Organizational- and system-level characteristics that influence implementation of shared decision-making and strategies to address them — a scoping review

We performed a scoping review rather than a systematic review due to the broad nature of our research questions, the young field of SDM research, and our anticipation of high variation in study designs and methodologies [22]. We used the definition of scoping review given by Colquhoun and colleagues: “a form of knowledge synthesis that addresses an exploratory research question aimed at mapping key concepts, types of evidence, and gaps in research related to a defined area or field by systematically searching, selecting, and synthesizing existing knowledge” [23].

We developed our protocol based on the Arksey and O’Malley framework [22], as well as on subsequently published guidance on how to conduct scoping reviews [24‐26]. The final version of the protocol can be found in Additional file 1.

We included publications that reported on the results of projects, quality improvement programs, or studies that aimed to implement SDM, decision aids (i.e., tools for use inside or outside the clinical encounter [27]) or other decision support interventions (i.e., mediated by more interactive or social technologies [27]) in routine healthcare through a certain implementation strategy or effort. To be included, these full texts also needed to report on organizational-level and/or system-level characteristics described to influence the implementation, and/or describe strategies that might address organizational-level and/or system-level characteristics. Opinion pieces, reviews, and study protocols were excluded, but reviews were used in the secondary search process, as described below. The full list of inclusion and exclusion criteria, specifying concepts and contexts of this scoping review, is displayed in Table 1.

We performed an electronic literature search in Medline, CINAHL, and Web of Science Core Collection. We included articles published between January 1997, the year in which Charles and colleagues described the concept of SDM in their seminal article [28], and October 10, 2016. The search was limited to articles published in English or German, as these were the only languages spoken by a minimum of two members of the review team. Details of the search strategies in the different databases can be found in Additional file 2.

Our primary electronic search was complemented by a comprehensive secondary search strategy. All records excluded through criterion E2 (systematic, scoping, and structured literature reviews) [10, 14, 15, 29‐42] were checked to see whether they reported on studies that could potentially be relevant for this scoping review. Subsequently, the reference lists of six of these reviews [10, 15, 29, 36, 39] were assessed for eligibility. Furthermore, two books were searched for chapters meeting the inclusion criteria [43, 44], and a gray literature search was conducted on a range of websites listed in Additional file 3.

We imported all identified records into reference management software (Endnote) and removed duplicates. First, IS and a second reviewer (PH, AL, or RPM) performed an independent title and abstract screening to check for potential inclusion of records. A record was included into the next step of full text assessment if at least one reviewer deemed it appropriate. Second, full text assessment was conducted. To ensure quality and consistency of full text assessments, the first 20% of randomly selected full texts were assessed by two team members (IS and PH or IS and AL). In 83% of the cases, the team members agreed on inclusion or exclusion. Discrepant assessments were subsequently discussed by the team members. This process led to minor revisions in the exact wording of the inclusion and exclusion criteria and an instruction of how to use the criteria. Then, another round of double assessment using another set of 10% of randomly selected full texts was conducted, leading to agreement in 93% of the cases. The subsequent assessment of the remaining 70% of full texts was conducted by one reviewer (IS) using a conservative approach. Whenever the single assessor (IS) was in slight doubt about whether to include or exclude a full text, a second reviewer was assigned to assess that full text, and final decision regarding inclusion was made by discussion. This procedure was done for a total of 14 full texts.

We extracted general information on each study and specific information related to the research questions. We extracted any information on experiences related to organizational- and system-level characteristics and potential strategies to address them. As we wanted to give a broad overview, we extracted all information on experiences reported in the publications, including experiences derived from results (empirical) and from the interpretation of results (opinion-based). The number of full texts identified and selected is described using the PRISMA flowchart. The initial data extraction sheet was developed by one team member (IS), based on experience from other reviews [12, 15, 45, 46]. It was pilot tested by IS and AL, using two included full texts [47, 48]. We compared the extracted data and found only very minor differences in the level of detail of the respective extractions. As a result, the extraction sheet was slightly revised (e.g., by adding definitions of what to extract). Further data extraction was conducted by one person (either AL or IS). Whenever one data extractor was in doubt regarding what to extract for a certain category, the second person checked the full text and both met to discuss agreement on what to extract.

We did not appraise the methodological quality or risk of bias of the included studies, which is consistent with guidance on the conduct of scoping reviews [22].

We conducted a descriptive analysis of characteristics of the included studies (e.g., types of study design, years of publication) as well as a qualitative thematic analysis of the organizational- and system-level characteristics identified in the studies. We decided to report what other studies reported as influential characteristics, rather than classify them as barriers or facilitators. This analysis drew on principles of qualitative content analysis described by Hsieh and Shannon [49] and consisted of the following steps: first, two researchers (AL and IS) read the entire set of extracted data to gain an overview. Second, one researcher (AL) coded the material (initial inductive coding). Third, comments by a second researcher (IS) led to adaptation of the coding system. Fourth, the revised codes were organized into a coding system using clusters and subcategories, agreed in discussion with two other team members (GE and SK). Fifth, the material was re-coded by one researcher (AL) using the established coding system. Sixth, the re-coded material was cross-checked by a second researcher (IS) and minimal changes were made in discussion (IS and AL). Potential strategies mentioned in the publications to address organizational- and system- level characteristics were synthesized and mapped onto identified characteristics in a team discussion (IS, AL, GE). No qualitative data analysis software was used. Analyses were conducted on the level of distinct implementation projects, i.e., publications reporting on the same implementation project were grouped under one single project ID.

After screening 7745 titles and abstracts for eligibility, and checking 354 full texts against the inclusion and exclusion criteria, we included 48 full texts (see Fig. 1). Reasons for exclusion of full texts are displayed in Table 1. The included full texts report on a total of 32 distinct implementation projects. While most projects were only reported in a single publication, several projects were described in two or more publications. Twenty-two projects were conducted in the USA, and 26 projects focused on the implementation of decision aids or other forms of decision support. Projects focused on various settings and a broad range of decisional contexts. Table 2 gives an overview on the included projects and publications.

Figure 2 gives an overview of the identified characteristics.

A range of possible strategies to address organizational- and system-level characteristics and thereby potentially foster SDM implementation were discussed in the publications and mapped onto the identified characteristics. They are displayed in Table 5. Similar to the results on experienced characteristics, most proposed strategies focused on the organizational level. Most studies identified workflow as an organizational-level characteristic influencing SDM implementation and also generated potential strategies to tackle that characteristic. Few strategies were suggested to change organizational culture [50‐52], which was also described in fewer studies. A large range of potential strategies were also described to promote leadership activities that might facilitate SDM implementation (see full list in Table 5). At the system-level, fewer strategies were described. Suggestions included changes in payment models [53‐55], legislation [51, 56, 57], and health professional education [51, 58‐60].

We described a broad range of organizational- and system-level characteristics that were experienced as influencing the implementation of SDM in routine care, as well as strategies to potentially address those characteristics. Included studies reported more often on characteristics influencing the organizational level than the health system level. The reported organizational characteristics are strongly influenced by health system characteristics; for example, the amount of time that a HCP has for a patient’s visit is linked to payment models, the organizational culture is influenced by the general culture of healthcare delivery, and the leadership decisions within an organization are affected by policies, payment models, and accreditation criteria. As the identified characteristics can be barriers, facilitators or both barriers and facilitators to SDM implementation, we described them in a value-neutral way.

We extracted reports from implementation studies described in any part of the included publications. Our analyses therefore cannot differentiate between experiences based on results and those reflecting interpretation of results. However, for a young research field, we believe this broad scoping review is an important first step to gaining an overview of the topic.

A second limitation is that the primary search was limited to three electronic databases, so we might have missed relevant publications. However, we prioritized sensitivity in our electronic search, which is reflected by the high number of screened abstracts, to identify most relevant work. Furthermore, we conducted an extensive secondary search, including gray literature to find more work not indexed in the electronic databases searched. Another limitation is that we did not conduct a full double assessment and double data extraction. However, we did our best to minimize error by consulting with a second reviewer whenever there was the slightest doubt. A main strength of this review is that it is the first of its kind to focus solely on the impact of organizational and system characteristics on the implementation of SDM. In previous work, the focus had mainly been on the individual clinician-patient level, and organizational- and system-level characteristics had not been examined in depth [10, 14]. Furthermore, it was conducted in an inter-professional and international team.

First, these findings need to be compared to previous work on SDM. Our results reinforce prior calls for better coordination of care, engagement of non-physician personnel, and the use of the electronic health record (EHR) to implement SDM in previous work [61]. The suggestions to use clinical practice guidelines, postgraduate training, and accreditation as means to better implement SDM [5] are also reflected in the data collected in this scoping review. Many of the characteristics identified in this review have been discussed in trials of SDM interventions or decision aids, in studies of clinicians’ perceptions, or in opinion pieces, but this is the first piece of work looking at characteristics experienced in actual implementation studies.

Second, the results need to be compared to more general work in healthcare implementation science, beyond the case of SDM as a particular innovation to implement. Implementation frameworks and conceptual models like the one postulated by Greenhalgh and colleagues [20] or the Consolidated Framework for Implementation Research (CFIR) [19] have described elements in the inner and outer settings to influence implementation. Our results found a range of very similar characteristics on the organizational level to the ones described in the inner setting, e.g., communication and culture within an organization, leadership engagement, resources, and priorities. However, some of the characteristics we found (e.g., workflows) were not described in the CFIR [19]. One could hypothesize that these aspects are more focused around decision aid implementation and therefore not included in a more general implementation framework. Similarly, several of our system-level characteristics map well onto the CFIR’s outer setting (i.e., aspects around policies, guidelines, and incentives), but the culture of the healthcare system and education and licensing of healthcare professionals cannot be found in the framework [19]. Furthermore, a systematic review on determinants of implementation of preventive interventions on patient handling identified a total of 45 environmental barriers and facilitators [62] that overlap with experienced organizational characteristics identified in our scoping review, particularly the availability of resources, leadership support, and the organization of workflows. Overall, our results in the field of SDM display many similarities with the characteristics described in implementation science frameworks and in other fields of health innovation. However, as we also identify characteristics less described in implementation science literature, we believe it is important to not to be restricted by such frameworks, but enrich them with derived empirical evidence.

Third, some of the strategies recommended by the included projects to intervene on characteristics influencing SDM implementation (Table 5) are vague and require further specification and tailoring to a specific context [63]. For example, some of the strategies that fall into the leadership category could benefit from distinguishing which level of leadership should take action for which strategy. While the people in a governing board of an organization might be the ones to revise mission statements, executive leadership, and departmental management might be the ones who create a culture that supports SDM [64]. Furthermore, all other categories identified as organizational-level characteristics, despite not specifying who should be in charge of making specific changes, imply that organizational leadership is the actor here. Although it is not specified, for example, who should implement multidisciplinary teams or create an SDM coordinator position, there is an implicit assumption that these are leadership tasks. Beyond looking at implementation literature, it might therefore be worthwhile for stakeholders working on SDM implementation to look into organizational theories in healthcare [65], e.g., on the effective organization of healthcare teams or on strategies to restructure healthcare organizations.

As healthcare systems are complex and composed of components that act nonlinearly [66], a certain identified characteristic can be a facilitator to one stakeholder and a barrier to another. Therefore, more work is needed to move beyond the descriptive stage of this review, especially as differences in the numbers of studies reporting on certain characteristics do not necessarily mean that those characteristics are the most important. Similarly to Koppelaar et al. [62], we believe there is a need to quantify the influence of the identified characteristics, especially as this scoping review’s broad nature is not distinguishing between experiences based on results of implementation studies and interpretation of those results. By evaluating the influencing characteristics in implementation studies, we could analyze interactions between characteristics and find out which of them predict implementation outcomes [18].

As the included studies were predominantly from the USA, future work needs to assess the importance of the identified characteristics in different healthcare systems with variation in financing, coverage, spending, utilization, capacity, and performance [67], as well as different fields of healthcare (e.g., cancer care, mental healthcare). This would help to gain a more specific insight that could foster prioritization of the most important characteristics in a particular setting and strategies to address them.