Systematic review on barriers and facilitators of complex interventions for residents with dementia in long-term care

ABSTRACT Objectives: Psychotropic drugs are frequently and sometimes inappropriately used for the treatment of neuropsychiatric symptoms of people with dementia, despite their limited efficacy and side effects. Interventions to address neuropsychiatric symptoms and psychotropic drug use are multifactorial and often multidisciplinary. Suboptimal implementation of these complex interventions often limits their effectiveness. This systematic review provides an overview of barriers and facilitators influencing the implementation of complex interventions targeting neuropsychiatric symptoms and psychotropic drug use in long-term care. Design: To identify relevant studies, the following electronic databases were searched between 28 May and 4 June: PubMed, Web of Science, PsycINFO, Cochrane, and CINAHL. Two reviewers systematically reviewed the literature, and the quality of the included studies was assessed using the Critical Appraisal Skills Programme qualitative checklist. The frequency of barriers and facilitators was addressed, followed by deductive thematic analysis describing their positive of negative influence. The Consolidated Framework for Implementation Research guided data synthesis. Results: Fifteen studies were included, using mostly a combination of intervention types and care programs, as well as different implementation strategies. Key factors to successful implementation included strong leadership and support of champions. Also, communication and coordination between disciplines, management support, sufficient resources, and culture (e.g. openness to change) influenced implementation positively. Barriers related mostly to unstable organizations, such as renovations to facility, changes toward self-directed teams, high staff turnover, and perceived work and time pressures. Conclusions: Implementation is complex and needs to be tailored to the specific needs and characteristics of the organization in question. Champions should be carefully chosen, and the application of learned actions and knowledge into practice is expected to further improve implementation.

NPSs are the result of interactions of biological, psychological, social, and physical environmental factors (Cohen-Mansfield, 2000;Steinberg et al., 2006;Zuidema et al., 2010). Complex, multicomponent interventions seem to be the most appropriate approach to address these, given the multifactorial origin of NPS. Complex interventions comprise multiple interacting components and are characterized by the number and difficulty of behaviors required by those delivering or receiving the intervention, the number of groups or organizational levels targeted by the intervention, the number and variability of outcomes, and the degree of flexibility or tailoring of the intervention permitted (Craig et al., 2013).
Although complex interventions have the potential to reduce inappropriate prescribing of antipsychotic drugs in NHs (Livingston et al., 2017;Thompson Coon et al., 2014), these interventions commonly show small to modest effects (O'Connor et al., 2009;Quasdorf et al., 2016;Zwijsen et al., 2014a), which often reflects suboptimal implementation rather than shortcomings of the implemented intervention (Anderson et al., 2013;Craig et al., 2013).
To examine barriers and facilitators influencing the implementation of complex interventions for people with dementia in long-term care, we reviewed literature on process evaluations, qualitative studies, and (cluster) randomized controlled trials targeting NPS and/or psychotropic drug use (PDU). By assembling knowledge about factors influencing implementation of complex interventions, effectiveness of interventions can be maximized, and translating results into practice is enabled which in turn enhances widespread implementation (Craig et al., 2013;Lawrence et al., 2012;Thompson Coon et al., 2014;Quasdorf et al., 2016;Zwijsen et al., 2014b).

Eligibility criteria
A predefined protocol was developed and registered on PROSPERO (CRD42018112731), on November 9, 2018, and is available in full on the National Institute for Health Research website: https:// www.crd.york.ac.uk/prospero/ (Groot Kormelinck et al., 2018).

Types of studies
We included process evaluations, qualitative studies (that may include quantitative process data), and (cluster) randomized controlled trial studies that reported barriers and facilitators affecting the implementation of complex interventions targeting NPS/ PDU for residents with dementia in long-term care. Systematic reviews or studies not being published in peer-reviewed journals were excluded.

Types of interventions
This review was limited to studies targeting implementation barriers and facilitators of complex interventions aimed at PDU (antipsychotics, anxiolytics, hypnotics, antidepressants, anticonvulsants, anti-dementia drugs) and/or NPS (umbrella term, or at least one symptom). We defined a complex intervention as introduced by Craig et al. (2013, p.588): "multiple interacting components, a certain number and difficulty of behavior of those delivering or receiving the intervention, the number of groups or organizational levels the intervention targets, the number and variability of outcomes and the degree of flexibility or tailoring of the intervention permitted."

Search
Electronic databases were searched to identify relevant studies. The search was applied to PubMed, Web of Science, PsycINFO, Cochrane, and CINAHL. Searches were run between 28 May and 4 June 2018. No publication date restrictions were imposed. Studies published in English, German, and French were eligible for inclusion. Key search terms related to institution, outcome (barriers, facilitators), and psychotropic drugs or NPS. For full search strategy, see Appendix A1, published as supplementary material online.

Study selection method
Two reviewers (CMGK and SIMJ) independently screened titles and abstracts and selected potentially relevant articles for full-text review. Duplicates were removed using reference manager software (Refworks), after which two reviewers independently reviewed the full text for in-or exclusion. Reviewer findings were compared during the screening process, with disagreements being resolved by involvement of a third reviewer.

Data extraction
We used a predesigned data extraction sheet, which was piloted on several articles before actual use and refined it accordingly. One reviewer extracted data (CMGK), which was checked by a second (SIMJ). Additional reviewers were involved to reach consensus in the case of disagreement. Data that were extracted included setting, study aim, type, content, and results of intervention, implementation method, data collection method, method of analysis, data collection moment, and implementation barriers and facilitators.

Study quality
The methodological quality of each study was assessed using the Critical Appraisal Skills Programme qualitative checklist (Critical Appraisal Skills Programme, 2017). The quality of the studies was appraised by one reviewer (CMGK) and scores were checked by a second (SIMJ). Disagreements were resolved by discussion. Papers were not excluded based on quality. Instead, quality of studies is addressed in the discussion section.

Data synthesis
Each barrier or facilitator was given a code, using Atlas.ti 8.3. The Consolidated Framework for Implementation Research (CFIR) was used to guide data synthesis, following a deductive approach. The CFIR is a comprehensive, "meta-theoretical" framework. The standardized list of constructs allows researchers to identify variables that are most relevant to a particular intervention (Damschroder et al., 2009). The codes were subdivided into the five domains of the CFIR framework: intervention characteristics, outer setting, inner setting, characteristics of individuals, and process. We kept in mind the possibility that codes might not fit the CFIR.
The importance of the barrier/facilitator was addressed by gaining insight into their frequency. Deductive thematic analysis was used to assess a factor's positive or negative influence (Elo and Kyngäs, 2008;Hsieh and Shannon, 2005).
Two reviewers (SIMJ and CMGK) independently coded four studies, and findings were compared and discussed. After this, one reviewer (CMGK) continued with coding the other studies. The coding of each study was discussed by both reviewers to reach agreement. The other reviewers were involved to obtain consensus in case of disagreements.

Study selection
The search of all the databases yielded 4734 records of which 15 studies were included. See Preferred Reporting Items for Systematic Reviews and

Systematic review on barriers and facilitators 875
Meta-analysis flow for application of eligibility criteria ( Figure 1). Table 1 presents the study characteristics. With the exception of one German study, all studies were published in English. Studies were carried out in Australia (n = 2), Canada (n = 2), the U.S.A. (n = 1), the U.K. (n = 3), Norway (n = 1), Germany (n = 2), and the Netherlands (n = 4). The majority of the studies were qualitative (process) evaluations, sometimes combined with quantitative data. Most studies pertained to residents with dementia in NHs, residential aged care facilities, or long-term care homes. We identified four types of interventions, often combined: (1) managing NPS by methodical and multidisciplinary collaboration (n = 10); (2) psychosocial interventions tailored to the resident or person-centered care (PCC) approaches (n = 9);

Study characteristics
(3) training and education (n = 2); and (4) an activity or exercise program (n = 2). Several implementation strategies were used, such as coaching on the job, follow-up meetings, sharing experiences, and telephone support. Multiple methods of data collection were used, among others questionnaires, focus groups, and individual interviews. Most studies applied triangulation to enhance credibility of findings. A range of stakeholders provided the data on implementation factors, mostly being staff, managers, and/or project coordinators. Table 2 provides a detailed overview of the quality assessments of the studies. On a scale from 0 to 10 (the higher the more quality), five studies scored 5 to 7 points (Borbasi et al., 2011;Kovach et al., 2008;McAiney et al., 2007;Stein-Parbury et al., 2012;Wingenfeld et al., 2011), and ten studies scored 8 to 10 points (Appelhof et al., 2018;Boersma et al., 2016;Bourbonnais et al., 2018;Ellard et al., 2014;Van Haeften-Van Dijk et al., 2015;Latham and Brooker, 2017;Lawrence et al., 2016;Mekki et al., 2017;Quasdorf et al., 2016;Zwijsen et al., 2014b).

Barriers and facilitators
The barriers and facilitators reported in the studies were grouped according to the five domains and 36 constructs of the CFIR. All codes fitted within the CFIR. Table 3 shows the frequency with which the CFIR constructs were addressed and provides an overview of the CFIR constructs pertaining to the individual studies. A short description of each construct can be found in Table S1, published as supplementary material online. Domain 1. Intervention characteristics Relative advantage was addressed in six articles (Appelhof et al., 2018;Boersma et al., 2016;Bourbonnais et al., 2018;Ellard et al., 2014;Van Haeften-Van Dijk et al., 2015;Lawrence et al., 2016). The added value of the intervention was having a shared method for multidisciplinary consultations (Boersma et al., 2016), and expected gains in care time led to increased implementation willingness and efforts of staff (Van Haeften-Van Dijk et al., 2015). Also, experiencing visible effects and positive reactions of residents were facilitators (Ellard et al., 2014;Van Haeften-Van Dijk et al., 2015;Boersma et al., 2016). Concerns about consequences of the intervention, such as how to deal with aggression when PDU is reduced, impeded implementation (Bourbonnais et al., 2018;Lawrence et al., 2016).
Adaptability was addressed by three articles as a facilitating factor (Bourbonnais et al., 2018;Van Haeften-Van Dijk et al., 2015;Mekki et al., 2017). For example, the transfer of information and knowledge was tailored to the local NH culture, which stimulated implementation (Bourbonnais et al., 2018).
Cost was addressed in four articles (Appelhof et al., 2018;Boersma et al., 2016;Van Haeften-Van Dijk et al., 2015;McAiney et al., 2007). Facilitators were sufficient funding for the proposed intervention (Van Haeften-Van Dijk et al., 2015), wards receiving extra budget from the NH (Appelhof et al., 2018), and inexpensive training, especially if a regular training budget exists that can be used to provide the intervention (Boersma et al., 2016). Pressures on financial resources such as budget cuts negatively affected the implementation process (Boersma et al., 2016;Van Haeften-Van Dijk et al., 2015;McAiney et al., 2007).    Four constructs within the domain intervention characteristics yielded no relevant factors affecting implementation in the included articles (see Table 3).

Domain 2. Outer setting
Only few studies reported about factors affecting implementation within this domain. The domain contains four constructs, of which cosmopolitanism and peer pressure were not represented in the reviewed articles (see Table S1 CFIR constructs with short definitions).
Patient needs and resources were addressed by one article. A lack of background information about the residents was a barrier for implementation (Boersma et al., 2016).  Facilitating factors were a well-functioning and stable team, a less hierarchical structure and flexible organizational structures, being specialized in dementia care (Quasdorf et al., 2016), and having a small-scale care setting and rural environment (Boersma et al., 2016). Barriers regarding high patient-to-caregiver ratios (Bourbonnais et al., 2018), and multiple levels of management made access to resources challenging (Latham and Brooker, 2017). Half of the articles found staff turnover/ absenteeism/fluctuations, shortages, and changing positions to be an impeding factor (Appelhof et al., 2018;Boersma et al., 2016;Bourbonnais et al., 2018;Ellard et al., 2014;Van Haeften-Van Dijk et al., 2015;Quasdorf et al., 2016;Zwijsen et al., 2014b). It might lead to hindering factors such as new staff not being informed about, or familiar with, the program (Appelhof et al., 2018;Bourbonnais et al., 2018;Zwijsen et al., 2014b), and new staff needing time to get acquainted with the intervention (Appelhof et al., 2018;Zwijsen et al., 2014b).
Culture was addressed in five articles (Boersma et al., 2016;Lawrence et al., 2016;Mekki et al., 2017;Quasdorf et al., 2016;Stein-Parbury et al., 2012). A more dementia friendly culture as expressed in staff attitudes and the physical environment was helpful (Quasdorf et al., 2016), as were mutual respect and reciprocity in relationships with residents (Lawrence et al., 2016), a positive team culture where people acknowledge each other (Mekki et al., 2017), and staff feeling able to voice opinions (Stein-Parbury et al., 2012). Staff with different cultural backgrounds and difficulties with the Dutch language were barriers (Boersma et al., 2016).
Implementation climate consists of six subconstructs, of which five were addressed (see Table 3) (1) Tension for change was reported in one article.
Pressure from peers to resist change negatively affected implementation (McAiney et al., 2007). (2) Compatibility was addressed by five articles (Appelhof et al., 2018;Boersma et al., 2016;Van Haeften-Van Dijk et al., 2015;Latham and Brooker, 2017;Zwijsen et al., 2014b). Interventions being consistent with care goals facilitated implementation (Van Haeften-Van Dijk et al., 2015), while a barrier was that the intervention as perceived by the care professionalsmay not necessarily be in line with the corporate imageas set by the management (Latham and Brooker, 2017). Overlap with current working was reported as a barrier in two studies. For example, an overlap with tools already available in the electronic health record led to staff being more inclined to keep working according to their old working routine (Appelhof et al., 2018).
(3) Relative priority was addressed by six articles (Appelhof et al., 2018;Boersma et al., 2016;Bourbonnais et al., 2018;Van Haeften-Van Dijk et al., 2015;Latham and Brooker, 2017;Zwijsen et al., 2014b). Limited involvement in research projects promoted implementation (Appelhof et al., 2018), while other innovations implemented at the same time were a barrier (Van Haeften-Van Dijk et al., 2015). Implementation of the care program was easier on wards that rarely initiated new projects, which helped staff to remain motivated. Being involved in several new projects seemed to interfere with implementation, since time was scarce (Zwijsen et al., 2014b). Ward issues such as renovations to the facility (Appelhof et al., 2018), transition toward self-directed teams (Appelhof et al., 2018;Boersma et al., 2016), staff turnover (Bourbonnais et al., 2018;Latham and Brooker, 2017), and changes in staff members' positions and management structure were barriers (Zwijsen et al., 2014b). (4) Goals and feedback were reported by one article.
Little or no feedback and collaboration with internal facilitators, and a low level of feedback and engagement within the team and on the individual level hindered implementation (Mekki et al., 2017). (5) Learning climate was addressed by eight articles (Appelhof et al., 2018;Boersma et al., 2016;Borbasi et al., 2011;Ellard et al., 2014;Latham and Brooker, 2017;Lawrence et al., 2016;Mekki et al., 2017;Zwijsen et al., 2014b). Openness to changing working routines facilitated implementation (Appelhof et al., 2018;Mekki et al., 2017), while an insufficient learning climate limited implementation (Boersma et al., 2016;Ellard et al., 2014). The degree of learning climate can depend on the ward. In one study, several wards were reluctant to alter routines, whereas wards that were enthusiastic to work with the care program seemed to have a more open attitude toward change and welcomed external input (Zwijsen et al., 2014b). Other facilitators were that the intervention team worked on the floor together with the staff and provided compliments and encouragement (Borbasi et al., 2011). Also, sufficient support and meetings to discuss events during the day and their negative and positive sides led to positive experiences (Latham and Brooker, 2017), as did reporting details of success stories and sharing strategies that worked (Borbasi et al., 2011;Mekki et al., 2017). Staff fearing criticism of the training team hindered implementation (Lawrence et al., 2016).
Readiness for implementation contains three subconstructs, of which two were addressed (see Table 3).

Domain 4. Characteristics of individuals
Knowledge and beliefs about the intervention were addressed in all but five articles (Borbasi et al., 2011;McAiney et al., 2007;Mekki et al., 2017;Latham and Brooker, 2017;Stein-Parbury et al., 2012). In one study, management had limited awareness of the added value of the intervention and some staff had critical attitudes. However, the expected gains in terms of care time and experienced positive effects on residents made staff enthusiastic to implement the intervention (Van Haeften-Van Dijk et al., 2015). Implementation of the program (Appelhof et al., 2018) or managing disruptive behaviors (Kovach et al., 2008) was time-consuming and increased stress and frustration. Repeatedly starting a functional analysis of behavior was perceived as discouraging (Bourbonnais et al., 2018), and interventions being perceived as childish or disrespectful to people with dementia hindered implementation (Boersma et al., 2016;Van Haeften-Van Dijk et al., 2015). Three articles addressed self-efficacy (Borbasi et al., 2011;Van Haeften-Van Dijk et al., 2015;Stein-Parbury et al., 2012). Staff working together coordination between disciplines, management support, sufficient resources, educated staff, and culture. Barriers related mostly to unstable organizations, such as renovations, changes toward self-directed teams, high staff turnover, perceived work and time pressures, and being involved in several projects.
Similar to our findings, other reviews demonstrated that lack of time, high staff turnover (Vlaeyen et al., 2017), and lack of organizational support (Beeber et al., 2010) can be barriers to implementation. In a review on implementation of evidencebased practice in community nursing, organizational changes such as decentralization were a barrier, while facilitators were the use of local champions, training being embedded in practice, actual or perceived skills, perceptions about usefulness and evidence that the intervention will positively impact the resident or caregiver (Mathieson et al, 2018). Despite the fact that these reviews took place in a different setting, the barriers and facilitating factors found are comparable to our findings, implying that some barriers and facilitators are generic in nature. However, several "setting specific" factors seem to affect implementation as well. For example, in a systematic review on fall prevention in residential care facilities, poor information transfer among care providers, staff, and family, and across shifts and lack of care plan communication were barriers (Vlaeyen et al., 2017). Similar barriers emerged in our review, implying that these "setting specific" factors should be taken into account in care innovations. As is suggested by Vlaeyen et al. (2017), we also underline that a focus on modifiable barriers and facilitators such as communication is needed in implementation projects in daily practice.
Other recently published papers in International Psychogeriatrics on implementation in long-term care had similar findings. A review on strategies for successful implementation of psychosocial (including multicomponent) interventions in daily residential dementia care, for instance, found that time required to learn and apply the intervention, having a learning culture, and putting knowledge into practice (such as on-the-job reinforcement of learning) were facilitators, whereas multiple projects running simultaneously impeded implementation (Boersma et al., 2015). The commitment of higher management and professionals were important factors in two studies (Boersma et al., 2015;Gerritsen et al., 2019), which is in line with our results. Our systematic review specifically focuses on the implementation of complex interventions targeting NPS/ PDU, while other studies focused on the implementation of guidelines for PCC in NHs (Vikström et al., 2015), implementation of the Meeting Centers Support Program (Van Mierlo et al., 2018), or implementing best practice dementia care in hospitals (Tropea et al., 2017), for example. Several barriers and facilitators identified in those studies are in line with our results, such as inadequate staffing levels (Tropea et al., 2017;Vikström et al., 2015), workload, insufficient time, communication difficulties within team and with family, and limited staff knowledge and skills of dementia (Tropea et al., 2017). In addition, the need for qualified and motivated staff, the presence of a project manager to guide the implementation, and the possibility to target the program to the needs of the target population were identified as facilitators (Van Mierlo et al., 2018). Although those studies had a different focus compared to our review, several barriers and facilitators were in line with our findings. Perhaps this implies that the barriers and facilitators identified in our review may account for different types of interventions and settings, beyond merely complex interventions targeting NPS/PDU.
To summarize, although some implementation factors are generic in nature, setting and organizational factors seem to play an important role in implementation. Our systematic review adds to this that the factors or issues that are perceived as impeding implementation in one care organization can be perceived as no barrier in another care organization. For instance, some organizations seemed to have more difficulties as a result of staff turnover than other organizations. In the study of Bourbonnais et al., (2018), for example, staff turnover did not negatively affect implementation, since other persons such as study coordinators continued to work actively with staff. Differences may even exist between wards of a care organization. In the study of Zwijsen et al. (2014b), for instance, the degree of learning climate depended on the ward. Several wards were reluctant to alter routines, while other wards had an open, enthusiastic attitude toward the care program. Hence, perhaps the most important recommendation is that we stress to take into account the local conditions and specific barriers and facilitators of a care organization by means of a tailored implementation plan.

Strengths and limitations
A strength is the use of a well-known, metatheoretical framework and the applied deductive thematic analysis to synthesize the results. Using the predefined codes of the CFIR provided the complex data with a clear direction (King, 2004). The coded data fitted the predefined constructs of the CFIR. Its standardized nature enhances comparison across studies. A limitation that warrants further consideration is that we did not exclude studies based on our qualitative appraisal. This Systematic review on barriers and facilitators 885 requires some caution in the interpretation of findings. Ten studies did not consider the relationship between researcher and participant, which potentially led to researcher bias (Critical Appraisal Skills Programme, 2017). Selection and recruitment of participants was also not thoroughly described, potentially leading to bias in the included studies, and consequently in our review. However, for the other categories, the quality of the included studies was generally considered sufficient. Also, the factors found in the included studies might not be the most important ones, but the ones focused on the most. Our results show that constructs within the domains "intervention characteristics," "outer setting," and "process" were less frequently addressed than the other domains. Apparently, several parts of the CFIR framework receive little research attention. This is contrary to a recent systematic review, which assessed the application of the CFIR in implementation research in a wide range of study aims and settings. In this review, all constructs were identified to a greater or lesser extent (Kirk et al., 2016). This difference might be explained by the fact that Kirk et al. (2016) only included studies that used the CFIR, while in our review, the included studies used different theories or frameworks to evaluate implementation. CFIR constructs were not used as a "checklist" of variables for consideration. Possible consequences are that relevant factors were not assessed.
Although it might be relevant to distinguish between barriers and facilitators related to the intervention and those related to the implementation strategy, the reviewed articles rarely present their results in this manner. Furthermore, several interventions incorporate elements, such as training (Smeets et al., 2013), that are considered implementation strategies by others (Gerritsen et al., 2011). Further research could explore the added value of this distinction.

Conclusions and implications
Our study showed that the engagement of champions can be an important facilitator, but their effectiveness relies on personal skills and relationships with colleagues. Consequently, we stress that champions should be carefully chosen. Translating learned actions and knowledge into practice by means of on-the-job reinforcement of learning or role modeling should be part of the implementation strategy for complex interventions by default. Caution should be employed while participating in several projects/ studies. The capacity of the involved key stakeholders should be leading. The current systematic review demonstrated that the implementation of complex interventions requires a lot of effort of the organizations and their staff members, and the degree of implementation is subject to many factors, which makes it complex. Our results indicate that some factors are generic in nature, but the setting and factors related to the organization such as staff turnover and reorganizations seem to influence implementation as well. The presence of factors and degree to which these are perceived as a barrier might differ between organizations and even between wards, depending on potential facilitating factors that can reduce the influence of the barrier and on the coping strategies of staff. Organization problems on the ward as such may be not necessarily barriers to successful implementation, but the coping mechanisms of the team could be of greater importance. Therefore, barriers and facilitators might be best examined at the organizational level, being for instance an NH, or even on the level of an NH ward. We underline that implementation needs to be adapted to the specific needs and characteristics of the organization in question and needs to focus on modifiable barriers and facilitators. To implement a complex intervention with several interacting components, in a complex and dynamic organization, with its own local characteristics and specific barriers and facilitators, is challenging and advocates for a tailored intervention and implementation plan. Assessing and addressing possible barriers and facilitators before and during implementation by means of tailored implementation can increase effectiveness (Baker et al., 2015).
Frameworks such as the CFIR can help identify which constructs have predictive ability, which can be manipulated to enhance implementation outcomes, and the situations in which specific constructs are salient.
Future studies could explore whether a focus on the "forgotten" constructs would be beneficial for implementation.