history of breast malignancy, cyst aspiration, or biopsy were eligible for participation. In accordance with age recommendations for regular mammography screening in the general population (27, 28), women younger than age 40 were not included. Women over the age 79 were also excluded because of a lack of consensus with regard to screening recommendations for older women (29, 30). Approvals of the institutional review boards of Yale University School of Medicine and each participating hospital were maintained throughout the study period.

Initially, 2,359 women were identified for participation; 1,451 were interviewed after excluding ineligible women (n = 171), those who could not be contacted (n = 206), and women who declined participation (n = 531). Participation differed across race group (African-American, 69%; White, 77%; P < 0.001) and marginally by age (ages 40-49, 76%; age ≥50, 72%; P = 0.052). Two interviews were conducted in this study: (a) a structured 45-min baseline telephone interview conducted ∼1 month after the index screening mammogram so as to allow time for receipt of mammography results (mean time to baseline interview, 1.5 months; SD, ±0.85 month; range, 1-6 months) and (b) a follow-up interview arranged at a minimum of 26 months after the index screening. The time interval between baseline and follow-up interview averaged 29.4 months (SD, ± 1.42 months), with a range of 27 to 41 months. Of the 1,451 women who participated in the baseline interviews, 1,249 (86%) completed follow-up interviews. Twenty women were excluded either due to a cancer diagnosis associated with the exam or in the time interval between interviews (n = 11) or due to inadequate information to determine adherence to mammography screening guidelines (n = 9). Women included in the analysis differed significantly from those excluded or lost to follow-up by race (African-American, 78%; White, 93%; P < 0.001) but not by age.

Nonadherence to Screening Mammography Guidelines. The American Cancer Society screening guidelines in effect at the onset of the data collection period (1996) of this study (30) were used to determine the main outcome, nonadherence to screening mammography guidelines. Women ages 40 to 49 were considered nonadherent if they did not obtain at least one mammogram within 2 years (+2 months) of the index exam. Women ages ≥50 were considered nonadherent if they did not obtain at least two screenings within 2 years (+2 months) of the index exam. The “+2 months” allowed for reasonable delays in scheduling appointments.

For 1,126 respondents (92%), the outcome was determined by self-report. The remaining 103 women (8%) did not provide sufficient self-reported information to ascertain the outcome (i.e., could not recall the month or year of at least one mammogram) but did consent to a review of their mammography records. For these women, we relied on radiology records to determine outcome status. These 103 women did not differ from women with self-reported data by recruitment site or family breast cancer history, but they were more likely to be African-American than White (55% versus 38%, respectively; P < 0.001) and to be age ≥ 50 [78% versus 63% (age < 50); P < 0.003].

Neighborhood-Level Socioeconomic Factors. Baseline residential addresses (97.8%; n = 1,420) were successfully geocoded to obtain 1990 census tracts. Census tract-level variables were downloaded from the U.S. Census (31) and then linked to the individual-level study data. Because the health effects of neighborhood are thought to be related in large part to factors associated with socioeconomic position (2, 32), six conceptual domains of SES that are relevant to neighborhood socioeconomic position were evaluated. These domains and corresponding census indicators are as follows: (a) occupational class (percentage of adults in working class positions and percentage of adults unemployed), (b) income (median household income), (c) poverty (percentage of persons with households below the U.S. poverty line), (d) wealth (high assets defined as percentage of owner-occupied homes valued at $300,000 or more), (e) education (percentage of adults without a high school education), and (f) crowding (percentage of persons living in crowded conditions defined as more than one person per room). Additional variables that were considered in preliminary analyses were percentage of households without a car, percentage of housing units rented, percentage of housing units boarded, and racial composition (percentage non-Whites). A composite neighborhood socioeconomic index (SEP Index) was also created, following the example of Krieger et al. (33), using a standardized z score combining data on percentage working class, percentage unemployed, percentage below the U.S. poverty line, percentage without a high school education, percentage of expensive homes, and median household income; a higher score indicates a higher degree of deprivation.

Categorization of each of the neighborhood SES measures addressed specific analytic objectives. (a) In the absence of a priori category considerations, quartiles of the neighborhood measures were used for initial analyses. (b) Because of major differences in SES distributions by race, resulting in insufficient racial/ethnic overlap in categories, race-specific quartiles were created to examine race-specific relationships, a strategy used by Diez Roux et al. and Borrell et al. (34-36). (c) In addition, following the example of Diez Roux et al. (34) and Borrell et al. (35), categories of the SEP Index and the neighborhood-level education variables were created to have the same relative positions within these distributions as their respective positions within the race-specific individual-level income distributions and individual-level education distributions. This approach allowed a meaningful comparison of the effect of neighborhood-level SES and individual-level SES within each race group. Thus, the neighborhood SEP categories were created based on the four-category individual-level income distribution. Likewise, categories of neighborhood-level education mimic the three-category individual-level education distribution.

Individual-Level Socioeconomic Measures. Three measures of individual-level SES were collected: (a) annual family income (<$14,999, $15,000-$29,999, $30,000-$49,999, ≥$50,000), (b) education level (<12 years, 12 years, >12 years), and (c) occupation [combined spouse pair score, adapted from the Duncan Socioeconomic Index (37, 38), categorized as quartiles plus a missing data category that included nonrespondents as well as women who reported no occupation for either themselves or a partner].

Bivariate associations were examined between the neighborhood SES measures, the outcome (nonadherence to screening mammography guidelines), race, individual-level SES, and additional covariates that were hypothesized to influence mammography screening. Statistical significance was determined by the χ² test (P < 0.05). Linear trend was tested using the Mantel-Haenszel χ² test (P < 0.05).

Multivariable logistic regression was used to determine the adjusted association between neighborhood SES and nonadherence to screening mammography guidelines; adjusted odds ratios (OR) with 95% confidence intervals (95% CI) are reported. To account for possible within-area correlations, models were analyzed using Monte Carlo Markov Chain methods in MLwiN (39), software specifically designed for multilevel data. However, the estimated variance of the neighborhood random effects was negligible and not statistically significant in any models (final model random effects variance variable estimate, 0.032; SE, 0.030). Thus, models without the random effects essentially remained unchanged from models with random effects included. Therefore, only fixed effects for neighborhood SES from logistic regression modeling using Statistical Analysis System 9.1 are presented. Likelihood ratio tests were calculated to identify variables that contributed significantly to the fit of the model (40). A wide range of potential confounders and variables with known associations with screening mammography or SES, as well as variables known to vary significantly by race/ethnicity, was considered for inclusion in the multivariable models. These included sociodemographic factors, variables that were specific to the experience of undergoing mammography screening, health status and behaviors, logistical barriers, interaction with provider and provider characteristics, psychosocial factors, and known breast cancer risk factors. In addition to variables that were independently associated with the outcome, covariates that changed the estimate of the relationship between neighborhood variables and nonadherence by ≥10% were retained in the final models. Multicollinearity between the multiple neighborhood-level measures of SES was assessed by examining variance inflation factors in regression models with two or more neighborhood-level SES measures included simultaneously. A variance inflation factor of <2.5 was considered acceptable (41). Because the variance inflation factors exceeded this criterion, multiple measures of neighborhood-level SES were not included in the same model (except when included as a one-variable composite index). All two-level interactions were tested, including race interactions. Due to the significant racial/ethnic variation in neighborhood-level SES distributions, resulting in insufficient racial/ethnic overlap in categories, race-specific models, using race-specific quartiles and categories, were also examined. All reported analyses were done with Statistical Analysis System software, version 9.1 (42).

Characteristics of the study population by race/ethnicity are presented in Table 1. More than 60% of the respondents were age ≥50, with no significant difference by race/ethnicity. African-American women were significantly more likely than White women to be single than married/living as married, to have lower annual family incomes, to have <12 years of education, and to be in the lowest occupational status quartiles. Considering the main predictor, African-American women were disproportionately represented in the most disadvantaged neighborhoods as measured by the SEP Index. Over two thirds of all participants reported having complete coverage for annual screening mammography, and the majority of women reported having a usual healthcare provider and received a recommendation from their provider to get a mammogram in the 2 years after the index screening (with no significant differences by race/ethnicity). African-American women were significantly less likely than White women to correctly identify age-appropriate screening mammography guidelines and less likely to report a family history of breast cancer (any first- or second-degree relative). African-American women were also more likely to report a history of nonadherence to mammography screening guidelines. With respect to the study outcome, 47.8% of the total study population was nonadherent to screening mammography guidelines. African-American women were more likely than White women to be nonadherent (African-American, 53.7%; White, 43.9%; OR, 1.48; 95% CI, 1.18-1.87).

The unadjusted associations between individual-level SES and individual neighborhood-level SES variables (i.e., the components of the SEP index) and nonadherence to mammography screening guidelines are not shown in tables but are briefly described. All three individual-level SES measures (income, education, and occupation) were statistically significantly associated with nonadherence to guidelines (with statistically significant linear trend). Further, all neighborhood-level SES measures (i.e., occupation, unemployment, income, poverty, assets, education, crowding, percent of households without a car, and percent of housing units boarded up) examined were significantly associated with nonadherence (comparing participants living in the most disadvantaged neighborhoods with the least disadvantaged neighborhoods). A significant linear trend was observed for all measures of neighborhood SES. In unadjusted analyses, SEP Index [OR (quartile 4 versus quartile 1), 2.26; 95% CI, 1.60-3.18] and percentage of adults without a high school education [OR (quartile 4 compared with quartile 1), 2.43; 95% CI, 1.73-3.42] were significantly associated with nonadherence.

Although multivariable logistic regression models were initially tested using the total study population (data not shown), the significantly different distributions of SES for African-Americans and Whites resulted in lack of racial/ethnic overlap in categories. As such, we present race-specific multivariable models with sequential multivariable adjustment (first for age, then adding individual-level measures of SES, next adding other known predictors of mammography screening, and finally adjusting for neighborhood racial composition). Table 2 shows results using race-specific quartiles for all of the neighborhood-level SES measures examined, and Table 3 shows results using categories based on race-specific individual-level income distributions for the composite neighborhood SEP Index and neighborhood-level education, with the corresponding individual-level SES results.

Table 2 shows that, for African-American women, neighborhood-level unemployment, neighborhood-level median household income, neighborhood-level poverty, neighborhood-level education, and the SEP Index (all presented in separate models due to multicollinearity) were significantly associated with nonadherence to mammography screening guidelines after adjustment for age (highest quartiles versus lowest quartiles). After further adjustment for individual-level SES and mammography-related covariables, neighborhood unemployment and neighborhood median household income and neighborhood-level poverty, were no longer significantly associated with nonadherence. The results for neighborhood education and the SEP index were attenuated after adjustment for individual-level measures of SES but remained statistically significant. These associations were further attenuated with adjustment for mammography-related covariables and were no longer statistically significant. However, when the models were further adjusted for neighborhood racial composition, both neighborhood education and the overall SEP Index were significantly independently associated with nonadherence [neighborhood-level education: OR (4th quartile versus 1st quartile), 3.56; 95% CI, 1.34-9.47; neighborhood-level SEP Index: OR (4th quartile versus 1st quartile), 3.13; 95% CI, 1.01-9.70]. Conversely, for White women, neighborhood-level unemployment, neighborhood-level assets, neighborhood-level education, neighborhood-level crowding, neighborhood-level car ownership, and the SEP Index were all predictive of nonadherence to mammography screening guidelines after adjustment for age (highest quartiles versus lowest quartiles). However, only neighborhood-level crowding was significantly associated with nonadherence after adjustment for individual-level SES and mammography-related covariables. After further adjustment for neighborhood racial composition, neighborhood-level crowding and neighborhood-level assets (marginally significant) were associated with nonadherence to mammography screening guidelines for White women [neighborhood-level crowding: OR (4th quartile versus 1st quartile), 3.56; 95% CI, 1.34-9.47; neighborhood-level assets: OR (4th quartile versus 1st quartile), 1.72 (1.07-2.74); 1.46 (1.00-2.12)].

The following results (Table 3) are based on models using neighborhood SEP variables categorized to mirror the race-specific individual-level SES distributions. For African-American women, neighborhood-level SES (SEP Index) was associated with nonadherence to mammography screening guidelines after each step of sequential adjustment (first for age, next for individual-level SES, then for mammography-related covariables, and finally for neighborhood racial composition). The OR for African-Americans living in the most disadvantaged neighborhoods (SEP Index) compared with those living in the least disadvantaged neighborhoods was 3.55 (95% CI, 1.33-9.51) in the fully adjusted model. Sequential adjustment for a model including neighborhood-level education (categorized based on African-American individual-level education distributions) was also undertaken with similar results. Neighborhood-level education was significantly associated with nonadherence in the fully adjusted model [OR, 3.21; 95% CI, 1.25-8.26 (category 3 versus category 1)]. The results for individual-level SES (individual-level income and individual-level education, respectively) are also shown for each model. For African-American women, neither individual-level income nor individual-level education significantly predicted nonadherence with the selected neighborhood-level SES variable in the model.

For White women (with categorization of the neighborhood SEP Index and neighborhood-level education reflective of individual-level income and education, respectively), both neighborhood SEP (SEP Index) and neighborhood-level education were associated with nonadherence after adjustment for age. However, with further sequential adjustment for individual-level measures of SES, mammography-related variables, and neighborhood racial composition, the associations remained only marginally significant [OR, 2.13; 95% CI, 0.97-4.67 (quartile 4 versus quartile 1)] and neighborhood-level education [OR, 2.31; 95% CI, 0.93-5.76 (category 3 versus category 1)] once finally adjusted for racial composition. Neither individual-level income nor individual-level education was independently associated with nonadherence in White women in their respective models. Please note that these race-specific models are not directly comparable with each other or with total study population models, as race-specific distributions were used for categorization, resulting in different categorizations of neighborhood-level SES for each race-specific model.

Using data from the Race Differences in the Screening Mammography Process Study, we have shown that measures of neighborhood SES were associated with regular mammography screening independent of individual-level SES and other predictors of regular mammography screening behavior for both African-American and White women. However, the specific neighborhood measures that significantly predicted nonadherence differed for each race. For African-American women, neighborhood-level education and the overall neighborhood SEP Index predicted nonadherence, whereas neighborhood-level crowding and neighborhood-level assets were significant predictors of nonadherence for White women. When examining models with the neighborhood measures categorized to allow for comparison with individual-level SES, the neighborhood-level SES effects were larger in magnitude than individual-level measures of SES for all women irrespective of race. Furthermore, after adjustment for percentage non-Whites in the census tract, the magnitude of the associations for neighborhood-level SES measures became larger due to negative confounding (i.e., racial composition was masking a stronger effect of neighborhood-level SES). This suggests that there is an aspect of neighborhood socioeconomic deprivation that influences adherence to mammography screening (health behavior) independent of racial composition.

Although there are many potential mechanisms by which neighborhood SES influences regular mammography screening, hypothesized pathways relate to access to material or social resources or exposure to stressors or harmful environmental conditions. Because social norms and attitudes are believed to differ by residential areas (43) and may influence individual health behavior, we examined several mammography-specific psychosocial variables (with a theoretical basis in the Health Belief Model and other health behavior models) that may be correlated with social norms and attitudes. These included perceived susceptibility to getting breast cancer, perceived usefulness of mammography, barriers to screening, and cues to action. Even after adjustment for these psychosocial variables, neighborhood-level SES remained significantly associated with nonadherence. Additionally, low-SES neighborhoods may offer limited resources in support of good health and health behaviors (6). Although not measured directly here, specific conditions of one's environment may influence one's ability to develop supportive relationships (44), as literature suggests that neighborhoods shape neighboring and social interaction patterns (45). Moreover, social resources such as social capital, social cohesion, or collective efficacy, which are associated with poor health outcomes (44, 46-55), may be lower in more deprived neighborhoods. Thus, although we were able to control for a range of covariates that were theoretically based on the major health behavior models, there is the possibility of confounding due to unmeasured factors.

A limited number of studies have examined the association between neighborhood SES and mammography use, with varying conclusions. Whereas some of the studies showed a significant, independent effect of area-level factors on mammography screening, one notable study, using data from the Black Women's Health Study, did not find an association between neighborhood SES and regular mammography use (7). However because that study was composed of relatively affluent, well-educated Black women (97% had graduated high school) with high rates of regular screening (∼80%), that study population is not likely to be representative of African-American women in the general U.S. population. With only two neighborhood characteristics used in that study, and because the women were generally of high SES, the variation in the types of neighborhoods in which they resided may have been limited. Comparatively, our study offers several important methodologic advantages. (a) Because of the socioeconomic variability of the study participants' individual characteristics (e.g., ∼70% of African-American women graduated from high school and ∼47% had annual household incomes of ≤$15,000) and of the neighborhoods in which they lived, we have sufficient variation and numbers in the SES strata to detect significant associations. In addition, this socioeconomic variation in our study population is comparable with the general Connecticut and U.S. populations (e.g., U.S. African-American high school graduation percentage, 72%; Connecticut African-American high school graduation percentage, 74%; refs. 21-26). (b) Nonadherence to mammography screening guidelines was measured prospectively, with predictors collected at baseline, allowing for correct temporal sequencing. (c) The outcome addressed regular mammography screening behavior rather than one-time or recent use, an outcome more closely related to receiving the full benefit of mammography (56). (d) In addition to the benefit of having a large sample size of both African-American and White women, extensive data on many aspects of the mammography process were collected, including mammography-specific variables and factors relevant to health disparities that may confound, mediate, or moderate the relationships between neighborhood-level SES and nonadherence to mammography screening guidelines.

A limitation that many studies of neighborhood effects share is the use of administrative census data as a proxy for neighborhood. It is unlikely that census boundaries directly coincide with any meaningful definition of “neighborhood” as defined by residents. However, there are several advantages to using census data, such as the systematic collection of data for the entire population and its accessibility. Another measurement issue in our study, potentially leading to some misclassification, stems from linking geocoded address information provided at baseline (1996-1998) to 1990 census tract characteristics. The 1990 census information was chosen over 2000 census information to maintain the correct temporal sequence, but we cannot rule out changes in neighborhood characteristics between the 1990 census and the study enrollment period. However, as neighborhoods generally do not change significantly over short times (57), this is unlikely to have had a significant effect on reported findings. As a check, models using 2000 census tract median income (as an example of a neighborhood-level SES measure) were examined and produced results that were consistent with those using the 1990 census data.

The outcome, adherence to mammography screening guidelines, measured after an index screening, is a more detailed assessment of mammography use than that which is generally reported from retrospective studies of mammography screening and national surveys. However, if women who were lost to follow-up (and were also nonadherent) or never presented for a mammogram (not assessed in this study) were more likely to have lived in even more disadvantaged neighborhoods than those represented in the study sample, it is possible that we underestimated the effect of neighborhood-level SES on mammography screening behavior. However, because all facilities chosen were situated in urban areas [where ∼98% of the African-American population in Connecticut live (20)], and due to the relatively high levels of residential segregation in Connecticut (58, 59), it is likely that African-American women who have never had a mammography screening live among women who have had at least one mammogram. Thus, it is possible that the neighborhood-level barriers are similar for both groups. However, because some (mostly White women) also receive mammograms in smaller, private facilities, limiting our comparison with larger, hospital-based facilities may have attenuated differences across race groups. That said, as in the general U.S. and Connecticut populations, we observed the usual racial/ethnic differences in SES variables (21-26). Finally, with only 15.9% of eligible women reporting that they had never had a mammogram in national survey data (2002; ref. 60), the findings of this study may be applicable to the majority of women for whom maintenance of this screening behavior is relevant.

These results underscore the importance of examining social context as well as individual factors in health outcomes, as shown in this analysis of neighborhood-level SES and mammography screening behavior. In this study, living in disadvantaged neighborhoods was associated with nonadherence to mammography screening guidelines. Neighborhood disadvantage is increasingly recognized as a determinant of health outcomes, and our findings suggest that the effect of neighborhood disadvantage may extend to health prevention behaviors.

Although some aspects of neighborhood conditions may be amenable to intervention efforts, further research is needed to fully understand the mechanisms involved to develop appropriate interventions that are likely to influence change in health prevention behaviors, such as adherence to mammography screening guidelines. Meanwhile, building up both economic and social resources in disadvantaged communities may contribute to the adoption of better health behaviors and eventually lead to the reduction of racial/ethnic and socioeconomic disparities in selected health outcomes.

We thank the hospitals in Connecticut that allowed access to their patients and medical records (Bridgeport Hospital, Lawrence and Memorial Hospital, St. Francis Hospital and Medical Center, Waterbury Hospital, and Yale-New Haven Hospital) and our project coordinator, Lisa Schlenk, for her assistance with data collection and management.