Key points are not available for this paper at this time.
BACKGROUND Background In the United States, diabetes is the eighth leading cause of death 1, with an estimated 38 million people living with diabetes 1. This chronic condition requires consistent care for effective management to help avoid poor health outcomes 2. In fact, it is estimated that people with diabetes (PWD) spend over 8,000 hours per year managing their diabetes outside of medical settings 3. Diabetes distress is the disruptive and demanding emotional response to these daily demands of living with diabetes 4. This emotional burden associated with diabetes is pervasive, with one in four people experiencing severe diabetes distress 4. DD is associated with negative impacts on engagement in self-care and self-management behaviors, medication adherence, and exacerbation of mental health conditions 5. Accordingly, the American Diabetes Association (ADA) recommends that diabetes care be delivered by an interdisciplinary team with a person-centered approach 6, and that it include regular screening for and monitoring of DD in routine diabetes care for PWD with treatment for DD to be provided by practitioners with specific training to address DD 5. Additionally, a recent white paper by the National Committee for Quality Assurance (NCQA) suggested having a variety of pathways to tailor DD treatment for individuals who screen positive for DD by involving relevant healthcare professionals and care modalities 7. Yet, in everyday clinical settings DD is infrequently identified and only a small number of PWD are asked how diabetes affects their life by their healthcare professionals 8. As primary care is where most people receive their diabetes care, it is a crucial setting within which to assess and address DD 9,10. However, there remains a lack of consistent screening for DD within primary care, which likely contributes to the emotional burden of diabetes going undetected and untreated 4,11. Thus, even though DD is highly prevalent and there exist well-validated measures to assess for DD, there is a significant knowledge gap in best practices for implementing DD screening and treatment interventions systematically in primary care. Clinical decision support systems (CDSS) have proven effective for prompting providers to deliver recommended care 12. In general, CDSS improve healthcare delivery through the utilization of technology to enhance clinical-decision making, sometimes even using data and observations that are normally unobtainable by providers alone 13. Clinical decision support technology leverages electronic health records, medical knowledge databases, and algorithms to provide patient-specific recommendations, thus enabling providers to make more informed decisions 14. Benefits can include a reduction in medical errors, enhanced patient safety, improved decision-making, and scalability 14. The recent integration of AI into healthcare technology has led to the classification of CDSS as either knowledge-based systems using traditional technology frameworks, or non-knowledge-based systems to indicate the utilization of AI to transform data into information for the user 15. Recent reviews of studies that focus on the implementation of non-knowledge-based CDSS in diabetes care have demonstrated significant improvements in patients' blood glucose, blood pressure, and lipid profiles in 71%, 67%, and 38% of the studies, respectively 15. While CDSS can promote diabetes care by facilitating patient self-management, it is hoped that further emerging technology will allow for more efficient and effective management for many people living with diabetes 15. OBJECTIVE Objective diagnosed with type 1 or type 2 diabetes; patient at the primary care clinic; able to understand English or Spanish; willing and able to sign the Informed Consent Form (ICF); willing to be contacted by the study team through the patient portal, phone, or text to complete study measures; ability to reliably send and receive text messages. Exclusion Criteria: Participation in another study that might interfere with participation in this study; unable to follow the study procedures for the duration of the study or is deemed unacceptable to participate in the study per PI judgment; participant or participant’s immediate family member is an employee of the healthcare chatbot company providing services for the study; planning to move in the next 6 months; planning to change primary care practices in the next 6 months. Data Collection Outcome Measures Both qualitative and quantitative methods will be used to assess study outcomes at the patient level, practice level, and technology system level. Patient-level health outcomes include DD, depression, and anxiety measured at baseline, 3 months, and 6 months using the T1-DDAS or T2-DDAS, PHQ-8, and GAD-7. User experience with technology will be assessed through administration of the UMUX-lite at 3 months, the System Usability Scale at 3 months, and a technology use assessment at baseline and 3 months. In addition, qualitative data from interviews and responses to open-ended survey items will be collected from practice staff, primary care physicians, and patients to determine the acceptability and feasibility of implementing screening and technology-supported treatment for diabetes distress. Further, PCPs will complete the Diabetes Distress Provider Time Survey to track activities and time spent on workflow tasks. Acceptability with the eConsults will be evaluated by standard data collection protocols in the EHR. eConsult data will capture the sent requests from PCPs and responses given by specialists. AI chatbot performance data will be collected from the chatbot system and used to evaluate engagement with the AI chatbot according to the People at the Center for Mobile Application Design (PACMAD) framework 23. Table 1 lists the measures to be collected for this study in detail. RESULTS Results Implementation Status The principal investigator met with clinic leadership to describe the project and obtained buy-in from the team. Workflows were developed that outline study design and patient flow (see Figure 1). The principal investigator conducted a team training with the clinic where this study will take place as well as met with the primary care physicians that will be participating in this study to provide education and training on diabetes distress, an overview of validated measures for diabetes distress (e.g., T1-DDAS and T2-DDAS) and conversational tools that can be used to support people with diabetes and diabetes distress. T1-DDAS and T2-DDAS scoring was incorporated in the EHR through an EHR build. Chatbot content specific to diabetes distress was developed with assistance from two leading diabetes psychologists with expertise in diabetes distress. The chatbot was field tested by the primary research team and patients with diabetes through a patient advisory committee and feedback was incorporated in refinements of the chatbot content. Recruitment Status Patient recruitment is anticipated to begin during late June-July 2024. Research Status IRB approval was obtained on March 15, 2024. We have signed contracts with RAs that will be performing the duties of providing outreach to eligible patients, obtaining informed consent from patient participants, administering screening tools at the 3 month post screening and 6 month post-screening timepoints. Additionally, RAs will ensure that data collection from patient screeners is complete and documented appropriately. Following 3 months post screening, research assistants will conduct the semi-structured interviews and administer the survey questions with both patients and clinical staff. At the 6 month study timepoint, the primary team will send the remaining screeners to patient participants. The primary research team has been meeting weekly since November 2023 to develop the research plan and discuss project tasks. Technology Status Licensing agreements and institutional risk assessment approval for the AI chatbot were obtained before patient recruitment. The AI chatbot was initially beta-tested by the primary research team and patients with diabetes to determine if messages could be delivered on a schedule and that the system could get replies back. The beta-test revealed that additional content related to suicidality and ‘hating having diabetes’ was needed, more training on the model to correctly match intents to the content library was needed, and that more of the intents had existing content in the chatbot library but that improving the link to these was still needed. Prior to study launch, the primary research team retested the chatbot system to ensure that the updates had been completed. eConsults are an active clinical care option for PCPs at our institution, fully integrated into the EHR and in use by over 28 specialties 25. As a result, they are a readily usable aspect of this study. For our study, specialists available by eConsults will include behavioral health, social work, care management, diabetes education, pharmacy, and endocrinology. T1-DDAS and T2-DDAS were created and incorporated into the EHR as flowsheets. The PHQ-8 and GAD-7 are already embedded in the EHR. The results of the diabetes distress screeners will be able to be pulled into visit documentation, facilitating care coordination with eConsulted providers. Additionally, flowsheet data can be tracked over time and will be easily accessible for providers to review during and after patient visits. Funding Status Funding for this study was secured from the Peer Mentored Care Collaborative (PMCC) in January 2024. CONCLUSIONS Conclusions Creating and disseminating workflows for screening and treating DD in primary care is an important component for delivering whole-person diabetes care. The use of an AI chatbot to deliver individualized treatment and support for DD and eConsults providing additional specialty support are expected to help increase support and treatment for DD without contributing to increased workload for primary care practices. This study is intended to help us to begin to understand how to implement diabetes distress screening and treatment in primary care settings in a scalable and real-world manner. CLINICALTRIAL N/A
Kostiuk et al. (Mon,) studied this question.