AcademyHealth’s Annual Research Meetings are always productive and insightful, and this year’s was no exception. Waymark’s Data Science team, led by Sadiq Patel and Parth Sheth, were in attendance, presenting three research studies on the applications of machine learning to predict preventative care utilization, the impacts of benefit-based approaches to care delivery on reductions in ER visits and avoidable hospitalizations, and an ML model that identifies pregnancy risk among patients receiving Medicaid benefits.

In addition to presenting this valuable work, the team was thrilled to spend the meeting learning from their peers about the impact of social determinants of health (SDOH) interventions on patient outcomes, new applications for ML learning models on patient care, and more. Here are their top takeaways:

AI Governance, Explainability and Workflow Fit

AI model transparency and explainability were top themes, especially in presentations focused on implementation in safety-net settings. Researchers emphasized the need for tools that are interpretable, auditable, and aligned with clinician workflows. To that end, several sessions showcased the growing adoption of ambient AI tools, such as AI scribes that listen to clinician–patient conversations and auto-generate notes. Early results presented by researchers in attendance showed a reduction in EHR documentation time and high clinician satisfaction.

Equity-First AI Design

Across sessions, there was a strong focus on ensuring AI reduces bias and improves care equity, especially through fairness audits and subgroup evaluation. There is a growing commitment to ensuring models don’t exacerbate disparities. One subtopic of this discussion was leveraging social context for the most accurate prediction models. Recent studies have emphasized the need for ML models to incorporate social risk factors into prediction models, which is consistent with findings the Waymark team has observed in their own research.

NLP for Social Risk Detection

The Waymark team also had the opportunity to join the conversation around the use of Natural Language Processing (NLP) tools that extract social risks – everything from housing and food security issues to loneliness – from unstructured provider notes. These tools can be used to supplement the structured patient data collected separately, which opens the door to more accurate, real-time social needs screening. 

Thank you to AcademyHealth for organizing such a great event, and thank you to the Waymark Data Science team for your work in advancing the evidence base in Medicaid care delivery. 

‍