SEARCH

SEARCH BY CITATION

Keywords:

  • mammography;
  • ethnicity;
  • cancer worry;
  • beliefs;
  • screening;
  • within-group differences

Abstract

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

BACKGROUND

Ethnic differences in breast cancer screening behaviors are well established. However, there is a lack of understanding regarding exactly what causes these differences and which characteristics in low-screening populations should be targeted in an effort to modify screening behavior.

METHODS

Stratified cluster sampling was used to recruit 1364 women (ages 50–70 years) from 6 ethnic groups: African-American women; U.S.-born white women; English-speaking Caribbean, Haitian, and Dominican women; and immigrant Eastern-European women. In interviews, respondents provided information concerning demographic and structural variables related to mammogram utilization (age, education, income, marital status, physician recommendation, access, and insurance) and a set of cognitive variables (fatalism, perception of personal risk, health beliefs concerning cancer) and socioemotional variables (stress, cancer worry, embarrassment, and pain).

RESULTS

For data analysis, the authors used a 2-step logistic regression with frequency of mammograms over a 10-year period (≤ 4 mammograms over 10 years or ≥ 5 mammograms over 10 years) as a dependent variable. U.S.-born African-American women and Dominican women were screened as frequently as European-American women, but the remaining minority groups were screened with less frequency. With one exception, ethnicity ceased to predict screening frequency once cognitive and emotional variables were controlled.

CONCLUSIONS

Although women from clearly operationalized ethnic groups continue to screen at rates substantially below those of the majority groups, these differences appear to be explained substantially by differences in psychologic variables. This is encouraging because, rather than targeting culture for intervention, variables can be targeted that are amenable to change, such as emotions and beliefs. Cancer 2004. © 2004 American Cancer Society.

Breast cancer is the second most common malignancy in women.1 Although there has been some recent controversy,2 a number of studies and reviews have shown that mammography reduces mortality,3, 4 and several authors have argued that early detection and screening represent our best means for decreasing breast cancer mortality.5–7

Across the last 2 decades, a large body of descriptive research has documented the background factors that are believed to act barriers to, and facilitators of, mammography. In general, greater age,8–13 lower income and education,14–16 single marital status,10 the absence of insurance,17 and lack of physician recommendation6, 18–20 all were reported to be predictive of poorer mammography utilization. Minority women also are less likely to have regular mammograms, a finding that is attributed commonly to lower education and income. What is of interest, however, is that ethnic differences in screening behavior persist even when variables such as socioeconomic status or education do not differ or are controlled statistically.7, 21, 22

A growing number of screening studies also have examined an array of cognitive, attitudinal, and emotional factors that act as barriers to regular screening: patient perceptions of vulnerability, perceptions of negative consequences, perceived benefits and costs of self-protective behaviors, and emotional predispositions. In general, poorer knowledge,23–26 a belief that cancer treatments do not work,27–30 lower estimations of personal risk,6, 30–32 embarrassment,33–36 and stress37 are associated with lower screening rates; whereas anxiety may act as a facilitator or a barrier, depending on a number of factors.38

Limiting the utility of these studies is the fact that much of the research in psychosocial oncology, particularly research examining psychologic factors, has been restricted to small, nonrandom, predominantly European-American, convenience samples. More recent studies have included African-American women, although researchers nonetheless continue to examine minority women within very broad ethnic categories—Asian, Caucasian, Hispanic, and African American.39–41 However, there are a number of reasons to consider ethnic groups with greater specificity. In the first instance, contemporary urban populations continue to grow more heterogeneous, a major demographic trend that begs for consideration. Perhaps more important, the practice of imprecisely defining cultural and ethnic groups creates the possibility that practitioners will misinterpret research findings and what they tell us. Historically, for example, it has been found that African-American or black samples employ mammography at significantly lower rates compared with the European-American or white majority, although more recent studies have suggested that the disparities may be declining.14, 42 However, there is no way to determine whether these data apply equally to U.S.-born African Americans and to immigrants of African descent from the Caribbean or from the African continent because the groups invariably are not well differentiated.43

In the current study, we examined screening practices in a large set of diverse ethnic subpopulations of adult women based on a stratified cluster-sampling plan. The size of the sample enabled us to test the contributions of large numbers of variables that were identified as important in earlier, small-scale studies at the same time, thus distinguishing their individual contributions. Previous studies typically have been of limited size and power and could not include the full range of variables that are implicated in screening, although there are a few important exceptions.44 Herein, we include not only demographic and structural variables, such as physician referral, but also health beliefs and socioemotional variables, which are linked strongly to culture45, 46 and are potentially amenable to modification through culturally tailored message framing. For the current study, we operationalized ethnicity at a high level of specificity, differentiating among three groups of African descent (U.S.-born African Americans, English-speaking Caribbean, and Haitians), one Hispanic group of Dominican women, and two additional European groups (U.S.-born European Americans and Eastern-European immigrants). The main hypotheses of the study were that 1) after ethnicity and background/structural variables are entered into regression analysis, the addition of cognitive and socioemotional variables would add significant variance in the prediction of screening; and 2) we also anticipated that, with the inclusion of the latter variables, ethnic effects would be reduced, if not eliminated.

MATERIALS AND METHODS

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

The participants in this study were 1364 community-dwelling adult women ages 50–70 years who had no prior history of breast cancer and were drawn on the basis of a cluster-sampling plan. Data regarding census blocks were gathered from the Household Income and Race Summary Tape File 3A of the 1990 Census files. Blocks were then stratified by ethnic group and on the basis of income (high, medium, and low). Random selection without replacement was used to choose samples of block groups from each stratum. Further random procedures below the level of block group were not entertained because of our attempt to gain sufficient interviews with very precisely defined, and rarely studied, ethnic groupings. Trained interviewers were sent to conduct interviews with respondents who lived within the selected blocks. Respondents were recruited for a women's health project and were paid $25 for their participation.

Data were collected during face-to-face, structured interviews that lasted approximately 1.5 hours and were conducted in the respondents' homes or at other locations of their choice, such as a senior center or church. Questionnaires that included the study's measures were administered in a standard order across respondents.

Measures

Background and structural variables were obtained using a demographic questionnaire (age, ethnic background, household income, level of education, and marital status). Six ethnic groups were differentiated: U.S.-born African Americans, women born in the English-speaking Caribbean (immigrants from Jamaica, Trinidad, and Barbados), Dominicans, Haitians, U.S.-born European Americans, and a sample of immigrant women from Eastern Europe (Slavic groups from Russia, Ukraine, and Belarus) who served as a white immigration control group. Also included were questions pertaining to breast cancer screening behaviors, physician recommendation, and insurance. Participants were asked to indicate the frequency with which they had had mammograms over the past 10 years. Although there may be possible biases in self-reports, validation studies consistently have indicated that self-reports correspond reasonably well with clinic charts.47–49 Respondents indicated whether mammographic services were accessible to them on a scale from 1 (not at all accessible) to 5 (extremely accessible), whether their physician had recommended that they have a mammogram, and whether they had health insurance.

Cognitive factors were measured in three ways: cancer fatalism, perception of personal risk, and health beliefs. Cancer fatalism, the belief that one is powerless to control the onset or progression of cancer, was measured by the Powe Fatalism Inventory, a 15-item scale with a yes/no format that has an internal reliability of 0.84.50 The α in the current study was 0.88, and scores were summed to form an aggregate measure. Similar to previous research,32 perception of personal risk was obtained by asking women to assess their personal risk of getting breast cancer relative to other women (higher than other women their age, the same, or lower than other women). Finally, beliefs about breast cancer were measured using items derived from Gregg and Curry's ethnographic study of low-income African-American women.51 This scale measures three beliefs about the causes of breast cancer and four beliefs about cancer treatments. Respondents indicated the extent to which they agreed with each of 6 items from 1 (strongly disagree) to 7 (strongly agree). In terms of beliefs regarding cause, 21% of women in that study believed that cancer originates from a bruise or a sore, 18% believed that cancer is caused by chemicals in food, and 25% believed that the course of cancer is governed by God. In terms of treatment, 33% of women believed that radiation, surgery, and chemotherapy treatments do almost as much damage to an individual's health as the disease itself; 60% believed that surgery makes cancer worse by causing it to spread; and 25% believed that surgery can be helpful but only if the cancer is caught in time. Another belief about breast cancer fatality was not included, because this aspect already was captured by the fatalism measure described above.

Socioemotional factors measured in this study included stress, cancer worry, and discomfort with mammograms. The National Survey of Black Americans52 is a stress measure that asks respondents to indicate the intensity of stress experienced from 1 (not at all) to 3 (very much) across 10 life-event domains: health, money, job, problems with family or marriage, problems with people outside the family, children, crime, police, love life, and racial conflict. Summing across the domains yielded an overall stress score (α = 0.81). Worry over cancer was measured by the Cancer Attitude Inventory,53 which is a four-item scale that taps four domains of cancer concern. Scores on each item range from 1 to 6. The items were added to form an aggregate variable (α = 0.73). Finally, discomfort with mammograms was measured by asking respondents to indicate whether they found having a mammogram 1) embarrassing or 2) painful; the items were coded 0 (no) and 1 (yes). Women who indicated that they had never had a mammogram were asked whether they believed that it probably would be embarrassing and/or painful.

Data Analysis

A two-step logistic regression analysis was conducted to predict mammography screening rates. The screening guidelines—recommended starting age and intervals between screenings—have varied over the past decade or so. Twelve years ago,54 the American Cancer Society (ACS) recommended that women start having mammograms every 1–2 years beginning at age 40 years and start annual mammography at age 50 years. The ACS guidelines were updated in 199755 and recommended starting annual screening at age 40 years: These guidelines were reinforced in the latest published guidelines.56 According to this standard, the women in our survey ideally should have reported that they had had at least 5 mammograms over the past 10 years, and optimally 10 mammograms in that period. However, given the range of ages in the current sample (50–70 years) and the changes in screening guidelines over the past decade, the current analysis differentiated between women who screened at a rate of ≤ 4 mammograms in 10 years (infrequent screeners) and women who screened at a rate of ≥ 5 mammograms in 10 years (frequent screeners). We first checked to determine whether women in the 50–60 years age range screened at a different rate than women in the 61–70 years age range. They did not (chi-square test, [1] = 0.05; P = 0.82).

The first step in the screening model assessed the contribution of the background/structural variables of age, education, income, marital status, physician recommendation, accessibility of mammograms, insurance, and ethnicity (U.S.-born African American, English Caribbean, Haitian, Dominican, and Eastern European were all coded 1, and a comparison group of U.S.-born European Americans was coded 0). The second step included the entry of the eight cognitive variables (fatalism, perception of personal risk, and the six health-belief items51) and the four socioemotional variables (stress, cancer worry, and discomfort with mammograms [embarrassment, pain]) as predictor variables.

RESULTS

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

The mean age of the sample was 59.3 years (standard deviation [SD], 6.5 years). The mean level of education was 12.8 years (SD, 12.1 years), and the mean household income was $10,915 (SD, $9834). Table 1 presents the means of the demographic/structural variables by ethnic group and results of chi-square tests, analyses of variance, and Hays tests.57 We also calculated the median household income for each group. The median household income for U.S.-born African Americans, Haitians, Dominicans, and Eastern Europeans was $24.0 K; the median for English Caribbeans was $31.5 K; and the median for European Americans was $38.5 K.

Table 1. Background Characteristics of the Sample Broken Down by Ethnic Group and Results of Chi-Square Tests or Analysis of Variance
VariableEthnic groupF or chi-square valuePost-hoc comparisons
African American (n = 295)English Caribbean (n = 299)Haitian (n = 305)Dominican (n = 160)Eastern European (n = 151)European American (n = 154)
  • SD: standard deviation; H: Haitian; EE: Eastern European; AA: African American; EC: English Caribbean; D: Dominican; EA: European American.

  • a

    P < 0.01.

  • b

    Insurance code: 1, yes; 0, no.

  • c

    The mean number (standard deviation) of mammograms over the previous 10 years.

Mean (SD) age (yrs)58.9 (6.2)58.4 (7.0)60.4 (6.5)58.2 (6.1)60.8 (6.1)59.4 (6.5)5.7aH,EE > AA,EC,D
Mean (SD) income ($K)30.0 (26.5)34.9 (20.3)22.5 (15.5)25.3 (14.5)30.1 (22.7)46.3 (34.6)26.5aEA > all; EC > H,D; EE > H
Mean (SD) yrs of education13.5 (12.1)13.3 (8.3)7.4 (8.1)7.2 (8.2)16.2 (13.0)14.3 (13.1)194.4aEE > EA > AA, EC > H
Married (%)21.031.137.051.958.342.283.1aEE > D > EA > H > EC > AA
Physician recommendation (%)b92.593.383.095.092.792.230.3aH < all
Accessibility1.8 (0.87)2.0 (0.77)2.5 (1.0)1.9 (1.3)1.7 (0.88)1.7 (0.94)23.10aH < all; EC < EE,EA
Insurance (yes)92.588.065.673.893.490.9119.1aH < all
 No insurance (%)7.512.034.426.36.69.1HaysAA,EC,EE,EA < H,D
 Medicare (% yes)12.28.09.24.41.321.4HaysD,EE < AA,EC,H,EA; EA > all
 Medicaid (% yes)25.414.425.945.664.27.8HaysEAEC < AA,EC,H,D; D,EE > AA,EC
 Private (% yes)54.965.630.523.827.861.7HaysAA,EC,EA > H,D,EE
Mean (SD) no. of mammogramsc6.6 (4.1)6.6 (4.1)3.6 (3.3)5.6 (3.7)4.5 (3.7)6.9 (4.0)26.1aEA,AA > EC,H,D,EE; EC,D > H
 % Never screened7.87.420.04.48.64.5HaysH > all; D,EA < AA,EC,EE
 % 1–428.545.246.940.646.427.9HaysAA,EA < all
 % 5–1061.445.831.853.843.062.3HaysAA,EA > all, H < all
 % > 102.41.71.31.32.05.2HaysEA > all

The frequency of mammography across the last 10 years ranged from 0–20 mammograms, with an overall mean ± SD of 5.3 ± 3.9 mammograms. Across groups, 90% of women indicated that their physician had recommended they undergo a mammogram, and 83% that they had some kind of insurance coverage.

For the logistic regression analysis, there was a good model fit on the basis of the 12 background variables (chi-square [12 variables; n = 1362 respondents] = 225.65; P < 0.01; Nagelkerke coefficient [R2] = 0.20). For Step 2, with entry of the 12 cognitive and socioemotional variables, yielded a chi-square (12 variables, n = 1362 respondents) = 65.15 (P < 0.01), and the overall model was significant (chi-square [24 variables; n = 1362 respondents] = 290.80); Nagelkerke R2 = 0.26). Thus, the addition of Step 2 variables allowed us to account for an additional 6% of the variance in screening behavior. After the entry of the second step, the correct classification rates were 71% for the regular screeners and 66% for the infrequent screeners, for an overall classification rate of 69%.

Table 2 shows the contribution of the individual Step 1 demographic and structural predictor variables to the model (Table 2, far left column). In Step 1, as indicated in the table, higher education was associated with a 61% greater likelihood having a mammogram. Physician recommendation and insurance had the largest impact on screening rates; women with physicians who recommended screening were 2.3 times more likely to have a mammogram, and women with insurance were 2.6 times more likely to have a mammogram. Deterrents to screening were single marital status (30% less likely to screen) and English Caribbean ethnicity (45% less likely to screen), Haitian ethnicity (55% less likely to screen), or Eastern European ethnicity (74% less likely to screen).

Table 2. Multivariate Logistic Regression Analysis of Mammography Screening as a Function of Demographic and Structural Variables (Step 1, Model 1) and Cognitive and Psychosocial Variables (Step 2, Model 2)a
Model 1No. of participantsDemographic/ structural variablesModel 2 (Step 2): Psychosocial variables
OR95% CIOR95% CI
  • OR: odds ratio; 95% CI: 95% confidence interval; Ref group: reference group.

  • a

    For Model 1, higher rates of screening were predicted by higher education, being married versus single, physician recommendation, greater ease of access, and having insurance. Lower screening was associated with English Caribbean, Haitian, or Eastern European ethnicity. For Model 2, higher rates of screening were predicted by higher education, physician recommendation, greater ease of access, having insurance, greater belief that cancer treatments do as much damage as the disease, greater belief that surgery helps if the cancer is caught early, lower stress, greater cancer worry, lower embarrassment over mammograms, and greater experience (expectation) of pain/discomfort. Lower rates of screening were associated with Eastern-European ethnicity and beliefs that bruising or sores cause cancer.

  • b

    P < 0.01.

  • c

    P < 0.05.

Step 1     
 Total participants1364    
  U.S.-born white1541.00Ref group1.00Ref group
  U.S.-born African American2951.000.64–1.561.240.77–2.0
  English Caribbean2990.55b0.36–0.860.870.53–1.43
  Haitian3050.45b0.28–0.720.880.50–1.55
  Dominican1600.810.48–1.371.440.79–2.61
  Eastern European1510.26b0.16–0.440.36b0.20–0.63
 Age 1.010.99–1.031.010.99–1.03
 Education 1.61b1.17–2.221.50c1.08–2.08
 Income 1.001.00–1.001.001.00–1.00
 Marital status 1.30c1.01–1.701.230.98–1.67
 Physician recommendation 2.29b1.42–3.692.49b1.51–4.09
 Accessibility 0.62b0.54–0.710.63b0.55–0.73
 Insurance 2.59b1.79–3.752.63b1.80–3.84
Step 2     
 Fatalism   1.020.99–1.06
 Personal risk   1.090.86–1.35
 Bruise or sore causes cancer   0.90b0.83–0.96
 Chemicals in food cause cancer   1.040.96–1.12
 Course of cancer governed by God   0.940.88–1.00
 Treatment does as much damage   1.11b1.03–1.20
 Surgery causes cancer to spread   0.950.88–1.01
 Surgery good if caught in time   1.10c1.00–1.22
 Stress   0.97b0.95–0.99
 Cancer worry   1.90b1.04–1.34
 Embarrassment   0.71c0.50–0.98
 Pain   1.54b1.17–2.03

Table 2 also shows the contribution of the Step 2 cognitive and socioemotional variables (Table 2, right columns). Seven of 10 variables entered in Step 2, as indicated, improved the prediction of screening with all variables in the model: three of the health-belief variables and all four of the socioemotional variables. Believing that cancer is caused by a bruise or sore and that conventional treatments for cancer (surgery, radiation, chemotherapy) cause as much damage as the disease were both associated negatively with screening; whereas the belief that surgery can be effective if the cancer is detected early was associated positively with screening. Stress and embarrassment over mammograms were associated negatively with screening, whereas cancer worry and pain with respect to mammograms were associated positively with screening. Note that the size of the effects associated with stress, worry, and pain were of a magnitude approximating that of physician recommendation, a classically strong predictor. Finally, it is important to note that, before the entry of the cognitive and socioemotional variables, three of the five dummy ethnic variables were significant. However, two of those variables dropped to nonsignificance and a third variable was reduced in magnitude once they were introduced.

DISCUSSION

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

The current study of facilitators and barriers to breast cancer screening in a large urban population provides what to our knowledge are some of the most ethnically specific, comprehensive, and up-to-date data concerning the mammography screening behavior of women ages 50–70 years—those who are at greatest risk for breast cancer. Similar to previous researchers, we found that the variables associated with higher rates of screening included higher education,14–16 being married,10 a physician who specifically recommended mammography,6, 18–20 access to facilities that provide mammograms, and possession of medical insurance.17 In contrast to research conducted in younger, less diverse samples, however, we did not find an association between age and mammography utilization.8–13

In previous research, physician recommendation and insurance were among the most robust factors associated with screening, and this also was confirmed in the current study. However, it is important to point out that only 9% of women reported that their physician had never recommended getting a mammogram (the majority of these were Haitian women) and only 17% were without insurance (again, primarily Haitian women). Thus, although the odds ratios for these variables were high, the clinical relevance and opportunities for intervention in ethnically diverse urban populations may not be as significant as these values suggest.

What the current study adds to the literature regarding the factors associated with mammography practices among minority women is the demonstration that beliefs, attitudes, and socioemotional factors contribute significant and unique variance to screening behaviors, even when other variables are controlled. There was a negative correlation found between screening behavior and the belief that bruises or sores cause cancer and a positive association found between screening and the belief that conventional treatments for cancer are as damaging as the disease itself and the belief that surgery can be helpful if the disease is caught in time.27–30 Inconsistent with previous research, however, there were no associations found with estimations of personal risk6, 30–32 or cancer fatalism,50 as other authors have reported. In terms of socioemotional factors, women who reported greater stress and greater embarrassment reported fewer mammograms;33–36 whereas women who reported greater cancer worry, which is a very specific type of anxiety, reported more mammograms.10, 58, 59 Given the complexity of the relations between fear and screening noted in previous research53, 60 and some previous failures to find this effect among African Americans,53, 61 demonstrating the generality of this effect across groups is an important contribution. Finally, reports of greater discomfort or pain associated with previous mammography predicted greater screening. Although it may be somewhat counterintuitive, we suspect that reports of pain are a result of greater familiarity with mammography procedures due to more frequent or recent screening, with a consequence that the association may reflect the connection between past and future behavior.

Perhaps the most encouraging finding was that the screening rates for African-American women were equal to those of European-American women.14, 62, 63 Somewhat surprisingly, however, Dominican women also screened at equivalent rates,64 a result that is inconsistent with some previous research.30, 65, 66 The rates of the remaining immigrant groups—two of African descent and one of European descent—had odds ratios that differed significantly from the European-American comparison group. Before the entry of the cognitive and socioemotional variables, women from the English-speaking territories of the Caribbean, women from Haiti, and women from Eastern Europe were 55–74% less likely to be screened regularly compared with U.S.-born European Americans.

However, it is interesting to note, and in accordance with our primary hypothesis, that when the cognitive and socioemotional variables were entered into the model, the ethnic effects of being English Caribbean or Haitian no longer were associated significantly with lower screening, and the odds ratio for the Eastern-European women had dropped in magnitude. This pattern of results suggests that the variance in mammography was accounted for more effectively by cognitive and socioemotional variables than by ethnicity per se, and these variables may represent at least some of the underlying factors that account for the associations between ethnicity and screening behavior. In contrast to background factors, beliefs and emotional elements all are tractable conditions and are eminently suitable for educational and psychotherapeutic interventions.

The literature concerning breast cancer screening suggests that there are multiple factors influencing the likelihood that a woman will undergo regular mammograms. However, the various factors rarely are studied within the same model, and therefore the relative contributions of these variables are difficult to evaluate. Data from the current study illustrate three major points. First, cognitive and emotional factors are associated with screening, notwithstanding variations in the classic background factors that may be less amenable to intervention. Second, mammography rates vary within major ethnic groupings, necessitating a rethinking in the way researchers operationalize their samples. Third, recent writings and data suggest that ethnicity, in itself, may not be a useful variable for explaining health disparities.43, 67 Instead, it appears that race and ethnicity are more usefully thought of as indexing cultural differences in health beliefs and attitudes and in emotional dispositions and emotion regulatory patterns. The fact that two of the ethnic variables ceased to predict screening once such variables were entered suggests that interventions designed to increase screening among minority women may be aimed more profitably toward modifiable cognitive beliefs and emotional experiences pertaining to mammogram utilization, rather than toward less mutable background characteristics, such as education or income.

REFERENCES

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