An ecological levels approach (31), as suggested in the IM guide (21), was used to search for important, but accessible actors on various levels (e.g., on the individual level, on the organizational level, etc.). Characteristics and assets were identified which were present in their social, physical, informational and policy/practice environments (31). Next, program goals were defined.

A problem modeling template from the IM guide (21) was used, based on PRECEDE (32, 33), a health program planning model. Although the elements of this model concern public health problems (33), various earlier studies in the field of insurance medicine [e.g., (34–37)] demonstrated that problems could be modeled and addressed with IM. Therefore, “WDPE problems” was used as substitute for the construct “health problems” and “social security outcomes” was used instead of “quality of life.”

Within the planning group, it was discussed what would be the opposite (desired outcomes) of the behavioral and environmental outcomes from the logic model of the problem. These desired outcomes marked the endpoints for the logic model of change. Next, for every actor the detailed behavior [e.g., who, what, when, how often, with whom (23)] was described that should achieve the desired outcomes. Although the clients are important stakeholders and the recipients of WDPE, they were considered not to be actors that should be directly subjected to behavior change strategies within the intervention. Most of their performance objectives (e.g., “client understands what kind of outcome and consequences WDPE may yield”) were thought to be achievable through the IPs. Therefore, they were not shown as separate intervention target in the logic model of change.

The aforementioned description of the desired behavioral outcome was converted into one or more performance objectives (such as “the IP takes a structured, evidence-based approach in WDPE”) and split into their sequential steps (e.g., participating in education, using provided tools).

As determinants, we used the 14 domains of the Theoretical Domains Framework (TDF) (38), because they are consistent with all known behavior change theory constructs (39) and include all key determinants form behavior theories (40–42). These domains (determinants) can be arranged under a uniform overarching model, COM-B (22), meaning that Behavior (B) is determined by the Capacity (C), Opportunity (O) and Motivation (M) for the individual or physician to perform it. The TDF-domains and their overarching COM-B components are illustrated in Supplementary Table S2. For every actor, each determinant’s contribution to WDPE problems as well as to WDPE change was worked out.

For each actor, a matrix of change was made, by crossing the performance objectives with each of the TDF determinants, thus providing change objectives.

The actual content of the WDPE working method was designed in parallel with the intervention program. We started out with the basic principles of EBM (10) and added information from stakeholder input, theories and evidence. These included, for example, the aspects worth considering in WDPE (17), content and design requirements [e.g., (2, 17, 43)] and outcome quality conditions and characteristics [e.g., (8, 16, 44)]. With these requirements and conditions and the findings from the formal IM steps, a concept WDPE working method was designed.

Program themes, components, scope and sequence were gradually specified in planning group meetings and stakeholder contacts. To select the most important determinants (e.g., ‘Knowledge’), it was explored for each of 14 TDF determinants whether a change would result in the desired behavior (23). For example, when the determinant “Knowledge” would change, physicians would know what quality and comprehensiveness is needed for high quality WDPE, how to substantiate their judgment with evidence and which approach to take. The behavior change objective “physician knows the ingredients of a good quality WDPE and is familiar with the way to obtain it” would then be met. The most relevant determinants, in terms of best meeting the change objectives, were selected.

The BCW offered a set of nine intervention functions (e.g., “Education” or “Restriction”) and provided the set of APEASE-criteria (Affordability, Practicability, Effectiveness, Acceptability, Side-effects/Safety and Equity). These can be used to determine which intervention function would best target the found determinants (22).

Next, the same criteria were used to identify which of the BCW’s set of seven policy categories (e.g., “Service provision” or “Guidelines”) would best carry out the intervention functions found (23). Then, behavior change techniques (45, 46) were sought which best targeted the determinants with the intervention functions (e.g., “Feedback provision” could be a technique to improve the determinant “Knowledge” and could be part of the intervention function “Education”). In order to provide a rationale for the problem and the intervention and to guide more detailed design choices, a suitable behavior change theory (45, 47) was searched for, which used the identified behavior change techniques as constructs (e.g., a well-founded theory that explains how “Feedback provision” could bring about a desired behavior change). A detailed scheme of the selection processes for determinants, intervention functions, policy categories, behavior change techniques and theory is depicted in Supplementary Table S3.

By applying the theory to each of the actors, practical applications were deduced. Then, intervention and its components were drafted. During design meetings with the planning group, it was checked whether the change objectives were still met by the various concepts of the intervention.

By considering the ecological levels around the physicians, influencing the WDPE quality, four important actors to address were identified: Physicians (P) performing WDPEs were the main actor on the individual level. On the interpersonal level, the client (C), a claimant undergoing WDPE was an important actor. The organization (O), which contracts the physicians and provides disability benefits and vocational rehabilitation assistance to clients, was identified on the organizational level. And on the community level, the professional community of the physicians (PC) was found as a relevant actor. Regarding the societal level, there was not a specific actor isolated to direct the intervention to. The intervention was developed within the societal level of the existing social security system in the Netherlands. An elaboration on the needs and contexts of these actors is shown in Supplementary Table S4. From this elaboration, program goals were distilled. For example, to ensure transparency and uniformity on the one hand, and to involve unique client and context characteristics on the other hand, an important goal was the development of a stepwise, but client-tailored WDPE working method. The limited time but meanwhile high quality demands, led to the adoption of optimal efficient WDPE EBM support as a goal.

The Logic Model of the Problem is presented in Figure 1. In the right panel, it shows that problems within WDPE could lead to missed chances for vocational rehabilitation, unsatisfied stakeholders and physicians experiencing professional insecurities. These can arise from an unstructured WDPE approach or low quality evidence base for the prognosis assessment. Behavioral and environmental factors from the various levels included, for example, a lack of feedback on the provided quality by the physician. The left part of the table shows the determinants underlying the behavioral and environmental factors.

An ideal outcome would be that WDPE consequences for benefit admission and assistance in vocational rehabilitation perfectly match the client’s health, contextual and personal factors, and that all actors are satisfied with the WDPE process and outcomes. Several requirements of WDPE quality and context can contribute to this outcome. One of these is a stronger evidence base, as this would increase quality and transparency (48). The setting in which WD(P) E is performed should offer an optimal knowledge infrastructure (49), to enable physicians to quickly identify and apply the latest scientific insights into their WDPEs. Also, a comprehensive set of personal or context related factors needs to be taken into account e.g., (17, 50, 51), to ensure optimal tailoring.

To achieve these outcomes, physicians must be regularly trained in EBM, and need to adhere to a stepwise protocol to guide them along the complex WDPE process. The organization needs to support the physicians with time, information, infrastructure, feedback and education. Their professional community needs to stimulate and assist them in knowledge exchange.

The identified performance objectives are shown in short, without all underlying sequential steps in Table 1. Because the intervention will not target the clients’ behavior directly, separate performance objectives and determinants for them are not listed in this table.

The contribution of each of 14 TDF determinants on WDPE problems was worked out as well as what needs to be changed into for an ideal outcome. As an example, the TDF determinant Memory, attention and decision processes refers to the possibility that insurance physicians may lack an overview of relevant prognostic factors in WDPE as they are distracted by organizational demands and time pressure. It should change into physicians easily performing the WDPE steps, because they have acquired a routine, a certain speed and are familiar with its structure. This enables them to cope with and include other stakeholder demands. When crossing each of these determinants with each of the performance objectives, change objectives were obtained. As an example, the TDF determinant Cognitive and interpersonal skills within physicians, when crossed with the performance objective of taking a structured and evidence-based approach in WDPE, resulted in the change objective: “Physician is skilled in following the WDPE working method’s steps, in overviewing WDPE aspects, in using evidence, in weighing information and in communicating a well-substantiated outcome with the client and involved others.”

The logic model of change is shown in Figure 2.

Based on all relevant aspects and needs for WDPE (17, 18), a five-step plan working method, accompanied by a training and a supporting tool was developed to meet the program goals. This working method, “Prognosable,” is shown in Supplementary Table S5. The overview as proposed in the “Prognosable” working method uses the International Classification of Functioning, Disability and Health (ICF) (52) in combination with the construct “Course” to also represent a time frame (26). For the construct of “Personal Factors” within the ICF, a set of ten main cognitions and perceptions influencing vocational rehabilitation previously identified by de Wit and colleagues, was used (53). To obtain a set of the most important aspects concerning “Environmental Factors” within the ICF, the ICF core sets of insurance and occupational medicine and the set from the Social Medicine Work Capacity instrument (51, 54–56) were used.

An overview of determinants, intervention functions and policy categories for each actor, which informed the WDPE method development, can be found in Supplementary Table S6. The importance of Education as an intervention function, combined with the policy category Service provision, led to the addition of an educational program to teach and train physicians in the WDPE method. The importance of Enablement as an intervention function, informed our choice to offer the WDPE method as a supporting software tool. Some derived practical applications regarding the educational program and the software tool are also briefly outlined in Supplementary Table S5.

Since 2007, various studies in occupational and insurance medicine have used IM for intervention development [e.g., (35–37, 60, 61)] and some have used the BCW (e.g. (62, 63),). Some of the studies using IM, reported advantages for intervention development such as meeting the specific needs of the actors and targeting specific behavioral determinants (e.g., “Knowledge”). However, a systematic review by Fassier et al. (34) on studies using IM for developing return-to-work interventions, demonstrated that the effectiveness of the implemented interventions was often limited. They reported that possible causes included a not truly participative planning group, a somewhat randomly chosen behavior change theory and not involving workplace actors in intervention development (34). In our study we therefore chose to include targeted actors in the planning group and to hold stakeholder meetings with others. Also, we chose to use a behavior change theory based on required constructs (covering the identified determinants and behavior change techniques) and specific insights (see Supplementary Table S3). Moreover, because we chose a multilevel approach with the ‘organization’ as an actor in itself, this workplace actor was integrated in the development of the intervention.

Several tools and algorithms exist to calculate prognostic outcomes or to suggest possible treatment interventions for certain prognostic risk factors. Examples in the field of disability evaluations include Louwerse et al. (29, 64), who developed a prediction rule to predict future changes in work ability. De Wit et al. (53, 65) presented an overview of interventions to alter negative cognitions and perceptions for return-to-work. In addition, Hesse et al. (27, 66) developed a list of indicators, which assist in determining acceptance for long-term disability benefits for people with psychiatric illnesses. None of these, however, described a complete working method, a generic work-up for disability prognosis evaluation, such as was developed in this study.

Because difficulties in prognosis assessment included early steps (such as gathering information) as well as late steps (such as a concrete and well-substantiated reporting of the evaluation), the entire prognosis evaluation process was captured within the working method. However, earlier studies on guidelines and prediction rules indicated that physicians tend to fear a loss of professional autonomy when these are too strict, whereas when they are too generic, they are not considered helpful (28). Therefore, the intervention with a stepwise working method should offer guidance and overview, while preserving all individual case-specific characteristics and professional freedom within the steps.

As knowledge was an important determinant, education was an important intervention function. Several other studies in the field of occupational or insurance medicine pursued physicians’ behavior change, for example toward guideline adherence. They all identified needs in knowledge and skills (29, 30, 37, 62), often leading to an educational program as an appropriate, acceptable and feasible intervention component. Moreover, stakeholders in our study mentioned that the only well-implemented interventions targeting physicians within this organization, were delivered through education. In addition, this education would need to be part of the mandatory training program for residents and highly recommended for the registered insurance physicians.

In this study, feedback provision appeared to be a very important behavior change technique, which was included into the educational program by means of exercises. For practicality and cost-effectiveness, part of the program can be offered by means of e-learning. De Leeuw and colleagues (67) described an evidence-and theory-based development strategy for digital medical education, which will be consulted. Its principles such as feedback provision, interactive elements, real-world translation, showing progress, reference provision and clear layout very well matched physicians’ needs and implementation strategies identified in this study. For the EBM education, a practice-integrated teaching method, as Coomarasamy and Khan (68) showed, would have beneficial outcomes on knowledge, skills, attitudes and behavior, whereas a standalone method might only improve knowledge. This will be realized by using realistic or actual cases and by using the working method and tool in the workplace setting.

The working method will be supported by a software tool, which guides the physicians through all WDPE steps. For literature resources, immediately accessible hyperlinks to preselected evidence will be offered, as well as search strategies for additional resources. An underlying database consists of linking tables for diagnostic codes and their search terms with reference sources and for prognostic aspects and reference sources. This will help meeting physicians’ needs in evidence identification and might overcome time barriers (2, 20).

Also, an overview of aspects will be offered, as physicians suggested this kind of support (17, 18), indicating a need for information structuring. A study involving a prediction rule (43) showed that easy graphical representations were preferred design choices, which we will adhere to in the software development.

Although physicians need to make evidence-based considerations [e.g., (49, 69)] and although they received EBM training (13, 19), EBM usage remains limited [e.g., (16, 58)]. They still consider EBM usage difficult (20, 58) and are in need of EBM assistance (2, 17, 18). Moreover, since most EBM questions involved the topic of prognosis (7), EBM steps (such as finding, appraising and applying evidence) were integrally included within the working method and the education. For practice, a digital tool will be provided, with help in acquiring, searching and appraising evidence, such as direct links to relevant selected EBM resources, filters, search strings and appraisal questions.