April 7-9, 2022 • Hyatt Regency • Lexington, KY
Communication Strategies to Promote Comprehensive Well-being
Communication Strategies to Promote Comprehensive Well-being
Abstract: Understanding Key Stakeholders’ Perceptions of Using Artificial Intelligence to Discover Patients’ Social Determinants of Health
Background: Social determinants of health (SDOH) are non-medical factors that influence health outcomes, such as where people are born, live, learn, work, worship, and age that affect health, quality-of-life, and risks. It is critical that clinicians are aware of patients’ SDOH, but such data are not captured consistently and can be difficult to decipher within clinical documentation. SDOH are blended in with unstructured, free-text clinical notes. As artificial intelligence (AI) capabilities expand, it is possible to mine SDOH information within electronic health records (EHR) using advanced natural language processing (NLP) methods. Our objective was to understand the feasibility, challenges, and benefits of developing an AI system to find SDOH through qualitative interviews with key stakeholders. Specifically, we aimed to discover: RQ1: How are SDOH interpreted? RQ2: What are the most/least important SDOH? RQ3: How are SDOH communicated and documented when interacting with patients? Method: Using the Delphi technique, a framework that elicits the opinions of a group of advisers who are knowledgeable about a certain topic, in-depth, semi-structured interviews were conducted with 1) clinicians, 2) data analysts, and 3) patient navigators. Interviews were recorded and professionally transcribed. Upon transcription, the research team performed a thematic analysis using the constant comparative method. Results: Fifteen stakeholders participated (8 navigators; 3 clinicians; 4 analysts). RQ1: The theme, “Familiarity,” was observed as participants from all three groups accurately articulated a definition of SDOH, although the phrase “SDOH” was unfamiliar to analysts and navigators. RQ2: The theme of “All SDOH are important” occurred because clinicians and navigators had difficulty identifying an unimportant SDOH. Social support and housing were deemed most important by clinicians and navigators, but analysts almost exclusively used zip code data, as it was easily accessible. RQ3: The theme, “SDOH affects health care delivery,” transpired as clinicians and navigators used SDOH to determine if laboratory tests could be administered at locations closer to the patient. Similarly, clinicians and navigators solicited SDOH information during conversations with patients, which helped form connections and contributed to understanding patients’ lifestyles. Patients were very willing to share information, but the EHR had a limited number of structured data fields to capture such information. Therefore, clinicians acknowledged that some SDOH may not be included in their notes due to manual documentation, but navigators would input their own note in the EHR if patient information were missing. All stakeholders acknowledged that interpreting medical jargon in the EHR made it difficult for patients to check the accuracy of information when reading notes. AI systems were endorsed as a way of synthesizing data to uncover additional SDOH information. Conclusions: SDOH data is extremely valuable for patient care, but can be difficult to access due to the way unstructured data is entered into the EHR. AI, using NLP, can ease the burden on clinicians and navigators by identifying hidden SDOH data within the EHR. Having additional SDOH would enable clinicians and navigators to provide better patient care, while data analysts would be able to use identified SDOH variables to perform health outcomes research.