Looking into mechanisms of change: Evaluation of a multifaceted implementation strategy to change pain management behavior of care workers

Background Underutilization of evidence-based pain management in nursing homes is common. Evidence towards effective approaches to improve adoption of evidence-based practices in nursing homes is limited. To get a better understanding of the impact of our multi-faceted implementation strategies, care worker training workshops and the introduction of trained pain champions, this study explored the underlying mechanisms of the implementation strategies using behavioral theory. We conducted a mixed-methods evaluation alongside an implementation- effectiveness study in four Swiss nursing homes. Based on an a priori contextual analysis in the participating homes implementation strategies were developed. Furthermore, we developed a conceptual framework describing hypotheses concerning determinants of implementation and mechanisms of change underlying our implementation strategies. Care workers’ questionnaire surveys were conducted at baseline (n=136), after three (n= 99) and six months (n=83) to assess self-ecacy in pain management and self-reported guideline adoption. We computed linear mixed-effect models to assess changes over time in self-ecacy and logistic regressions to assess associations between self-ecacy and guideline adoption. Concurrently, we conducted focus groups with care workers (n=8) to explore their response to the implementation strategies and to gain a deeper understanding of the potential mechanisms. After transcription, interview data was analyzed using content analysis. Overall, there was a signicant increase in self-ecacy after three and six months (p<0.001). Self-reported adoption of guideline components ranged between 44% and 73% depending on the component. We found signicant associations between self-ecacy and adoption of two guideline components, i.e. performing a comprehensive pain assessment and using an observational pain assessment tool in cognitively impaired residents. Qualitative ndings showed that the training workshops and pain champions were received positively by care workers. Focus group participants reported to be more attentive to residents` pain experience and to assess and document pain more frequently and with more detail than before. Our ndings highlight the importance of continuous commitment of an implementation facilitator, e.g., a pain champion, within an organization. Regarding persistent implementation challenges, a theory-based conceptual model contributes to the overall understanding. to understand the survey questions. We conducted a questionnaire survey collecting data at baseline (T0), three (T1) and six months (T2) after start of the intervention, lasting from November 2017 to November 2018. Local coordinators (e.g., director of nursing) were responsible for distributing questionnaires in the NHs. Participation was voluntary and anonymous. A pre-stamped envelope was provided with each questionnaire to ensure condentiality.

Generating and testing of theory in the development and evaluation of implementation strategies contributes to an improved understanding how change can be affected in a speci c context; This study showcases a theory-driven mixed-method evaluation approach of an implementation intervention highlighting different responses to the implementation strategies between certi ed and uncerti ed care workers Background Implementation of evidence-based interventions in health care organizations has been recognized to be a challenging endeavor (1). To increase the uptake of new practices, current literature recommends to systematically select and tailor implementation strategies with regard to needs of the implementation context (2). An a priori analysis of the context to identify barriers and facilitators to use the new practice, is a pivotal rst step to inform the development of appropriate implementation strategies (3). Although there is some evidence that strategies tailored to determinants of change are more likely to change practice (4), little is known about the mechanisms of how implementation strategies affect change in practice (2,5,6).
To facilitate understanding of mechanisms of change in a speci c context it has been recommended to specify implementation strategies with regard to their active components. In a next step, hypothesized mechanisms of change can be generated based on an existing program theory (5). To date, implementation science literature is vastly lacking theory about underlying mechanisms of implementation strategies, only few studies tested mechanisms empirically (6,7).
To advance the theoretical understanding of mechanisms, a current synthesis of 277 behavior change interventions used constructs of behavioral theories to link mechanisms of actions to behavior change techniques identi ed in the included studies (8). The most frequently identi ed mechanism "Beliefs about Capabilities", originates from Bandura's theory of Self-E cacy (9). The theory describes "mastery experience", "vicarious experience", and "verbal persuasion" as strategies to improve self-e cacy, which can be translated into behavior change techniques (8). Implementation strategies that incorporate these behavior change techniques, e.g., modeling, monitoring and feedback on the behavior, hence might be able to increase self-e cacy related to the intended behavior.
This study reports the evaluation of a multifaceted implementation strategy comprising interactive training workshops and the introduction of trained pain champions to facilitate the adoption of a pain management guideline in nursing homes (NHs).
In the eld of nursing home care, pain management is a critical topic with an considerable knowledge to practice gap (10). Although international guidelines for geriatric pain management are available, their adoption into daily practice of NHs is often inadequate (11,12). Internationally, between 40-85% of NH residents report pain (13,14). Insu cient pain management can result in severe consequences for NH residents' health and quality of life (15). Residents with cognitive impairment are at particular risk for unrecognized and undertreated pain due to their inability of communicating pain (16).
In a rst phase of this project we conducted a comprehensive analysis of the implementation context to identify determinants of behavior change with regard to pain management practice (XXXblinded for review). Based on these ndings we developed a multifaceted implementation strategy and tested it in a sample of NHs showing that our approach was effective in improving NH residents' pain outcomes (XXXblinded for review). The overall aim of this mixed-methods study was to gain an understanding of the underlying mechanisms of the multifaceted implementation strategy with regard to behavior change in care workers. Our speci c aims were (1) to determine changes in the hypothesized central mechanism, care workers' self-e cacy, in pain management; (2) to assess associations of self-e cacy and care workers' self-reported adoption of the pain management guideline and (3) to explore care workers' response to the implementation strategies with regard to pain management practice.

Conceptual model
In the planning phase of this study we developed a conceptual model, hypothesizing how our implementation strategies might affect change in pain management practice. To underpin our hypothesis we de ned the active components of the strategies using the behavior change taxonomy (17). With regard to the central strategies, training workshops and pain champions we hypothesized that 'demonstration of the behavior', 'verbal persuasion about capability' and 'feedback on the behavior' would increase self-e cacy as suggested by Bandura's theory of self-e cacy (9) ( Table 1). In relation to his theory, we assumed that in a rst step it would be necessary to increase general awareness and address negative beliefs towards pain in older people. Based on this precondition, an increase of self-e cacy in pain management can be achieved and ultimately lead to the adoption of the guideline in daily practice as depicted in Figure 1.
Design A mixed-methods evaluation using quantitative and qualitative data from care workers participating in an implementation-effectiveness study (hybrid II) to improve pain management in NHs.

Sample/ Setting
This study is part of an implementation study which was conducted in a convenience sample of four NHs located in the German-speaking part of Switzerland.
All institutions belong to a privately-owned NH group, which is part of a large European operator of long-term care facilities.

Intervention
A protocol for this study describing the intervention and implementation strategies in more detail was published previously (XXXblinded-for-review). We developed a pain management guideline based on international recommendations for the management of geriatric pain as in Switzerland currently no national guideline for the management of geriatric pain is available (11,12,18). The nal guideline was adapted to the local context in collaboration with the participating NHs. Core components targeted in this study comprised (A) comprehensive pain assessment; (B) use of observational pain assessment scales, e.g. Pain Assessment in Advanced Dementia Scale (PAINAD) (19) for residents with severe cognitive impairment; (C) pain assessments and re-evaluation after treatment on a regular base and (D) standardized documentation.

Implementation strategies
The implementation strategies have been developed based on a comprehensive contextual analysis involving perspectives of multiple stakeholders (XXXblinded-for-review). An overview of the implementation strategies is displayed in Table 1. We itemized the overall implementation strategies into discrete strategies according to the Expert Recommendations for Implementing Change (ERIC) (20). Further, we determined corresponding behavior change techniques and their hypothesized mechanisms of change (17).

Quantitative part
Sampling and data collection For our quantitative data collection, we included care workers from all educational backgrounds, i.e. registered nurses (RNs), licensed practical nurses (LPNs) and nursing assistants (NAs) who worked in direct resident care, had been employed for at least one month and were su ciently uent of German to understand the survey questions. We conducted a questionnaire survey collecting data at baseline (T0), three (T1) and six months (T2) after start of the intervention, lasting from November 2017 to November 2018. Local coordinators (e.g., director of nursing) were responsible for distributing questionnaires in the NHs. Participation was voluntary and anonymous. A pre-stamped envelope was provided with each questionnaire to ensure con dentiality.

Variables and measurement
Self-e cacy in pain management was assessed at all data collection points with a self-developed 13-item scale. At the time of data collection, no measure assessing this construct was available. In developing the scale we followed guidance on creating self-e cacy scales (21). Care workers were asked to rate items assessing how con dent they feel in tasks related to pain management, e.g. pain assessment, documentation and non-pharmacological treatment on an 11-point rating scale (0= not con dent at all -100= very con dent). An acceptable value for the scalesreliabilitywasestablished(Cronbachs α = 0.94), for evaluation a mean score of all 13 items was calculated.
A dichotomous indicator of self-reported adoption of each of the four guideline components was built from survey items assessed in the third data collection (T2). A detailed description of the indicator was provided in our previous study (XXXblinded). Further items of the care workers' questionnaire assessing sociodemographic data comprise age, sex, educational level (RN/LPN or nursing assistant), years of work experience and tenure in NH.

Data analysis
Descriptive statistics were computed to explore means, medians, distribution and con dence intervals of the data. To determine changes in self-e cacy over time we used unadjusted linear mixed-effect models. Indicators for the time of data collection (T0, T1, T2) were added as xed effects. To account for the nested data structure, care workers nested in NHs and the repeated measures of individuals, we added the NHs and individual IDs as random effects. We conducted a sensitivity analysis based on a sample of care workers who have been participating in all three data collections.
To describe the self-reported adoption of core components, we built different sub-samples, considering only data from respondents who indicated to have experienced relevant situations corresponding to the core component. Further, for the indicators "comprehensive pain assessment" and "use of PAINAD", we excluded data from nursing assistants since these components are not part of their scope of practice (XXXblinded for review). In a next step, we determined associations between adherence to guideline components and self-e cacy items by calculating several simple logistic regressions based on these subsamples.
Statistical analysis was performed using R statistical computing software (22). Linear mixed-models were computed with the LME4 package (23). Statistical signi cance was assigned at the P < .05 level.

Data collection
Qualitative data was collected via focus groups conducted three months after the start of the intervention with a convenience sample of care workers in each participating NH (March-August 2018). The semi-structured interview guides included questions towards the experience of the training workshops and pain champions and perceived changes in pain management practice. Focus group discussions were moderated by the rst author (XXXblinded-for-review), additionally a research assistant took notes of the discussion and summarized main points subsequent to each topic to check with the participants. Interviews have been audio-recorded and lasted between 45 and 60 minutes each.

Data analysis
After verbatim transcription of the audio data, transcripts were re-read for familiarization with the data. For data analysis a deductive approach following the directed content analysis described by Hsieh and Shannon (24) was used. In a rst step, data was coded using pre-de ned categories based on the implementation strategies and different intervention components. In the subsequent analysis steps, subcategories and new categories emerging from the data were added to the coding scheme. Finally, data was summarized in the pre-de ned categories using subcategories and exemplary quotes to comprehensively describe the ndings.

Integration
Findings of the quantitative and qualitative part of this study will be integrated in the discussion section of this paper.

Results
Quantitative data Sample size of care workers differed between the three waves of data collection due to care worker turnover and partial decline of response rates, resulting in 136 respondents at baseline (average response rate: 84%), 99 respondents at T1 (69 %) and 83 respondents at T2 (59 %). A sub-sample of 41 care workers participated at each time point. An overview of the baseline's sample characteristics is displayed in Table 2. The characteristics of respondents did not differ signi cantly between the time points.

Changes of self-e cacy in pain management
In LPNs and RNs the mean score of self-e cacy increased from 69.6 (SD 14.6) at baseline to 74.2 (SD 15.2) at T1 and 76.8 (SD 14.7) at T2. In nursing assistants, the mean score changed from 64.3 (SD 15.1) at baseline to 72.4 (SD 12.1) and 69.2 (SD 12.4) at T2. Mixed models overall con rmed a signi cant increase (p< .01) of self-e cacy between baseline and T1/T2 for all educational levels (Table 3).
Associations with self-reported pain management behavior Self-reported adherence to guideline components ranged between 44% and 73% depending on the component, detailed results were reported in an earlier paper (XXXblinded for review). We found signi cant associations between care workers' adherence to core component I (conducting a comprehensive assessment) and II (using PAINAD scale) and corresponding self-e cacy items. However, we did not nd signi cant associations between the other two core components and corresponding self-e cacy items. An overview of the associations is displayed in Table 4.

Qualitative data
In total, we conducted eight focus groups in the four participating NHs, including 30 care workers (15 RNs/LPNs and 15 NAs). Overall, participants were mainly female (80%) and the average age was 36.6 years (SD = 8.6). The majority of care workers had participated in both parts of the workshop, few participated only in the rst part due to logistical reasons within their NHs. Overall, care workers' perception of the project three months after the start were positive. The participants appreciated the project's focus on pain management, a topic which was recognized to be of high importance for residents' quality of life.

General experience with workshops
In general, NAs were more satis ed with the workshops than RNs and LPNs, several NAs highlighted that they have never received a training speci cally targeted at them before. RNs and LPNs on the other hand, partly perceived the workshops' content to be redundant to what they already knew before (e.g. with regard to pain assessment instruments). Some RNs remarked that they would have preferred a focus on pharmacological treatment options and less on general principles of pain management. Despite some reservation towards the content, the participants agreed that the workshops were overall helpful to raise the care team`s awareness for residents' pain situations. One aspect that was highlighted particularly by the NAs was the recognition that pain almost always goes beyond its physical component and can be related to psychosocial or spiritual factors. Though this was not new knowledge for most participantsre ecting this aspect in depth during the workshop was perceived as a major learning moment. Participants of all job levels reported to be more sensitized to pain when talking to residents or observing their behavior than before.
"Because we could delve into the topic of pain, so that we became more aware of what pain really means in that sense. That is a matter of opinion for everybody and that we react more conscious when people complain about pain" (RN).
In particular the perception of residents with cognitive impairment or other conditions limiting communicative abilities has changed. Participants acknowledged that they sometimes forgot that these residents might also perceive pain from time to time though they do not actively express it.
"Well, I was really shocked about the residents with dementia, I just realized: Hey, you have to keep an eye on them, they are not only having dementia and have no other issues" (LPN)

Perception of the pain champions
In general, the idea of having a pain champion on the ward was perceived as a gain by most participants. Similar to the workshops, NA expressed more positive reactions to the introduction of this new role as compared to RNs and LPNs. During the focus group discussions, most NAs agreed that they could easily approach the new pain champion in case of any questions related to resident spa ∈ . Theyap ≺ iated → haveadesignatedgo -→ persononthewardwhoexpla ∈ scomp ≤ xmaers, e. g. pa ∈ assessment ∈ struments, using pain seriously.
"Yes, you just know… simply to whom you can go. And then you don't have to [think] who should I.. And they are also more attentive, I think, and trained, you know?" (NA) Discussions with RNs and LPNs on the other hand, showed some variation between different NHs. RNs in one NH questioned the added value of having a nurse pain champion because ultimately the physician would decide about a resident`s medication.
"I think, for me the physician is always the pain expert, in the end, he is telling us when we don't know what to do with the medication and whatever, in the end the physician is telling us what to do" (RN) In contrast, RNs and LPNs in another NH reported to see the pain champion as a connector with the responsible physician to advocate residents` needs with regard to appropriate medication.
"The [physician] has a different understanding of applying pain killers, and I think, this is where [the pain champion] can be kind of a connector with the physician" (LPN)

Observing changes in pain management
Overall, there was agreement in most focus groups that during the past three months, changes in the care teams` pain management behavior were noticeable. However, perceptions about the extent of changes differed in parts between professional groups (NAs vs. LPNs/RNs).
One behavior, most participants highlighted to have experienced change is the assessment of residents pa ∈ , ∈ partica ̲ rwithregard → theequency and theuseofm or ed ⇔ erentiatedassessmentquestions. SeveralNAsrep or tedtûntilrecentlyth pain experience, except when a resident has verbally expressed pain. In contrast, the focus group participants reported that they now ask residents actively about their pain if they see signs of discomfort or if the resident has reported pain before.
"Well, after the course, I .. tried harder to observe the resident and I took more time to look at the mimic and yes, well, I.. I was very glad about the course. And now, I pay more attention, or I ask the resident more "Do you have pain?", which I've done less before." (NA) Another aspect that was brought up in this regard, was the newly implemented pain report in the resident documentation. Before, residents` pain was exclusively reported in the overall documentation, focus group participants acknowledged that they used to document only that a resident experiences pain without being speci c about the details. However, in the new form there are elds asking for the different qualities of pain e.g., intensity, location, etc. With this new pain report form NAs felt reminded to be more detailed when describing pain situations and also to ask residents more speci cally about their experience. "Or, for example you can't just write "Mrs. X has pain". Now you have to describe, where is the pain, what kind of pain, where .. and you have to take this to pieces, in detail." (NA) Furthermore, with regard to the use of speci c assessment instruments, NA reported to have tried out different self-report scales with residents. Assessing pain in residents with severe cognitive impairment, still was perceived as a challenge by NAs, however, they reported to be more aware of signs of discomfort and changes in the behavior than before. Likewise, LPNs and RNs reported to be challenged by assessing pain in these residents. For most care workers, using the observational pain assessment scale was perceived to facilitate assessment, however, some participants reported challenges applying the scale correctly in the beginning. In addition to changes in assessing pain, participants of all job levels agreed that NAs reports of residents` pain to the charge nurse were more timely. In the mornings, they would inform the responsible nurse as soon as possible instead of waiting till the designated report time in the late morning.
Aside from changes in NAs` behavior, RNs/LPNs reported that the pain champion has motivated them to go through each resident's medication scheme to critically review if the medication is still appropriate in light of the current pain situation.
"Here, I am working together with the pain champion and we are discussing the residents, what we could do or not. Or if we should reduce something or not." (RN) Based on their critical review, they themselves or the pain champion talked with the responsible physicians about their suggestions. Participants reported that based on the reviews they were able to adapt or reduce unnecessary pain medications in several residents.

Discussion
The current study found that interactive training workshops and introduction of a trained pain champion could signi cantly increase care workers' self-e cacy related to pain management. Furthermore, we could show signi cant associations between care workers' self-e cacy and adoption of two core components: 'conducting a comprehensive assessment' and 'use of PAINAD', however there was no signi cant relation with the components 'documentation' and 'reevaluation'. Our qualitative ndings showed that the training workshops and pain champions were generally received positive. Care workers reported to be more attentive to residents` pain experience and to assess and document pain more frequently and with more detail. Furthermore, focus group participants reported changes in the pain reporting behavior and the pharmacological management.
This study was the rst to look at the implementation of a pain management guideline in NHs using a behavioral perspective. In general, effects of previous studies in this eld were mixed (25,26), yet most evaluation approaches did not allow to differentiate between the effectiveness of the intervention and the implementation efforts. Much of the previous research was based on quality improvement approaches and did not integrate any theoretical underpinning.
Guidelines for pain management generally consist of a set of recommended actions to assess and treat pain. Pain assessment in NH residents depicts a challenge for care workers since pain is a highly subjective symptom and the ability to self-report pain is often limited in residents. In order to increase care workers adherence → evncebasedpa ∈ assessmentitwas ∴ crucial → raiseaware ≠ ssaboutpotentialpa ∈ ∈ resnts and → addressprevail ∈ gaitu ndings it can be assumed that our implementation strategies were effective in sensitizing care workers for resident pain and furthermore motivated care workers to question their own beliefs concerning pain. The two components "conducting a comprehensive pain assessment (with newly admitted residents and those reporting new pain)" and "using an observational assessment tool (in residents with severe cognitive impairment)" require certain clinical skills and competencies beyond an increased awareness. Bandura`s theory of self-e cacy states that in addition to beliefs and attitudes, con dence in the own ability to perform a speci c behaviors can determine adoption of the new behavior (9). The two other components 'documentation' and 're-evaluation', on the other hand, require less advanced competencies, however, with regard to their regular performance, other factors, such as memory or motivation, play a more important role. One possible explanation why we were not able to nd an association between these components and the self-e cacy items might have been the wording of the items. The self-e cacy items focused mainly on the con dence in performing speci c pain management behaviors, and less on remembering these behaviors in daily practice. Changing existing routines in the daily care practice and remembering to perform the new behavior is challenging, in this context a recent systematic review and meta-analysis highlighted the potential in uence of habits on healthcare professionals` behavior (27). Habits can be de ned as "a process by which a stimulus generates an impulse to act as a result of a learned stimulus-response association" (28).
With our approach to introduce a pain champion on the ward, we intended to provide care workers with a role model, who helps to memorize speci c behaviors, e.g. documentation, use of PAINAD and to eventually form new habits facilitates with regular behavioral cues and the provision of performance feedback. Champions have been used in other studies to improve quality of pain management in NHs (29)(30)(31)(32). The type of strategies used by the champions reported in Kaasalainen, Brazil (31) are comparable to our approach. However, the authors did not directly assess changes at the care worker level, instead they evaluated process indicators based on the residents' documentation. Information retrieved from chart reviews can partly be informative concerning the documentation behavior, however with regard to other behaviors related to pain management they are only crude indicators of changes occurring on the level of care workers.
This was the rst study in the eld of pain management in NHs to use behavioral theory to gain an understanding of the underlying implementation mechanisms. A clear strength of this study was the systematic speci cation of our implementation strategies into behavior change techniques and the use of a conceptual model. The speci cation of strategies allowed the generation of hypotheses with the aim to increase our understanding of the underlying mechanisms. Clear de nitions of implementation strategies enhance the comparability of studies and thus facilitate the generation of transferable knowledge. Furthermore, the mixed-methods design allowed to validate our initial ndings with a set of focus groups. Besides its strengths, there are also some limitations to this study. First of all, this study was based on a quasi-experimental, uncontrolled design limiting our ability to draw direct conclusions about the effectiveness of our implementation strategies and precluding causal inferences. Secondly, the study was based on care workers' self-reports of behavior which might have introduced bias due to care workers' ability of recalling behavior or by social desirability.

Conclusions
The purpose of this current study was to explore the underlying mechanisms related to the implementation of a pain management guideline in Swiss NHs using training workshops of care workers and trained pain champions. We could show a signi cant increase in care workers' self-e cacy in pain management, increased awareness of care workers towards residents` pain and changes in the pain management behavior. Our ndings highlight that continuous commitment of pain champions or similar implementation facilitators is pivotal to the embedding of new routines in care workers' practice.
Increasing the adoption of evidence-based pain management guidelines in NHs remains of crucial importance to improve management of residents' pain and ultimately their quality of life. Future studies in the eld of pain management in NHs should therefore make use of behavioral theory to understand and tackle such implementation challenges. Availability of data and material The datasets generated during and/or analysed during the current study are not publicly available due to the con dentiality of respondents but are available from the corresponding author on reasonable request.

Competing interest
The authors declare that they have no competing interests.

Funding
This work was supported by Senevita AG. The sponsor had no role in study concept or design, methods, subject recruitment, data collection, analysis, and preparation of the manuscript.
Author Contributions: All authors made substantial contributions to conception and design, acquisition of data, or analysis and interpretation of data. All were involved in drafting the manuscript or revising it critically for important intellectual content; All authors gave nal approval of the version to be published.  Note. RN= registered nurse, LPN= licensed practical nurse