Value Proposition of AI for Social Care Integration in Health Care and Human Services: Qualitative Insights from Social Care Providers in the Safety Net in Michigan, US (Preprint)
BACKGROUND There are increasing interests in developing AI tools to identify and address individual-level social determinants of health in both health care and human service settings. These activities are part of social care integration, which at the individual level involves identifying individuals with social risks (awareness) and connecting them with relevant social care resources (assistance). Social care providers such as community health workers and social workers are deemed critical stakeholders in both settings. Chatbots have shown feasibility and acceptability for social risk screening in emergency departments and primary care centers. However, we do not know if a screening chatbot is worth developing for the safety net health care and human service settings, where there are visitors with greater social needs and less organizational resources. OBJECTIVE The study aims to investigate the perceived value proposition of an AI-based chatbot for social risk screening from the perspectives of social care providers in safety net health care and human service organizations. Providers’ perceived value propositions of other AI-based applications for social care integration were also examined. METHODS We conducted semi-structured interviews with 19 social care providers who have experience with awareness and/or assistance from 16 safety net health care and human service organizations in Michigan. Interview questions focused on their experiences and challenges regarding awareness and assistance when applicable. A simulated screening chatbot based on ChatGPT-4o was also used to solicit their feedback on the technology. The nonadoption, abandonment, scale-up, spread, and sustainability (NASSS) framework was used to guide data analysis. Interview transcripts were first coded deductively, then inductively. RESULTS Social care providers perceived the screening chatbot as offering limited value. This is mainly because many participants engaged in assistance activities and they noted addressing social needs is a multi-step process requiring follow-up that screening chatbots do not provide. In addition, they valued cultivating trust as many patients/clients have high social needs and lack trust in the health care system, and felt chatbots present new challenges for maintaining essential trust and care quality with clients/patients. Instead of a screening chatbot, we identified that technologies to reduce documentation burden could improve providers’ efficiency and potentially increase time spent with patients/clients. This is because social risks and needs documentation generates administrative burden for providers. We also found that technologies to improve referral accuracy and engagement could improve providers’ effectiveness, as study participants have overall limited access to technologies that typically support referral-related activities. CONCLUSIONS Social care providers in the safety net preferred AI-based applications for addressing documentation burden and social needs assistance rather than for social risk screening. Future strategies to develop AI tools for social care integration should align with social care providers’ professional values and focus on equity-centered care.