Epidemiology of Colon Cancer Screening in a Safety-Net Setting.

INTRODUCTION
Colorectal cancer screening is a high-priority preventive intervention with favorable clinical benefit and cost effectiveness.1,2 Over the past 3 decades, colorectal cancer incidence and mortality have declined.3 During this period, the uptake of screening increased,4 the sensitivity of stool-based screening tests improved,5 and a multimodal approach to colorectal cancer screening was adopted by the U.S. Preventive Services Task Force (USPSTF).6 The Healthy People 2030 target for self-reported colorectal cancer screening is 74.4%, higher than the 2018 rate (65.2%) reported in the National Health Interview Study.7
Despite the benefits of colorectal cancer screening,8,9 rates are lower than for cervical or breast cancer screening.4 There are important disparities in terms of incidence and mortality from colorectal cancer among African Americans and Native Americans5 and less penetrance of screening among those with less education, lower income, and lack of insurance or a regular healthcare provider.10
To reach target colorectal cancer screening rates, how best to operationalize screening programs needs to be understood. Although provider communication increases screening,11 physician recommendation alone is an impractical approach given the overwhelming clinical and preventive demands on primary care providers.12 The following approaches can augment provider communication: team-based care with responsibilities shared among providers, nurses, and medical assistants13, 14, 15; patient navigators to guide patients through screening, diagnosis, and therapy16,17; and centralizing screening through a health system–wide prevention program.18
Each approach relies on registries to track a patient’s status, including lost to follow-up.19 Registries have supported care management since the 1980s. Initially, registries were developed for chronic disease management and supported focused assessment and provision of evidence-based clinical interventions.20 The use of registries extended to preventive services, including colon cancer screening,19,21, 22, 23 and their importance is widely acknowledged.18,24
Beginning in 2010, the authors initiated a colon cancer screening program, applying the Chronic Care Model25 and a population approach within hospital-based ambulatory clinics of Cook County Health. The authors used 2008 USPSTF guidelines as their standard of care.26 The authors worked with the American Cancer Society to develop patient education incorporated into visits. During the entire project period, the authors’ institution used the same electronic health record (EHR) system (Cerner, Inc., Kansas City, MO). The authors developed a web-based registry, integrated with the EHR with unidirectional flow of data from the EHR to the registry. The registry was designed to capture data on demographics, clinic appointments, and fecal occult blood tests (guaiac-based and immunochemical testing) and results and to generate a daily list of patients coming to the clinic who were screen eligible and due for colon cancer screening. Registry updates for results of guaiac fecal occult blood test (G-FOBT) or fecal immunochemical test (FIT) were enabled through an electronic interface to the database. Colonoscopy, flexible sigmoidoscopy, and pathology reports were unavailable; volunteer students reviewed the EMR and manually inputted these results. For screen-eligible patients, the authors offered available modalities of screening. In 2012, 24,300 unique patients visited the general medicine clinic (GMC), which is a primary care internal medicine clinic. During clinic visits, volunteer students (trained as educators) educated patients during their previsit waiting time. The educators discussed screening options and risks and benefits of each option. This partnership gave patients the opportunity to choose their preferred option. The educators provided FIT kits with instructions and helped schedule flexible sigmoidoscopy or combination of FIT and flexible sigmoidoscopy or colonoscopy according to patient preference. They updated the patient’s provider about their preferred screening option and entered data into the registry. It is established that mailing stool test for colon cancer screening outreach has been found effective,27, 28, 29, 30, 31, 32 but owing to limited resources, the authors were not able to mail the FIT kits and distributed them during their clinic visits.
In addition, the authors used the registry for case management of patients due for diagnostic colonoscopies. EHR prompts were not used because the registry was a stand-alone project database, and integration of the detailed colon cancer screening status and logic into the EHR was not considered feasible.
In 2014, the authors undertook a retrospective assessment of the registry looking at 2 time points spanning 28 months—January 1, 2012, and April 30, 2014—to observe how often categorization of the screening eligibility of patients changed between these 2 time points. The authors describe their findings and the implications for development of a registry to support colon cancer screening.

METHODS

Population
In 2012, the authors enrolled patients aged ≥18 years empaneled in the GMC, a primary care internal medicine clinic of Cook County Health, the safety-net provider for Cook County. To monitor the screening-eligible population, a web-based registry was built that interfaced with a research data warehouse that comprehensively captured clinical data from the EHR.33 The registry was prepopulated manually with patients, after which it autopopulated daily with demographics, appointments to primary care clinics, and results of G-FOBT or FIT. The interface captured performance of endoscopies. The status of individual patients with respect to colon cancer screening was incomplete until there was a manual chart review focused on results of colonoscopy, flexible sigmoidoscopy, and pathology reports from the EHR.
After chart review and manual registry updates, the screening status of each patient was categorized (Table 1). The authors defined active clinic patients as any patient aged 49–78 years with at least 2 clinic visits in the past 24 months, each visit separated by at least 60 days and at least 1 visit in the past 12 months. The authors registered patients aged <50 years to include all patients when they turned age 50 years. Similarly, patients aged 78 years were selected because they were less than 75 years when their information was entered into the registry in 2012.
The registry was populated on January 1, 2012, with the addition of patients who newly met the criteria of active patients (Ingress). The authors removed patients who no longer met the clinic visit criteria for active patients, who no longer met the age criteria for screening, or who had died (egress patients). The authors could not capture the specific reasons for egress owing to resource constraints. The authors studied the active patient population (those enrolled on project and initiation and those who entered the cohort [ingress]) for a 28-month period from January 1, 2012, to April 30, 2014; all patients in the sampled population were present at the time of project termination.

Measures
The authors adapted 2008 USPSTF recommendations to determine the up-to-date status of the screening population in their clinic.34 Screening was considered up to date if the patient had completed (1) either G-FOBT or FIT in the past 1 year, (2) combination FIT (past 3 years) and screening flexible sigmoidoscopy (past 5 years), or (3) screening colonoscopy (past 10 years).
The medicine clinic population was categorized into screening eligible, nonscreening eligible, and refused, following the recommendations of 2008 USPSTF guidelines26 and American College of Gastroenterology 2008 Practice Guidelines.35 The screening-eligible category was divided into screening up to date and screening not up to date. The non-screening-eligible patients were further divided into the following 5 groups: surveillance, work up in progress, poor health status, indeterminate, and personal history of colon cancer (Table 1).

Statistical Analysis
To validate registry data and reliably capture screening status as reflected in the registry at 2 points (T0 and T28) in time, the authors selected a 10% simple random sample from the registry population. The authors grouped the random sample of active patients into screening categories at the time that they became an active patient (T0) and, again, at the time of the 28-month follow-up period (T28). Screening status in the registry was reviewed and validated independently by 6 staff members: 2 physicians, 2 research assistants, 1 nurse clinician, and 1 data analyst.
The authors performed descriptive analyses to evaluate how often patients’ colon cancer screening status changed over time. They categorized patients into the mutually exclusive colon cancer screening categories mentioned earlier at times T0 and T28. For people who moved between categories over the 28-month interval, the authors describe the relative frequency of these transitions and graphically display the changes among categories over time. Data were imported into IBM SPSS (Version 25.0) for all analyses. The statistical analysis was started in February 2016. The data were analyzed, results were obtained, and analysis was completed in November 2017. Generation of the Sankey diagram demonstrating the flow between initial and final screening status categorizations was performed in Jupyter Notebooks using the plotly.graphobjects and pandas libraries in Python (Version 3.9.13).

RESULTS
In 2012, there were approximately 15,000 unique screening-aged GMC patients, of whom 13,361 were in the registry. At the beginning of the study—January 1, 2012—3,544 patients who met the definition of active GMC patients were added to the registry (ingress). The authors removed 6,259 patients from the registry because they had lost their status as active GMC patients (egress).
After accounting for egress and ingress, the registry had 10,646 patients, from whom the authors sampled ∼10% (1,074 patients) for chart review (Figure 1). Of the 1,074 randomly sampled patients, 716 were already active in the registry, and 358 were new patients who were added to the registry (ingress) at T0 on the basis of either their age or they were new to the clinic and met the definition for active GMC patient. The mean age for the study populations at follow-up was 61 years, most were aged <75 years, and the majority were female and African Americans (Table 2).
In the initial EMR review at T0, most patients were screening eligible, and few had refused screening; by the end of the study period, there was a 5% absolute increase in the screening up-to-date category (Table 3). Few screening-eligible patients at baseline (T0) had moved into different categories at T28, but among those who changed categories, it was most commonly to surveillance, followed by workup in progress, refused to be screened, a personal history of colon cancer; indeterminate; and poor health (Figure 2). Although some individuals who refused at T0 became screen eligible, by T28, there was a small overall net increase in refusals (Figure 2). The largest numerical change in categories was from screening eligible at T0 to surveillance at T28. Among nonscreening eligible, the majority were in the surveillance category owing to higher risk of colon cancer or prior pathologic colonoscopy findings; fewer common reasons for being ineligible for screening in descending order were work-up in progress, personal history of colon cancer, poor health, and indeterminate (Table 3). Over 28 months, a change in category occurred in 125 of the 1,074 patients (12%) (Figure 2).
Other changes between the major categories for colon cancer screening status were less common (Figure 2). Notably, of the 41 (4%) patients who refused to be screened initially, a substantial minority (15%) subsequently agreed to colon cancer screening. Of these 6 patients, 1 moved into the surveillance category, and 5 became screening eligible.
Among the 363 patients who were screening eligible and up to date at T0, after 28 months, a total of 549 screenings had been performed, as follows: colonoscopy (n=273), FIT testing (n=142), G-FOBT (n=101), and flexible sigmoidoscopy (n=33).

DISCUSSION
The technology implemented in this study enabled systematic approach to colon cancer screening and facilitated the management of colon cancer screening program in primary care clinics. The authors defined a screening-eligible population and used the registry as a platform for case management and monitoring of the population's screening status. They developed a categorization scheme for colon cancer screening, which was clinically meaningful and comprehensive. To improve the rate of colon cancer screening, initially, the authors focused on screen-eligible patients; they offered individualized brief colon cancer education and options for screening modalities at intervals recommended by USPTF. The authors used this registry to case manage patients due for diagnostic evaluations on the basis of screening results and to resolve the indeterminate category. The authors did not use the registry to case manage those in the surveillance category; however, it would be important to also intervene on these patients to ensure timely follow-up evaluations. Prior to the authors’ initiative, colon cancer screening was provider driven, highly variable, and less reliable. Patient navigation and education play a very important role in screening uptake.29,36,37 During the time frame of this report, the authors demonstrated an increase in the rate of screening in their clinical setting from 42% to 47%, which exceeded screening rates reported by Federally Quality Health Centers.38,39
Applying the definition for active GMC patients, the authors identified substantial ingress and egress to the registry over the 24-month evaluation period. Some ingress and egress into a colon cancer screening registry are expected as patients age in, age out, have lapses in primary care, transfer care to another healthcare facility, or die. Because most persons age into Medicare during the time period for colon cancer screening, changes in insurance status often leads to changes in access and site of care. In addition, high rates of patient migration between medical sites are common throughout the U.S. owing to insurance and job changes, geographic relocation, or dissatisfaction in care. As an urban public hospital, the authors provide care for a unique population who may not be representative of other care settings. The substantial ingress and egress reported in this study may have been in part due to the launch of the Affordable Care Act (ACA) in 2010; some newly insured patients had more options to choose their site of medical care. There exists a complex relationship between work and health insurance that along with how healthcare policy such as ACA can alter patient–provider relationships.40 The ACA reforms altered the structure and sites of healthcare delivery.15
Although it is not common for older patients to change their primary care physicians, changes in the healthcare system might force them to change their primary care physicians.41 The substantial ingress (33%) and egress (47%) to the registry during the evaluation period are notable. Because these proportions were determined retrospectively, it was beyond the scope of the project to collect data and quantify the reasons why people entered or left the clinical site. A separate study reported that 11% of patients with a positive FIT transferred to a different nonstudy site or had a change in their medical condition that prevented a colonoscopy.29 Although the observed egress and ingress may not generalize to other time periods or healthcare settings, it reflects that movement between healthcare sites and/or instability in primary care follow-up is common. When patients change their site of care, there is loss of information on screening status, delays in care, and risk for unnecessary and redundant testing, sometimes resulting from incomplete availability of medical records.42,43 This fragmentation of care results in increased cost and safety and satisfaction concerns.42, 43, 44
One potential solution is to have a regional data warehouse specific for endoscopies and associated pathologic findings to aggregate and share cancer screening data. Although the number of successful implementations is limited owing to obstacles faced with interoperability among major EMR vendors, advanced health information exchanges are valuable and could provide a centralized data repository for patients eligible for colon cancer screening.45, 46, 47, 48, 49 A centralized repository would reduce barriers of obtaining medical records from outside institutions and may address the problem of indeterminate screening status due to transitions of care. Such an endeavor, although complicated by pathology and colonoscopy reports being compiled into formats difficult to parse and need for harmonization of clinical records, would be a critical advantage to improving colon cancer screening. There exist efforts to integrate longitudinal data across systems for cancer patient populations, but the authors are not aware of any for cancer screening populations.50,51
As part of development of a colon cancer screening electronic registry, having a clinically meaningful and comprehensive categorization scheme is necessary. The authors developed such a scheme (Table 1) and believe that it offers a unique contribution to the literature. The authors included a definition for screen eligible, limited to patients who were due for screening or had prior normal screens. Having a defined screen eligible population is critical for case finding and tracking of screening rates. Substantial effort was needed to establish screen eligible status by eliminating those the authors considered ineligible because of prior colon cancer or colonoscopy findings more appropriate for surveillance colonoscopy, pending diagnostic colonoscopy due to a positive hemoccult test or clinical concerns outside the screening pathway, or inappropriate for screening due to poor health status or refusal. It is important to know accurate screening rates and to screen relevant populations to reach the colon cancer screening target of 74.4% set by Healthy people 2030.7
When the authors established the registry, most age-appropriate patients were eligible for screening. After 28 months, eligibility decreased primarily owing to performance of screening colonoscopy, but some patients left the screen-eligible category owing to refusal to continue screening or being deemed in poor health. Overall, approximately 1 of 8 patients changed screening categories. Most commonly, screen-eligible patients moved into the surveillance or work-up-in-progress (awaiting a diagnostic colonoscopy) categories. Not surprisingly, graduation from poor health status to screening eligible was rare. Some patients moved from initial refusal to participation in the active screening population, supporting encouraging those who initially refuse to reconsider screening while also respecting their decisions. Although 100% of surveillance patients remained in this category, none were diagnosed with cancer; it was beyond the scope of the study to capture whether these patients received timely follow-up. Had the authors more actively case managed the surveillance category, it is possible that additional patients would have changed their screening category.
Without case management or medical record innovations, the accuracy of a colon cancer screening registry will rapidly degrade. Although an EHR can facilitate automated identification of patients eligible for colon cancer screening and engage patients electronically through a patient portal,52 the system reported in this study required intensive case management to review colonoscopy and pathology reports to classify the screening status of patients after colonoscopy.
Tracking the changes in categorization within the registry is essential but challenging. Within the random sample, the authors documented a 12% change in category over 28 months, an ∼5% change per year. This substantial recategorization reflects a rapid decay in the designation obtained during initial registry enrollment. Using the current EHR system, registry maintenance required investments in personnel to manually review charts and reports. Bias in estimating screening colonoscopies exists; there can be underestimation due to lack of capturing colonoscopy or pathology reports from outside institutions or overestimation through erroneous self-reporting of colonoscopy completion.53 Alternatively, thoughtful structural changes in documentation of colonoscopy and attendant pathology reports could improve automation of registries, reduce manual effort, and increase adoption.

Limitations
Limitations are that this study was completed in a single safety-net institution and may not generalize. For example, the authors had relatively limited access to colonoscopy for screening. Therefore, the findings may have less relevance to institutions in which colonoscopy plays a larger role in colorectal cancer screening. Another limitation is that the project was completed in 2014. Despite the program being implemented 10 years prior to manuscript submission, the types of screening options offered have not changed significantly; the authors have not eliminated any general screening modalities (biochemical and/or procedural). Most importantly, the categories that the authors defined for colon cancer screening status remains unchanged and are broadly applicable to screening programs for other cancers. The authors believe that the dynamic nature of cancer screening status will be similar to what the authors developed. It is critical that quality-improvement programs recognize the challenges of establishing an accurate denominator in the context of highly dynamic categorizations.
Another limitation is that the authors did not collect data on the reasons for patient egress and ingress into the registry. This study was conducted during the time frame of roll out of ACA, which likely affected the proportion who ingress and egress patients. Finally, the registry did not capture potentially important covariates (e.g., education level, income, and insurance status) that could have affected the acceptance of colon cancer screening.

CONCLUSIONS
The authors raise 2 major concerns that impacted the building and maintenance of a colon cancer screening registry. One was the movement of patients into and out of the health system and the resulting concern that information about their screening status is lost in transfer of care. To address this, the authors recommend a health information exchange that shares screening records, including colonoscopy findings and pathology reports, across institutions. The second concern is that the screening categorization changed over time, mostly owing to colonoscopy findings and secondarily owing to changes in patients’ health status or their preference to not participate in colon cancer screening. To promote patient-level case management and population-level metrics, the authors recommend adoption of structured colonoscopy reports to allow registries to capture procedure dates and findings critical for determining screening status. By improving data capture, the registry will need less manual maintenance and can be more readily harnessed to guide improvement in screening, case management, and surveillance.