Using concept mapping in the knowledge-to-action process to compare stakeholder opinions on barriers to use of cancer screening among South Asians

The multiple phases of our project are guided by the KTA process that conceptualizes a relationship between knowledge creation and an action cycle to translate research into ‘real-world’ settings. KTA views knowledge creation as occurring through a funnel that includes the multitude of primary studies or information on a topic at the widest end, knowledge synthesis in the middle, and knowledge products (e.g., EBIs) at the most narrow end. The action cycle is the process that leads to the implementation of EBIs. Based on numerous theories and frameworks of implementation, the action cycle consists of the following activities: identify a problem; identify the EBI relevant to the problem; adapt the identified EBI to the local context; assess barriers to using the EBI; select, tailor, and implement strategies to promote use of the EBI; monitor EBI use; evaluate the outcomes of using the EBI; and sustain ongoing use of EBI. The KTA process is dynamic and is accomplished through iterative exchanges between researchers and the end-users of research [37, 42].

Following an adapted KTA process, our project consists of a pre-implementation phase and three phases of implementation (Figure 1) [37]. In the pre-implementation phase, we identified low rates of cancer screening among South Asians in Ontario and implementation strategies that are effective to increase use of mammograms, Pap tests, and FOBTs. We developed relationships with three key stakeholders in our target setting including Cancer Care Ontario, the provincial authority for cancer screening programs in Ontario; the Medical Officer of Health in the region of Peel, an area of 1.2 million residents with cancer screening rates lower than other regions of Ontario [11], and a high concentration of immigrants from South Asia [43]; and the Executive Director of Punjabi Community Health Services, a community service organization that delivers culturally tailored health promotion services to South Asians in Mississauga and Brampton, the largest cities in the Peel region. These organizations were our initial community partners from the provincial, regional, and local levels respectively. Following funding from the Canadian Institutes of Health Research, we completed phase one (concept mapping) and launched phase two of the study. Knowledge gained from concept mapping is described in this paper. Future papers will describe findings from our current activities (phase two) and future activities (phase three).

Concept mapping is a participatory research method that has been widely used for program planning [41, 44‐46]. This mixed methods approach uses qualitative procedures to generate data and quantitative methods to analyze data [41, 44]. Concept mapping involves six steps: preparation, brainstorming, sorting and rating, analysis, interpretation, and utilization [47]. We conducted the preparation step in the pre-implementation phase of our project in collaboration with our initial community partners. During preparation we clarified the core issue to be addressed through concept mapping, developed the focus statements for brainstorming and rating sessions, and identified potential participants. The brainstorming, sorting, rating, analysis, and interpretation steps in concept mapping were conducted in phase one, and are described in this paper. The utilization step of concept mapping in which we are using concept mapping results to guide the selection, tailoring and implementation of interventions is being conducted in phase two of our study (Figure 1). Study activities have been approved by the St. Michael’s Hospital Research Ethic Board.

To achieve a broad sampling of ideas about barriers to cancer screening for South Asians, we recruited 53 participants for brainstorming including potential decision makers, program implementers, and program participants from Brampton and Mississauga using a snowball sampling process that was initiated by our community collaborators. Punjabi Community Health Services recruited 24 South Asian immigrants to Canada through personal networks and existing health promotion programs. We recruited residents who spoke English, Punjabi or Urdu, the most common languages spoken among South Asians in Peel [43]. Translated invitation letters and consent forms were used for non-English speaking participants. South Asian resident participants ranged in age from 18 to 49 years (66%) to 50 to 69 years (34%), were male (36%) and female (64%), and spoke English (45%), Punjabi (38%) and Urdu (17%) as a primary language. Residents’ religious beliefs included Muslim (n = 4), Sikh (n = 14), Hindu (n = 5) and Christian (n = 1). Cancer Care Ontario’s Regional Primary Care Lead helped recruit 10 South Asian primary care physicians. A total of 13 organizations participated in brainstorming. Of the 13, seven were community service organizations (entities that routinely provide outreach and education to South Asian residents for the purpose of relocation assistance, health promotion, etc. but do not plan or provide cancer screening services), and six were health service organizations (public health, provincial health service program delivery, local clinical service delivery). Representatives from community service (seven senior managers, two health promoters, one case manager, one volunteer outreach coordinator) and health services organizations (four senior managers, two public health nurses, one project coordinator, one diversity support specialist) were identified through a collaborative effort by all partners.

From the group of participants in the brainstorming sessions, we invited residents that spoke English, primary care providers, and organizations to participate in sorting and rating. We also extended sorting and rating invitations to potential participants that were referred to us by participants in the brainstorming sessions. The 46 community members that participated in sorting and rating included South Asian residents (n = 15, eight men and seven women), five primary care providers and 17 organizations (11 community service, 6 health service). Representatives from community service organizations included: nine senior managers, two settlement counselors, one community services coordinator) and representatives from health service organizations included: eight senior managers, two health promoters, one community services coordinator, one project coordinator, one diversity support specialist, one public health nurse). They ranged in age from 18 to 49 years (70%) to 50 to 69 years (30%), and were female (70%) and male (30%). The lower number of participants in sorting and rating was due to lower participation by primary care providers and our decision to not recruit Punjabi and Urdu speaking residents because of budget constraints.

During brainstorming, participants worked in groups to generate statements in response to the focus prompt, ‘A barrier to use of mammograms, Pap tests, or fecal occult blood tests among South Asians in Peel is ______________?.’ We conducted ten brainstorming sessions. On the recommendation of our community partners, we held separate sessions for male and female residents, led by facilitators representing the same gender as participants, to minimize the discomfort of discussing personal health issues in a group. Brainstorming sessions in Urdu and Punjabi were led by lay facilitators from the community who had been trained by research staff. Following these sessions, the statements collected in Urdu and Punjabi were translated to English by qualified translators. The session with representatives from community organizations was held separately from the session with representatives from health service and public health organizations. The brainstorming sessions generated 290 statements. Two authors (RL and AP) used an independent review process with iterative meetings for comparison analyses to synthesize statements that were ascertained from multiple brainstorming sessions, reduce ideas to eliminate redundancy, and edit statements for clarity. The final list of statements included 45 unique barriers to use of mammograms, Pap tests, or FOBTs among South Asians in Peel.

The majority of participants completed sorting and rating in-person with the exception of eight representatives from organizations who completed sorting and rating using the web-based Concept Systems software version 4.0175, Concept Systems, Inc. (Ithaca, NY). During sorting and rating sessions, the participants worked on an individual basis to sort statements into conceptually similar groups. Following the sorting activity, we administered four rating surveys to participants. Three surveys asked participants to rate each barrier based on, ‘How likely is it that addressing this barrier would increase the use of [specific test] among South Asians?,’ where the specific tests were mammograms, Pap tests, and FOBTs. A fourth survey measured the feasibility of addressing each barrier based on, ‘How strongly do you agree with the statement, It would be easy for the Peel community to remove this barrier within two to three years?.’ Response options were on a continuous scale for surveys one to three (1. extremely unlikely – 6. extremely likely) and survey four (1. strongly disagree – 6. strongly agree).

The analysis for sorting, rating, and comparison of ratings was also performed using the Concept Systems software. The software uses multi-dimensional scaling to create a point map based on sorting data. The point map shows the spatial relationships among statements with the goodness of fit indicated by a stress value. The stress value associated with our analysis was 0.2641, a value within acceptable limits for goodness of fit (<0.365) [16]. Details of the multi-dimensional scaling analysis are described in detail elsewhere [44, 45]. Next, the software uses hierarchical cluster analysis to create a cluster map that partitions the statements on the point map into conceptual domains [44, 45, 47]. ‘No simple mathematical criterion is available by which a final number of clusters can be selected’ [45], p.13, because the ‘best’ number of clusters ‘depends on the level of specificity desired and the context at hand’ [48], p.316. Instead of using a statistical criterion to determine the final number of clusters, we asked experts in cancer control planning to examine different cluster solutions to interpret the best number of clusters and grouping of statements. This approach is standard for concept mapping [44, 45, 47]. It is important to note that the spatial position of the statements never change in the cluster map because the statement’s position is determined by the multi-dimensional scaling. However, the clusters, formed by circles around the statements, can be influenced by asking the concept system software for a specific solution (e.g., ten-cluster, nine-cluster, etc.) or by using the software to place a specific statement in a specific cluster. The flexibility with deciding the final groupings for the statements invites the community to take ownership of the data and create information from the data that is meaningful to them [44, 45, 47]. Using the software, we created a visual display of the clusters (concept maps) with the statement numbers assigned by the concept mapping system. We computed the average ratings for each barrier and each cluster of barriers, and estimated the simple linear correlation in average ratings for groups using the Pearson correlation coefficient (r²). We report average ratings across subgroups of residents, and representatives from community service and health service organizations.

We used a research-processing stage and participant-processing stage to interpret the final number of clusters and grouping of statements [47]. The research-processing stage consisted of two study investigators (RL and AP) examining different cluster solutions and observing the clusters that were merged from the upper limit of clusters (n = 10) to the lowest limit (n = 5) [44, 48]. The study investigators decided to present a seven-cluster solution to stakeholders in the participant-processing stage because the statements within the clusters were conceptually similar to each other and discrete from statements in other clusters (Figure 2a) [44].

The participant-processing stage consisted of an interpretation session during which study investigators (RL and AP) presented the seven-cluster solution to nine community leaders (one primary care physician, four representatives from community service organizations, and four representatives from public health and health service organizations). Four residents who attended earlier sessions were also invited to the interpretation session but were not able to attend due to work or school related commitments. Consistent with other concept mapping studies, community leaders examined each statement and discussed whether the number of clusters and statements within the clusters were most appropriate for program planning [44, 45, 47]. The community leaders agreed with the seven-cluster solution and the names for the clusters. Although, they expressed the opinion that some barriers could be representative of more than one cluster and suggested that we move two barriers to other clusters. As a result of this suggestion, we moved the barrier ‘education programs do not offer materials that are well translated and culturally appropriate’ (statement 34) from the cluster labeled ‘health education programs’ to ‘ethno-cultural discordance with the health system,’ and the barrier ‘patient is concerned about the cost associated with tests’ (statement 11) from ‘patient’s beliefs, fears, lack of social support’ to the cluster labeled ‘costs.’ By moving these statements the final concept map (Figure 2b) had overlapping clusters. While concept mapping ideally strives for a clustering solution that does not have overlapping clusters [44], the researchers felt it was more important to have the final map represent the community leaders’ viewpoints since they were ultimately the ones who would use the data.

The final list of barriers, statement numbers, and cluster descriptions are provided in Table 1. The clusters included: patient’s beliefs, fears, and lack of social support (eleven statements); cost (three statements); limited knowledge among residents (seven statements); ethno-cultural discordance with the health system (four statements); limited knowledge among physicians (seven statements); health education programs (four statements); and the health system (nine statements). The concept map (Figure 2b) shows the relationships among these clusters. The close proximity of health system, ethno-cultural discordance, and health education programs clusters on the concept map shows how barriers within these cluster are more related to each other than to barriers in clusters that are further away (limited knowledge among physicians or residents; cost; patients beliefs, fears, lack of social support). The smaller size of the limited knowledge among physicians, limited knowledge among residents and health education clusters indicates that barriers in these clusters were more frequently sorted together than the barriers in larger clusters (health system; ethno-cultural discordance; patient beliefs, fears lack of social support). The overlap in two clusters reflects the community leaders’ opinions on the conceptual overlap among statements in the ethno-cultural discordance and health education clusters.

Overall, three clusters consistently ranked highest for all surveys (importance for each of mammography, Pap test, and FOBT and feasibility): ‘limited knowledge among residents,’ ‘ethno-cultural discordance,’ ‘health education programs.’ Clusters of barriers related to ‘cost’ and ‘patient’s beliefs, fears, lack of social support’ consistently ranked as lowest in importance. There were a few instances when subgroups of participants differed in opinions on the three clusters with the highest ratings. ‘Cost’ ranked second among residents for the FOBT survey and ‘limited knowledge among physicians’ ranked third among representatives from health service organizations for the feasibility survey. Residents ranked ‘limited knowledge among physicians’ lowest for all surveys but this cluster was ranked third for the feasibility survey and fourth for all other surveys among representatives from health service organizations (Table 2).

We found the correlation in ratings for clusters among residents and representatives from health service organizations was substantially weaker compared to the correlation in ratings for clusters for other bivariate comparisons (Table 3). The weakest correlation in ratings for clusters was for residents and representatives from health service organizations for the feasibility survey (r² = 0.24). Correlations in ratings for clusters among residents and representatives from community service organizations were strong for the cancer screening surveys (r² = 0.80-0.84), and relatively weak for the feasibility survey (r² = 0.54). The correlation in ratings for clusters among representatives from community service and health service organizations was r² = 0.77 or higher for all surveys.