Factors Associated with Racial/Ethnic Differences in Colorectal Cancer Screening

Patients and Setting

Patients were recruited from a University-based family medicine clinic in Southeast Texas during a 16-month period in 2004 and 2005. Although surveys conducted over large scale regions offer information about a larger population, a primary care setting was chosen as the site for this study because multiple topics could be handled in-depth and most CRC screening delivery is initiated in outpatient primary care settings.

The clinic serves a diverse mix of racial/ethnic groups from both urban and semirural areas; there is an excess of 40, 000 visits per year. The racial/ethnic distribution of patients over the age of 50 years is 66% non-Hispanic white, 24% African-American, and 10% Hispanic. More than 90% are insured. To be eligible, patients had to be 50 to 80 years of age and of non-Hispanic white, African-American, or Hispanic race/ethnicity. Individuals with a history of CRC or high risk of CRC (familial adenomatous polyposis syndrome, hereditary nonpolyposis CRC, or ulcerative colitis) were excluded.

A stratified sampling scheme, balanced by race/ethnicity; age (<65, ≥65); and sex was instituted to increase the statistical power for comparisons across racial/ethnic groups and the older age group. Interviewers recruited patients for each stratum until the target number was reached. Interviewers were bilingual (Spanish and English). Interviewers approached patients attending the clinic for any reason and invited them to participate in the study. The study was approved by the University of Texas Medical Branch Institutional Review Board and informed written consent was obtained from each patient. Interviews were conducted in a private room around the time of the doctor visit and lasted approximately 45 minutes.

Survey Development

Variables were chosen for inclusion in the study if they had been correlated with CRC screening in previous studies or were found to be important in our previous qualitative work in the same population⁴⁴ and if they were relevant for the practice setting. Items were developed to be culturally and linguistically appropriate, and a Spanish language version of the items was developed using standard methods.⁴⁵ The final instrument consisted of items organized into the following categories: (1) sociodemographic characteristics, (2) knowledge and beliefs about CRC and CRC screening, (3) medical history, and (4) health care experience variables. The outcome variable was self-reported history of CRC screening.

Measures

The outcome variable, self-reported CRC screening, was assessed with validated items adapted from Vernon et al.⁴⁶ A detailed description of each test preceded each question; descriptions were pilot tested for comprehension and feedback was incorporated into the final version. Current screening was determined by whether the patient reported that they had undergone any of the recommended tests, for any reason according to guidelines current at the time, as follows: annual home fecal occult blood testing or flexible sigmoidoscopy every 5 years; or annual fecal occult blood testing plus flexible sigmoidoscopy every 5 years; or double contrast barium enema every 5 years; or colonoscopy every 10 years.^6,7 The sociodemographic items were adapted from a national survey⁴⁷ and elicited information about the patient's age, educational level, sex, income, and insurance type. Race and ethnicity was self-reported and elicited with a 2-part question consistent with Federal criteria.⁴⁸

A 21-item knowledge test was developed covering CRC prevalence (1 item), symptoms (5 items), risk factors (13 items), screening test availability (1 item), and treatment (1 item). These items were developed de novo based on the findings from prior studies and qualitative interviews in this population.^44,49,50 An additional question assessed awareness of screening. We utilized the Health Belief model to examine attitudes and beliefs about CRC screening because it provides a useful organizing framework for explaining screening behavior.^{15,18,27,29–35,50–54} Scales were developed specifically for this multiethnic population and were tested to assess the constructs of perceived susceptibility (4 items), perceived benefits (10 items), and perceived barriers (11 items). All response categories were 3- to 5-point rating scales, ranging from strongly agree to strongly disagree. We included a 15-item fatalism scale because it is recognized as being an additional barrier in minority groups.⁵⁵ Medical history items included overall perceived health status, whether there was a family history of CRC in a first-degree relative, and health behaviors such as whether they had ever had an annual health examination or had previous testing for CRC. Health care experience variables included whether patients named a regular primary care doctor, whether they had ever received a doctor recommendation for a CRC screening test, and their satisfaction with their regular doctor or the previous provider, which was assessed with 3 items.^27,29 A summary of the scale measures is included in the Appendix.

Analysis

Descriptive statistics were used to describe patient demographic characteristics, and reliability on the knowledge and belief scales (Cronbach's alpha) was checked. A higher scale score indicated greater perception of that variable, except for perceived barriers, which were reverse coded. Analyses were conducted with weighting to make the sample representative of the clinic population from which it was drawn. We examined the bivariate relationship between potential explanatory variables (socioeconomic factors, knowledge and belief scales, medical history, and health care experience) with both race/ethnicity and CRC screening using χ² or analysis of variance as appropriate, and using 2-sided tests and α = .05. We then conducted series of nested logistic regression models and the scales were dichotomized at the mean score for ease of interpretation of the odds ratios. The first model included race/ethnicity with age, sex, and education as covariates. A second model evaluated knowledge and beliefs; a third model evaluated health care experience and medical history variables (without the knowledge and belief scales), and a final model included all variables together. For each model the effect of the variables on the odds ratios between race/ethnicity and current CRC screening was ascertained. Statistical analyses were conducted using SAS version 9.1.3 (SAS Institute, Inc., Cary, NC) and SUDAAN version 9.0.1 (Research Triangle Institue, Research Triangle Park, NC).

Demographic characteristics

Of the 1079 eligible patients approached for the study, 133 refused and 344 agreed but could not be scheduled at a mutually convenient time, giving an overall response rate of 55.7%. The first 30 interviews were used for piloting purposes and 12 subsequent surveys were incomplete; this left a total of 560 surveys complete for analysis. Thirty surveys were completed in Spanish. The rate of missing responses was less than 1% for each variable with the exception of income (3%); all missing responses were excluded from analysis. The final sample was 204 non-Hispanic white, 194 African-American, and 162 Hispanic patients, and almost all had health insurance (96.9%). We recruited slightly fewer Hispanics than planned and so we tested for statistical bias and observed no statistical differences between the respondents and nonrespondents by race/ethnicity, age, or sex. Further, the weighted sample was identical in profile to the clinic population aged 50 or older. The sample fell at the low end of the socioeconomic scale, with more than half of respondents reporting a total annual income of less than $25,000 (see Table 1).

Race/Ethnicity and Socioeconomic Factors, Knowledge and Beliefs, Medical History, and Health Care Experience Variables

Race/ethnicity was strongly associated with educational level (P < .001) and income (P < .001), and educational level and income were associated with each other (rho = .50; P < .05). A large percentage (96.9%) of patients had health insurance, no age or sex differences were noted among racial/ethnic subgroups (Table 1).

The knowledge and belief scales showed good reliability (Appendix). Minority groups perceived fewer benefits to screening (mean scores: non-Hispanic whites, 32.82; African-Americans, 31.43; and Hispanics, 30.52; P < .0001); were more fatalistic (mean scores: non-Hispanic whites, 18.27; African-Americans, 19.97; and Hispanics, 20.99; P < .0001); and had lower knowledge scores compared with non-Hispanic whites (mean scores: non-Hispanic whites, 9.87; African-Americans, 8.24; and Hispanics, 8.39; P < .0001). Hispanics, in particular, were least likely to be aware of screening for CRC. There were no racial/ethnic differences in perceived susceptibility or barriers to CRC screening. Of the medical history and health care experience variables, minority groups reported poorer overall health status, lower rates of previous testing and annual health exams, and receipt of a doctor recommendation for CRC screening (Table 1).

Current CRC Screening and Sociodemographic Variables, Beliefs, Medical History, and Health Care Experience Variables

Overall, 62.5% were current with CRC screening according to recommended guidelines. CRC screening rates were lowest in the minority groups: 67.5% in non-Hispanic whites, 54.3% in African-Americans, and 48.6% in Hispanics (P < .001). Colonoscopy was the most prevalent test but was reported least often by African-Americans and Hispanics; double contrast barium enema was reported least often by Hispanics (Table 2).

Older age and non-Hispanic white race/ethnicity were associated with higher rates of current CRC screening in bivariate analyses (Table 3). Of the knowledge and beliefs, fewer perceived barriers (P < .0001), less fatalistic attitudes (P < .05), and awareness of screening (P < .001) were associated with screening. Of the medical history and health care experience variables, having a regular doctor, having annual exams, having a doctor recommendation, and previous testing were all significantly associated with current CRC screening (Table 3).

The first multivariate model tested the effect of race/ethnicity while controlling for other sociodemographic factors (education, age, and sex). Income was excluded from further analyses because of its redundancy with education (rho = .50), and because of missing values (3%). Insurance was excluded because the majority (96.9%) was insured. Minority groups were significantly less likely to be current with screening even after controlling for sociodemographic factors. Furthermore, younger age and lower educational level were also associated with lower screening levels. This suggests that socioeconomic factors do not entirely explain racial/ethnic differences in screening (Table 4).

In a second model, where knowledge and belief variables were added to the above sociodemographic factors, the association of race/ethnicity with screening did not change meaningfully, although higher perceived susceptibility, fewer barriers, and an awareness of screening were significant. Addition of the health care experience and medical history variables in the third model tended to reduce the association between race/ethnicity and screening in both minority groups; this change was entirely attributable to doctor recommendation, which was the only significant variable in this set. Educational level and age retained their independent effect on screening.

When all potential explanatory variables were included in the final model, racial/ethnic differences became nonsignificant in both groups, but a greater reduction in the odds ratio was observed for Hispanics. Older age, higher educational level, a doctor's recommendation for CRC screening, awareness of screening, and perceived susceptibility remained significantly associated with screening. We also checked for and found no interactions between race/ethnicity and all significant variables in the last 2 models.

Strengths and Limitations

The study has a number of limitations. First, it was a clinic-based population that was of lower socioeconomic status but predominantly insured, which affects its generalizability to other groups. Although insurance coverage is often confounded with socioeconomic status (poor or less educated are less likely to have insurance) and age (those 65 and older have Medicare), in this study, controlling for insurance enabled us to examine the effects of race, education, and age independent of insurance coverage. Our findings can be generalized to insured patients that make visits to outpatient primary care practices in the United States, which account for over 384 million outpatient visits to primary care practices with non-Federally employed physicians.⁶⁹ Secondly, CRC screening was self-reported and was not verified by medical record abstraction, and although studies do suggest that there is a good correlation between self-report and actual screening⁷⁰ there is conflicting evidence about whether the accuracy of self-report varies by sociodemographic group.^58,71 Third, this was a cross-sectional study so we cannot draw conclusions about causal relationships. Fourth, we did not distinguish between tests done for screening versus diagnosis of symptoms because it is not known how accurately patients can distinguish between the two; this may have affected the findings. Also, although our response rate was 55.7%, a major reason for this was difficulty in scheduling patients who agreed to participate; these patients were considered to be nonrespondents. However, we did determine that respondents were no different to nonrespondents and were representative of the clinic population. Strengths of the study include the recruiting of African-Americans and Hispanic patients in sufficient numbers to make comparisons across groups and that our psychosocial instrument was specifically developed, based on a previous study in the same population,⁴⁴ to be culturally and linguistically appropriate; it displayed excellent psychometric properties.

Overall, our study makes an important contribution to the literature because it enhances our understanding of the reasons underlying racial/ethnic differences in CRC screening. It seems that what matters most when demographic, attitudinal, and health experience variables are considered together is that the health experience variables seem to be most important. This suggests that future strategies to address racial/ethnic disparities should focus predominantly on the health care environment and less on patient attitudes and beliefs.