The TREC program is situated in residential LTC in the 4 western Canadian provinces (British Columbia, Alberta, Saskatchewan, and Manitoba) where we maintain a cohort of 94 participating LTC homes. The cohort sample is a stratified (health region, owner-operator model, and bed size) random sample [61] of urban LTC homes in 5 health regions of participating provinces.

We have longitudinal data (Table 1) from 2 main sources: (1) TREC surveys (collected in 5 waves since 2007) of LTC homes, care units, and all levels of staff (regulated and unregulated nursing staff, allied health providers, specialists and educators, and managers) and (2) administrative data collected using the Resident Assessment Instrument – Minimum Dataset (RAI-MDS) 2.0 [62]. The RAI-MDS 2.0 is a routinely collected (quarterly, annually) and standard mandated assessment of clinical and functional outcomes for LTC residents [62]. Beyond these observational data from our ongoing cohort study, we have data from pilot studies, clinical trials, and case studies. All research data are housed in the Health Research Data Repository (University of Alberta), which provides virtual data access for team members within a highly secure environment. Data are extensively processed to exceed Canadian Institute for Health Information standards for RAI-MDS 2.0 data. Extensive quality assurance (during data collection and post data collection) is carried out with survey data, including assigning each resident to a single unit within a single LTC home [63].

Data captured in the TREC survey include (1) structural characteristics: LTC home surveys completed by the LTC home administrator and unit surveys completed by the unit care manager and (2) individual measures from managers, regulated nursing staff, allied health staff, and unregulated care aides (nursing assistants).

Examples of structural characteristics are size and the owner-operator model of LTC homes and type and staffing level of care units. Individual surveys are a suite of instruments and questions that capture demographic characteristics, best practice use, quality of work life, and organizational context. Quality of work life variables are captured using validated measures. They include burnout (Maslach Burnout Inventory) [70], physical and mental health (Short Form-8) [71], work engagement (Utrecht work engagement scale) [72], psychological empowerment (Psychological Empowerment Scale) [73], organizational citizenship behavior [74], job satisfaction (positively phrased version of Michigan Organizational Assessment Questionnaire Job Satisfaction Scale) [74], and responsive behaviors of residents toward staff [75].

Organizational context is captured using the Alberta Context Tool (ACT) [54,55]. We developed the ACT based on the PARIHS initial conceptualization of context, which includes culture, leadership, and evaluation (feedback of data to end users). We added constructs substantiated in health services literature [76], such as social capital, formal and informal interactions (2 communication concepts), organizational slack, and resources [55]. The ACT has measured context at the LTC home, unit, and group levels [27,28], but it is designed specifically for the clinical microsystem or resident care unit level [26-29]. It has had extensive psychometric assessments [54,55,77-79].

We have access to deidentified resident data from the RAI-MDS 2.0 for each of the 94 participating LTC homes. From our 5 large waves of data collection, we have over 500,000 resident data records. Further, we can link these data to external Continuing Care Reporting System data nationally (eg, Canadian Institute for Health Information [80]) and provincially (eg, Alberta Health [81]), and to other Canadian administrative databases (eg, Discharge Abstract Database [82]).

Table 2 provides an overview of data from our 2 completed clinical trials: Improving Nursing home care through Feedback On performance (INFORM) data [50,83,84] and Safer Care for Older Persons [in residential] Environments (SCOPE) [56,85,86]. The INFORM trial supported LTC home managers as they used our feedback from survey findings. INFORM included 143 care units from 58 LTC homes. It gathered quantitative data from surveys, reports, and workshop and rating evaluations, along with qualitative data from focus groups, interviews, and reports (process evaluation data).

The SCOPE trial examined the effects of empowering care aides to lead quality improvement strategies in their care unit. SCOPE included 408 participants from 45 LTC homes. It gathered quantitative data as rich survey data from individuals, teams, and leadership, and qualitative data from interviews, focus groups, and observation (process evaluation data).

We have data from 3 extensive ethnographic case studies completed in TREC phase 1 (2007-2012) [87,88]. We obtained 70 interviews and 22 sets of field notes from nurses, care aides, managers, allied health personnel, and family of residents from 3 LTC homes in Alberta, Saskatchewan, and Manitoba. In-person interviews (2008-2010) were semistructured. Ethnographic observations (2008-2009) were written as field notes by research associates. The purpose of the original ethnographic case studies was to explore how organizational context mediates staff use of evidence-based best practices in LTC homes [87].

Facilitation, implementation success, and improvement success are not directly available in our data, but we will derive the variables from our INFORM and SCOPE trial data. Many quality improvement interventions in health care settings target health care providers’ adoption of evidence-based best practices [89]. Our team has used survey data on staff adoption of best practices as a proxy for implementation success [56,85,90,91]. Specifically, we measured conceptual use of best practices [90,92] and instrumental use of best practices (applying best practice knowledge) [37,92]. In this study, we will draw on our earlier work to derive variables of implementation and improvement success using INFORM and SCOPE trial data [50,57]. We will rank the sites in each of our 2 trials on the basis of success, then derive a “success” variable, and experiment to find an optimal derivation.

We have published one conceptualization of facilitation [48]. Other data derivations are possible from examining role or process. We will rank facilitation effectiveness as we rank implementation and improvement success to derive a facilitation score by site.

We will examine qualitative data from our case studies and INFORM and SCOPE trials (process evaluation data). For the case study data, we will search for evidence on context elements that most strongly influence the association between facilitation and staff quality of work life, and between facilitation and improvements in resident outcomes. We will propose hypotheses for quantitative analysis that will contribute toward aims 2 and 3. For the trial data, we will look for evidence on the context elements that most strongly drive better trial intervention delivery, enactment, and receipt, and on improvement success of the trial. This will contribute toward our first aim of understanding implementation success. We will use ATLAS.ti software [93] to support data management and analysis and visualization of findings.

We will examine the data first with the lens of the PARIHS framework (eg, context and facilitation) and then augment with other key theoretical perspectives (eg, adaptive leadership and sense-making) [94,95]. We will use directed coding [96], guided by PARIHS concepts, and open coding [96] to capture concepts that are not part of PARIHS or that describe relevant staff and resident outcomes. Two coders will read and code each transcript and write a summary. Senior qualitative researchers on the coding team will review summaries. Codes will be added and refined, and summaries updated as needed.

Coders will then use the summaries to synthesize findings into a matrix to illustrate the concepts of interest and potential relationships among them. They will include sample quotes to exemplify the concepts. Matrices will be examined by senior team members and refinements made as needed. We will use the matrices to create network figures, using the network feature of ATLAS.ti software to depict relationships identified in the data. Networks will be reviewed by the full team and refined. Matrices and networks will be examined across cases to identify patterns and areas of contradiction. Questions from the full team will be explored by the coding team, who will look back at the original source data and examine the audit trail of the initial coding, summaries, matrices, and networks.

Our quantitative working group will analyze quantitative data from TREC surveys and RAI-MDS 2.0. We will be able to measure many constructs derived from qualitative analyses. TREC data are of high quality (few missing data and high response rates) and have been collected, cleaned, and processed through a rigorous process [97]. We will use a variety of statistical software packages for this work, such as SAS (SAS Institute), SPSS (IBM Corp), R (R Foundation for Statistical Computing), and STATA (StataCorp).

We will conduct basic descriptive and bivariate analyses, followed by more advanced statistical modeling using; for example, general estimating equations, hierarchical linear mixed models, and generalized linear mixed models. These enable us to adjust for complex nested structures of our data [26,38]. We will also explore the use of structural equation modeling and configurational modeling [98,99]. We will adjust our analyses as appropriate for different outcomes, including resident or care staff characteristics (depending on outcome), care unit characteristics, LTC home characteristics, and regional features depending on the addressed aim. We will begin with an approach that examines a standard set of context influences and assess it across staff and resident outcomes at unit and individual levels (aims 1 and 2). Based on these analyses, with the input of our qualitative team and panels, we will identify patterns and trends over time.

We will adopt convergent and sequential mixed methods models to obtain different but complementary data on the same problem for a more complete understanding [100]. In a convergent model, we will first analyze qualitative and quantitative data separately and independently, then integrate results in matrices, and interpret how the 2 sources of results converge or diverge [60,101]. In sequential modeling, results of qualitative analysis will inform the quantitative analysis and vice versa [60].

As an example, we will complement the PARIHS framework with other conceptual frameworks (eg, adaptive leadership and sense-making) [94,95,102] and form hypotheses related to our aims, including hypotheses on how context factors influence implementation outcomes. We will then test the hypotheses by analyzing both the quantitative and qualitative data and explore support or nonsupport of interrelationships. Next, we will focus on interactions among the context factors and interaction between context and facilitation based on our qualitative analyses, forming additional hypotheses. We will test these additional hypotheses with quantitative analysis. When unanticipated results emerge from the quantitative analysis, we will use qualitative analysis to explore processes and mechanisms explaining the results.

We will work iteratively with our stakeholder and expert panels to further inform and refine analyses. We will hold 2 virtual facilitated meetings with all panels (Figure 1). Before meetings, work groups will prepare summary reports of research findings and a focused set of guidelines and questions that are tailored to each panel. During meetings, the external expert panel will comment on summary reports and advise through scientific and theoretical lenses on early findings presented, most promising and important avenues to pursue, and areas not yet identified that our data would support for further exploration. The decision- and policy maker and LTC home manager panels will comment and advise on these same aspects through a lens of system management and utility of findings. The direct care staff panel will provide a frontline perspective on whether and how our findings could be useful to them and suggest areas we may not have considered. The citizen engagement panel will also comment and advise on whether and how our findings could be useful to them and suggest areas we may not have considered, from the perspective of findings important to their constituencies. This panel approach assists us with direction and focus, to keep our results both scientifically and practically relevant.