The Colorectal Cancer Screening Process in Community Settings: A Conceptual Model for the Population-Based Research Optimizing Screening through Personalized Regimens Consortium

In the United States, colorectal cancer is the second leading cause of cancer deaths despite availability of screening tests that could greatly reduce both colorectal cancer incidence and mortality (1–3). Guidelines recommend three screening strategies: guaiac-based fecal occult blood testing (gFOBT) or fecal immunochemical testing (FIT) every year, flexible sigmoidoscopy every 5 years with gFOBT/FIT every 3 years, or colonoscopy every 10 years (4–6). Most intervention research has focused on one-time participation in guideline-recommended colorectal cancer screening (7, 8). However, the colorectal cancer screening process is complex and involves delivery of several steps of care beyond one-time screening, including diagnostic evaluation following an abnormal test result, treatment of detected lesions, surveillance after abnormal results or—for those with normal results—repeat screening at guideline-appropriate intervals. Completion of the colorectal cancer screening process requires coordination of care and transfer of responsibilities among healthcare teams (physicians, nurses, and staff) in primary care, gastroenterology, and oncology (9). For screening to achieve greater impact on colorectal cancer incidence and mortality, teams and delivery systems must implement efficient, effective interventions that address the full colorectal cancer screening process (10, 11).

There are gaps in our knowledge about effective delivery of colorectal cancer screening including the best ways to measure breakdowns in the screening process and the impact of implementing evidence-based interventions on screening outcomes in community settings (7, 12). Given the complexity of the colorectal cancer screening process, more research is needed on the multiple factors acting at the patient-, provider-, and system levels that dynamically influence optimal screening test use. This knowledge will inform future interventions and policies to improve colorectal cancer screening in real-world settings.

To address these gaps, the NCI established in 2011 the Population-Based Research Optimizing Screening through Personalized Regimens (PROSPR) consortium. PROSPR's overall aim is to promote multisite, coordinated, transdisciplinary research that evaluates and improves breast, cervical, and colorectal cancer screening in community settings (13). A key PROSPR goal is to create a central data repository on the entire cancer screening process. Recognizing the numerous factors influencing delivery, PROSPR is collecting and organizing data at the multiple levels in which individuals are nested (e.g., healthcare systems, facilities, providers, and patients) for a large, geographically and demographically diverse patient cohort. Common data elements, measured at the patient-, provider-, facility-, and system level, include information on tests and procedures ordered and performed, cancer outcomes, and characteristics empirically associated with the screening process (e.g., patient sociodemographics and healthcare coverage, provider specialty and patient volume, facility and systems policies and incentives to promote screening). Before annual data submission, investigators refine operational definitions and expand the list of required elements to broaden comprehensiveness and robustness of the dataset. Data pooled across participating sites (described below) will facilitate research comparing screening performance across different patient populations, provider types, and delivery systems.

The PROSPR consortium has established seven research centers and one statistical coordinating center. Three research centers are focused on studying and improving the colorectal cancer screening process—Group Health Research Institute (Group Health), Kaiser Foundation Research Institute [Kaiser Permanente Northern California (KPNC) and Kaiser Permanente Southern California (KPSC)], and the Parkland Health & Hospital System/University of Texas-Southwestern Medical Center (Parkland-UTSW).

Here, we (i) describe the conceptual model depicting the colorectal cancer screening process that is guiding data collection; (ii) highlight colorectal cancer–specific research questions that will be addressed using pooled data and some of the methodologic challenges to these analyses; and (iii) provide an overview of the research centers with respect to study populations, settings, and policies, procedures, and programs to facilitate completion of the colorectal cancer screening process.

Each colorectal cancer research center has two principal aims. The first aim is to execute three research projects around a central theme that addresses key factors affecting the colorectal cancer screening process. Group Health, an integrated healthcare system located in Seattle, is assessing the effectiveness of different colorectal cancer screening strategies and identifying low-risk groups for whom a less-intensive screening approach might be appropriate. KPNC/KPSC is examining screening delivery among its enrollees, with the aim of optimizing FIT and colonoscopy screening strategies. Parkland-UTSW is evaluating screening delivery within an integrated safety-net system, including use of personalized approaches and whether these approaches improve outcomes in a low-income, uninsured population.

The second aim is to contribute data to the central repository and collaborate on analyses both specific to colorectal cancer and across cancer types: breast, cervical, and colorectal. Research using pooled data requires establishing a common conceptual and methodologic model (e.g., similar operational definitions), so that data can be validly integrated for comparative studies of the screening process.

Our colorectal cancer screening process model (Fig. 1) expands on the general continuum of cancer care developed by Taplin and colleagues (10, 11, 14–17). The model illustrates challenges specific to delivering colorectal cancer screening, including the multiple test modalities involved, the many activities that require coordination between primary and specialty care teams, and the need to identify patients with symptoms who enter the process at the diagnostic step. This model is intended to represent the screening process for patients at average colorectal cancer risk because it allows for completion of the process with any of the three guideline-recommended screening modalities. Patients at elevated colorectal cancer risk due to personal history (e.g., adenoma) and/or family history (e.g., first-degree relative diagnosed with colorectal cancer) should follow surveillance guidelines recommending colonoscopy at more frequent intervals and/or an earlier starting age (18, 19).

We used Taplin and Zapka's typology for classifying aspects of the colorectal cancer screening process (9). Figure 1 highlights in gray the four types of care that are critical to accomplishing key goals of the colorectal cancer screening process—risk assessment, detection, diagnostic evaluation, and treatment. Steps, represented by boxes, are medical encounters or actions required of patients, providers, or clinical staff. Steps with dashed borders depict interfaces where information and/or responsibility must be transferred among different healthcare teams (e.g., primary care and specialist) to progress to the next step. Transitions are sets of multiple steps and interfaces necessary to move from one type to the next (e.g., detection to diagnosis).

Our model depicts all steps that may occur during one screening episode. The process can begin in three ways: (i) a provider recommends screening to a patient during a clinical encounter; (ii) the healthcare organization systematically invites eligible patients to screen; or (iii) a patient requests screening (Fig. 1, Transition #1). Which screening test modality is used depends on patient and provider preferences, resources of the patient, and policies of the organization. If results are normal, the process for this episode ends and guideline-based recommendations for repeat screening are communicated to the patient and relevant providers. Patients are referred for diagnostic colonoscopy if the screening process began with other tests (e.g., gFOBT, FIT, sigmoidoscopy, or CT colonography) with abnormal findings. Symptomatic patients (e.g., iron deficiency anemia, gastrointestinal bleeding) enter the process at referral for diagnostic colonoscopy (Fig. 1, Transition #2). Because colonoscopy can be performed for screening or diagnostic purposes, defining and documenting the exam's indication is particularly important, yet challenging (20–24). Also, critically important is reporting of colonoscopy results, to help providers and organizations identify patients at elevated risk who may require a more frequent screening or surveillance regimen. There are several interfaces within each type of care and during care transitions that require coordination between primary care and specialty providers; these interfaces may be particularly vulnerable to breakdown (10, 15). Unless cancer is detected or surveillance is recommended at the end of the screening episode, the process begins again and the patient re-enters the model based on guidelines and patient and provider preferences. PROSPR is uniquely positioned to address questions raised by our conceptual model.

Pooled PROSPR data offer unique opportunities to assess the colorectal cancer screening process in diverse, community settings. PROSPR colorectal cancer investigators intend to examine a wide range of clinical and policy-relevant research questions addressing:

• What are the best strategies to promote appropriate use of colorectal cancer screening and follow-up, in terms of patient preference, patient adherence, local healthcare systems, and specific populations (25)? For example, Holden and colleagues (7) found strong evidence that elimination of structural barriers, one-on-one interactions, patient reminders, and system-level interventions improved screening rates; however, what is the impact of deploying these evidence-based strategies in real-world settings? What contextual factors hamper their effectiveness (e.g., characteristics of the patient population, providers, and/or facilities)?

• What steps in the screening process (Fig. 1) are vulnerable to failure (26, 27)? What types of patients are at particular risk for failures (8, 28) and what are the factors at the provider-, facility-, and system levels that contribute to failures (11, 15, 29–33)?

• What are the performance characteristics (sensitivity, specificity, positive predictive value) of colorectal cancer screening tests conducted in the diverse health systems represented in PROSPR (34, 35)?

• To what extent do underuse, misuse, and overuse of colorectal cancer screening occur (7)? For example, how often are FITs given to patients who had a recent colonoscopy (less than 10 years) with normal results? How often do physicians recommend an FIT to patients who are in colonoscopy surveillance program due to prior adenomas? What strategies ameliorate misuse and overuse patterns (7)?

• How does colorectal cancer risk change following screening (36, 37)? Can a risk-based management strategy be used for screening and surveillance (38, 39)?

• What is the comparative effectiveness (i.e., balance of benefits and harms) of FIT versus colonoscopy screening regimens in the diverse health systems and patient populations represented in PROSPR, and the generalizability of these organized screening approaches to resource-limited settings (40–42)?

Findings will inform policy and clinical recommendations issued by experts, as well as design and implementation of population-based screening programs such as those supported by the Centers for Disease Control and Prevention (43).

To generate pooled data, colorectal cancer PROSPR investigators are establishing operational definitions for a wide array of common data elements characterizing the colorectal cancer screening process (e.g., benefits and harms of screening, colonoscopy indication) and are harnessing relevant clinical information systems, including the electronic medical record (EMR) and local clinical and administrative databases. Research centers submit data on an annual basis, and each year the required list of common data elements is reviewed, refined, and expanded to ensure the comprehensiveness and robustness of the pooled dataset. Research centers are linking screening process data with cancer outcomes available from regional and local cancer registries and considering linkages with pertinent contextual data to derive measures of community-level health status, resources, and utilization (e.g., U.S. Census, Health Resources and Services Administration's Area Resource File; refs. 44–46).

Although pooled PROSPR data provide a unique opportunity to address important questions about colorectal cancer screening, they also pose several challenges for analysis. Accounting for potential confounders and heterogeneity across systems are two key challenges. Screening rates and performance characteristics are likely to vary across research centers due to differences in preferred test modality, approaches to facilitate screening, and characteristics of the screen-eligible patient populations. To minimize bias, investigators are using methods to account for confounding when combining information across the research centers. For example, instrumental variable analysis (47, 48) or propensity score adjustment (49, 50) are approaches used to minimize selection bias (51, 52). Also, system-level variability can be addressed using multilevel modeling (53). Another challenge is accurate measurement of patient-level risk factors for prediction of short- and long-term outcomes, because risk factor information may change at any step in the screening process. Accuracy of information affect whether risk prediction algorithms can be used to develop personalized screening regimens and improve population-level screening outcomes. We are also assessing patient comorbidity to examine competing risks. Although conducting comparative effectiveness analyses of different colorectal cancer screening strategies with PROSPR's observational data is complex, validity can be greatly improved by carefully formulating research questions, implementing appropriate study designs, and applying sophisticated statistical methods.

The three colorectal cancer research centers are uniquely positioned to answer the key research questions described above. The research centers are comprised of four U.S. integrated healthcare systems covering four geographic regions. Integrated systems have been recognized as ideal settings for assessing the entire screening process in contrast to loosely affiliated or unaffiliated primary care practices that refer to specialty providers. This is because integrated systems can systematically track data at multiple levels through a comprehensive EMR(54, 55) and care is delivered through a single, coordinated organization of primary and specialty practices (56–59).

Group Health is a mixed-model, nonprofit, health plan serving nearly 625,000 members in Washington State. The Kaiser Permanente Research Center is comprised of two healthcare systems delivering care to about 7 million members in Northern (KPNC) and Southern (KPSC) California. Parkland-UTSW is the sole safety-net provider for under-insured and uninsured Dallas County residents living at ≤200% of federal poverty level; approximately 64,000 adults each year use Parkland's 11 primary care clinics. These four systems have a combined population of over 7.7 million people from which a pooled, dynamic colorectal cancer cohort has been derived.

We have developed cohort definitions and eligibility criteria to facilitate pooling of the screening process data (Table 1). The Group Health and Kaiser Permanente cohorts consist of health plan members ages 50 to 89 years on or after January 1, 2010 with new cohort members added as they become eligible for colorectal cancer screening (i.e., enters cohort on 50th birthday). For the Kaiser Permanente cohort, one year of prior membership is also required for cohort inclusion. Membership eligibility criteria were used to provide new enrolled Kaiser Permanente patients with a sufficient period of time for providers to document colorectal cancer screening history, determine screening eligibility, and offer screening services. For Parkland-UTSW, Dallas County residents ages 50 to 64 years with a visit at one of Parkland's 11 primary care clinics on or after January 1, 2010, are included into the cohort on the date of their visit. These latter eligibility criteria were applied for two reasons: (i) access to colorectal cancer screening services at Parkland is limited to patients seen in primary care (as opposed to those seen only in the emergency department or at a specialty clinic); and (ii) Parkland is obligated to care for all uninsured Dallas County residents but residents who become Medicare-eligible at age 65 may opt to receive care outside the Parkland system. All research centers excluded patients diagnosed with colorectal cancer or with a partial or total colectomy before cohort entry date.

Table 2 describes sociodemographic characteristics and screening process outcomes of cohort members who entered in 2010 for each of the four systems. Age distributions of screen-eligible patients at Group Health, KPNC, and KPSC are very similar, with approximately 75% of patients in the fifth and sixth decade of life and 25% in the seventh and eighth decades. Because of the restricted age range of the Parkland-UTSW cohort, almost 72% of the patients are in their 50s'. More than half of the patients are female, with the highest proportion of females at Parkland-UTSW (64%). There is substantial racial and ethnic variation across the systems. Group Health members are predominantly White (72.1%) and, although the proportions are lower at KPNC (58.4%) and KPSC (45.7%), non-Hispanic Whites still comprise the majority. Nearly a quarter of the KPSC cohort is Hispanic, compared with 11.6% at KPNC and 3.3% at GH. Only 18.4% of Parkland-UTSW patients are White; most are either non-Hispanic Black (38.9%) or Hispanic (36.1%). Of the four systems, KPNC has the highest proportion of Asians and Pacific Islanders (14.0%).

Almost all patients across the four systems live in metropolitan areas (60, 61). Group Health, KPNC, and KPSC are very similar in insurance coverage distribution—most patients are privately insured (54.9%–66.6%) or have Medicare (32.4%–44.3%), and very small proportion have Medicaid or are dual-eligible for Medicare and Medicaid. In contrast, 68.4% of Parkland-UTSW patients are uninsured; their medical care costs are covered by Dallas County's medical assistance program financed by property taxes. About 9% of Parkland-UTSW patients have Medicare coverage; 7% are covered by Medicaid. Overall, significant heterogeneity of the pooled cohort with respect to race/ethnicity, age, and insurance coverage is evident.

FIT comprised the vast majority of stool-based tests performed in 2010 at KPNC, KPSC, and Parkland-UTSW. In contrast, Group Health used gFOBT during 2010 and transitioned to FIT in late 2011. More than 30,000 abnormal stool tests required diagnostic evaluation within the four systems during 2010. Across the four systems, colonoscopy was the most common endoscopy procedure. We are currently cross-validating methods to define indication as none of the systems' EMR has a discrete field for identifying reason for colonoscopy; this effort will enable definition of a screening episode and identification of misuse and overuse patterns. In 2010, more than 2,300 colorectal cancer cases were identified by the systems.

Table 3 describes the four systems' healthcare delivery resources. There is substantial range in number of primary care physicians who care for adult patients (total > 3,700) and number of endoscopy providers (total > 250). Group Health and Parkland-UTSW use mid-level providers such as nurse practitioners and physician assistants in primary care. KPSC and Parkland-UTSW are teaching institutions that employ gastroenterology fellows; therefore, attending physicians and fellows perform endoscopy procedures. To meet increasing capacity needs over a wide geographic area, Group Health contracts with outside endoscopy providers to perform some colonoscopies; the contracted providers use various clinical information systems for storing data and are not all universally linked to Group Health members' EMR.

Endoscopy procedures are performed in three distinct facility types—ambulatory surgical centers, medical offices, and hospitals. These types are subject to different regulatory requirements (62, 63). These facilities may be owned and managed by the healthcare system or may have a formal contract with the system to provide services. The nature of the relationship may affect interfaces (i.e., transfer of clinical information and responsibility) between primary care and endoscopy providers and increase difficulty of extracting endoscopy results because scanned reports are stored as images and require more resources to extract than text information stored in the EMR. We are also cross-validating natural language processing and algorithm-based methods to systematically extract text data from endoscopy reports given the wide variation in how results are documented.

All four healthcare systems use a common and comprehensive EPIC platform-based EMR. The EPIC EMR can provide information on electronic orders, referrals, scheduling, test results, treatment, and billing for both inpatient and outpatient care. KPNC, KPSC, and Parkland-UTSW have a Cerner-based clinical information system for reporting pathology results; Group Health uses Progeny. Group Health, KPNC, and KPSC carry out centralized processing of stool-based tests. In contrast, each of the 11 Parkland-UTSW primary care clinics process FITs at local on-site laboratories. Pathology processing for endoscopy specimens is centralized for Parkland-UTSW and Group Health–performed procedures; for KPNC and KPSC and some contracted providers for Group Health, pathology processing is local to each endoscopy facility. The heterogeneous nature of the three colorectal cancer research centers not only contributes to the robust nature of the data collected in the PROSPR consortium but will also make findings from PROSPR studies more generalizable across populations and delivery settings.

All colorectal cancer research centers have clinic- and system-level policies procedures, and programs targeted toward asymptomatic, average-risk individuals to facilitate completion of screening process steps, transitions between types of care, and interfaces among providers (Table 4; refs. 9, 15, 16).

To initiate the screening process, Group Health, KPNC, and KPSC have organized programs that systematically invite all average-risk members ages 50 to 75 years to participate in colorectal cancer screening. Electronic clinical information systems identify and generate lists of eligible patients. Invitation letters are mailed and timing of delivery may be based on birth month or anniversary of prior screening. Stool blood testing is the preferred screening modality, but endoscopy is also available as an option based on patient/provider preference. At KPNC and KPSC, the FIT kit is mailed with the invitation letter; if endoscopy is chosen, a primary care provider must place an order or request a consult. At Group Health, patients can self-refer for colonoscopy. Response to invitations is monitored, and either a reminder is sent to nonadherent individuals or a patient's primary care team is notified. Timing and delivery mode of reminders varies across the systems. We are currently working to obtain data from these clinical information systems and expand our common data elements to identify moderators of program effectiveness.

At Parkland-UTSW, screening is available to all primary care patients 50 years of age or older opportunistically during in-person clinic visits. The screening modality offered is based on provider and patient preferences.

Across all four systems, screening results are reported to primary care providers via EMR in-basket. The systems have at least three methods for notifying patients about results (e.g., information is posted in a secure website, letter is mailed, or a provider calls), and delivery mode depends on the test modality and findings. Group Health, KPNC, and KPSC also have system- and practice-level strategies to encourage patients with abnormal test results to complete diagnostic evaluation. Finally, all four systems generate population-level reports based on data stored in their clinical information systems to monitor adherence to colorectal cancer screening guidelines.

Collaboration among the PROSPR colorectal cancer research centers will advance knowledge about the screening process by providing detailed, longitudinal, multilevel, population-based data about steps, transitions, and interfaces along the colorectal cancer screening process. Background information about the cohort, integrated healthcare systems, and colorectal cancer–specific policies and procedures illustrates that PROSPR is poised to comprehensively compare and systematically evaluate the colorectal cancer screening process in diverse metropolitan community settings.

There are a few limitations to acknowledge. Our main data source for the pooled dataset is the EMR. Because each system's EMR differs in its documentation of colorectal cancer screening process, we have simplified some of our common data elements. We refine operational definitions each year and expand the list of required elements to broaden the comprehensiveness and robustness of the pooled dataset. Also, the EMR does not systematically record patients' and providers' beliefs, intentions, or motivation for colorectal cancer screening; thus, we will not have information about those important variables. There are two limitations that affect generalizability of PROSPR colorectal cancer pooled dataset. We will have little data from patients living in rural areas. Data are from integrated systems, and their ability to coordinate services along the screening process may differ substantially from other practice models. Therefore, we encourage researchers with access to rural populations and different practice settings to use our conceptual model to collect similar data addressing research questions detailed above. PROSPR colorectal cancer investigators are interested in collaborating to broaden our collective understanding of colorectal cancer screening delivery. In accordance with the NIH's data sharing policy, we are currently creating a web portal to receive proposals from researchers interested in collaborating with PROSPR investigators and use the pooled data.

Analysis of patient-, provider-, practice-, and system-level factors that influence screening process outcomes is critical to inform the design and evaluation of multilevel interventions and policies to optimize delivery of colorectal cancer screening in healthcare systems serving diverse populations. The ultimate goal of the PROSPR consortium is to optimize screening delivery so that it is consistent with the Institute of Medicine's goals of effective, efficient, coordinated, timely, and safe health care (64).

S. Gupta has received commercial research support from Polymedco. No potential conflicts of interest were disclosed by the other authors.

Conception and design: J.A. Tiro, A. Kamineni, T.R. Levin, Y. Zheng, J.S. Schottinger, C.M. Rutter, C.S. Skinner, C.A. Doubeni, C. Klabunde

Development of methodology: J.A. Tiro, T.R. Levin, C.S. Skinner, C.A. Doubeni, E.A. Halm, C. Klabunde

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.A. Tiro, T.R. Levin, J.S. Schottinger, D.A. Corley, C.A. Doubeni

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.A. Tiro, A. Kamineni, T.R. Levin, D.A. Corley, J. Chubak, E.A. Halm, C. Klabunde

Writing, review, and/or revision of the manuscript: J.A. Tiro, A. Kamineni, T.R. Levin, Y. Zheng, J.S. Schottinger, C.M. Rutter, D.A. Corley, C.S. Skinner, J. Chubak, C.A. Doubeni, E.A. Halm, S. Gupta, K.J. Wernli, C. Klabunde

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): D.A. Corley, C.A. Doubeni

Study supervision: J.A. Tiro, C.S. Skinner, C. Klabunde

The authors thank Dr. Ann Geiger, Dr. Virginia Quinn, Dr. Nirupa Ghai, Dr. Christopher Jensen, Dr. Noel Santini, Katharine McCallister, Winifred Apraku, Adam Loewen, Arvind Ramaprasan, and Susan Carol Bradford for their assistance.

This research was funded by the NCI. J.A. Tiro, C.S. Skinner, E.A. Halm, and S. Gupta were supported by grant 5U54CA163308. A. Kamineni, C.M. Rutter, J. Chubak, and K.J. Wernli were supported by 5U54CA163261. T.R. Levin, J.S. Schottinger, D.A. Corley, and C.A. Doubeni were supported by 5U54CA163262. Y. Zheng was supported by 5U01CA163304.