Data Work for AI-Supported Clinical Tools: Showing Value to Encourage Patient Engagement in Providing SBDOH Data
Event Type
Oral Presentations
TimeWednesday, April 144:10pm - 4:30pm EDT
LocationDigital Health
Data related to patient social and behavioral determinants of health (SBDOH) must be current, accurate, and complete to be the most useful for clinicians who use it in their treatment of patients. This is particularly true for patient SBDOH data that would be used in an EHR-based clinical decision support (CDS) application supported by artificial intelligence (AI) technologies. It can take time, however, to collect the amount and quality of data that is fundamental to such an endeavor. Patients can become overwhelmed and fatigued from providing SBDOH data for use in such systems. The presenters take the view that the challenge of eliciting and sustaining patient engagement in providing SBDOH data for AI technologies under development potentially could be addressed by clearly communicating the value of sharing such data and providing immediate payoffs to contributing patients.

Social and behavioral determinants of health (SBDOH) (e.g., diet, physical activity, sociodemographic characteristics, housing, food security, transportation access, social connectedness, etc.), while not biomedical measures, are linked to clinical health outcomes. As such, SBDOH data can be used to tailor interventions for patients to improve their health outcomes. SBDOH-informed interventions potentially could be highly effective in that they could take into consideration the specific needs of each patient. Given that the EHR is viewed as an effective tool for supporting healthcare providers as they work toward integrating SBDOH data into patient treatment, it could be particularly useful to integrate SBDOH data into an EHR-based clinical decision support (CDS) application for healthcare providers who want to support their patients making health behavior changes.

One approach to consider is for an AI technology-supported CDS application to use SBDOH data from patient questionnaires (among other sources) to identify which SBDOH factors may be clinically relevant for a patient. Once relevant SBDOH data is identified, patients would be categorized (e.g., resident of neighborhood with low walkability, limited financial resources, food insecurity, etc.). Based on those categorizations, the CDS application would provide care plan options incorporating helpful resources for the patient to overcome their barriers. Furthermore, care plan options and recommended resources could be evaluated by users, thereby allowing the CDS application to provide increasingly more effective options and resources tailored to the individual characteristics of a patient.

As promising as such an AI technology-supported CDS application may be, it relies heavily on patient SBDOH questionnaires and as a result is subject to the—at times—tenuous nature of collecting such data.

Based on experience working with clinicians and patients, the presenters have come to learn that while clinicians find patients to be the best source of SBDOH data for themselves, clinicians are often concerned about whether data from patient SBDOH-related questionnaires are current, accurate, and complete.

It is important to note that the AI system that underlies the CDS application requires a great amount of high-quality data to begin making sense of the universe of SBDOH data. It is a long and intensive process to gather and clean the amount of data that is required to be analyzed to derive understanding, knowledge, and reasoning throughout the development of algorithms that deliver valuable insights. As a result, current, accurate, and complete patient SBDOH data is needed over a sustained period of time as the application learns and develops.

The need for current, accurate, and complete data, as well as given how AI technologies are developed suggest that engaging patients in the SBDOH data collection process is highly critical for an AI technology-supported CDS application. A unique challenge that tends to be found in collecting data (including that related to SBDOH) from patients for the type of AI system that supports the envisioned CDS application is a lack of visual feedback, progress, and future value communicated to the patient.

Interactions with people, processes, and artifacts for a routine visit with a primary care physician’s (PCP) are rife with opportunities for patient confusion and frustration. As such, there is the risk of reduced patient engagement in providing SBDOH data; the system has demanded large amounts of the patient’s data over a long period of time and provided no communication about the value of the patient’s contribution of SBDOH data via questionnaire responses.
Once the AI system underlying the CDS application is mature and has the ability to surface value and elicit feedback from the patient, the risk of abandonment will lessen. The patient, however, must be engaged strategically in the early, less mature stages of the AI system. Such engagement should occur before identified points of confusion and frustration.
The presenters take the view that engagement could potentially occur through the use of transparency, feedback loops, and setting appropriate expectations related to how long the process will take and what insights could be surfaced at each stage of the AI system’s maturity. Additionally, patients should be made aware of why their SBDOH data is important, and should be provided immediate payoffs for sharing data.

Sustained contribution of SBDOH data is critical for the development of CDS applications that use AI technologies to support clinicians who are helping their patients enact health behavior changes. Challenges associated with ensuring current, accurate, and complete patient SBDOH data via questionnaires for use in the AI system could potentially be addressed by showing early on the value of contributing SBDOH data to motivate patients to continue doing so over time. The presenters encourage the HF/E community to conduct further research to better understand the nature of confusion and frustration that patients may experience when asked to share SBDOH data, 2) identify how to best communicate value to patients, and 3) determine what kinds of immediate payoffs are most appealing to patients.
Lead Product Designer