Study design
The project will be grounded in the pragmatism paradigm which relies on the assumption that the finality of knowledge is to address concrete problems and provide answers or direction to progress [33]. We will conduct a two-phase mixed methods multiple case study (sequential explanatory design) [34, 35], starting with a quantitative phase (phase 1) to answer objective 1, followed by a qualitative phase (phase 2) for objectives 2 and 3. This design is well suited to answer research questions addressing complex systems in varied and dynamic contexts, allowing an in-depth analysis of each case, and offering opportunities for comparison between cases (objective 4) [36]. We will work with six cases in three provinces in Canada (Quebec, New Brunswick, and Newfoundland and Labrador), each case being the actual care transitions across community, primary care and hospital settings [37]. Each case will be called a health network. According to the conceptual model of factors affecting care transitions, presented thereafter [38], three levels of analysis will allow an in-depth understanding of each case: 1) the patient level; 2) the provider level; and 3) the healthcare system level.
Researcher and knowledge user collaboration throughout the research process is a strong predictor that research findings will be used [39]. In addition to traditional knowledge translation (KT) at the end of the project, we will carry out integrated KT by engaging knowledge users from each audience (patient partner, providers, decision-makers/managers) in the study’s Steering Committee. They will participate in key decisions throughout the study to ensure findings are useful to them in their respective contexts.
Conceptual model
We used the conceptual model of factors affecting care transitions [38] to identify relevant independent variables to measure with questionnaires in phase 1, and to develop interview guides in phase 2. We chose this model because of its three-level structure of factors affecting care transitions and the relevance of the factors at each level: 1) the patient level (severity of illness, factors of vulnerability, self-management ability, social support); 2) the provider level (accountability, clarification of roles); and 3) the healthcare system level (access, coordination).
Phase 1 (objective 1)
Sampling of the cases
Each case will be the actual care transitions across community, primary care, and hospital settings in each of the six health networks. To identify the cases, we targeted six emergency departments (ED), two per province, which are already engaged to participate in the study. The EDs were identified using a purposeful sampling strategy [40], to represent real-world differences [41] in terms of provinces, geographic area (rural, semi-urban, and urban), and both official languages (French and English). The inclusion of multiple cases capitalizes on organizational variation in care transitions to develop a more informed understanding. It also allows for observation of similar or singular care transitions, and draws conclusions that could be transferable to other contexts [42]. It is recommended that four to ten cases be considered [43] in the multiple case study logic of theoretical replication [42].
Sampling of patients with complex needs
Identifying patients with complex needs is a challenge because it does not depend on a precise diagnosis. We know that patients with complex needs frequently use many health and social services, and that ED visits are a good proxy of this use [44]. For screening, we will thus use our COmplex NEeds Case-finding Tool – 6 (CONECT-6) [45]. We validated this 6-question tool among patients at their third or more visit to the ED within 12 months, to screen those with complex needs (INTERMED Self-Assessment positive), with a sensitivity of 90% and a specificity of 66% with a threshold of two or more positive answers [45]. Patients screened as having complex needs with CONECT-6 will be invited to confirm their complex needs with INTERMED Self-Assessment (IMSA) [46]. IMSA is a self-reported version of the INTERMED questionnaire, taking 15 min to complete and measuring the complexity of adult needs. The first version of INTERMED was developed in the 1990s by an international team that combined their research expertise on complexity. Its psychometric qualities are well documented [47,48,49,50,51]. IMSA includes 20 questions subdivided into four domains: biological, psychological, social, and health system. Every domain is divided into three segments: history, current state, and prognosis. French and English-language versions are available. Patients with a score of 19 or higher will be invited to participate in the study, since this threshold confirms complex needs [52].
During the first year of the study, research assistants will be present in each ED four days a week to identify adults (≥ 18 years) at their third or more visit to the ED within 12 months, using the information system of the ED. Relying on a previous study [45], we estimate this number of patients at six per day in each ED (more in bigger EDs and fewer less in smaller EDs). Research assistants will invite those patients to answer CONECT-6 (two minutes). Approximately 30% of these patients will score positive on CONECT-6 [45] (n = 2 per day in each ED), and the research assistants will administer the IMSA questionnaire to those patients. CONECT-6 has a positive predictive value of 50% [45]; therefore, one patient will score complex on IMSA per day in each ED. Estimating an acceptance rate of 50%, we will recruit two patients per week for the project in each ED. We estimate a percentage of 60% of women and 40% of men [45]. It will then take about six months to recruit 180 women and nine months to recruit 180 men.
Data collection
At baseline, questionnaires with good psychometric properties in English and French will be administered by the research assistants to all participants, preferably during waiting at the ED, or by telephone (without affecting validity) [53] within two weeks of their ED visit. The questionnaires will collect information on age, sex, gender, indigenous identity, ethnicity, language, marital status, education, occupation, income, housing conditions, residential address, food security, social support, health literacy, alcohol, and drug use, multimorbidity, and self-management. These variables were identified based on the conceptual model [39]. Required time to complete the questionnaires is about 30 min. Age, language, marital status, education, occupation and income will be measured with questions from the Canadian Community Health Survey (CCHS) [54], sex and gender with the Statistics Canada census questions [55], ethnicity and indigenous identity with the Tri-Agency self-identification Equity and Diversity Questionnaire [56] (2 items), housing conditions with the Housing Satisfaction Question [57] (1 item), food security with the U.S. Household Food Security Survey Short Form [58] (6 items), social support with the Medical Outcomes Study Social Support Survey [59] (8 items), health literacy with the Brief Health Literacy Screening (BHLS) questionnaire [60, 61] (3 items), alcohol and drug use with the Alcohol, Smoking and Substance Involvement Screening Test (ASSIST) questionnaire [62] (9 items), multimorbidity with the Disease Burden Morbidity Assessment (DBMA) [63, 64] (21 items), and self-management with the Partners in Health Scale (PIH) [65, 66] (12 items). Environmental data such as neighborhood deprivation, gentrification, and marginalization will come from the Canadian Urban Environmental Health Research Consortium (CANUE) [67], using patients’ 6-digit postal codes.
At six months, we will administer a questionnaire by telephone to measure the experience of care transitions in the previous six months taking into consideration the patient’s holistic experience. The Patient Experience of Integrated Care Scale (12 items) [68], which we developed and validated from a set of items proposed by the Picker Institute Europe and the University of Oxford [69], will allow us to focus on the global experience of care transitions. The 12 items will result in a continuous score ranging from 0 to 48 where a higher score indicates better care integration.
Data analysis
Descriptive statistical analyses will be performed. The association between baseline independent variables and patient experience of care transitions (dependent variable measured at six months) will be assessed with multilevel linear regression models using SPSS V.24. Bivariate analyses (separate for each sex) will first be conducted using a cut-off level of α = 0.10 for inclusion in the multivariable model. The multilevel regression models (one for each sex) will then be reduced with backward elimination (α = 0.05) [70]. For all analyses, allowing for random intercepts and slopes with multilevel modeling will allow us to take into account the possible clustering of care transitions outcomes within the same health network. We estimate the intra-cluster correlation to be around 0.01, which is the median intra-cluster correlations estimated from Adams et al. [71] who examined 1039 variables from 31 studies in primary care. For each model, at the 5% significance level, a sample size of around 117 has 80% power for an expected medium effect size (R2 = 0,15) [72] with up to 10 predictors in the final multivariable models (G*Power 3.1.9.4) [73]. We find that 144 participants clustered within six cases are statistically equivalent to 117 independent participants, considering the design effect of (1 + [nm-1])*0.01) [74] where n = 24 women (or men) per case (144/6). We therefore need 288 patients (144 women and 144 men) completing the questionnaire at 6 months. Estimating a loss to follow-up of 20% [75, 76], 360 patients (180 women and 180 men) will be recruited at baseline in the EDs. For data security and privacy, all data will be hosted on hospital-grade internal servers. All data will be stripped of personally identifying information, and only the principal investigators from each province will have access to the key to identify individual participants within their province.
Phase 2 (objectives 2 and 3)
Data collection
In person or virtual individual semi-structured interviews will be conducted to capture the richness of the perspectives [77, 78] with eight patients and family members per case (total n = 48), with a diversity of gender using a purposive sampling [79] among participants with the lowest and highest results for care transitions from phase 1. After providing written informed consent, each participant will do a one-hour interview conducted by a research assistant trained in qualitative research methods with a semi-structured interview guide composed of open-ended questions on their experience of care transitions. The interview guide developed for the interview with the patients and family members is provided as Additional file 1. The interviewer will take time to clearly explain the concept of care transitions with examples at the beginning. A few examples of questions are: What is working well in your care transitions (explain more if required) and why? Can you provide an example of a transition you felt good about and why? What is the most difficult in these transitions and why? Can you provide specific examples of transitions that were more challenging? How do you think the healthcare system could improve the way you experience these transitions? The research team’s patient partners will help refine and test the interview guides. The interviews will be digitally recorded and transcribed verbatim. We will aim for data saturation while expecting a certain variability among the cases [80], so the number of participants will be adjusted iteratively. About 40 interviews are usually needed to reach saturation in multisite studies [81].
In person or virtual focus groups (FG) of six providers (good balance between women and men) will be conducted in each case: one FG of family physicians, nurses’ practitioners, and specialists; one FG of other professionals including social workers; and one FG of community pharmacists and community organization partners. Key informants [82] and a snowball technique [83] will be used to identify providers who could share their experience to better understand care transitions of this population. Results of phase 1 will be presented to all FG participants to contextualize the discussion that will be facilitated with a semi-structured interview guide composed of open-ended questions on their experience of care transitions with this population. The interview guide developed for the focus group with the providers is supplied as Additional file 2. A few examples of questions are: What is going well, what is more difficult, and what should be done to improve these transitions? The FG will be digitally recorded and transcribed verbatim. We estimated the number of groups (3 groups per case) to reach data saturation [80, 81] for each category of providers. The optimal number of groups will be determined iteratively depending on the variability among the cases.
To explore the healthcare system level of our conceptual model, in person or virtual individual semi-structured interviews will be conducted with eight (good balance between women and men) health managers per case (total n = 48), working in different settings (hospitals, primary care, etc.), identified with key informants [82] and a snowball technique [83]. Results of phase 1 will be presented to health managers to contextualize the discussion that will be facilitated with an interview guide composed of open-ended questions on their experience. The interview guide developed for the interview with the managers is provided as Additional file 3. A few examples of questions are: What is going well, what is difficult, and what should be done at your managerial level to improve these transitions? The FG will be digitally recorded and transcribed verbatim. We estimated the number of interviews (eight per case) to reach data saturation [80, 81]. The optimal number of groups will be determined iteratively depending on the variability among the cases.
Data analysis
Two team members from different professional backgrounds will read the transcripts and iteratively analyze them using a deductive (themes based on the conceptual model of factors affecting care transitions) [38] and inductive (themes emerging from the data) thematic analyses [84]. Qualitative data will be managed using NVivo V.12 server software (QSR International Pty). To ensure credibility and minimize the effect of researcher subjectivity, the two team members will share and discuss results of the analysis with the team to confirm and enrich the findings. We will encourage pair debriefing and triangulation of researcher backgrounds and of collaborators’ expertise [85]. Transparency in analysis and reporting will be achieved by providing extensive deidentified verbatim quotes and a detailed description of the contexts, which will also help promote transferability to similar contexts.
Integration of quantitative and qualitative results (objective 4)
Two types of integration will be performed [86]. First, qualitative and quantitative results will be compared. Then for each case, qualitative and quantitative data will be merged [42]. A case summary will be reported (synthesizing merged data), and the six case summaries will be used to compare cases by means of a descriptive and interpretative matrix, allowing systematic comparisons among cases and analysis units (patient, provider, and health manager levels). Different analytical techniques will be used such as pattern comparison, research of competing explanations, and construction of explanations [42]. Management, reduction, and comparisons will be conducted with NVivo V.12 software. Knowledge users on the steering committee of this project will participate in key steps of the analysis to ensure meaningful interpretation [87, 88]. Deidentified case summaries could be used as “vignettes” in the knowledge translation plan and the web interactive learning module to illustrate good or poor care transitions. Knowledge users, including patient partners, will be involved in the knowledge translation plan.