Using concept mapping to prioritize barriers to diabetes care and self-management for those who experience homelessness

Background Diabetes is a chronic medical condition which demands that patients engage in self-management to achieve optimal glycemic control and avoid severe complications. Individuals who have diabetes and are experiencing homelessness are more likely to have chronic hyperglycemia and adverse outcomes. Our objective was to collaborate with individuals experiencing homelessness and care providers to understand the barriers they face in managing diabetes, as a first step in identifying solutions for enhancing diabetes management in this population. Methods We recruited individuals with lived experience of homelessness and diabetes (i.e. clients; n = 32) from Toronto and health and social care providers working in the areas of diabetes and/or homelessness (i.e. providers; n = 96) from across Canada. We used concept mapping, a participatory research method, to engage participants in brainstorming barriers to diabetes management, which were subsequently categorized into clusters, using the Concept Systems Global MAX software, and rated based on their perceived impact on diabetes management. The ratings were standardized for each participant group, and the average cluster ratings for the clients and providers were compared using t-tests. Results The brainstorming identified 43 unique barriers to diabetes management. The clients’ map featured 9 clusters of barriers: Challenges to getting healthy food, Inadequate income, Navigating services, Not having a place of your own, Relationships with professionals, Diabetes education, Emotional wellbeing, Competing priorities, and Weather-related issues. The providers’ map had 7 clusters: Access to healthy food, Dietary choices in the context of homelessness, Limited finances, Lack of stable, private housing, Navigating the health and social sectors, Emotional distress and competing priorities, and Mental health and addictions. The highest-rated clusters were Challenges to getting healthy food (clients) and Mental health and addictions (providers). Challenges to getting healthy food was rated significantly higher by clients (p = 0.01) and Competing priorities was rated significantly higher by providers (p = 0.03). Conclusions Experiencing homelessness poses numerous barriers to managing diabetes, the greatest of which according to clients, is challenges to getting healthy food. This study showed that the way clients and providers perceive these barriers differs considerably, which highlights the importance of including clients’ insights when assessing needs and designing effective solutions. Supplementary Information The online version contains supplementary material available at 10.1186/s12939-021-01494-3.

many PWLEH do not have a xed mailing address or reliable access to a phone, issues contacting patients are also common (8,15), making it di cult to schedule and notify patients of appointments (16). Perceived feelings of being unwelcome in clinical settings can also prevent people who are experiencing homelessness from seeking health care services (17). Other priorities, namely, obtaining the necessities of life (food, shelter, etc...), drug and alcohol use, and mental health or cognitive issues, can also interfere with diabetes self-management (7,14,18).
While there are clearly many barriers and challenges to diabetes management, the importance of these in relation to one another is not well known, nor is it well known if these priorities differ between providers and PWLEH. Other studies that have compared the perspectives of service users with those of service providers have found differences of opinions between the two groups (19,20). One such study asking community members and service providers about which social and mental health services should be made available at a new medical clinic in the community, found that community members and service providers had different views regarding which services were most needed (20). Another study aimed to identify existing gaps in services for youth, where the opinions of the youth differed from those of the service providers regarding importance, but not in terms of 'what to do rst' (19).
Given the complexity of diabetes management, myriad barriers that can affect adherence, and the potential for differences in perspectives between providers and PWLEH, we sought to collaborate with people who had lived experience of homelessness and diabetes, as well as health and social care providers who work with people experiencing homelessness and/or people with diabetes, to prioritize the challenges faced by this population in managing diabetes. The objective of this study was to determine the perceived relative impact of barriers to diabetes management and whether the perceptions about the impact of those barriers differed between providers and PWLEH.

Study Design
We used a semi-quantitative, participatory methodology known as concept mapping. We used concept mapping because it can be used to create visual representations that depict the combined thoughts of a group, by integrating input from multiple people using quantitative data analysis techniques (21). This methodology enabled us to gather insights from a group of PWLEH who have diabetes (herein referred to as the clients) and from various providers representing different areas of health and social care (herein referred to as the providers), and to quantitatively compare their collective feedback.
Our study was approved by the research ethics boards of the University of Calgary and Unity Health Toronto/St. Michael's Hospital.

Study Participants & Recruitment
Clients. The clients for this study were all recruited in Toronto, Ontario -Canada's largest city. To be eligible to participate, they had to: be older than 18 years, have a history of diabetes with a duration greater than two years, and have experienced homelessness within the previous two years. Homelessness in this study was de ned similarly to the de nition offered by the Canadian Observatory on Homelessness; we included participants whose living situations could be described as unsheltered, emergency sheltered, or provisionally sheltered (22). Colleagues and advocates in the homeless-serving sector were consulted to determine the best places for recruiting the target population. Posters were put up in shelters, at homeless-serving agencies, and health clinics catering to the homeless population for recruitment. Established programs serving people who were experiencing homelessness or housing instability also helped with recruitment by advertising an information session to their program attendees.
We explained the study to interested parties and obtained informed consent.
Eight of the clients in this study were recruited as part of a larger community-based participatory research project designed to understand what it is like to live with diabetes while experiencing homelessness or housing instability, and to propose and develop potential interventions for this population (citation removed). We then recruited a number of other clients to participate in this concept mapping exercise to increase the sample size.
Providers. The service providers who took part in this study had previously been recruited as participants in another study of diabetes care for PWLEH (citation removed). They comprised professionals from four main categories of care: diabetes care professionals who worked primarily with inner-city or homeless populations, other health care professionals who worked primarily with inner-city or homeless populations, endocrinologists or diabetes care providers who did not speci cally focus on inner-city or homeless populations, and other stakeholders, including frontline staff in shelters and social care providers. The providers were recruited from ve major Canadian cities (Toronto, Ottawa, Edmonton, Calgary, and Vancouver), and they participated in this study remotely.

Data Collection
The concept mapping process involves six steps, three of which relate to data collection: preparation, generation of statements, and structuring of statements (23).
In the rst step, preparation, the researchers must determine what the focus of the study will be. Usually, this involves two separate focus statements. The rst statement is referred to as the focus prompt or the brainstorming focus, and it is developed to guide the brainstorming exercise in the second step. The second focus statement, called the rating focus, de nes the factor(s) upon which the brainstormed statements will be rated in the rating exercise of the third step (23,24). The focus prompt for the brainstorming exercise in this study was, 'What are some ways, good or bad, that diabetes might be affected by homelessness?'. The rating prompt was: 'What impact does [that factor] have on diabetes care and self-management (on a scale from no impact to high impact), speci cally for people who are struggling with homelessness or housing instability?'.
Generation involves a brainstorming exercise, in which the participants are provided with the focus prompt and asked to brainstorm statements in response (23). The brainstorming exercise is done individually by each participant, and eventually the ideas are combined and re ned to form one set of statements (24). Eight clients did the brainstorming exercise during an in-person group meeting. The providers had been individually interviewed by members of the research team, who then shared the providers' ideas with the clients during their brainstorming session. The researchers recorded the ideas from the clients and the providers and re ned the list with the clients.
Structuring involved two different exercises, sorting and rating. For the sorting exercise, each participant was asked to group individual statements into categories based on similarity, in a way that was meaningful to that individual (21,25). The statements could not all be sorted into one pile, nor could they all be in their own pile (24). "Miscellaneous" piles were also discouraged. The clients completed this exercise manually with statements printed on individual cue cards, while the providers completed this exercise using a tailor-made online platform. For the rating exercise, a Likert scale ranging from zero to four was used to rate each statement. A rating of zero meant the statement had no impact on diabetes control and self-management, and a rating of four meant it had an extreme impact on diabetes control and self-management. This exercise was done on paper forms by the clients and online by the providers.

Data Analysis
The nal three steps of the concept mapping process relate to data analysis: representation of statements, interpretation of maps, and utilization of maps (23).
In the representation step, sorting and rating data were analyzed using concept mapping software (Concept Systems Global Max, 2020, Ithaca NY) (25,26). The representation step was done separately for the clients and the providers: the clients' data was analyzed to create a clients' cluster map and the providers' data was analyzed to produce a providers' cluster map. By having two cluster maps, rather than one, we were able to see how the number of clusters, the names of clusters, and the sorting of the statements, differed between the two participant groups.
The sorting data were analyzed using multidimensional scaling analysis, a technique that plots each statement as a point on a map. Statements that are closer together on the map were grouped together during the sorting exercise more often than statements that are farther apart. Multidimensional scaling was used to generate a matrix for each participant with one row and one column representing each of the statements (24). In this study, the matrix had 43 rows and 43 columns. Each cell in the matrix corresponds to two statements, the row statement and the column statement. If those two statements were sorted into the same pile by the participant, the cell is given a value of one, and if they were not sorted into the same pile, the cell is assigned a value of zero (21,24). This is done with each participant's data so that there are as many separate matrices as there are participants in the study. All of the individual matrices are then summed to create a similarity matrix indicating how many participants sorted each pair of statements into the same pile (24). This similarity data is used to create a two-dimensional point map, where the distance between two points on the map corresponds to their similarity (21).
These points were then grouped together to form clusters through hierarchical cluster analysis.
Hierarchical cluster analysis uses the data from the point map to separate the points into nonoverlapping clusters, developing a cluster map. The number of clusters in the nal cluster map is determined by the participants and the researchers so that the map is presented in a way that is meaningful and useful to them (21). This analysis begins by considering each statement to be in a cluster of its own, and it combines two clusters at a time, reducing the number of overall clusters until all the statements are in the same cluster. The statements are combined based on proximity, meaning that the two points that are closest together on the point map would be the rst two points to form a cluster together (24). The reliability, or goodness-of-t of the map, is determined by calculating the nal stress of the model (27). Lower stress values suggest a better t. Stress values in concept mapping data tend to be higher than in other multidimensional scaling analyses (28), therefore there is no accepted standard, however, it has been reported that an acceptable value for studies of this nature is less than 0.30 (29).
An average rating is then calculated for each cluster using the ratings for each of the individual statements in that cluster.
Upon reviewing the average cluster ratings, it was clear that the clients and providers had rated the statements very differently, such that the clients had generally given lower ratings whereas the providers had given higher ratings, which resulted in lower average cluster ratings for the clients' clusters and higher average cluster ratings for the providers' clusters.
To determine whether there was in fact a difference in the ratings between the two groups, the ratings were standardized and compared using unpaired two-sided t-tests, with an alpha of < .05. To standardize, the participants' ratings for each statement were subtracted from the average rating of all the statements for their participant group (i.e., the clients' average rating or the providers' average rating).
Both groups had different numbers of clusters with different numbers of statements in them, which would have been di cult to compare, so it was necessary to use the same cluster structure for both groups. We chose to compare based on the clients' cluster structure because we wanted the analysis to be clientcentred and because the clients had reviewed the clusters and the statements that were in each cluster, so the nal structure re ected their feedback. We used unpaired t-tests to quantitatively compare the standardized cluster ratings to determine whether there was a statistically signi cant difference between the ratings of the clients and those of the providers for each cluster in the patient cluster map.
In the interpretation step, the maps and reports generated in the previous step are used to help frame the interpretation of the data. We assessed point maps, cluster maps, cluster reports with statements, and raw ratings reports. In this step, some of the clients were reconvened for an in-person meeting to discuss any changes that needed to be made to the maps and to provide input on the names of each of the clusters. The providers were not able to come together for a meeting, so their clusters were named by the researchers.
In the nal step, utilization, the concept maps are reviewed to determine how they can be utilized to answer the study question (25).

Results
There were 59 participants who took part in this study: 32 clients and 27 providers. This sample size is consistent with what has typically been reported in other concept mapping projects (29).

Characteristics of the Clients
The demographic characteristics of the clients are presented in Table 1. All of the clients were over the age of 25 and just over half identi ed as men. None of the clients were sleeping rough at the time of this study, while a quarter of the clients were in private housing with the remainder in shelter or tenuous, transitional, or community housing. Collectively, they had experienced homelessness or housing instability for an average of 4.8 years and had been living with diabetes for an average of nine years. A majority of the clients had experienced some diabetes-related complications and had physical or mental health comorbidities as well. The most commonly reported method of managing diabetes was the use of oral medications, which were utilized by three-quarters of the clients, although approximately one-third of the clients used injectable agents (including insulin). The clients also saw a variety of care providers for treatment of their diabetes; about three-quarters saw a family physician, one-fth saw a pharmacist, and one-third saw a medical specialist, a diabetes nurse, and a diabetes dietitian, respectively.

Characteristics of the Providers
Only eight of the providers completed the demographic survey. Most (6/8) were women, half were between the ages of 55 and 64, and 5/8 had 20 or more years of experience working with patients that have diabetes and/or complex social needs, while the rest had 15-19 years of experience. The providers also had a variety of different quali cations and titles including, registered dietitian, registered nurse, certi ed diabetes educator, family physician, program administrator/manager, and executive director. They also worked in a variety of settings such as, community/private family medicine practices, specialty practices, community health centres, community pharmacies, academic/public family medicine practices, and homeless shelters.

Concept Mapping
During the brainstorming exercise, the participants generated a list of statements representing barriers to self-management, starting with a large list, which was then re ned by removing duplicates and combining similar statements. The resulting, nal version of the list contained 43 statements (Table 2). Table 2. List of unranked and unsorted statements from the brainstorming exercise.
Using the data from the sorting exercises, the statements were plotted in relation to one another and presented as point maps through multidimensional scaling, and the points were then grouped to form cluster maps (Figs. 1a and 1b) through hierarchical cluster analysis. The clients' cluster map is based on sorting data from eight participants and it consists of nine clusters, with a nal stress value of 0.2987. Those clusters were named: 1) Challenges to getting healthy food, 2) Inadequate income, 3) Navigating services, 4) Not having a place of your own, 5) Relationships with professionals, 6) Diabetes education, 7) Emotional well-being, 8) Competing priorities, and 9) Weather-related issues.
The providers' cluster map had seven clusters, made using sorting data from 16 participants, and had a nal stress value of 0.2521. The clusters were named: 1) Access to healthy food, 2) Dietary choices in the context of homelessness, 3) Limited nances, 4) Lack of stable, private housing, 5) Navigating the health and social sectors, 6) Emotional distress and competing priorities, and 7) Mental health and addictions. Appendix A (see Additional File 1) contains tables that list which statements were in each cluster for both the clients and the providers. Using the grouping of clusters de ned by clients, we calculated standardized cluster averages for the two groups (Fig. 2). There were two clusters for which the standardized averages differed in a statistically signi cant manner, Challenges to getting healthy food (standardized difference = 0.6396; p = .013) and Competing priorities (standardized difference=-0.5250; p = .026). The clients had a higher rating for Challenges to getting healthy food than the providers and the providers had a higher rating for Competing priorities. While there were trends suggesting potential differences in the other clusters, these were the only two that reached statistical signi cance, due to a relatively small sample size for this quantitative analysis.

Discussion
The clients and providers identi ed a plethora of barriers to diabetes management, ultimately resulting in a re ned list of 43 unique barriers. The clients' and providers' sorting data resulted in concept maps with distinct cluster names and con gurations representing differing perceptions of barriers to managing diabetes while experiencing homelessness.
Both the clients' and the providers' clusters represented themes related to access to healthy food; nancial limitations; housing; health and social care; and psychosocial wellbeing. The clients chose to have a cluster titled Relationships with professionals, whereas the providers' map included those barriers in the Navigating the health and social sectors and Lack of stable, private housing clusters. While the clients' map has two clusters representing similar themes: Navigating services and Not having a place of your own, the clients saw the barriers related to relationships as being distinct from those related to navigation issues or housing issues and chose to have a separate cluster for them. These ndings con rm the notion that the perspectives of clients/patients/service users and service providers regarding barriers to diabetes management may be different, and that the reasoning behind patients' differing perspectives may not be evident to the providers. Traditionally, the identi cation of barriers and the creation of solutions has not meaningfully considered input from the individuals who have the most at stake, the patients/clients. This highlights the importance of considering the patient perspective when designing solutions for enhancing diabetes management to address the barriers which have the greatest impact, according to patients.
Comparisons of the cluster ratings indicated that there were signi cant differences between the clients and the providers in the perceived impact of the barriers faced in diabetes management. The cluster related to challenges to accessing healthy food was the most in uential from the client perspective, and it was rated signi cantly higher by the clients than the providers. The cluster representing priorities that compete with diabetes management was thought to be the most impactful by providers, and the rating for this cluster was signi cantly higher for the providers compared to the clients.
There are many reasons why accessing healthy food can be challenging for PWLEH. Many may rely on shelters or community kitchens for food so they must eat whatever is available, even if they have been advised to avoid such food by their healthcare providers (30). The meals in shelters often have high amounts of sugar, starch and fat, and there are few fruits and vegetables available, which results in diets that are likely to be inappropriate for people with diabetes (7). These shelters and community kitchens, however, also have limited resources and must often rely on food that is donated to them (31). Sometimes, PWLEH may be unable to get three meals in a day so they are often eating when they can and hoarding extra food, so they have something to eat later. Alternatively, if they had access to affordable, nutritious food, they would not be worried about where their next meal is going to come from (32) and they would have greater diabetes self-management self-e cacy (33). The lack of access to food is especially concerning for people who are using insulin, as they may use less than the prescribed amount of insulin for fear of developing hypoglycemia, if they are not able to predict mealtimes reliably (16). It is likely that because PWLEH need to negotiate their dietary intake on a daily basis, this rose to the top of the priority list for them. Providers may bene t from knowing what PWLEH nd to be the most troubling aspect of diabetes self-management.
With regards to competing priorities, studies have noted that PWLEH tend to have many demands and prioritize things such as food, shelter, and employment above diabetes care and self-management practices (14). When they continually face di culties in meeting these basic needs, people may forego preventative care or sacri ce self-care (34). This may partially explain why the providers believed that competing priorities have a great effect on diabetes management. As for why this cluster of factors was rated lower by clients, we hypothesize that this discrepancy may relate to lower health-related selfawareness in this population (35), or because clients did not fully grasp how their other issues may affect their diabetes, likely due to the fact that their social networks are often comprised of others who face similar issues. This is consistent with literature documenting that lay people are typically less aware of the impact of the social determinants of health (36). By contrast, healthcare providers are more likely to be informed about the social determinants of health and how diabetes care is affected signi cantly by the complete picture of patients' lives (37). It is also possible that barriers related to competing priorities were deemed less impactful by the clients because their greatest competing priority is accessing food, which had a separate cluster of its own, and had the highest average rating for the clients. The clients may also have felt that accessing healthy food is more important than other aspects of self-management if the diabetes education they received emphasized the importance of diet above all else.
The second highest rated cluster for the clients was focused on issues related to the weather. Cold-related injuries are very common amongst PWLEH in Canada and they often result in emergency department visits (38). One of the many complications of diabetes is reduced circulation, especially in the feet, so in cold weather foot care becomes even more important, as the winter weather can increase risk for infections, frostbite and amputations (39,40).
The providers' third highest-rated cluster focused on a lack of stable, secure housing and while the clients had a similar cluster, for them, it received the second lowest rating. It is understandable that providers would consider this an important barrier, given that there is much in the literature describing the need to treat homelessness as a health issue (41) and suggesting that it is important for providers to help address housing concerns (16). It is surprising, however, that this cluster was given such a low rating relative to the other clusters by the clients because participants in other qualitative studies have described housing as a foundational need that affects diabetes self-management in numerous ways (30,42). In one study, participants reported that being unstably housed is emotionally and physically draining, which makes it di cult to prioritize diabetes, and when the need for shelter is not met, there is no foundation from which they can pursue their health goals (42). This view is in accordance with Maslow's hierarchy of needs, which places basic needs such as food and shelter ahead of health (8,16,43). Not having housing also means not having a secure place to store diabetes testing supplies such as glucometers or medications and insulin, and there is a fear that they may be stolen in a communal living arrangement such as an emergency shelter (30). Furthermore, stable housing can provide a sense of consistency and control that can help with the routinization of diabetes care and allow some control over diet, while high housing costs can compete with the cost of diabetes care (42).
One of the strengths of this study is its participatory nature. We gathered input from PWLEH as well as a variety of providers who work in health and social settings, which ensured that we had diverse perspectives represented.
The clients were able to review the concept maps and name the clusters, which meant that the nal maps re ected their perspectives rather than the opinions of the researchers. This is important because the participants may see the value of having certain themes or representing the barriers a different way than the researchers. Concept mapping incorporates both qualitative and quantitative analyses in one process, which enables complex ideas to be explored in a short period of time, and the output of the quantitative analyses supplements and enhances the qualitative interpretations. Furthermore, the creation of visual representations through this combination of analyses provides structure and credibility to the results (25). Additionally, it includes both individual and group activities, and the process in which these activities occur avoids some of the issues that are commonly experienced when using qualitative methodologies such as, the monopolization of group discussion time by one or two individuals, the likelihood of conformity bias, or the need for individuals to publicly discuss their personal opinions or experiences (25).
Another strength is that the participants were involved in the analysis of the data and they were able to interpret the concept maps that were created using their data.
This methodology ensures that the thoughts of the participants are accurately re ected (23,25). The visual concept maps allowed us to display the associations between multiple themes and the rating data enabled comparison of the relative importance of each theme (23,25).
There are also limitations to this study, one of which is that only eight clients completed the sorting exercise. This was due to the time commitment required of participants and the complexity of the task.
Ideally, participants must take part in the brainstorming, sorting, and rating exercises, and then review the results and provide feedback, but it may be di cult to keep participants engaged throughout the whole process. Another limitation of this study is the lack of available demographic data about the providers, as only a small proportion of the providers completed the demographic survey. While it is unfortunate that we are unable to fully describe the group of providers who participated in this study, we do know that the roles they had and the settings they worked in varied considerably. Additionally, the providers did not have the opportunity to re ect on the concept map that was produced with their data. The research team made the decision to nalize the cluster map on their own because the providers were from ve different cities, and it was not possible to plan a meeting for all of the providers.

Page 13/19
The participants in our study identi ed many of the same barriers that are described in other studies. However, this study is unique in that it allowed participants to rank these barriers from most to least impactful on diabetes management -and compared these rankings between PWLEH and providers. The results show that clients and providers differ in their understanding of barriers and the impact they have on diabetes management. Given that the clients in this study have indicated that di culties with accessing healthy food are the greatest barriers to managing diabetes, future research and interventions aimed at improving diabetes management among PWLEH who have diabetes could focus on determining how to improve access to diabetes-appropriate food for this population.

Consent for Publication
Not applicable.

Availability of Data and Materials
The data pertaining to this study may be available upon request from the corresponding author.

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

Funding
This study received funding from the O'Brien Institute for Public Health and the Cal Wenzel Cardiometabolic Fund.

Authors' Contributions
The study was conceived by DJTC with meaningful contributions from RBC, GLB, KAM, SWH, and PO.
Data collection and analysis were conducted by EKG, RBC and DJTC. The initial draft of the manuscript was written by EKG and DJTC, and was revised with feedback from RBC, GLB, KAM, SWH, and PO. * one participant chose 2 housing options, therefore n sums to 33 Table 2. List of unranked and unsorted statements from the brainstorming exercise.