A Systematic Review of Implementation Science Frameworks in Cancer Prevention Interventions: An Exploratory Health Disparities Analysis

Background: A growing number of studies have used implementation science (IS) frameworks, such as RE-AIM, to inform and evaluate the implementation of evidence-based cancer prevention services (e.g., cancer screening); however, the impact of such applications is not well understood, including whether the use of an IS framework can lead to reductions in health disparities. The purpose of this systematic review is to explore how IS frameworks can guide adaptations to cancer prevention services to specically address health disparities. Methods: This systematic review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines and registered with PROSPERO. Searches were conducted in Ovid MEDLINE, PubMed, PsycINFO, CINAHL, and EMBASE. Search strategies used a combination of terms related to implementation science frameworks, cancer prevention, and/or intervention. All searches were conducted between January to May 2020. The QATSDD tool was used to assess the quality of studies included in the review. Results: A total of 1,025 titles and abstracts were screened, and 84 were deemed eligible for full-text screening. After full-text screening, n=27 articles were included for data abstraction and synthesis. Of the 27 studies that were included, an overwhelming majority (n=19, 70.3%) were based in the United States, utilized mixed-methods (n=12, 44.4%), focused only on a single cancer site (n=21, 81.5%), and took place in a health system (n=18, 66.7%). Approximately half of the studies (n=13, 48.2%) used an IS framework for post-implementation evaluation. Most notably, only one-third of studies (n=9, 33.3%) used an IS framework to address cancer-related health disparities. Of those nine studies, six (66.7%) of them used the Consolidated Framework for Implementation Research (CFIR). Other IS frameworks that were used to inform a health disparities adaptation were Diffusion of Innovations, Knowledge-To-Action, and RE-AIM. Most studies were at moderate risk of bias (n=19, 70.4%). Conclusion: Across the various cancer prevention studies that have been implemented, CFIR has been the all health systems. In the year before STOP HCC implementation, data from four health systems showed that 110 of 13,216 baby boomers (0.8%) had been screened for HCV (pre-intervention). After 29 to 43 months of STOP HCC (depending on site), 13,334 of 27,700 eligible baby boomers (48.1%) were screened, varying by health systems from 19.8–71.3%. By comparison, only 8.3% of 60,722 patients in a national study of community health centers were screened for HCV from 2010 to 2013, and 17.3% of baby boomers were screened nationally in 2017 according to the National Health Interview Survey.

]. Yet, little research exists on whether the speci c tailoring or adaptations identi ed and implemented through IS frameworks in cancer prevention EBIs can lead to any signi cant reduction in cancer-related disparities. Discussions on how IS frameworks can address health inequities have only recently emerged within both national and global health contexts [20]. Central to these conversations is how to appropriately identify and adapt evidence-based practices to the context, culture, and acceptance levels of key stakeholders (implementing organizations, intended program recipients) [14,[21][22][23]. The primary aim of this review is to characterize the state of how IS frameworks have been used across the implementation continuum within cancer prevention studies globally. The exploratory aim is to examine how a subset of these studies may have used IS frameworks to address or adapt intervention designs to t within the context of a population(s) experiencing cancer-related health disparities.

Methods
This systematic review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [24] (see Supplementary Material Tables 1 and 2 for completed checklists) and registered with PROSPERO (CRD42020171970) [25]. Studies were identi ed through searches from ve bibliographic databases: (1) Ovid MEDLINE; (2) PubMed; (3) PsycINFO; (4) CINAHL; and (5) EMBASE. Table 1 Characteristics of studies included in this review (n = 27) a

Screening and Study Selection
Inclusion criteria for studies included: any type of study design (including randomized controlled trials (RCTs), non-RCTs, intervention and non-intervention studies, qualitative, quantitative, and mixed-methods); published in peer-reviewed journals with full-text available; English-only publications; published between 2001-2020; and explicitly used an IS framework. Geographic limitations were not applied in order to capture implementation science applications across global health initiatives (i.e., as some IS frameworks have been developed in non-US contexts). Studies were excluded if they were: pilot/formative studies leading to the development of an IS framework; case studies; systematic reviews or meta-analyses of cancer prevention interventions; did not address cancer prevention and/or detection; included the prevention or relapse of subsequent cancer tumors (i.e., cancer survivorship); did not use an IS framework; and were not empirical studies (involving primary data collection) or research protocols.

Data Extraction and Analysis
Two reviewers (SX, HIA) performed title and abstract screening for inclusion and exclusion criteria. All con icts were resolved by a third reviewer (RP). Kappa statistics were calculated to assess inter-rater reliability agreement for both inclusion and exclusion processes. Both reviewers performed full-text screening for inclusion and exclusion criteria, and the same third reviewer also resolved con icts. Consensus coding and resolution of disputes were conducted between the two reviewers, with the third reviewer arbitrating.

Study Outcomes
Speci c outcomes collected for each study included: (1) author name and year of publication; (2) study sample (age, race/ethnicity, sex); (3) study characteristics (study design, study dates, analytic sample, type of cancer/cancer site, intervention description); and (4) implementation science characteristics (phase of implementation, use of IS framework -veri ed with the Dissemination & Implementation Models in Health Research & Practice hub [27], name of IS framework) and (5) magnitude of bias.

Study Characteristics
For each included study, the following characteristics were collected: (1) Study design, which referred to the type of design employed in the study. Four coding options were available for this indicator: "Quantitative" was coded for any study that employed quantitative data (e.g., EMR, survey). "Qualitative" was coded for any study that used qualitative methods (e.g., focus groups, interviews). "Mixed-Methods" was coded for any study that used a combination of quantitative and qualitative methods (e.g., survey & focus groups; claims data & interview). Finally, "RCT" was coded for any study that employed a randomized controlled trial design. (2) Study dates referred to the year(s) in which data collection for the study took place (e.g., 2009-2010).
(3) Analytic sample was de ned as the sample size used to analyze cancer-related outcomes. (4) Type of Cancer/Cancer Site referred to the type of cancer(s)/cancer site(s) the intervention or study targeted. (5) Study/Intervention Description re ected a short description of the study/intervention. For example, "a patient navigation cancer screening program" or a "mobile-based tobacco cessation program."

IS Characteristics
The IS characteristics coded for each study included the phase of implementation and the name of the IS framework used in the study. Phase of Implementation referred to the stage in which the implementation science framework was employed and included three coding options. "Pre-Implementation" was coded for any study that used an IS framework to inform the implementation of an intervention (i.e., before implementation of EBI). "Post-Implementation" was coded for any study that used an IS framework to evaluate the implementation of an intervention. "All Phase" was coded for any study that incorporated an IS framework throughout all phases (pre and post) of implementation. Name of IS Framework referred to the IS framework(s) used in the study and was veri ed with the D&I Models in Health Research & Practice hub [27].

Exploratory Aim Outcomes
When applicable, each study was assessed for whether it targeted a known or documented population impacted by cancer-related health disparities as de ned by the National Institutes of Health [28]. This de nition was also applied to global health contexts, where there are known and ongoing health disparities in low-and middle-income countries [29]. Information on this outcome was abstracted from the background context in each study (e.g., explicit statement that this intervention is aimed at increasing screening rates within the Hispanic/Latino population), and was either coded as "Yes" if the study met this criterion or "No" if the study did not meet this criterion. If the implementation of a cancer prevention EBI was adapted to meet the needs of a health disparities population, this health disparities application was further categorized according to when, for whom, and how the adaptation(s) occurred. The Timing of Adaptation referred to when the adaptation(s) of an EBI took place and can be coded as either "by happenstance" (as the implementation was ongoing) or "by design" (if the adaptations were incorporated as part of the design of the study). In addition to capturing when the adaptation(s) were considered and implemented, who initiated and decided on the adaptation(s) was also abstracted (Decision-Makers of Adaptation). The decision to modify an intervention with adaptations was classi ed either as by "the core research team" or by a "team of researchers and community members" (e.g., community-engaged studies). The type of adaptations was coded as either contextual or content modi cations using the coding scheme adapted by Stirman and colleagues [30]. For any study, it was possible to have both contextual and content modi cations. A narrative highlighting the study overview and ndings are also presented for all studies included within this exploratory aim.

Study Quality Assessment
The Quality Assessment Tool for Studies with Diverse Designs (QATSDD) tool established and validated by Sirriyeh and colleagues was used to assess the quality of studies included in the review [31]. The QATSDD tool consists of 16 items (with some items only applicable to quantitative and qualitative studies), scored on a Likert scale from 0 = "high risk of bias" to 3 = "minimal risk of bias", and has strong reliability and validity in scoring studies with various (mixed) designs [32]. A QATSDD overall percentage score was calculated for each included study. A study with "low risk of bias" has an overall QATSDD percentage score greater than or equal to 75%. A study at "moderate risk of bias" has an overall QATSDD percentage score between 50-74%, meanwhile a study with "high risk of bias" has an overall QATSDD score between 0-49% [31].

Results
A total of 1,025 article titles and abstracts were screened, and 84 were deemed eligible for full-text screening (Figure). After full-text screening, n = 27 articles were included for data abstraction and synthesis (Table 1). Inter-rater reliability was moderate to strong between both reviewers during inclusion and exclusion screening processes (Κ = 0.728, Κ = 0.866, respectively). Of the 27 included studies in the nal sample, more than half of them (n = 19, 70.3%) were based in the US. Two studies were based in Argentina and conducted by the same research team [33,34]. The remaining individual studies were implemented in the

Study Design, Cancer Types, & Study Settings
Most studies utilized mixed-methods (n = 12, 44.4%) or qualitative methods (n = 12, 44.4%) to conduct their data collection and analyses. Only three studies exclusively used quantitative data, which were heavily drawn from electronic medical records (EMR) [34,41,42]. More than 80% of the studies (n = 21, 81.5%) focused on a single cancer site, with cervical (n = 8, 29.6%) and colorectal (n = 8, 29.6%) being the most targeted cancer types. Most studies implemented a cancer screening intervention; only two studies were non-intervention or pre-planning studies [43,44]. Of the 27 studies, 66.7% (n = 18) took place in a health system, 22.2% took place in a community setting, and three studies were implemented in both contexts [34,43,45].

Exploratory Findings
Only one-third of studies (n = 9, 33.3%) used an IS framework to address cancer-related health disparities ( In a qualitative study conducted by VanDevanter and colleagues [40], the CFIR was used to identify barriers, facilitators, and modi cations to implementing a tobacco cessation program. As a formative evaluation (non-intervention study), VanDevanter and colleagues were interested in how to translate an existing tobacco use treatment program created and tested in a high-income developed country (HIDC) to the local context of a low-income developed country (LIDC) in this case Vietnam. The original intervention included training patients and providers with a toolkit, and a reminder system to prompt providers to identify eligible patients for screening and brief counseling. The adaptation of the intervention to the Vietnamese context was to include a village health worker (VHW), who would provide patients with more intensive cessation counseling. Before implementing this adapted intervention, VanDevanter and colleagues wanted to examine if any barriers and facilitators may exist and if further adaptations were needed.
They identi ed four potential facilitators within the following CFIR domains: (1) a greater advantage of the intervention compared to existing practice (intervention characteristics), (2) a need to address the burden of tobacco use in the population (outer setting), (3) a demand to increase training, skill-building and leadership engagement (inner setting), (4) and a strong collective e cacy to provide services for tobacco cessation (individual characteristics).
Conversely, the following CFIR barriers were uncovered: the perception that the intervention was complex (intervention characteristic) and not necessarily compatible with current work ows (inner setting); and that the Ministry of Health (MOH) has not historically prioritized tobacco cessation and control, and therefore, external resources were lacking (outer setting). VanDevanter and colleagues also identi ed additional modi cations to the intervention, including: (1) lengthening the initial training session with providers and VHWs, and adding a booster session to provide opportunities for these personnels to continually re ect and build capacity; (2) training VHWs as a team rather than individually; and (3)  In 2011, Lobb and colleagues [43] used the KTA framework to help inform barriers to implementing cancer EBIs (e.g., FOBTs, mammograms, Pap tests) in Canadian health systems and community organizations. Overall, 45 unique barriers to use of mammograms, Pap tests, and FOBTs were identi ed with limited knowledge among residents; etho-cultural discordance; and health education programs ranked highest for all surveys. Barriers related to cost, patient beliefs, fears, and lack of social support.
Turner and colleagues [49] used the RE-AIM framework to guide and evaluate the implementation of STOP HCC (Screen, Treat, Or Prevent HCC) -a health system-based program designed to promote Hepatitis C Virus (HCV) screening in safety-net primary care practices with large populations of Hispanic patients.
STOP HCC was adapted from an intervention that included built-in EMR reminders, posters and handouts about HCV screening, re ex HCV RNA testing, and inperson counseling by a community health worker (CHW) about HCV and follow-up care. STOP HCC adopted the majority of these components with some modi cations.
Turner and colleagues used the RE-AIM framework to consider contextual (format, personnel) and content (tailoring/tweaking/re ning, adding elements) adaptations for STOP HCC. In STOP HCC, the format was adapted by having CHWs provide remote navigation to patients instead of in-person. In addition to CHWs, STOP HCC had other personnel to assist patients, including nurses, pharmacists, and social services workers. Regarding content adaptations, HCV screening materials were translated in Spanish (tailoring/tweaking/re ning), and a mobile application was created and added to educate patients about HCV, HCV-associated stigma and risk factors, and curative options.

Discussion
This is the rst review that employed a mixed-methods synthesis to examine how IS frameworks have been used in cancer prevention studies, with an exploratory aim of how some of these frameworks have been applied in the context of health disparities. Cancer prevention studies using IS frameworks were uncovered across both high-income developed countries (HIDC) (e.g., Canada) ands low-income developed countries (LIDC) (e.g., Vietnam). A resounding number of these IS studies were based within health systems in the US, which may re ect the growing trend and value of developing learning health systems in this country [50]. The sizeable amount of LIDC-based studies also demonstrate a promising expanse and application of implementation science within the international context. However, the number of non-US studies remains disproportionate to US-based studies, a rming the concern raised in Rabin and colleagues' 2012 IS harmonization campaign that outreach to and engagement with the international IS research community is limited [51]. To continue advancing IS, cross-fertilization and collaboration with the global research community will be needed to maintain the standardization and rigor of the eld.
Our review also found that IS frameworks were applied with variable frequency in cancer prevention studies, with the CFIR most commonly used. The high prevalence of CFIR use may be a function of its earlier establishment, advancements, and ubiquitous applications compared to other frameworks [52]. Across all of these studies, most researchers employed IS frameworks within either the pre-implementation phase or the post-implementation phase. As a corollary of this, many of these studies were not designed to assess, measure, and report implementation outcomes, prohibiting any empirical assessment of whether applications of IS frameworks can in uence these outcomes. The limited number of studies that used an IS framework throughout the implementation continuum also suggest that these frameworks continue to be applied prescriptively, either as a determinant framework or an evaluation framework [12]. This prescriptive nature highlights two potential implications: 1) that current IS frameworks may be limited in their ability to incorporate the dynamic and cyclical nature of implementation, and 2) that IS researchers and practitioners may need to incorporate IS frameworks throughout the implementation continuum.
Although new frameworks, such as the Dynamic Sustainability Framework, are emerging, they are yet to be widely used and tested within the context of cancer prevention [53]. Given the deluge of available IS frameworks, selecting an appropriate framework also continues to be a challenge for IS researchers [54,55]. Although several decision tools have been created to support IS researchers [8, 54], a growing appreciation and application of these tools within the cancer prevention eld may facilitate greater transparency and standardization around the identi cation, purpose, and elucidation of how an IS framework may mechanistically in uence implementation outcomes.

Exploratory Aim Implications
Within the exploratory aim, very few studies purposely used IS frameworks to tackle cancer-related health disparities. While it is unclear why this is, we surmise several possible explanations. First, the application of a health equity lens within the eld of IS is still relatively new [3,56]. Discussion of this topic at organizations like the National Cancer Institute's Implementation Science Consortium in Cancer (ISCC) [57] and National Implementation Research Network [58] only materialized in the late 2010s and may have not yet entered into the radar of cancer prevention IS researchers. Secondly, as this sub eld of IS has only been recently recognized, myriad opportunities exist to further advance it, including developing and re ning IS models and frameworks that intentionally integrate a health equity lens [3]. Finally, as this sub eld continues to grow, we anticipate future cancer prevention EBIs will employ IS frameworks to explicitly address health disparities.
Within this subset of studies, CFIR remained the predominantly applied framework. Mirroring the primary ndings, IS studies on health disparities also primarily employed IS frameworks for pre-implementation, as an analytical framework to identify potential facilitators and barriers (i.e., adaptations) pertinent to a health disparities population. When applied in a pre-implementation context, CFIR was most commonly used to describe "organizational" barriers and facilitators to implementation, which then informed content and contextual adaptations of the intervention of interest. Therefore, the domains of most interest within these studies were the inner setting, outer setting, and individuals involved. Only two studies employed the CFIR throughout the implementation continuum, using it to identify pre-implementation adaptations and to characterize additional and future adaptations post-implementation [47,48]. Furthermore, all but one study [48] identi ed adaptations by design or during pre-implementation, and most adaptations were exclusively identi ed by the core research team and implementation staff. These adaptation processes highlight two signi cant implications. First, the identi cation and tracking of adaptations are occurring systematically at the beginning of studies (i.e., proactively and planned) and are primarily driven by researchers. Second, the reporting of reactive (i.e., unplanned) adaptations and the consideration of adaptations driven by input from stakeholders outside the research team are almost non-existent. This nding points to a critical need to employ community-engaged research approaches to democratize the adaptation process in implementation studies [59,60]. Community members outside of the research team must be involved with the implementation process so that they can provide any potential adaptations to optimize the intervention's t and acceptability. Encouraging and activating more participatory research approaches, therefore, is necessary to achieve the growing call for more equitable IS [61].
As illustrated in the primary ndings, implementation outcomes were also not measured, assessed, and reported in this exploratory subset. However, two health disparities studies did note how the application of an IS framework led to the identi cation and implementation of adaptations, and how those framework-informed adaptations in uenced patient-reported outcomes [42,49]. In both studies, the application of an IS framework led to a small but signi cant increase in recipient-level outcomes, such as increased cancer screening rates within the targeted health disparities populations. These modest ndings suggest that IS frameworks infused with a health equity lens can potentially reduce cancer-related disparities. Conversely, more empirical research is needed, to evaluate and elucidate how the varying components of an IS framework may in uence patient-reported outcomes and link to different implementation outcomes.

Strengths and Limitations
This review had several strengths. First, we included studies with diverse designs (e.g., mixed-methods), searched multiple databases with additional handsearching, and considered studies in all locations. Despite our vast search strategies, we still observed little regional diversity. Most studies were conducted in the US combined with a few studies coming from other HIDCs (e.g., Canada) and some LIDCs (e.g., Vietnam). Further research should seek to capture countryspeci c vernacular around implementation science and include studies published in other languages. Secondly, with the interplay between cancer sites and complex interventions, various outcomes were assessed and reported in the included studies. As a result, we could not perform a meta-analysis due to the sparse effect estimates available for a random-effects model. A more comprehensive systematic review and meta-analysis of the use of implementation science frameworks across various public health interventions may be valuable for discerning the effect of applying IS frameworks and its impact on implementation-and patient-reported outcomes. We also did not include cultural adaptation models as part of the search strategy for IS frameworks. This limitation may be an artifact of the Nilsen taxonomy of IS frameworks, which currently does not account for cultural adaptation models as explicit IS frameworks [12]. Consequently, many studies that have used cultural adaptation models to inform the implementation of a cancer prevention EBI to address a speci c health disparity may not have been included our review.

Conclusion
This study suggests that the application of IS frameworks to cancer prevention studies is still in an embryonic stage. Recent clarion calls about infusing more equity within the practice of IS highlight that cancer prevention researchers must do more to intentionally design interventions that center health equity at the core. While no speci c health equity IS frameworks were identi ed in any cancer prevention studies in this review, there is promising growth and development of novel theories and models that infuse a health equity lens within cancer treatment and other disciplines. One notable framework includes Woodward and colleagues' "Health Equity Implementation Framework", which draws on two conceptual frameworks -the Health Care Disparities Framework and the i-PARIHS -that explicitly highlights health determinant factors within a healthcare delivery system that may in uence disparate health outcomes (i.e., clinical encounter) [62]. A recent addition to this work by Woodward also highlighted some core domains to consider when addressing implementation-related disparities, such as 1) cultural factors of recipients, 2) clinical encounter, or patient-provider interaction, and 3) societal context (including but not limited to social determinants of health).
[63] However, health equity frameworks are not just limited those mentioned earlier; an expanse of IS frameworks, speci cally, cultural adaptation frameworks and models exists and are currently being re ned to t within speci c health and implementation disparities contexts [64].
Exciting developments, such as the Transcreation framework [65], is aiming to create more equitable and participatory approaches within implementation science to address health disparities. Since its inception a decade ago, the eld of IS has grown tremendously. Along with this growth comes a wide variety of tools and frameworks that are now available to practitioners of the eld to take up and explicitly address health inequities. It is incumbent upon IS researchers, including cancer prevention scientists, to implement more equitable interventions that will ultimately reduce cancer disparities.

Consent for publication
Not applicable.
Availability of data and materials The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Competing interests
All authors have indicated they have no potential con icts of interest to disclose. Authors' contributions RP conceived of the idea for the review and oversaw the project. SX, HIA, RG, SK, DL, SM, and RP constructed and re ned the search strategies. SX conducted, acquired, and managed the data. SX and HIA were involved in the screening of studies, data extraction, coding and analysis. SX drafted, re ned, and revised the manuscript, and is the guarantor of this paper. All authors, edited, read, and approved the nal manuscript. Figure 1 PRISMA Flowchart

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. ReviewofISFrameworksinCancerPreventionStudiesSupp.docx