Theory-driven development of a medication adherence intervention delivered by eHealth and transplant team in allogeneic stem cell transplantation: the SMILe implementation science project

Background Medication adherence to immunosuppressants in allogeneic stem cell transplantation (alloSCT) is essential to achieve favorable clinical outcomes (e.g. control of Graft-versus-Host Disease). Over 600 apps supporting medication adherence exist, yet they lack successful implementation and sustainable use likely because of lack of end-user involvement and theoretical underpinnings in their development and insufficient attention to implementation methods to support their use in real-life settings. Medication adherence has three phases: initiation, implementation and persistence. We report the theory-driven development of an intervention module to support medication adherence (implementation and persistence phase) in alloSCT outpatients as a first step for future digitization and implementation in clinical setting within the SMILe project (Development, implementation and testing of an integrated care model in allogeneic SteM cell transplantatIon faciLitated by eHealth). Methods We applied Michie’s Behavior Change Wheel (BCW) and the Capability-Opportunity-Motivation and Behavior (COM-B) model using three suggested stages followed by one stage added by our team regarding preparation for digitization of the intervention: (I) Defining the problem in behavioral terms; (II) Identifying intervention options; (III) Identifying content and implementation options; (IV) SMILe Care Model Prototype Development. Scientific evidence, data from a contextual analysis and patients’/caregivers’ and clinical experts’ inputs were compiled to work through these steps. Results (I) Correct immunosuppressant taking and timing were defined as target behaviors. The intervention’s focus was determined within the COM-B dimensions Capability (lack of knowledge, lack of routine), Opportunity (lack of cues, interruptions in daily routine) and Motivation (lack of problem solving, trivialization). (II) Five intervention functions were chosen, i.e. education, training, modelling, persuasion and enablement. (III) Twenty-four behavior change techniques were selected, e.g., goal setting, action planning and problem solving. (IV) Finally, seventeen user stories were developed to guide the SMILeApp’s software development process. Conclusion Our example on the theory-driven development of an intervention module in alloSCT delivered by eHealth and transplant team using a rigorous 3 + 1-stage approach based on BCW, COM-B and agile software development techniques, can be used as methodological guidance for other eHealth intervention developers. Our approach has the potential to enhance successful implementation and sustained use of eHealth solutions in real-life settings.


Background
Allogeneic stem cell transplant (alloSCT) is a wellestablished, potentially curative treatment modality for various malignant and non-malignant hematological diseases in which a patient's diseased blood-producing system is replaced by that of a healthy person [1]. These patients' complex care needs demand innovation in care delivery [2]. As information technology widely available through smartphones has the potential to support patients in adopting and developing essential behaviors such as those surrounding medication adherence [2,3], the National Institutes of Health Hematopoietic Cell Transplantation Late Effects Initiative [2] recommend that related interventions include the support of eHealth solutions.
Medication adherence is defined as "The process by which patients take their medications as prescribed, composed of initiation, implementation and discontinuation" (Fig. 1) ( [5], p. 697). Initiation means that the patient starts taking a prescribed medication. Implementation means the correspondence of a patient's actual dosing compared to the prescription. Implementation 1 errors can take the form of late, skipped, extra, or reduced doses or 'drug holidays' (skipping several doses in a row). If the patient quits the prescribed medication for any reason(s), this is called discontinuation. The time span between the first and last dose is called persistence [5].
Medication adherence is vital regarding clinical outcomes in patients following alloSCT [6,7]: as many as 80% of alloSCT patients develop an acute or chronic Graft-versus-Host Disease (GvHD) [8,9], a serious complication in which the donor's immune cells attack the recipient's body [1] leading to increased morbidity and mortality [8][9][10][11]. Although there is limited evidence and the few existing studies did not distinguish between the three medication adherence phases (initiation, implementation, persistence), we found medication adherence to immunosuppressants in alloSCT essential to prevent or treat complications such as acute and chronic GvHD [12,13]. The correlation is extremely compelling: In patients without chronic GvHD only 15.6% showed medication non-adherence to immunosuppressants while in patients with mild GvHD already 66.7% were non-adherent. In patients with moderate GvHD 74.1% showed medication non-adherence and in patients with severe GvHD even 88.9% were non-adherent [12]. As alloSCT receive their first doses of immunosuppressants during their hospitalization, the initiation phase of adherence is not an issue [6,7]; adherence enhancing interventions need to focus on post-discharge implementation and persistence of medication intake [12]. While available evidence on medication adherence in alloSCT populations is limited, we know that the prevalence of overall nonadherence to immunosuppressants in adult alloSCT patients is 64.6% with 33.3% taking, 61.2% timing and 4.1% dosing non-adherence, 3.2% drug holidays and 3.1% discontinuation [12].
Moreover, compared to the wealth of findings emerging from trials, comparatively little information is available on successful implementation and subsequent evaluation of medication adherence interventions in general and eHealth supported behavioral interventions in particular in real-world settings. A 2018 review on high quality medication adherence intervention trials in different patient populations found that most of them did not report on essential implementation research elements, hindering the adaptation and implementation of trial findings into real-world setting [21]. Another 2018 review paper found 681 apps to support medication adherence in the Apple App and Google Play Stores [22]. However, only 84 (12.3%) of these were developed by or in collaboration with health care professionals, only eight (1.2%) were theory-based, and none reported patient involvement in their development process [22]. And regardless of whether they were developed based on a specific theory, roughly two-thirds contained zero to two BCTs. Of apps including BCTs, 96% included reminders which are often unneeded or unwanted by patients [23]. Some higher-value BCTs such as self-monitoring or feedback on behavior were included in a minority (36-39%), but techniques such as social support or information about the health consequences were very rare (1-2%) [23]. Combined with developers' general omission of end-user involvement (especially from patients) [22,24], general shortages both of theoretical underpinnings [2,25] and of contextual information to adapt eHealth interventions for implementation in specific settings [26][27][28] very likely contribute to two major concerns: 1) More than two-thirds of patients who use health-related apps abandon them within 3 months [24]; and 2) Very few health behavior support interventions are ever successfully implemented in real-world settings [15,18,19,[29][30][31][32].
In order to support post-alloSCT medication adherence effectively and sustainably, eHealth supported medication adherence interventions need to be planned from start to finish with a clear focus on one overarching outcome: implementation in daily clinical alloSCT practice. Accordingly, our major goal with this paper is to report the development of an eHealth facilitated care medication adherence module (implementation and persistence phase) for alloSCT. To do so, we have employed the Behavioral Change Wheel (BCW) and the Capability-Opportunity-Motivation and Behavior (COM-B) modelboth of which, unlike other behavioral theories (e.g. Theory of Planned Behavior and Health Belief Model [33]), provide multilevel perspectives and include emotional factors [34]. In addition, we have developed user stories that facilitate the translation of research findings first to digitalization of the intervention, then to standard practice among both clinicians and patients.

The SMILe project
This work is part of the international, interdisciplinary, multi-phase, multicenter SMILe project, which aims at reengineering follow-up care to patients with alloSCT into an integrated care model first at the University Medical Center Freiburg, Germany and in a second step at the University Hospital of Basel, Switzerland (https:// nursing.unibas.ch/de/forschungsprojekte/forschung/forschung/smile/). The SMILe integrated care model builds on the eHealth Enhanced Chronic Care Model (eCCM) [3]. Its major aim is to introduce follow-up based on a chronic care model by providing self-management support, optimizing delivery system design (e.g., increasing continuity of care), clinical decision support and clinical information systems.
As these functions involve considerable time and expertise, we recommend the introduction of a Carecoordinator (CC) to perform them, along with the use of a SMILe e-platform including the SMILeApp and the SMILeCare monitoring component for the CC. For the Fig. 1 Concept of medication adherence with initiation, implementation and persistence [4], based on [5]. first year post-alloSCT, in addition to connecting patients to the CC in the transplant center, the app will allow patients to assess relevant biomedical and behavioral parameters on a continuous basis.
Four intervention modules are integrated to support self-management and promote alloSCT patients' target health behaviors: Monitoring and Follow-Up, Infection Prevention, Physical Activity and Medication Adherence. This paper reports on the development of the medication adherence module. This patient-centered, theorybased development approach will increase the likelihood of adoption and implementation in practice.

Guiding methodology
Our intervention development process used an iterative approach applying the three stages of the BCW and one additional stage for the SMILe Care Model Prototype Development (Fig. 2), with each stage of execution reviewed by the entire SMILe research team [34,35]. The various stages and steps will be further described in the next section ("The BCW"). Intervention development was informed by data from our previously performed multi-methods contextual analysis and evidence. Contextual analysis refers to the mapping of relevant multilevel contextual factors for the implementation of an intervention, e.g. local infrastructure, leadership, motivation of the stakeholders [27]. This means the inquiry of a specific context (in our case alloSCT follow-up care at the University Hospital of Freiburg, Germany), practice patterns as well as the attitudes and behaviors of all parties involved. Our preliminary work included such a contextual analysis using a cross-sectional quantitative survey of 60 alloSCT patients and five clinicians working in alloSCT. That analysis also included qualitative data, namely individual interviews with another ten alloSCT patients and three alloSCT clinician focus groups. The detailed description of the methods and results of this contextual analysis were recently published elsewhere [36]. It studied all relevant contextual aspects of the proposed integrated model of care, e.g., socio-cultural aspects at the micro (i.e., patient) and meso (i.e., transplant center) levels. Further, we examined the feedback of 21 stem cell transplant patients and eleven caregivers collected during three feedback rounds in Freiburg im Breisgau (Germany). Both, allogeneic and autologous stem cell transplant recipients were invited to give feedback as they participated a shared self-help group. We presented them our preliminary results and asked them to rank e.g. the importance of barriers to medication adherence or whether the proposed BCTs were feasible for them. The patients' and caregivers' feedback was included in the further development of the intervention. In addition, we reviewed existing evidence on 1) definition, prevalence and consequences of medication non-adherence in alloSCT (quantitative, appendix search diagram 1), 2) barriers and facilitators of medication adherence in alloSCT and solid organ transplantation (qualitative, appendix search diagram 2), and 3) existing interventions to support medication adherence in alloSCT and solid organ transplantation (quantitative, appendix search diagram 3). Because evidence in alloSCT is very limited, but medication adherence tasks are similar in alloSCT and solid organ transplant patients, we also included pertinent literature from solid organ transplantation.

The BCW
One particularly rigorous and widely useful system of developing and implementing behavior change interventions is the BCW [37,38]. Combining 19 behavior change frameworks, the BCW is used to identify, understand and explain behaviors and their influencing factors. It consists of three stages: 1) understanding the behavior, 2) identifying intervention options and 3) identifying content and implementation options, which are further divided into eight steps (Fig. 2) [34].

BCW stage 1: understand the behavior
Step 1: define the problem in behavioral terms Before attempting to modify a behavior it is necessary to understand it in detail and to identify possible starting points for change [34]. Therefore, the first step of the BCW is to define the problem in behavioral terms using three leading questions: What is the behavior?; Where does it occur?; and Who is involved?
In order to answer these questions, we first deepened our understanding of the problem by performing a scoping review of studies reporting definitions, prevalences and consequences of medication non-adherence in alloSCT (appendix, search diagram 1). From the twelve studies we identified, we extracted information on definition and relevance of medication non-adherence.
Results. Particularly regarding immunosuppressive regimes we found medication adherence essential in alloSCT to prevent or treat complications such as acute and chronic GvHD [12,13]. However, almost 65% of participating patients struggled with medication intake, especially immunosuppressants taking and timing (implementation phase of medication adherence) [12] ( Table 1).
The contextual analysis data of our preliminary work (see Guiding methodology section) [36] indicated that alloSCT patients were aware of the importance of following the medication regimen closely, especially when family, friends and caregivers were involved in medication management. According to clinicians, medication adherence was not systematically assessed at the studied transplant center. They also acknowledged a need for a person such as an Advanced Practice Nurse (APN) to coordinate follow-up, i.e., a CC [36]. The insights obtained from the literature review and contextual analysis enabled us to define the problem in behavioral terms (Table 1).

BCW step 2: select target behavior
The second step is to create a list of behavioral components known to influence the problem. From this list, behaviors are selected based on their expected impact (when improved) on target outcomes, the likelihood that they can be changed, the extent of expected spillover effects on other health behaviors and their measurability [34].
Results. In the literature search we identified 12 quantitative studies about prevalence and consequences of different behavioral components (appendix, search diagram 1). Based on the information from that literature review, we compiled a list of potential target behaviors that influence implementation and persistence adherence. For example, taking, timing and dosing immunosuppressants correctly (three behaviors), following food considerations concerning immunosuppressants and not taking drug holidays (implementation dimension). From this list, we selected correct taking and timing of immunosuppressants as the two most Table 1 Definition of the problem in behavioral terms

Leading question
Possible answer What is the problem/ behavior?
Medication non-adherence, which is associated with poor clinical and economic outcomes [20,39], pervades all chronically ill patient populations [12,13,29,40]. Post-alloSCT patients are chronically ill. With a mean of 12 medications [41], their medication regimen is complex. Most important medications in the 1st year post-alloSCT are immunosuppressants and anti-infectious medications. Co-medications (e.g., antibiotics, antivirals, fungicide) are essential to prevent and treat infections [42]. Using the ABC taxonomy, medication adherence consists of three interrelated phases: initiation, implementation and persistence [5], the latter two being relevant after alloSCT. Implementation adherence includes correct taking, timing, dosing, no drug holidays and correct food considerations (e. g., no grapefruit juice) [5,12]. While available evidence on medication adherence in alloSCT populations is limited, we know that the prevalence of overall non-adherence to immunosuppressants in adult alloSCT patients is 64.6%: 33.3% taking non-adherence, 61.2% timing non-adherence, 4.1% dosing non-adherence, 3.2% drug holidays and 3.1% discontinuation [12]. Non-adherence to immunosuppressants is strongly associated with GvHD [12]. Less is known in view of co-medication. And while only 57% of the adult alloSCT patients report perfect medication adherence to all prescribed drugs after alloSCT [43], non-adherence can be quite selective, e.g., in one study 17% of subjects discontinued antifungal prophylaxis prematurely [44]. Therefore, it is crucial to optimize adherence, especially to immunosuppressants, in adults after alloSCT (implementation and persistence).
Where does it occur? After alloSCT, non-adherence occurs at the patients' homes and / or where they are at the scheduled time of medication intake.
Our contextual analysis showed that alloSCT patients within the target setting understood the importance of following their medication regimen. According to clinicians, though, medication adherence was not systematically assessed at the target transplant center. If assessed, asking the patients for intake, monitoring blood-levels or checking for rejection signs were reported to be the most used practices. Clinicians also noted a need for a qualified person, e.g., an Advanced Practice Nurse or dedicated CC, to coordinate follow-up [36].
Who is involved? Who is our target group?
The entire health and health-related network surrounding adult patients after alloSCT (family, friends, health care professionals, ...) [36]. Community dwelling adult post-alloSCT patients alloSCT Allogeneic stem cell transplantation, GvHD Graft-versus-Host Disease, CC Care-coordinator prevalent and important target behaviors. This selection was guided by a systematic rationale: changing these behaviors was expected to have the best overall combination of direct impact on clinical outcomes, spillover effects on other behaviors, e.g., taking no drug holidays, and ease of measurement (Table 2). We based our rating system (i.e., ++ very promising, + promising, ± not promising but worth considering, − unacceptable) on evidence from the literature and the research team's clinical expertise.

BCW step 3: specify the target behavior
As the third step is to examine the selected behaviors (in our case, two) both from the patient's perspective and within the context of the surrounding system [34], we specified each in terms of who, what, when, where, how, how often and with whom they occurred. This choice of details was based on previously evaluated evidence from the literature [12,13,41], but mainly on our contextual analysis [36] and our research team's clinical expertise.
Results. The resulting details (Table 3) allowed identification of target behaviors [34].

BCW step 4: identify what needs to change
The fourth step is to identify what needs to change by analyzing risk factors not only at the individual but also higher in the system. For this step, the COM-B model (Fig. 3)the center of the BCW (Fig. 4) supports behavioral analysis very well, as it acknowledges that behaviors such as medication non-adherence are influenced by the physical and psychological capability, physical and social opportunity as well as reflective and automatic motivation. Capability is the capacity to engage in a certain health behavior (e.g., medication intake); opportunity refers to factors external to the individual that make behavior possible; and Motivation refers to the brain processeswhether reflective or unconsciousthat direct behavior. Two common examples of capability are cognitive functionality (psychological capability) and the physical capability to swallow medication. Opportunities may lie in the accessibility of the medication (physical opportunity) or support from a partner (social opportunity); and motivational aspects would include attitudes about the medication, such as beliefs about its efficacy (reflective motivation) or treatment fatigue (automatic motivation).
Of course, capability and opportunity also affect motivation, and the behavior itself influences all three adherence components (Fig. 3). As an ecologic model (i.e., involving factors on the patient, health care provider, health care organization and health care system level), the BCW is helpful in understanding behaviors' determinants and emotional drivers and provides the basis for the following steps.
In addition, the COM-B can be combined with the Theoretical Domains Framework (TDF) [34], which synthesizes multiple behavioral theories and includes 14 domains of behavioral influence (Fig. 4). Each domain represents theoretical constructs such as knowledge, skills or beliefs and relates to one COM-B component. Used together, the COM-B and the TDF allow a behavioral diagnosis, which facilitates the choice of effective behavior change interventions [34].
Where exactly to initiate behavior change was determined partly by the results of the contextual analysis and partly by qualitative evidence. We located the latter via a scoping review for qualitative findings on alloSCT patients' and clinicians' perspectives, barriers and facilitators to medication adherence. Due to limited evidence Performing no drug holidays ++ (↓ risk of GvHD [12]) ++ + (adherence to comedication) ++ [45] Following food considerations ++ (↓ risk of GvHD [12]) + + (considerations for comedication) - [45] EM Electronic monitoring; GvHD Graft-versus-Host Disease (++ very promising) (+ promising) (± not promising but worth considering) (− unacceptable) in the alloSCT population we also included literature from the solid organ transplant population (appendix, search diagram 2).
Results. According to the findings of the 16 identified qualitative studies and the contextual analysis, implementation of medication adherence is influenced by a broad set of multilevel factors, which we organized within the COM-B model (Table 4). For instance, lack of knowledge and lack of routine (Capability), lack of cues and interruptions in daily routine (Opportunity), lack of problem solving, trivialization and denial (Motivation) were identified as possible barriers to medication adherence implementation.
We crosschecked and compared our review's results with those of two recent quantitative studies on multilevel determinants of medication non-adherence in heart transplantation [62]. In addition, we analyzed data from a long-term bio-psychosocial follow-up in solid organ transplantation [64] in which our research group was involved. The results supported our previous findings.
According to the findings of the 16 identified qualitative studies [48-61, 63, 65], the results of our contextual analysis and patient feedback, we selected the highest-priority barriers. The behavioral diagnosis findings for each domain and necessary changes are listed in Table 4.

BCW stage 2: identify intervention options BCW step 5: identify intervention functions
After identifying the target needs, the relevant intervention functions (Fig. 4) must be identified. The BCW lists nine of these (education, persuasion, incentivization, coercion, training, restriction, environmental restructuring, modeling, and enablement), each of which is a set of effective intervention categories with the potential to mitigate barriers to behavior change identified by the COM-B and/or TDF models.
To guide our selection, we applied the APEASE criteria (test of Affordability, Practicability, Effectiveness and cost-effectiveness, Acceptability, Side effects/safety, and Equity) [66] on each intervention function ( Table 5). The identified components were rated (i.e., ++ very promising, + promising, ± not promising but worth considering, − unacceptable) with consideration for available evidence, the research team's clinical expertise, our contextual analysis and patient feedback. The APEASE criteria were also applied in further steps, always using the same data sources mentioned here. To search for existing interventions that support medication adherence in alloSCT, we performed a systematic literature search (appendix, search diagram 3). Due to limited evidence in the alloSCT setting, we also included findings from the solid organ transplant population.
Results. According to eleven quantitative studies we identified on medication adherence-enhancing interventions, combinations of education, training, enablement, modeling, environmental restructuring and persuasion were effective functions to support medication adherence [15,[18][19][20]. After considering the feasibility of each according to our contextual analysis, we excluded environmental restructuring, as this was not feasible within the SMILe project's context. We then presented the selected functions to alloSCT patients and caregivers to discuss their feasibility and usefulness. The research team applied the APEASE criteria to the functions favored by the patient and caregiver group. They found that education, training, enablement, modeling and alloSCT Allogeneic stem cell transplantation Fig. 3 The COM-B framework to understand a behavior [34] (open access figure) persuasion to have the greatest potential to improve medication adherence in post-alloSCT patients ( Table  5).

BCW step 6: identify policy categories
In this step, policy categories are identified (Fig. 4). The BCW lists seven such categories (communication/marketing, guidelines, fiscal measures, regulation, legislation, environmental/social planning, service provision), from which researchers can choose those that will best support delivery of the intervention functions selected in step 5 [34]. We based our selection of policy categories on the results of our contextual analysis.
Results. After applying the APEASE criteria once again we selected two policy categoriesregulation and service provisionas those most congruent with our chosen intervention functions (Table 6).

BCW stage 3: identify content and implementation options BCW step 7: identify behavioral change techniques
The seventh step is to identify BCTs, which are the active elements of an intervention. The BCT Taxonomy describes 93 BCTs, classified in 19 categories (e.g., goal setting, action planning, problem solving, information about health consequences, self−/monitoring of behavior, self-reward, punishment) to change a behavior. Thanks to the BCT Taxonomy, uniform terminology can also be used to describe intervention components, allowing replication of a study [34]. Based on the findings of the previous BCW steps and the data sources explained above, we selected the BCTs that we judged would best support our chosen intervention functions.
Results. In the literature search, we identified 11 quantitative studies on interventions (appendix, search diagram 3). According to the results, cognitiveeducational interventions (e.g., information and instructions) were frequent, but showed inconsistent results in enhancing medication adherence [18,20]. Behavioral interventions (e.g., counseling, reminders, self-monitoring and feedback on medication intake) led to significant improvements in medication adherence [15,[18][19][20]. While combinations of cognitive-educational, behavioral and psychological-affective interventions showed the best improvements, even these results were inconsistent [18,19]. To select suitable BCTs for our intervention, we applied the APEASE criteria to those we judged most promising (Table 7).
Based on these findings, we prepared a draft of the intervention, i.e. visualized the selected BCTs (Table 7) including a short description and a self-drawn representation of an app on a PowerPoint presentation. We showed the presentation to stem cell transplant patients and caregivers and discussed the various parts' feasibility and usefulness. The participants had the possibility to vote for the different BCTs via a real-time smartphonebased survey (using mentmeter.com) as well as via the accompanying oral group discussion. The patients' feedback indicated that they saw the intervention as both feasible and supportive. With consideration for all of the   information thus far compiled, we selected 24 patientlevel BCTs, including goal setting, action planning and self-monitoring of behavior. All selected BCTs are listed in Table 8.

BCW step 8: identify mode of delivery
The eighth and last of the BCW's core processes is to identify the most suitable mode of delivery for the selected BCTs. This determines how the intervention should be delivered to the alloSCT patients and their caregivers, i.e., the end-users. A wide variety of delivery modes can support an intervention's implementation and effectiveness, ranging face-to-face individual or group sessions to mobile phone apps [34]. The optimal delivery mode was determined using information generated from the previous steps, the contextual analysis and end-user feedback [34].
Results. Our contextual analysis [36] showed that alloSCT patients were open to technological assistance but emphasized that eHealth support should not replace personal contact with the health care team. Accordingly, of all available possibilities for face-to-face (delivered by the CC) or distance (technology assisted) interventions, we selected the most suitable mode of delivery for each of our intervention's BCTs ( Table 8).
The resulting draft of the SMILe Care Model Prototype integrates a CC (an APN with specialization in oncology) and the SMILeApp ("prototype" means the whole care model, not only the app component). The BCTs to support medication adherence will be delivered to the patients during two face-to-face visits with the CC, e.g. demonstration how to perform the behavior (e.g. prepare the medication) or habit formation. Depending on the patient's condition, the care protocols allow for a step-up approach including more intensive tailored interventions. Between face-to-face visits, we planned to support patients with several BCTs delivered by the SMILeApp, e.g., selfmonitoring of the behavior ( Table 8).
The intervention draft was discussed with alloSCT patients and caregivers to prioritize functionalities. This process will later guide the order of the software and intervention development. Patients agreed unanimously that the SMILeApp should ideally include a current and complete medication plan that could be automatically updated after every change to the medication regimen. Patients considered this as a feasible way to confirm their medication intake after every intake. Most participants also considered it acceptable and helpful to receive a reminder for data entry once a day at a user-defined time, with graphical feedback for entered values.
All findings arising from steps 4 to 8 are presented in Table 8 which also indicates which COM-B and TDF domains are influenced by each intervention function and policy category, as well as the resulting BCTs and modes of delivery.

Additional stage 4: SMILe care model prototype development
The translation of the SMILe intervention into an eHealth solution will enable delivery of important BCTs, e.g., self-monitoring and feedback on target behaviors to patients anywhere (e.g., in their homes or workplaces) at any time [2,3]. Therefore, after working through the three stages and eight core steps of the BCW, we  2). Developed to facilitate early integration within the target setting, a highly iterative methodology and the inclusion of end-user feedback in the development process ensures the software's usefulness to patients, caregivers and clinicians alike [35].

Additional step 9: prepare digitalization
To prepare for digitization, step 9 deals with the formulation of user stories [35]. User stories are a means to capture requirements that should be delivered by a software product. This makes their stories a valuable basis for discussion between software developers and intervention developers [35]. User stories are an opportunity for researchers to provide key information to software developers about important functionalities that are needed in the app. They will not directly appear in the app, only by means of the functionality implemented. The goal of user stories it to provide key information between researchers and software developers and, thus, are not communicated to patients. While user stories can be structured in numerous ways, the role-feature-reason format is most popular [35]. This begins with a structured sentence (As a … (person, e.g., allo-SCT patient), I want … (action, e.g., keep track of my medication intake), so that … (expected outcome, e.g., I know if I take my medication correctly)) describing a possible software function. Structuring the essential elements in a standard form eases the translation of the corresponding BCTs into the SMILeApp [35]. The concrete realization of a user story is decided in close consultation with the software development team.
Results. To ensure the translation of the most effective BCTs into our eHealth component, we wrote 17 user stories to the software developers in the role-featurereason format based on the previous findings (Table 9). Following the principles of agile software development, continuous end-user testing and user-centered design, the needs and priorities these stories suggest will help the designers first to develop mock-ups of new modules for early user testing, then to ensure that later versions of the app meet user needs as fully as possible [35]. A separate paper will describe the development of the technology aspects in detail (in preparation).

Additional step 10: prepare CC intervention
Based on the previous steps of the BCW, we wrote a comprehensive intervention protocol for the intervention's face-to-face CC visits. This protocol describes every face-to-face visit in terms of when, by whom and to whom it should be delivered (patient ± caregiver(s)), as well as which BCTs should be applied in what order   and which methods should be used. To ensure standardized, reproducible intervention delivery, descriptions and terminology adhered closely to the BCT taxonomy. The intervention protocol was written in alignment with implementation science lens, meaning, that the intervention has been designed using the previously described implementation science methods (e.g. contextual analysis, stakeholder involvement) in such a way that it can be implemented and used in real world settings in the future.
Results. The complete protocol consisted of 70 pages for all modules with the detailed description of the content to be used for education, training and supervision of the CC. We also wrote a short version with 49 pages for all modules to be used by the CC during the face-toface visits as a checklist.
We initially planned to implement and test the whole medication adherence module via both, the CC (face-to-face visits) and the SMILeApp in an RCT. However, due to lack of time resources, the current version of the SMILeApp does not include medication adherence support yet. For this reason, in the current RCT, which started at the beginning of 2020 at the University Hospital of Freiburg im Breisgau (FiB), only the faceto-face visits of the medication adherence module are implemented and tested. This downsized version of the care model is called the SMILe-V1 Care Model Prototype-FiB as it is tested at the University Hospital of FiB. Our prepared user stories will be prioritized and translated into the next version of the SMILeApp to be implemented and evaluated in combination with the CC's face-to-face visits in an RCT at the University Hospital of Basel (USB). This full version of the care model will be called the SMILe-V2 Care Model Prototype-USB.

Discussion
To our knowledge, this paper is the first to provide an example of how to develop a theory-based medication adherence intervention for translation into an eHealth system. Our work, embedded in an implementation science approach, applied the principles recommended by the BCW, and particularly of the COM-B model at its center. This framework's major strengths are its Goals will be set together with patient to take the medication correctly with a deviation < 2 h.
As a patient I want to be reminded of my set goals (which were set during visit) on a self-determined interval (e.g. daily / once a week) so that I am aware of my goal and know what to target at.
1.4. Actionplanning Patients will be encouraged to prepare a plan how to deal with barriers (e.g. have a travel set of their medication prepared and to take it with them when leaving home).
As a patient I want to have reliable information how to plan my expected actions (e.g. leaving home, travelling, eating outside) so that I do not forget the necessary preparations.

Self-monitoring of behavior
Instructions on checking the medication plan daily and confirming medication intake daily.
Instructions on how to control this monitoring on their own.
As a patient I want to self-monitor whether I take my medication as prescribed so that I know whether I take the drugs correctly.

Feedback on behavior
Give patients feedback about on how many days they correctly managed their medication intake.
Give patients feedback about on how many days their time of medication intake was correct.
As a patient I want to get feedback whether I take my medication sufficiently as prescribed so that I can be sure that I take the medication correctly

Instruction on how to perform behavior
Training on how to read the medication plan and prepare their medication correctly accordingly.
As a patient I want to know how to prepare my medication so that I can do it on my own correctly

Information about health consequences
Oral and written information about effects, side-effects of medication.
Oral and written information about consequences of non −/adherence.
As a patient I want to find information on what my medication is for and will happen, if I do (not) take it as prescribed (incl. wrong time) so that I know the importance of doing it correctly.

Demonstration of behavior
Demonstrate how to read medication plan and prepare the medication.
As a patient I want to get explained how to use the electronic medication plan so I can check when I have forgotten.

Electronic prompts/cues
Storage of medication in clear visible places (e.g., next to the coffee machine, TV), to prompt medication intake. Use of an electronic reminder (for medication intake, goal) on the App with a preferred signal (e.g., alarm tone, picture).
As a patient I want to get a reminder when I need to take my medication and don't forget to take it. As a patient I want to customize my app (e.g. different tones, signals, colours, pictures) so that I connect medication intake with a positive feeling.

Problem solving
Identify together with patient what could be barriers to take the medication correctly, and discuss ways in which they could help overcome them. Learn about most common side-effects and fitting interventions to take.
As a patient I want to have the opportunity to call a qualified health care provider if there are unexpected barriers (which were not discussed face-to-face) so that I get support in challenging situations. As a patient I want to be able to signal the CC that she/he should call back when she/he has time so that I get help without disturbing the CC in an unsuitable situation.

Discrepancy between behavior & goal
Point out if the recorded number / time of medication intake does not fit to the goal set.
As a patient I want to get a signal if the recorded number / time of medication intake does not fit to the goal set so that I realize that I have to change my behavior.

Electronic prompts/cues
Medication plan in app As a patient I want to have my medication plan in the app so that I can look up my medication on my smartphone.
As a patient I want to be able to update the medication plan in the app when the prescription of physician changes so that I have a current medication plan on my smartphone.

Message given by followed person
Present a speech / statement / message given by followed patients or recognized transplant professional to emphasize the importance of taking the medication correctly at the correct time.
As a patient I want to learn from a qualified health care provider or a peer why the medication intake and timing is so important so that I am aware of its importance.

Written persuasion about capability
Motivation when confirming medication intake / closing the app / pop-up As a patient I want to get a motivational feedback that I can successfully perform the behavior so that I feel capable to manage the correct medication intake.

Focus on past success
Explore with patients difficult circumstances in which patients nevertheless managed medication intake.
As a patient I want to be able to record occasions with correct medication intake in the App so that I feel confident to be successful again.
multilevel perspective and the explicit inclusion of emotional factors, which tend to be less prominent or absent in other behavioral theories [34]. While other authors have used the BCW to develop eHealth facilitated behavior change interventions, none have described their development process (including the translation of an intervention into an eHealth application) in this detail, and many have not included patients or incorporated them only at a late stage of intervention development [37,38,67]. Worldwide, health care services and providers are moving in the direction of digitization [68]. The number of newly released health related apps is currently growing by over 200 per day: between 2015 and 2018, the number available in the top app stores almost doubled to nearly 320′000 [24]. Many aim to support patients in medication adherence [22] and some have showed promising results in RCTs in various populations [14,16,17]. In the development of an eHealth facilitated intervention, it is crucial to rely on the theory, consider the most recent evidence, integrate information on the context where the intervention or app will be applied and involve all major stakeholdersespecially patientsfrom the earliest stages [24,28,69,70]. And while theory-based interventions are most likely to improve medication adherence [28,69], extremely few apps are developed following these recommendations.
To exacerbate the problem, producers typically supply limited or no information on the processes either of their apps' implementation or of their results in real-life settings [16,29]. This paper addresses these shortcomings using the example of the SMILe integrated care model's medication adherence intervention module. Unlike many existing eHealth tools, our intervention development incorporated extensive end-user involvement (i.e., of patients, caregivers, clinicians), enhancing its relevance regarding implementation in real-life settings. Our goal was not simple to develop the best possible intervention to provide the basis for digitalization and medication adherence interventional components, but also to maximize the likelihood of its successful implementation and sustainability in daily clinical alloSCT follow-up care.
This paper also presents a first step towards bridging the gap between trial and real-world contexts in the development and implementation of a medication adherence module: it describes how to combine empirical evidence with contextual data [36] as the foundation of an eHealth facilitated intervention. The involvement of all relevant stakeholders was indispensible for this module's durable implementation into clinical practice [27,34].
In addition, by developing user stories based on enduser needs and feedback, we combined the central principles of Implementation Sciencesensitivity to context, building on an existing evidence base, extensive stakeholder involvementwith the principles of agile software development and user-centered design [35,70]. With the resultan iterative, incremental, usercentered approach with early connection to the target contextwe intend to speed the translation of cuttingedge research findings into routine use not only by clinicians but also by the target (alloSCT patient) population [24,70]. By involving the various stakeholder groups at each appropriate stage, we ensured that the intervention would fit the needs of end-users and be feasible for sustainable use in clinical practice. This fact alone sets our eHealth tools apart from others, the vast majority of which are typically developed by software designers with little or no input from health care research teams [24].
The BCW is a relatively novel multilevel behavioral framework that both explains and provides a stable framework for intervention development. Building for our work upon the evaluation of correlates/determinants of medication adherence intervention development, not only does it include cognitive patient-level factors, it also explicitly includes emotional factors that are not prominent in currently prevalent health behavior models, e.g., the Integrated Model of Behavioral prediction [34,71].
By assisting in the selection of appropriate behavior change interventions [34], the BCW provides invaluable guidance for intervention development. And by drawing from diverse information sources it allows the combination of quantitative and qualitative evidence, the results of contextual analysis and stakeholder involvement [34], and the researchers' clinical expertise. The full range of these sources informed our intervention development.
Integral to the BCW, the BCT taxonomy [72] provides standardized language to label even the smallest units of behavioral change interventions. In addition to enhancing the reporting and communication of complex interventions' content and enhancing their replicability, this level of standardization facilitates meta-analysis, especially as it is used to detect the best-performing intervention components.
Although the BCW is arguably an excellent foundation for intervention development, we added one stage to it at the end of our development process. To speed and simplify translation of the medication adherence module into an eHealth application, we recommend and describe the formulation of user stories as a bridge between intervention developers and software developers. Congruent with the BCW's aims, agile software development focuses both on stakeholder involvement and on the adaption of the software according to the needs of its end-users [35].
One notable challenge regarding the use of the BCW is that, even while its developers provide a step-by-step process for intervention development, they tend to provide only brief descriptions of the links between those steps, which can make following their recommendations rather challenging especially for those who use the framework for the first time. Regarding applying COM-B to our specific research question, we faced the issue that there is very few evidence regarding medication adherence in alloSCT. Therefore, we had to expand our searches on similar populations. Additionally, following the steps left us with more potential options (i.e., BCTs) than could possibly be realized from a resource and logistical perspective. Future refinement of the BCW could help to overcome this challenge by providing more guidance on which of the possible BCTs would be most appropriate. In our example we based the selection and reduction to the most important and promising options on the feedback of the stakeholders to promote feasible implementation: We applied the APEASE criteria to each possible BCT and discussed which possible app functionalities would specifically help stem cell transplant patients and their caregivers with decision-making and prioritization. This approach ensured that the intervention would fit the needs and preferences of the end-users.
Application of theory is facilitated by examples. While our example focuses directly on the development of behavioral intervention and eHealth app, it and its underlying theory can also be applied to the adaptation of existing interventions and apps to new contexts. Nevertheless, a strong knowledge of the theory is crucial and needs to be firmly in place before starting the development process. To bring all the needed competencies and perspectives to the table, an interdisciplinary team and stakeholder involvement are essential.
Our international multidisciplinary research team consists of 31 researchers and clinicians, two of whom were members of the BCW development team for the medication adherence module. Several of our researchers also went to London to follow summer courses offered by the group led by Susan Michie, who developed the BCW. In all, development of the SMILe intervention modules required an investment of more than 1 year. Therefore, we advise other researchers that, to prepare adequately for this task, they should combine readings of papers and books with formal training. Another valuable option would be to collaborate with more experienced researchers to peer support this process.
It is still unknown, whether a theory-based developed intervention module such as ours is able to improve medication adherence in alloSCT. In addition, it is unclear whether an analogue or digital intervention, or even the combination of both is the most successful which needs further investigation. Our study's results will support agile technology development of the SMILe medication adherence module as part of the SMILe integrated care model. The SMILe-V1 Care Model Prototype-FiB (without medication adherence in the SMILeApp) and SMILe-V2 Care Model Prototype-USB (with medication adherence in the SMILeApp) will subsequently be implemented and tested as part of an implementation science study to address these knowledge gaps (https://smile.nursing.unibas.ch/). Once the SMILe Care Model Prototype has proven its effectiveness, it can be sustainably implemented in clinical practice as it was developed under consideration of implementation science aspects (e.g. feasibility and acceptability).

Limitations
Our development of the eHealth component of our care model was subject to certain restrictions. For example, as contextual factors such as data security legislation are very strict in Switzerland and the EU, certain proposed app functions could not be included. Even if qualitative literature and the involved patients had certain priorities regarding the functionality of the SMILeApp, other factors such as the legal framework could force different sequences of the app development process. However, the whole medication adherence module is more than just the eHealth part, why a first testing of the intervention is still possible.

Conclusion
As intended, this paper describes the theory-driven development, based on the BCW, of a medication adherence intervention module as part of the SMILe integrated care model for alloSCT patients, which includes a patient-centered app and introduces one new care team role: a CC. The associated study's results are currently in use for agile technology development, employing a user-centered design approach, of the innovative SMILe integrated care model's medication adherence module. The overall care model will be implemented and tested as part of a planned implementation science study. While the methods described are applied to this particular medication adherence intervention, as they follow BCW recommendations, they can be adapted to the development of a wide range of behaviortargeted eHealth applications.