Comprehensive, longitudinal data material showed that the level of the fidelity of this complex intervention generally was high. A total of 16 of the 18 intervention components were always or most often delivered according to the original plan. However, some non-adherence was also observed, including components that were not delivered, were modified, and were added to the original. The different moderating factors in the Conceptual Framework for Implementation Fidelity all affected the fidelity in a complex, interrelated way. The effects of the moderating factors on fidelity also changed over time, which further illustrates the challenges of evaluating impact of factors influencing fidelity. The Conceptual Framework for Implementation Fidelity [6, 22] was in general found to be empirically useful. The strengths, limitations, and the future use of the framework are discussed below.
Measurement of the four dimensions of adherence (content, frequency, duration, and coverage) included in the Conceptual Framework for Implementation Fidelity was found to be extensive and challenging, but also useful. First, some flexibility existed in the interpretation of the intervention components and delivery descriptions, which complicated the evaluation of adherence. Continuous discussions needed to be carried out in order to clarify each component. Standardization of core components and their delivery has also been emphasized by other authors . It is challenging to describe content and delivery of several components so that no unclearness exists. Perhaps future studies could take into consideration the four adherence dimensions when formulating descriptions of intervention components and delivery. This could help to specify content, frequency, and duration for each component.
The last adherence dimension, coverage, was especially useful in the present study because almost half of the potential participants declined to participate. Many prior studies have not evaluated coverage , which makes it difficult to determine to what population the findings are generalizable. The moderating factor, recruitment, was found useful because it provided information on factors affecting coverage. Another challenge concerning the evaluation of adherence was the fact that no standards exist for what is the optimal degree of adherence. We considered high adherence only when the components were always or most often delivered as planned concerning content, frequency, and duration. There is also no agreement on whether and how to weight fidelity of the different intervention components, i.e., whether high fidelity for core component compensates for low fidelity for less important components. It is recommended that further studies discuss and define acceptable levels of adherence for the four adherence dimensions.
The findings also showed that non-adherence also dealt with components that were added to the model. Therefore, all measurements of adherence, such as fidelity protocols, should also include categories for additional components. Our analysis showed that staff did not reflect and recover components they had added, which could make it difficult to capture these in a protocol or interview. Therefore, it is strongly recommended that observations be used repeatedly to measure adherence and added components.
We found that staff enthusiasm for the project (responsiveness) was high, and this seemed to be a reason for adding components to the intervention. These additional components were in line with the theoretical ideas of the intervention, and no contradictory components were added. It seems that a desire to give the best possible care for the participants was a driving force for adding components. This is in line with Fraser et al. , suggesting that a desire to improve program results can be a reason for local intervention adaptations. Fixsen et al.  highlighted that understanding the principles of intervention core components may allow for flexibility in form without sacrificing the function associated with the components. We also found that some contextual factors in terms of merging services with concurrent projects and additional resources enabled the staff to add components to the present intervention. Thus, contextual factors enabled the additional components, but high staff responsiveness determined that the components were actually added. Staff with lower enthusiasm would perhaps not have added the components although contextual factors made that possible. Some authors have suggested that local additions to an original model tended to enhance effectiveness . Effectiveness of additional components was not the focus of the present analysis, but we suggest that future studies should investigate the possible positive (and/or negative) impact of staff responsiveness and added components on program outcomes.
Contextual factors such as organizational routines were often reasons for not delivering or modifying components. For instance, the formal support for the relatives at the municipality was not targeting relatives of the present project, and therefore no formal support for relatives could be offered. In addition, staff enthusiasm about the project made them add components, but contextual factors such as increased workloads made them remove these in order to focus on the original components. This is a classical situation in organizational intervention research where interventions are not conducted in a vacuum. The longitudinal analysis revealed how the staff strived to strike a balance between resources and workloads on the one hand and staff willingness to deliver high quality care on the other hand. Fixsen et al.  suggested that high fidelity practices is best achieved when implementation is well-supported by strong organizational structures and cultures. This is most certainly valid, but difficult to achieve in practice when dealing with complex organizational interventions during a longer time period. In our case, the project had strong leadership support and the content of the intervention was developed in collaboration with the participating practices in order to develop a program that would suit the local context . Participating organizations change leaders, reorganize their units, and get involved in new projects, and these actions make it difficult to plan in advance. This further emphasized the importance of longitudinal, systematic analysis of implementation fidelity in connection with an intervention study.
We also found that participant responsiveness,' i.e., elderly peoples' preferences, was a reason for not delivering components. The CM was supposed to contact all participants at least once a month to check their status. However, some participants wanted to contact the CM by themselves and as often as they wished. Prior studies have shown that intervention components that are not in line with recipients' wishes are most often not delivered .
Most of the respondents experienced the intervention as a complex program, the description of the intervention as vague, and the initial facilitation as limited. This is in line with prior studies reporting initial confusion in project work . While some described the lack of clear descriptions in the initial intervention phase as frustrating and hindering, others experienced it as positive, because it gave them the possibility for individual interpretations of the intervention. Especially the municipality staff, which had long experience of working in similar projects, reported that they took a more active role and enjoyed the freedom to act according to their own judgment. It seems that the experiences of complexity and lack of initial facilitation did not impact fidelity, which is contradictory to prior studies suggesting that simple interventions and interventions with detailed descriptions are more likely to be implemented with high fidelity . In this study, the staff was highly responsive to the intervention, which may have functioned as a driving force for them to solve complicated practical issues and take a more active role in the implementation. It is possible that unmotivated staff would not have made the same efforts if they had experienced limited facilitation. Some prior studies have reported staff to be more engaged, motivated, and effective when they feel they are exercising their judgment and expertise [40, 41]. With this approach, the staff is not expected to follow process protocols exactly, but rather work according to their own judgments of what fits with the client characteristics and context and the program theory . Implementation components, such as training, need to be standardized, but also flexibly adapted to different provider levels of experience . In line with that approach, our findings suggest that individual and organizational differences in prior experiences and responsiveness to the intervention are important to consider by those delivering an intervention when developing work descriptions and planning facilitation activities.
Our findings emphasize the interrelationship that the moderating factors can have with each other and the fidelity. For instance, staff experiences of prior similar projects were to a great degree affecting their responsiveness to the present project, which in turn influenced preferred level of details in work descriptions and facilitation strategies. Many contextual factors also hindered and facilitated the work of the project staff, while the impact of these factors on fidelity seemed to be modified by the other moderating factors such as staff responsiveness. Fixsen et al.  suggested that the interactive implementation drivers compensate for one another so that a weakness in one component can be overcome by strengths in other components. Based on the results of the present study, it seems that staff willingness to deliver the program with high fidelity and participants' willingness to receive the components were the fundamental conditions for implementation of the program. Factors of particular importance for fidelity were staff and participants' responsiveness to the intervention on one hand, and the enabling and hindering contextual factors on the other hand. Implementation fidelity was shaped by the staff's commitment to the intervention program, as well as their ability to perform its content within the resources at hand. A staff with high responsiveness was also willing to overcome potential obstacles, such as contextual factors. These factors are recommended as first steps for evaluation of factors affecting fidelity. It is suggested that more research is needed for investigating the relationship between the moderating factors and fidelity.
The previously proposed [6, 22] Conceptual Framework for Implementation Fidelity was a useful tool for organizing the data collection of adherence and moderating factors. It covered factors causing non-adherence, suggesting that these factors are comprehensive measures of factors affecting fidelity. However, the framework does not provide any guidance for how to investigate the interrelations between the moderating factors. It is suggested that the framework be further developed or used together with other models to examine the relative impact of the moderating factors on each other and fidelity longitudinally.
The main strengths of the study were the use of three different data collection methods and the longitudinal design. In line with suggestions from other authors [44, 45], the different data sources complemented each other and offered reliable results. The direct observations were especially valuable. A longitudinal analysis allows the researcher to track the development of the program over time, providing a more thorough understanding . Some authors  have suggested that fidelity also needs to be measured in control groups. In the present study there was no possibility that the intervention components could have been delivered to controls due to organizational routines. The control group received care planning at the hospital and did not have any CM or a multi-professional team to contact in the municipality. Thus, after a careful evaluation, a decision was made that the research resources were not to be put into the evaluation of the control group. One limitation is also that elderly participants were not interviewed because their respondent burden was considered too high. Finally, the intervention was conducted in local practice, but in a research context. Thus, it is possible that the factors affecting fidelity in this project are not totally comparable to real-life situations, because support from researchers was offered. Nonetheless, as Dane and Schneider  point out, understanding fidelity under the research conditions is a first step to understanding program fidelity. The next step would be to study the implementation of the intervention after the research program.