Health Insurance–Related Disparities in Colorectal Cancer Screening in Virginia

Colorectal cancer (CRC) is the third most common cancer and the second leading cause of cancer-related death in American men and women (1). Several screening methods have been shown to reduce mortality from CRC (2) and recommendations in favor of screening have been made by the U.S. Preventive Services Task Force (3) and many societies (4-6). However, despite these recommendations, the proportion of Americans being screened for CRC remains low and screening practices have been shown to differ by gender (7), age (8), race and ethnicity (9), type of health care coverage (10), incomes (11), and body mass index (12). Relatively few studies have thoroughly evaluated the association of insurance status and CRC screening (11, 13-16). Improving CRC screening rates will require a thorough understanding of factors associated with screening. Using data from the 2005 Virginia Behavioral Risk Factor Surveillance System (BRFSS), we examined whether self-reported insurance coverage was associated with report of having undergone CRC screening (fecal occult blood testing within the previous year and/or lower endoscopy within the last 5 years) and what factors modified this association.

The BRFSS, a joint project between the Center for Disease Control and Prevention and state health departments, collects data about health risk behaviors, clinical preventive practices, and health care using a random-digit dialed telephone cross-sectional methodology. A disproportionate stratified random sampling method is used to ensure that a representative sample is collected from noninstitutionalized, U.S. civilian population ages ≥18 years of each state. To assess the association between self-reported insurance coverage and report of having been screened for CRC, our data analysis was limited to residents living in Virginia in 2005 who were ages ≥50 years, the age group for which CRC screening is recommended when at average risk for CRC. The University of North Carolina Institutional Review Board approved this study.

The main predictor, self-reported health care insurance status, was defined as having any kind of health care coverage (health insurance, prepaid plans such as Health Maintenance Organization, or government plans such as Medicare). The outcome measure was self-reported CRC screening defined as fecal occult blood testing within the past year and/or lower endoscopy (flexible sigmoidoscopy or colonoscopy) within the previous 5 years. Because the BRFSS does not differentiate between sigmoidoscopy and colonoscopy in its questionnaire, it is impossible to know which of these exams were done. In Virginia, a state law was introduced in 2000 requiring all insurance providers to offer comprehensive CRC screening coverage (17). Consequently, we considered any respondent who reported having undergone a lower endoscopy within the previous 5 years to have been screened.

Based on prior studies, the following covariates, shown to be associated with health insurance and CRC screening, were considered in our analyses: gender, age (50-54, 55-59, 60-64, ≥65 years), level of education attained (high school graduate or less, attended or graduated from college/technical school), annual household income (<$25,000, $25,000 to <$50,000, ≥$50,000), race/ethnicity (White, non-White), employment status (employed for wages, self-employed, unemployed/unable to work, homemaker/student, retired), and having seen a physician within the past year. Categories were chosen based on subject matter literature (age, income, education) or to ensure that the sample size in each category would be large enough (race/ethnicity, employment).

SURVEY procedures from Statistical Analysis System version 9.1 (SAS Institute) were used in all the analyses to account for the complex sampling design with stratification and unequal weighting used to collect the 2005 Virginia data. Multivariable logistic regression was used to estimate prevalence odds ratios (POR) and 95% confidence intervals (95% CI). The likelihood ratio test, which compares the main-effects, no-interaction model with the fully parameterized model containing the interaction term, was used to evaluate the potential effect measure modifier(s). If the interaction term significantly improved the model fit at α = 0.15, it was left in the final model. Confounding was assessed using a change in estimate strategy with a backward elimination approach. Variables were retained in the model as confounders if adjusting for them resulted in at least a 10% change in the effect estimate.

The distribution of the study subjects by characteristics and outcome overall and according to self-reported health insurance status is shown in Table 1. The study population was largely White, insured, and educated. There were slightly more females than males. The majority of participants (59%) reported having been screened for CRC.

The unadjusted bivariate analysis showed that participants who reported having health insurance coverage were more likely to report having been screened for CRC than those who reported not having health insurance coverage (POR, 2.16; 95% CI, 1.26-3.68).

As depicted in Table 2, the association between self-reported insurance coverage and CRC screening differed by gender (P = 0.10). The association was stronger for males (POR, 3.37; 95% CI, 1.63-6.96) than it was for females (POR, 1.46; 95% CI, 0.74-2.89). After adjusting for confounders (age and annual household income), males who reported having insurance coverage were still more likely to report having been screened for CRC compared with males without insurance coverage (POR, 2.02; 95% CI, 0.96-4.23). For females, there seemed to be no effect (POR, 0.81; 95% CI, 0.34-1.93).

Participants who reported having health insurance were more than twice as likely to report being screened for CRC compared with those who reported not having insurance coverage, but this relationship was stronger for men than for women. After adjusting for confounding, the relationship remained positive for men but not for women.

Studies evaluating the association of insurance status and CRC screening are relatively uncommon (11, 13-16, 18-21). Moreover, differences in patient populations, the variable definitions, and the statistical analyses of these studies make comparisons of the results challenging. In general, most of these studies found that having insurance coverage was positively associated with CRC screening (11, 13-16, 19-21).

To our knowledge, our study is the first that assesses potential effect measure modification of several covariates on the association between self-reported insurance status and report of having CRC screening. We observed that only gender affects this relationship. This gender-related difference has not been reported previously, and verification is necessary. We did, however, find one study that looked at the association between self-reported health insurance and report of being screened for CRC stratified by gender (21). Although potential confounders were not taken into account in their analyses, the authors found that having insurance was positively associated with being screened in both males and females.

We were surprised by the lack of association between report of having insurance coverage and self-reported CRC screening among women. One possible explanation for this result is uncontrolled confounding. Because this is an observational study, it is possible that this effect estimate was biased by an unmeasured confounder. For example, if an unmeasured factor was more common among uninsured women than those insured and if this factor was associated with an increased odds of being screened for CRC, it may have confounded the relation between insurance coverage and CRC screening in females. In addition, in spite of the large size of this study, there were relatively few uninsured participants resulting in quite imprecise effect estimates, particularly in women.

We acknowledge that this study has some limitations. First, because of the cross-sectional study design, a specific causative relation between insurance coverage and CRC screening may not necessarily be inferred. Second, because the BRFSS does not differentiate between sigmoidoscopy and colonoscopy in its questionnaire, we considered that a respondent who reported having undergone a lower endoscopy within the previous 5 years had been screened for CRC. Because several associations (4, 6) recommend that screening colonoscopy be done every 10 years in people ages ≥50 who are at average risk for CRC, our results might underestimate the actual proportion of CRC screening according to recommendations. On the other hand, the results might also be overestimated considering that the survey does not specify why the lower endoscopy was done (screening versus diagnostic use). Nonetheless, the association between self-reported health insurance coverage and report of CRC screening did not substantially differ when we considered lower endoscopy in the previous 5 or 10 years. Third, BRFSS data are based on self-report. Although self-reported data might be less exact than information obtained from medical charts (22), self-report of CRC screening is reasonably accurate (23). Fourth, BRFSS data are obtained by calling household landline telephone numbers. Although this can lead to undercoverage or noncoverage of certain subpopulations, in Virginia only 2.1% of households are without telephone service (24), which should have a minimal effect on our estimates. Finally, because Virginia requires insurance providers to cover preventive CRC screening (17), the proportion of those screened in our study is higher than national figures (25). Thus, the generalizability of our findings may be limited to states with demographic characteristics and insurance practices similar to those of Virginia.

In conclusion, this study showed that men who report having insurance coverage were more likely to report being screened for CRC. This association was not observed for women. To our knowledge, this gender difference has not been previously reported and additional investigation is necessary. These findings, if confirmed, could be of importance in the design and implementation of programs to improve CRC screening. The effect of insurance coverage may require gender-specific considerations: whereas improving insurance coverage may be expected to improve CRC screening in men, this may not necessarily be the case in women, who may require other interventions. Prospective studies, in more diverse populations, with different insurance requirements could provide more insights to our findings.