Our results highlight the challenges facing safety-net health systems seeking to institute systematic programs for colorectal cancer screening. The size of the potential screen-eligible population at our health network is large—more than 28,000. The historical rate of participation in a colorectal cancer screening test was low, at less than 1 in 4 individuals. Although comparison of the CRC screening rate reported here to that of other safety-net health systems is limited by a lack of published literature on similar populations, our findings suggest that access to care (defined by having health insurance or ≥2 outpatient visits in 2006) plays a critical role in determining whether individuals complete CRC screening in this setting.
When placed in context of national screening recommendations from the US Multisociety Task Force on Colorectal Cancer(10
) and the US Preventive Services Task Force (USPSTF)(11
), our findings raise several important questions. Multisociety guidelines emphasize that the goal of cancer screening should be to “diagnose and prevent” cancer, and recommend use of tests that mainly identify polyps and cancer (e.g. colonoscopy, flexible sigmoidoscopy, barium enema, and computed tomographic colonography) over those it claims mainly identify cancers (such as guaiac FOBT and FIT). USPSTF guidelines do not make these distinctions, and suggest screening with highly sensitive FOBT, flexible sigmoidoscopy, or colonoscopy based on modeling that demonstrated similar life-years gained for all three approaches compared to no screening(11
For our health system, taking the Multisociety approach would greatly strain both our financial and manpower resources. We estimate employing this approach to try to screen the approximate 16,000 individuals who have not had screening would require an estimated 800% increase in the number of combined colonoscopies, sigmoidoscopies, and barium enemas performed, assuming the number of unscreened individuals has remained constant. A primary FOBT based screening approach, as would be acceptable under USPSTF guidelines, would also require a substantial increase in resources. If every unscreened individual were to receive a FOBT even once, assuming a 5 to 10% positivity rate, 100% adherence to follow up colonoscopy for positive FOBT, and no increase in the population in need of screening, we would require a 150% increase in the capacity to provide colonoscopy over 5 years. Thus, the sheer scale of need and associated resources required for screening mandate that we determine whether more CRC-associated death will be prevented by programmatically offering a resource-expensive test such as colonoscopy to a subset of our target population versus a more economical test such as guaiac FOBT or FIT that is less sensitive for cancer or polyps to a larger group. Similar considerations are important for other safety-net health systems. Indeed, if the goal is to maximize the public health impact of screening, the question of whether offering programmatic screening with invasive tests will achieve greater population benefits than offering programmatic screening with noninvasive tests (e.g. FIT) should be addressed by future comparative effectiveness research.
Beyond raising questions as to the optimal test(s) to use for programmatic screening, our data support the concept that access to care (i.e., having health insurance and being able to see a health care provider on a regular basis) may be the most important requisite for permitting preventive care such as CRC screening(2
). In our analysis, participation in the local county health medical assistance program was associated with rates of screening comparable for that observed for individuals with other insurance such as Medicare. It is significant that a medical assistance program supported by local taxpayers can be associated with rates of completion of a preventive service such as colon cancer screening similar to other types of insurance. Our observation, if replicable, suggests that even if the scope of national health care initiatives were limited, substantial benefits with respect to CRC prevention could be achieved.
It is important to note that in this study population, neither African-American race nor Hispanic ethnicity was associated with decreased rates of participation. Indeed, screening participation among Hispanic participants was higher than for Whites. This is in contrast to prior reports, which have observed disparities in screening rates for African-Americans and Hispanics even after adjusting for factors such as socioeconomic status and insurance status(3
). We speculate that race and ethnic based disparities were not observed because of otherwise similar geography, insurance status, socioeconomic status, and access to care in this study population. Our findings may complement other observations that racial and ethnic disparities in U.S. health delivery may be surmountable when access to care is enabled(14
There are several potential limitations to our study. First our estimates and conclusions are based on administrative data, thus misclassifications in completion of a screening test and estimates of predictors of screening could have occurred(17
). However, preliminary results from an ongoing ancillary study, in which the paper and electronic medical charts for a random sample of 100 individuals each with administrative record of FOBT, colonoscopy, flexible sigmoidoscopy, barium enema, or “no screening,” have been systematically reviewed suggest that the agreement of administrative coding with individual chart record is substantial for all administrative coding overall (kappa=0.63, 95%CI: 0.55–0.71), and almost perfect for the most common tests completed (colonoscopy and FOBT). Further, just 4% of individuals with no administrative coding for screening had record of screening on chart review, the sensitivity of a positive administrative code for test completion ranged 91 to 99%, and the specificity ranged 55 to 92%. Thus, imprecise administrative coding may have had limited impact on our estimates of screening prevalence. Second, JPS is not a closed health system; therefore some individuals, particularly those with insurance, may have had screening performed at other health facilities, leading to underestimates of screening participation. Third, only colonoscopy data for 5 rather than 10 years were available. Though an individual who had a colonoscopy between 1997 and 2006 would be guideline adherent to CRC screening based on current definitions(10
), lack of electronic record of colonoscopy procedures prior to 2002 precluded analysis based on this criterion. Colonoscopy was not routinely recommended for primary screening in our system prior to this period, perhaps minimizing underestimation of screening completion based on this factor. Nonetheless, we estimate that even taking into account a possible 4% rate of false negative assessment of screening by administrative data, and a doubling of the colonoscopy rate due to unmeasured examinations occurring more than 5 years remote to cohort inception would increase our estimate of the prevalence of screening to only 29%(data not shown). Fourth, examinations performed for symptoms such as hematochezia are reflexive actions rather than preventive measures, and may not optimize the goal of finding early-stage cancerous or polypoid lesions in asymptomatic patients, whereas benefits of screening have been most clearly demonstrated in randomized controlled trials of asymptomatic patients(10
). Indeed, some investigators may characterize the present analysis as one of colorectal cancer testing or test use rather screening because indications were not abstracted (3
). Fifth, zip code linkage to ZCTA associated census data may be an imprecise estimate of the relationship of measures of socioeconomic deprivation and health outcomes(24
). Future investigation of any relationship between measures of socioeconomic deprivation and screening outcomes using more precise measures employing census tract or block measures is warranted. Sixth, some potential confounding factors for predictors of screening completion, such as potential confounding of the association between frequency of outpatient visits and screening completion by burden of comorbid illness, were not studied. Lastly, individuals seen in urgent care and emergency room settings in our analysis are generally not recruited for screening, and some might suggest that these individuals should not be included in our study. On sensitivity analyses, restriction of the study population to individuals seen ≥2 times, with at least one non-urgent care/non-emgency room visit in 2006, resulted in a modest increase in the estimate of prevalence of screening to 27%, and did not substantially change estimates for predictors of screening on multivariate analyses (see Supplementary Appendix 3
). Further, from the perspective of a safety-net health system, our true study base and target population includes all individuals in Tarrant County, TX, who, if symptomatic colorectal cancer developed due to lack fo screening, would present to our emergency department and clinics for treatment of later stage disease. From our local public health perspective, it is this population that requires identification and specific interventions. In conclusion, we have demonstrated that the size of the screen-eligible population, and the number who go unscreened, pose significant challenges to our safety-net health system. If our data are representative of other safety-net systems, specific and potentially modifiable variables (such as insurance status and access to a medical provider) deserve further study in order to overcome the challenges posed. Further, short and long term screening guidelines and policy efforts must take into account the feasibility and potential costs of proposed interventions. Substantial resources for near and long term population-based screening (including comparative effectiveness research into the best manner to provide screening to large populations, improving access to care, and promoting screening outside of traditional health visit settings) may be required to provide the immense potential benefit of CRC screening to individuals served by safety-net systems.