Implementing universal Lynch syndrome screening (IMPULSS): protocol for a multi-site study to identify strategies to implement, adapt, and sustain genomic medicine programs in different organizational contexts

The goal of precision medicine is to improve health outcomes by tailoring healthcare based on an individual’s genomic and other relevant information, including patient preferences [1]. One example is systematic screening of colorectal cancer (CRC) tumors to identify all patients whose CRC may be related to Lynch syndrome (LS) [2, 3]. LS is the most common form of inherited CRC and is also associated with significant risk for endometrial, ovarian, gastric, small bowel, and renal cancers, among others [4, 5]. LS most commonly results from the inactivation of the DNA mismatch repair (MMR) system. A first step in screening for LS in patients with cancer involves testing tumor tissue for signs of MMR deficiency in one of four genes (MLH1, MLH2, MLH6, and PMS2) [2]. Following a positive tumor screening test, additional testing for the presence of a germline pathogenic variant in one of the MMR genes is needed to confirm diagnosis of LS. A diagnosis of LS influences cancer screening and surveillance guidelines to be followed and treatment options for those already diagnosed with cancer. Patients with CRC who have LS benefit from treatment with immunotherapy [6] and have the option for more extensive colonic surgery to decrease the risk of metachronous malignancy, which is more common in LS [4, 5]. Additionally, prophylactic hysterectomy, and salpingoophorectomy can reduce risk of endometrial and ovarian cancer (90–100%) in women with LS [7].

Approximately one million people in the US have LS, of which only about 2% are aware [8, 9]. Therefore, most are not receiving potentially life-saving surveillance and treatment. In addition, at-risk family members, who are at 50% risk to also have LS, are not being identified. Cost-effective, evidence-based systematic screening strategies to identify cancer patients with LS are available and have broad recommendation for implementation [2, 10‐16], yet LS screening is inconsistently applied within healthcare systems, if at all [17‐19]. Systematic (also called, “universal”) screening of all newly diagnosed cases of CRC for LS is cost-effective by most measures [20, 21] and is classified by the Centers for Disease Control (CDC) as having top-tier evidence for reducing cancer morbidity and mortality and improving quality of life [11, 22]. LS screening was first recommended by the Evaluation of Genetic Application in Practice and Prevention (EGAPP) working group in 2009, and is currently recommended by multiple professional organizations, including the National Comprehensive Cancer Network (NCCN), and the Healthy People 2020 initiative [3, 10, 12‐16, 23]. LS screening was also recently recommended by the Blue Ribbon Panel Precision Prevention and Early Detection Working Group to meet the goals of the Cancer Moonshot [9].

Contextual factors such as organization mission, patient population, and costs may all influence decisions to implement genomic technologies in healthcare systems [17, 24‐27]. However, there is poor understanding of how these and other contextual factors impede or facilitate LS screening implementation in healthcare systems and under what circumstances [17, 18]. Factors specific to LS screening implementation may include: variable involvement of multiple key stakeholders and champions, availability of genetic counseling, genetic testing costs, and perception of value to the healthcare system [18, 19, 28]. Previous research has shown that institutional testing costs, number of CRC patients diagnosed per year, local prevalence of LS, and perceived cost of screening older cancer patients are key determinants for organizational decision makers and likely key barriers to LS screening implementation [28‐30]. Additionally, there are multiple evidence-based protocols that provide different recommendations for LS screening [4, 10, 31]. These conflicting recommendations are not surprising in the face of evolving evidence, but they are likely contributors to variability in implementation. For instance, the NCCN guidelines suggest that if all CRCs cannot be tested, then testing should occur on tumors of patients diagnosed under age 70 and for patients with CRC over age 70 who meet complex criteria using family history or tumor characteristics associated with increased LS risks [4, 15]. This conflicts with evidence demonstrating that imposing age restrictions can result in substantive loss of LS index case finding, is challenging to implement, and assessment of family history and other risk factors often fails to identify individuals who are appropriate candidates for LS testing [8, 32].

However, LS screening is only successful if patients who screen positive (by tumor tissue testing) are informed and follow-through with confirmatory germline testing so that patients and their family members can access appropriate cancer treatment, screening, and prevention options. A functional tracking system and/or other processes to ensure program success is therefore an additional component of critical importance to LS screening. A study of several medical centers found that positive results were often not disclosed and poor rates of patient follow-through to germline confirmation occurred until such procedures were included in LS screening programs [19]. Finally, healthcare systems vary in implementation across a spectrum from no organized LS screening to optimal LS screening. The latter is defined as including implementation or maintenance of cost-effective tumor screening protocols, procedures to ensure disclosure of positive screening, and active processes to facilitate patient follow-through to germline genetic testing for diagnosis of LS [33‐35]. An additional file shows a typical optimal LS screening program design [see Additional file 1].

Due to the complexity of LS screening and the variability of healthcare systems in which screening is to be implemented, a systematic approach based on the principles of implementation science is needed to ensure the success and sustainability of LS screening programs. The Consolidated Framework for Implementation Research (CFIR) has been used successfully to evaluate variation in program implementation [36, 37] and to identify conditions necessary for patient follow-through to confirmatory gene sequencing in LS screening programs [19]. The CFIR describes several processes that may influence implementation success and guides assessment of contextual conditions that may serve as implementation barriers or facilitators at the individual, organizational, and external levels. The CFIR also guides data gathering and structuring for conducting configurational comparative analysis to identify which CFIR constructs and LS screening program-specific procedures are minimally sufficient and necessary for optimal implementation, and under which circumstances in complex health care delivery systems [19, 38, 39].

The goal of this project is to utilize the CFIR and other tools from implementation science to describe, compare, and explain variations in LS screening across multiple healthcare systems and create a comprehensive, customizable organizational toolkit for implementing and improving LS screening programs. To achieve this goal, this study has four specific aims:

Aim 3: Determine the relative effectiveness and costs of different LS screening protocols using decision analytic models developed from previous work [29, 30] and data specific to each healthcare system to demonstrate the relative effectiveness and efficiency of various LS screening protocols used by healthcare systems based on their local data.

This study combines multiple methods of exploring implementation in complex healthcare systems to determine which implementation strategies are likely to work best in different organizational contexts. We will use traditional case-based, in-depth analyses of conditions that may serve as barriers and facilitators of LS screening (Aim 1), followed by cross-case and configurational comparative analysis to determine minimally sufficient and necessary conditions for optimal LS screening (Aim 2). These data will be used to populate economic evaluation decision analytic models to demonstrate how organizational context, available resources, and screening protocol impact organizational costs and program success (Aim 3). A toolkit will be created and disseminated to sites to guide implementation of new programs that are aligned with the organizational context and costs or to optimize existing programs (Aim 4).

The unit of analysis is the clinical site through which LS screening is or can be implemented. We will study the contextual factors of 21 clinical sites across 7 healthcare systems purposively selected to maximize the number of clinical sites in various stages of implementing LS screening, as well as to maximize diversity of location, system structures, and patient populations (Table 1). Six healthcare systems (Kaiser Permanente Northwest and Colorado, HealthPartners, Harvard Pilgrim Healthcare, Meyers Primary Care Institute/Reliant Medical Group, Sutter Health-Palo Alto Medical Research Foundation, and Geisinger) are part of the Healthcare Systems Research Network (HCSRN), and one is a national non-profit faith-based system with locations in 18 states (Catholic Health Initiatives). Of the six HCSRN sites, Kaiser Permanente sites are integrated healthcare systems where patients are members, while the other sites are “open” systems where patients can be members, patients, or patient members (e.g. members receiving care in the system or outside the system, or patients receiving care who are not members of the health plan).

A comprehensive case description of each organization, including practice variation and factors influencing implementation, evaluation, maintenance, and improvement of LS screening, will be created through qualitative interviews of organizational stakeholders and patients with newly diagnosed CRC. For sites with LS screening programs, additional interviews will be conducted with patients diagnosed with LS. Interviews of organizational stakeholders will be conducted centrally by trained staff at Geisinger. Interviews of patients will be conducted centrally by trained staff at Kaiser Permanente Northwest (KPNW). Consent will be obtained at the start of the interview process and patient interviewees will receive a $25 gift card upon completion of the interview.

We will use CFIR and the case summaries created for each site to guide cross-case comparisons with the purpose of describing contextual variation associated with where, when, and under which conditions different processes for implementing or improving LS screening might be successful [47]. The result will be a descriptive summary of patterns in variations across sites, which will provide the basis for selecting conditions for inclusion in the analyses for Aim 2.

Configurational comparative methods (CCM), such as Qualitative Comparative Analysis (QCA) and coincidence analysis (CNA) are particularly suitable analytic techniques for studying causal complexity in organizational implementation (e.g. multiple conditions may combine in various ways to cause the same outcome) [19, 38, 48‐50]. CNA was selected for use in our study because, unlike QCA, it uses a different minimization algorithm that can identify underlying causal chains [51]. This is useful because we anticipate the presence or absence of some contextual conditions may impact one or more combinations of other contextual or implementation conditions that are minimally sufficient and necessary for the outcome under investigation. Data analyses for Aim 2 will consist of 4 main steps [38]: 1) code data for context and implementation conditions (CFIR constructs), 2) code implementation outcomes, 3) calibrate conditions and outcomes to create data matrix and 4) conduct CNA and interpret solutions to create a model of minimally sufficient and necessary conditions for defined outcomes (Fig. 1).

In Aim 3 we will calculate expected outcomes central to evaluation of effectiveness and efficiency of LS screening in CRC and endometrial cancer (EC) patient populations. Previously developed models [29, 30] will be modified and updated based on recent developments, current evidence and guidelines. Models will be populated using local data to reflect site-specific conditions as determined from data collected from study site stakeholder interviews in Aim 1.

We will construct decision tree-type analytic models in Excel© software (Microsoft, Inc., Redmond, Wash.), with the @Risk© software add-on for Excel (Palisade, Inc., Ithaca, NY) to enable various complex analytic procedures. The decision analytic team will construct multiple decision tree models to represent all LS screening protocols the team considers currently viable. Model variables include estimates of annual incident CRC and EC cases by age strata, LS prevalence or the assumption of an equivalent rate for all populations, and cost of tests included in screening protocols from each site to populate decision analytic models. When institutional cost is not available, alternative methods such as the average test cost based on costs reported by other participating clinical sites, regional test cost figures if publicly available from testing companies, or Medicare reimbursement rates will be used. Finally, reliable estimates of site-specific LS prevalence may not be available; therefore, this model parameter may be estimated from the most current estimates for U.S. populations [56]. Because clinical sites, and scientific evidence are dynamic, we do not expect organizational resources, testing costs, or LS screening guidelines to stay static. Therefore, the decision analytic models developed will be designed in such a way to facilitate ongoing use and modification by organizations.

For base-case and sensitivity analyses, the decision analytic models will calculate, for each site and in CRC and EC virtual cohorts (e.g. 500 cases per year): (1) sensitivity and specificity of each LS screening protocol in identifying LS cases including number of LS cases expected to be identified, (2) total costs for each screening protocol, (3) cost per CRC and EC case-screened, (4) cost-per LS index case identified, (5) incremental differences between guideline-based screening protocols including costs and LS cases identified, (6) and patient adherence to each screening protocol including number of CRC and EC cases lost to follow up. Additional modeling of different LS screening age cut-off policies using local-level data will be conducted for each site. Models will provide objective metrics, driven by local data, of the impact of applying age-cutoffs in LS screening implementation.

Data from Aims 1, 2, and 3 will be used to generate a working organizational toolkit to guide implementation, maintenance, and optimization of LS screening. Additional information will be collected over the entire project period from monthly project meetings, communications from site investigators, pertinent data regarding site-specific screening changes, and external evidence or guideline changes for LS screening. These data will be recorded in a project specific database created for tracking such information related to implementation [57]. This tracking database will provide important information for the toolkit development should any sites begin to implement LS screening based on being interviewed for Aim 1, but prior to receiving the toolkit.

The organizational toolkit will be created based on the CFIR conceptual framework, the in-depth knowledge of LS screening programs and contextual factors of healthcare systems from Aim 1, the cross-site comparison and CNA results from Aim 2, and economic evaluation by decision analytic modeling with local data from Aim 3. This toolkit will be disseminated to all sites through site investigators and the tracking database will record to whom it is distributed, questions asked by those receiving the toolkit, and any immediate actions taken by the site.

Six months after distribution of the toolkit, additional qualitative interviews will be conducted with up to 5 organizational stakeholders at each site using the same processes described for Aim 1. Stakeholders from sites without LS screening and those with sub-optimal implementation will be interviewed about the utility of the tool to facilitate implementation and improvement. Stakeholders from sites with optimally implemented programs will be interviewed about the utility of the tool for improvement or adaptation to emerging evidence. The interview guide for this aim will be developed by the research team in parallel with the development of the toolkit.

Interviews will be conducted, transcribed, and coded as described for Aim 1. Interviews will be coded for information on to whom the organizational toolkit was distributed at each site, questions that were asked by key stakeholders, and whether and how the toolkit was used by organizational decision makers to facilitate LS screening implementation and/or improvement.

The final organizational toolkit for LS screening implementation, maintenance, and improvement will be modified based on the information learned throughout the study and from the pilot distribution to all study sites. The final product will include descriptions of most commonly included stakeholders and general processes needed for optimal LS screening, directions for processes and protocols that are more likely to work by contextual factors identified, and a generic decision analytic model for costs and effectiveness related to available LS screening protocols. This organizational toolkit will have an intuitive interface that allows for the input of local parameters for use by organizational decision-makers. This toolkit will be distributed through the Lynch Syndrome Screening Network (LSSN; www.​lynchscreening.​net) and through other channels that may be identified.

Through in-depth assessment of contextual conditions impacting LS screening implementation across an unprecedented number of sites representing diverse healthcare systems, geographies, and patient populations served, this study will provide significant information for the Precision Prevention and Early Detection Working Group of the Blue Ribbon Panel to successfully address the Cancer Moonshot [9], and for the working group of National Academies of Sciences, Engineering, and Medicine (NASEM) to successfully facilitate their goal of broadly implementing LS screening [58]. This study will also contribute to implementation science more generally, as it combines multiple methods to extend beyond the typical “lessons learned” approach and utilizes a relatively new analytic approach to illuminate minimally necessary and sufficient conditions for LS screening implementation in different organizational contexts. Another innovation of this study is our ability to provider tailored information to each site by combining key stakeholder information with business case decision analytic models that can be populated with local, real-world data. The relevance of general societal cost to organizational decision-making has been questioned [59, 60] and prior studies indicate that organization-specific costs to screen and to detect LS cases for different protocols is critical information for health systems to make decisions about LS screening implementation [28‐30].

To our knowledge, no studies have synthesized in-depth cross-site comparison of context, barriers and facilitators with local business case analyses into a comprehensive toolkit for organizations implementing and optimizing LS screening programs. Most studies to date have focused on strategies for initial implementation, rather than maintenance and improvement in the face of organizational context or changes in evidence. This study will fill that gap and provide insights into successful implementation in the era of precision medicine where evidence is evolving and costs and opportunities for genomic testing are rapidly changing.

Healthcare systems and evidence are dynamic; therefore, it is possible that any of the participating sites may implement LS screening during the course of the study, or that sites currently implementing LS screening may optimize their programs prior to receiving the toolkit. This possibility may decrease the diversity of our sites in contextual or implementation conditions or outcomes; thus, changes will be tracked to ensure the integrity of the analyses. Additionally, new evidence [61] and declining costs of sequencing may lead to the replacement of currently recommended LS screening protocols with tumor sequencing or even germline sequencing of all newly diagnosed cancer patients. Because economic evaluation is an integral part of this study and direct sequencing cost comparison will be included in the organizational decision-making guide, this changing evidence can easily be re-evaluated by each organization’s decision-makers over time. Because this study is specifically designed to collect cost and contextual information throughout the course of the study, we will utilize changes made at any site over the course of the study as data to inform implementation, maintenance, and adaptation broadly. Therefore, this study will not only result in a toolkit to guide LS screening implementation in an era of dynamic and changing evidence and costs, but may also broadly demonstrate for the field of implementation science methods for conducting research in the real-world that are focused on optimization, maintenance, and adaptation to new evidence.

This study addresses a major unmet need identified by the Blue Ribbon Panel to achieve the goals of the Cancer Moonshot and to improve our understanding of clinical implementation of complex interventions [9]. The overarching goal of this study is to determine factors associated with variations in LS screening implementation across multiple healthcare systems and to provide the organizational decision tools, or “recipes,” for implementation success in light of different organizational contexts and costs. The organizational toolkit that we produce will help health systems to maintain and optimize LS screening programs in the face of changing costs and evidence. Our toolkit based on implementation science may be adapted to help in implementing evidence-based screening for other genomic conditions or for implementing other types of complex genomics interventions into healthcare systems.