This project will be carried out in two main steps. First, a factorial survey experiment (FSE) design —also known as a vignette experiment—will be applied to test how professionals carry out decision-making processes while considering multiple factors that are likely to affect their decisions. This step will be preceded by a pilot study consisting of group interviews where relevant tensions for decision-making will be identified for each group of health care professionals to inform the content of the vignettes.
In the second step, we will develop a conceptual framework for professionals’ decision-making concerning de-implementation and prepare corresponding intervention suggestions. The FSE results, together with the results from relevant literature reviews [14,22] and group interviews, and a theoretical approach (i.e., theory of street-level bureaucracy) will be used to gradually shift from an intensive focus on how de-implementation decisions are made to the potential solutions to improve the process.
The research will be conducted within primary health care in Sweden. Swedish primary health care is part of the tax-funded health care system and is governed by 21 regions . Several different professional groups work at Swedish primary care centers, which makes it a promising environment to conduct this research. Potential respondents will consist of the following professional groups in primary care: nurses, physicians, physiotherapists and psychologists. The recruitment of research participants across the two steps will be performed in collaboration with an R&D partner from Stockholm: Academic Primary Care Center (APC). The APC operates in close collaboration with primary care, as well as universities and research institutes. It aims to contribute to quality assurance and development of primary care for staff and students in the Stockholm Region. This collaboration will ensure that the research meets the needs of the primary care and that the project will be well anchored in the organization.
Additional file 1 presents the STROBE checklist  of items included in this protocol.
A cross-sectional factorial survey experiment will be conducted to investigate the decision-making process of health care professionals concerning potential LVC practices. This approach combines survey research with experimental research [21,25]. Here, the professionals are asked to make multiple decisions based on random generation of the vignettes (i.e., concrete, fictive clinical case descriptions ). An FSE allows analyses of professionals’ judgments and is particularly suitable for investigating complex situations with multiple stimuli. This approach has the unique ability to simultaneously measure the independent effects of respondents and situational factors without the need to give participants every possible combination of variables. This method is widely used to study health care professionals’ decision-making, e.g., [26–29]. FSE is a suitable method to study the often, and somewhat, sensitive issue of LVC because it provides a less personal and therefore less threatening way of exploring the topic . Furthermore, professionals’ decision-making is often challenging to study with conventional social science methods such as interviews and observations . Although the decisions made in an FSE are hypothetical and do not replicate real life, it is considered suitable for providing an in-depth insight into decision-making processes, which this project aims to do. Apart from manipulating the factors in the vignettes, we will collect relevant data about health care professionals, which may affect how they make their decisions. Following the previous literature [14,30], we will consider their length of clinical experience, fear of possible litigation, and a tendency to worry that they may miss a certain diagnosis. For the sensitivity analysis, we will control gender.
Based on a recent scoping review summarizing factors affecting de-implementation , we will identify potential tensions in decision-making processes among health care professionals. Tensions consist of situational attributes that together constitute a potential dilemma for a professional. For instance, a tension can be a situation where a low-value practice is strongly requested by a patient and there are low costs related to providing it, but the evidence is clear that it has no clinical benefits. The factors identified in the literature will be discussed with the R&D partner to validate the ecological validity of the material. Thereafter, we will create drafts for vignettes, i.e., descriptions that comprise a series of sentences in a fixed order containing the relevant dimensions for a certain decision (see Table 1 for examples of dimensions and their values).
The level or presence of a certain dimension (e.g., patient wishes for a certain practice) is randomly varied among the vignettes and respondents (see Figure 1). These vignettes present a story that is relevant for a specific profession, although the tensions included can be similar among professional groups (i.e., patient requests and costs in relation to evidence).
Following the recommendations for FSEs, we will use approximately 7 ±2 dimensions with 2-3 levels for each of these [21,25]. In this way, we will be able to empirically test how differences in the dimensions impact the participants’ decisions (e.g., a low-value practice is strongly OR not at all required by a patient, low OR medium OR high costs are related to providing it, and the evidence is clear OR unclear for the lack of clinical benefits). Judging several similar, but not identical, situations by each respondent allows the FSE to decompose the structure of the individuals’ answer behaviors and thereby uncover the impact of the different factors . The dependent variable will be the respondent’s decision to provide the practice described in the vignette, using a visual analog scale from 0% (extremely unlikely) to 100 % (extremely likely).
The vignette universe will be generated by crossing all of the possible combinations (Cartesian product) of the vignette dimensions’ categories to ensure orthogonality across the factors . The experimental design (i.e., vignette universe) will be divided into different decks (blocks) presented to different respondents. This way, we will be able to use a larger overall number of vignettes to enhance statistical efficiency . For blocking to decks we will use deliberate blocking techniques based on design efficiency (D-efficiency) . A minimum of five respondents will assess each deck . We will use the SAS macro ‘%mktblock’ to ensure a randomized distribution of the whole vignette universe over the decks in order to attain maximum statistical efficiency. By distributing the different questionnaires as evenly as possible we will ensure that the correlations between dimensions are close to 0 and not significant.
The drafts for vignettes will be pilot-tested and further developed in the group interviews with a selection of professionals from the participating primary care centers . Thus, the clinical relevance of the tensions and factors will be further tested. This is a crucial step as the strength of the methodology used in the next project step relies on how well the respondents identify with the situations described in the vignettes . Use of professionals’ expertise is a common way to ensure that the vignettes represent the realities these are portraying and, in this way, increase the validity of the methodology . We will use a purposeful sample that varies in terms of gender and length of profession. This diversity seeks to increase the relevance of the vignettes. The group interviews will start with a presentation of the vignette drafts, which thereafter will be discussed by the participants. The moderator will pose questions about the formulation of the vignettes, their relevance and other potential factors that might impact decision-making. The group interviews will continue until saturation is achieved within each professional group. We estimate that approximately 2–3 group discussion for each professional group is needed, for a total of 8–12 group interviews. To finalize the vignettes, a smaller number of professionals will be consulted for a cognitive pretest based on think-aloud techniques . They will be asked to read the vignettes with the different dimensions and levels to check whether these are easy to read and understand.
Participants and Procedure
First, managers for each primary care center will be approached to anchor the project and to obtain approval for the professionals’ participation. Based on our previous experiences conducting surveys with an organization, we estimate that at least 60% of the centers will participate (i.e., at least 42 centers). The number of professionals at each center varies, commonly from 15 to 20. This estimate would give us a population of professionals that is large enough for the planned survey (at least 630-840 individuals, estimated response rate 60%) , although a more thorough power calculation will be performed when the vignettes have been finalized, i.e., the number of factors and dimensions is determined. SAS software will be used for the power analysis. We plan to use quota sampling as our sampling technique and multiple regression analyses for the data analysis. Although simple randomization is common, the current FSE literature proposes that quota sampling has advantages in understanding confounded parameters. We will, therefore, based on the results of the literature review, investigate whether a specific characteristic in the respondent population (e.g., gender) can be an appropriate factor for stratification in a quota sampling. Otherwise, we will use random sampling and carefully target the potential limitations of this method.
Each survey respondent will answer 9-12 vignettes. This number was determined based on prior research showing that health care professionals are able to provide reliable answers to this number of vignettes [21,26,29]. The number of vignettes will be based on the time the participant is estimated to be able to spend on the survey (max. 15 minutes), the difficulty and complexity of the vignettes and statistical considerations (e.g., power) [27, 30, 33].
Because each respondent will evaluate multiple vignettes, this implies a hierarchical data structure in which the responses (decisions to provide the LVC practice in each scenario) are nested within respondents (healthcare professionals). To address this violation of the classical regression assumption of uncorrelated error terms, mixed-effects models will be performed with vignettes’ dimensions as level 1 and respondents as level 2. First, a random intercept-only model with no predictors (i.e., null model) will be performed to calculate an intraclass correlation coefficient (ICC) and to benchmark model fit. Then, vignette dimensions variables and respondent variables, as well as their interactions, will be added sequentially. In the case of low ICC coefficients, regular multivariable regressions will be performed alongside multilevel analyses for a sensitivity test.
Participants and Procedure
The knowledge gained about how professionals make decisions when dealing with low-value practices will be utilized for creating practical suggestions that can be valuable for key actors involved in health care organizations and steering the de-implementation of LVC. Specifically, we will develop a conceptual framework for professionals’ decision-making concerning de-implementation based on the results from the FSE (Study 1), relevant literature reviews [14,22] and group interviews, and a theoretical approach (i.e., theory of street-level bureaucracy). These sources will be used to gradually shift from an intensive focus on how de-implementation decisions are made to the potential solutions to improve the process. It will be iteratively tested and further developed with the professionals, the local R&D partner, and other researchers in scientific conferences in order to obtain both optimal scientific rigor and practical usefulness.
Based on this framework, intervention suggestions (e.g., checklists and decision support tools) for facilitating professionals’ decision-making concerning the de-implementation of low-value practices will be developed. Suggestions will be relevant for all levels in the health care system (e.g., national, regional and organizational) and will focus on facilitating professionals’ decision to engage in de-implementation. The intervention suggestions will be developed in workshops with professionals and other stakeholders involved in organizations for de-implementing LVC.
The conceptual framework will be presented to guide participants’ efforts to propose suitable interventions. We will use an interactive, 2-step process  in which participants are first asked to brainstorm potential interventions by answering a specific question (e.g., What can facilitate professionals’ decision-making about the provision of LVC?). Thereafter, these suggestions are rated by the participants for their potential impact, feasibility as well as strengths and weaknesses in health care. Based on previous experience with this method , approximately 3–6 workshops will be conducted.