Delivery of person-centered care and optimization of these outcomes can be facilitated with the use of PtDAs. A systematic review of 105 studies found that PtDAs improved decision quality and process and reduced decisional conflict but had no impact on quality of life [15]. For trials evaluating PtDAs for diabetes decisions, patients were more likely to change their medication. With the evolution of technology, PtDAs have expanded from static pamphlets, booklets, or videos to include interactive video- and computer-based programs; the latter enable individualized content tailored to the patient’s characteristics and needs [16,17].

Given the complexity of diabetes care and multiple competing priorities, decision aids that support goal setting are particularly relevant. To date, one randomized controlled trial (RCT) of a goal-setting and SDM aid (which offered individually tailored risk information and treatment options for multiple risk factors to help patients prioritize between clinical issues) in 344 patients with uncomplicated type 2 diabetes found no difference in patient empowerment for setting and achieving goals [18]. However, this intervention neither addressed patient-important priorities and preferences specifically nor used a provider-specific point-of-care tool at the time of consultation.

To address this gap, we developed MyDiabetesPlan and a multicomponent PtDA toolkit, which includes patient-directed, provider-directed, and point-of-care tools. MyDiabetesPlan is an interactive SDM and goal-setting PtDA designed to help individualize care priorities and support an interprofessional approach to SDM, in the context of complex guideline recommendations for patients with type 1 or type 2 diabetes and other comorbidities. The overall aim of this study was to assess the impact of MyDiabetesPlan on decisional conflict, diabetes distress, health-related quality of life, and patient assessment of chronic illness care in individual patients in primary care practice groups randomized to MyDiabetesPlan.

We previously reported on how the development and refinement of MyDiabetesPlan, an interprofessional shared decision-making (IPSDM) toolkit, following the principles of user-centered design [19,20]. In this paper, we describe our assessment of the effectiveness of MyDiabetesPlan through a two-step cluster RCT followed by individual interviews. We used the Consolidated Standards of Reporting Trials (CONSORT) checklist (CONSORT-eHealth [21] and CONSORT extension for cluster trials [22]) to report this paper (Multimedia Appendix 1).

The study protocol and methods are described in previous studies [19,23]. We conducted a two-step, parallel, cluster RCT with a 1:1 allocation ratio. We selected a clustered design and randomized at the level of primary care practice groups to avoid contamination (eg, clinicians using an IPSDM approach with control patients). In brief, the first step was provider-directed (MyDiabetesPlan was delivered to physicians, nurses, dietitians, or pharmacists), whereas the second step (occurring 6 months later) was provider- and patient-directed (patients were asked to use MyDiabetesPlan by themselves before the appointment; this was then reviewed by the provider team). We chose a two-step approach because a prior feasibility study [20] identified that patients required clinician assistance for completing their initial MyDiabetesPlan. Outcome measures were administered at the first step (baseline), second step (6 months later), and follow-up (12 months later).

All primary care practice groups in Southern Ontario that had interprofessional staff (eg, nurse, dietitian, or pharmacist) and electronic medical records (EMRs, to identify patients with diabetes) were invited to participate via email, telephone, and in-person or virtual presentation to the executive or medical director; groups without interprofessional staff or EMRs were excluded. All primary care physicians in these group practices were invited to participate. A research coordinator identified patients with diabetes (type 1 or type 2) and 2 other comorbidities (heart disease, stroke, hypertension, cancer, chronic lung disease, arthritis, inflammatory bowel disorders, and urinary incontinence) from each consenting physician’s practice using a combination of keywords, International Classification of Diseases and billing codes. Patients were excluded if they did not speak English, had documented cognitive deficits, were unable to provide informed consent, had limited life expectancy (<1 year), or were not available for follow-up. Potentially eligible patients were identified via EMR query, and eligibility was further confirmed by chart review; from this group, participants were randomly selected and invited to participate and provided consent by telephone.

MyDiabetesPlan was described previously [19]. It is a web-based PtDA in which patients populate their cardiometabolic and psychosocial profiles and general care priorities; MyDiabetesPlan then generates individualized diabetes-specific goals and strategies based on these inputs that the patients then select, resulting in an action plan. After randomization, at study onset, clinicians at intervention sites underwent a one-on-one 60-min tutorial in their clinic room by the research coordinator, with access to a one-page how-to guide and 2-min video. During subsequent clinical encounters, a member of the interprofessional team (nurse or dietitian) logged into MyDiabetesPlan and completed it with the patient; the physician subsequently reviewed the resultant action plan with the patient. At 6 months, patients at intervention sites were provided with a patient-directed how-to guide and video and directed to update MyDiabetesPlan according to their progress before the appointment. The research coordinator followed up with participants by email and telephone at study onset, followed by quarterly debriefing sessions, in both individual and group formats.

Clinicians in the control sites received paper copies of the executive summary of the Diabetes Canada clinical practice guidelines [24] and a postcard outlining web-based clinical information resources. After 6 months, patients in the control sites received a Diabetes Canada patient education pamphlet [25] regarding diabetes self-management and a postcard outlining web-based additional patient resources.

The primary outcome was decisional conflict [26]; secondary outcomes were diabetes distress [27], health-related quality of life [28], chronic illness care [29], and intention to engage in IPSDM [30] (Table 1). Decisional conflict was measured using the Decisional Conflict Scale (DCS), which can predict individuals’ intentions and subsequent behavior. It has a test-retest coefficient of 0.81 and internal consistency coefficients ranging from 0.78 to 0.92 [7]. These outcomes were assessed at the individual participant level, at baseline, and at 6 months and 12 months (after an appointment) through a web-based survey or by mail.

With at least 40 patients per physician, 50% patient participation rate, and an anticipated patient attrition rate of 25% [38], we estimated that approximately 15 patients per practice would be able to participate. On the basis of a previous study using the DCS [7], we selected a standardized effect size of 0.4 with an SD of 0.6, α of .05, and β of .10. Previous data have shown that ρ (intraclass coefficient) for decisional conflict for patients living with diabetes clustered within primary care physicians is 0.013 [39]. Therefore, accounting for clustering, 56 patients per intervention/control group, or 4 sites per intervention/control group would be required.

Practices were simultaneously randomized and allocated by a biostatistician to either intervention or control using computer-generated randomization in a 1:1 ratio. Each practice was given a code, and the biostatistician was blinded to the allocation status. After assignment, investigators, research coordinators, and trial participants were no longer blinded to group allocation owing to the nature of the intervention. The list of all eligible patients from each cluster was randomly ordered; patients were recruited from this list until the target sample size was met.

The practice and sociodemographic characteristics of clinicians and patients were obtained at baseline. Outcome data were collected using participant questionnaires [23] distributed through web or by mail according to patient preference for pragmatic reasons at baseline, 6 months, and 12 months.

A modified intent-to-treat analysis was conducted. For the primary and secondary outcomes, a linear mixed effect model was used to analyze the total score for each scale where site was the random effect with adjustment for baseline value. To account for missing data, we conducted a fully adjusted mixed effect model using repeated measurements [40-42]. The baseline variables we adjusted for were baseline DCS score, age, sex, ethnicity, educational attainment, employment, and living arrangements as well as a history of cancer and heart, musculoskeletal, respiratory, mental health, kidney, eye, and nerve disease.

The impact of sociodemographic variables on these outcomes (decisional conflict, diabetes distress, health-related quality of life, and chronic illness care) was assessed. Specifically, we fit a main effects model that adjusted for age, sex, ethnicity, education, employment status, and living arrangements as well as a fully adjusted model that included all interactions between treatment and the preceding variables. As these subgroup analyses were explorative, these were performed on the complete case data because the use of repeated measure data would necessitate the need for interactions with time as well. P values for the treatment effect in the baseline adjusted models used Satterthwaite approximation for the denominator degrees of freedom, whereas the tests of interactions (subgroup effects) used likelihood ratio tests from a full maximum likelihood estimation [43]. Irrespective of the test result on subgroups, the treatment effects were then shown by subgroup, estimated from the second model specified above, along with 95% CI and a P value that tested each interaction in this fully adjusted model. Descriptive statistics were used to describe the intervention effect by MyDiabetesPlan use. Analysis of variance was used to assess differences in the clinician’s intention to practice IPSDM scores between the intervention and control groups at baseline, 6 months, and 12 months. Analysis was performed in R version 3.5.2 [44], and the packages lme4 (version 1.1-21) and lmerTest (version 3.1-0) [43] were used to fit and report the mixed effect models.

The study was approved by the Research Ethics Boards of Bridgepoint Health (15-009-BP), Markham Stouffville Hospital (Canadian Institutes of Health Research [CIHR] protocol, v1, January 2013), Michael Garron Hospital (609-1410-Mis-245), North York General Hospital (13-0265), Southlake Regional Health Centre (0055-1314), St. Michael’s Hospital (13-014; includes Humber River Regional Hospital), Sunnybrook Health Sciences Health Centre (345-2013), University Health Network (16-6044), and Women’s College Hospital (2013-2058, 2014-0043-B).

The study was funded by the CIHR Knowledge to Action Operating Grant (Funding reference number KAL 290086), which had no role in the design, collection, management, analysis or interpretation of data, or in the writing or publication of the manuscript. SS is supported by a Tier 1 Canada Research Chair. NI was supported by New Investigator Awards from CIHR and from the Department of Family and Community Medicine, University of Toronto.

A total of 10 primary care practice groups were recruited from December 2014 to November 2015; patients were recruited from December 2015 to September 2016, followed by October 2016 to September 2017. Recruitment metrics and the CONSORT flow diagram are shown in Figure 1. The practice and sociodemographic characteristics of clinicians and patients are shown in Table 2. In the intervention group, 50.0% (51/102) and 46% (33/72) of participants completed the questionnaire via web-based platform for time points 1 and 3, respectively. In the control group, 47.7% (53/111) and 51% (40/79) of participants completed the questionnaire via web-based platform for time points 1 and 3, respectively (Multimedia Appendix 1).

A total of 62 patients withdrew or were lost to follow-up. Of these patients, we had demographic data on 34 patients (control, n=10; intervention, n=24; Multimedia Appendix 1). There were proportionately more non–English-speaking patients (9/34, 26%) with high school education (11/34, 32%) who withdrew, compared with those who remained in the study.

Total decisional conflict was modestly reduced in the intervention group at 12 months compared with the control group (−3.5 of a total score of 100, 95% CI −7.4 to 0.4, P=.08; Table 3). At 12 months, the Uninformed subscale was reduced in the intervention group (−3.9, 95% CI −8.8 to −1.02, P=.11). Similarly, at 12 months, the Unclear Values subscale was reduced in the intervention group (−3.6, 95% CI −9.6 to 2.28, P=.21).

We found that MyDiabetesPlan, an interprofessional goal-setting decision aid for people living with diabetes, resulted in a modest reduction in decisional conflict (specifically the uninformed subscale) and increased patient assessment of chronic illness care but had no impact on diabetes distress or health-related quality of life. MyDiabetesPlan reduced decisional conflict and diabetes distress most prominently in participants older than 65 years. There was no impact of MyDiabetesPlan use on clinicians’ intention to practice SDM.

Our finding regarding the impact of MyDiabetesPlan on decisional conflict is generally consistent with the literature, although our results did not meet statistical significance. A 2017 systematic review found that decision aids reduced decisional conflict related to uncertainty caused by unclear values (−8.81/100; 95% CI −11.99 to −5.63; 23 studies; n=5068; high-quality evidence) and feeling uninformed (−9.28/100; 95% CI −12.20 to −6.36; 27 studies; n=5707; high-quality evidence) [45]. In our study, we found smaller reductions: a 3.5-point reduction in the total scale, a 3.9-point reduction in the uninformed subscale, and a 3.7-point reduction in the unclear values subscale. This discrepancy may be due to the goal-setting nature of MyDiabetesPlan: rather than offering risks and benefits of a single discrete decision, it offered risks and benefits of multiple potential strategies. Thus, answering a question in this context, such as “I am clear about which benefits matter most to me” may be more challenging.

Consistent with a previous review [45], we also found no impact of MyDiabetesPlan on general or condition-specific health-related quality of life such as diabetes distress (though there was a signal for an effect in participants older than 65 years). Authors have postulated that this may be owing to the fact that decision aids are often used in situations where the options have no clear health outcome advantage [45]. A subanalysis of this review identified 11 studies involving 2684 patients that examined the impact of PtDA on health-related quality of life [46]. Of the 11 trials, 6 trials neither reported difference in health-related quality of life between PtDA and control nor over time. This confirmed the lack of impact, suggesting that health-related quality of life may be an uninformative end point unless a specific hypothesis for its impact can be made [46]. Another potential reason for this is that the outcome is too distant from the act of using MyDiabetesPlan: patients may select a goal (such as increasing physical activity) but may not enact the goal (ie, actually increasing their physical activity), and thus may not experience any change in health-related quality of life. Although MyDiabetesPlan may have reduced decisional conflict and increased patient activation and thus increased goal setting, there may be a gap between goal setting and goal attainment; bridging this gap with additional behavioral supports to optimize goal attainment may then result in improvements in health-related quality of life.

Our finding that MyDiabetesPlan reduced decisional conflict and diabetes distress, particularly in participants older than 65 years, is consistent with the literature [47]. This systematic review of 22 studies examining the impact of PtDa in adults aged 65 years and older found that people exposed to a decision aid had greater knowledge (5 studies; mean difference 6.5, 95% CI 0.76 to 12.25) and reduced decisional conflict (11 studies; mean difference −3.17 out of 100, 95% CI −4.44 to 1.90). These findings are reassuring, particularly in light of concerns of social inequities and the digital health divide [48], and provide support that a digital health innovation such as MyDiabetesPlan is appropriate for a complex older population with multiple comorbidities.

The strengths of this study include rigorous adherence to RCT methodology (including central randomization, similar intervention and control groups at baseline, appropriate length of follow-up, use of validated scales, intention-to-treat analysis), its finding of reducing decisional conflict and improving chronic care delivery, especially in those older than 65 years, and its generalizability. We designed this study to assess the feasibility of implementing MyDiabetesPlan in interprofessional primary care settings; as such, it is primarily a pragmatic trial along the explanatory-pragmatic continuum [49]. Thus, our study results mainly reflect what would happen if MyDiabetesPlan would be implemented in the usual clinical practice of interprofessional primary care.

Study limitations include the lack of blinding of participants (patients and clinicians) owing to the nature of the intervention, use of both paper- and web-based data collection methods, attrition rate of 29%, and less-than-anticipated MyDiabetesPlan use, resulting in potential bias, and its lack of clinical outcome measures. Although different data collection methods may introduce respondent bias, recent literature has shown that response rates do not differ between paper- and web-based respondents [50] and that there was no difference in socioeconomic characteristics between paper- and web-based respondents [51]. Our attrition rate and reduced MyDiabetesPlan use were amplified by our complex study population, many of whom either withdrew owing to competing health concerns or were lost to follow-up, presumably for similar reasons. Challenges to conducting research in this population are well documented, and solutions for future studies include providing transportation compensation, conducting home visits, and encouraging greater engagement of family members [52]. Although we did not assess clinical outcome measures, we assessed proximal patient-reported outcomes appropriate for a decision-aid intervention, outcomes that have typically been underused in the literature [53], and outcomes that have been associated with clinical outcomes such as A_1c [12-14]. Assessment of researcher-selected clinical outcomes such as A_1c in a trial where patients are encouraged to select their own personalized goals is incongruent with the principles of person-centered outcomes research [54]. Future studies may consider the use of goal attainment scaling (which has some demonstrated validity evidence in the geriatric care setting) [55] or composite clinical outcomes, though this is not without its limitations [56].

MyDiabetesPlan modestly reduced decisional conflict and increased patient assessment of chronic illness care but had no impact on diabetes distress or health-related quality of life. The next steps in this research program are to engage with knowledge users (patients and their caregivers, clinicians, managers, and policy makers) to discuss the implications of these findings, modify MyDiabetesPlan and its mode of delivery, and consideration of clinical, organizational, and health care system contexts to plan scale-up to an implementation trial.