All timing below is shown in Atlantic Daylight Time.
Sunday, May 22 — Day 1
Tutorial & Sandbox: Intro to FHIR
Introduction to FHIR session will provide an overview of technical standards; a jump-start to building web and mobile apps; an overview of the current deployment landscape, along with future directions in policy and technology development. We’re planning for an interactive agenda driven by the interests of participants. Rough agenda:
- Overview of underlying technical standards including FHIR, SMART, and CDS Hooks
- Jump-start to building web and mobile apps, including overview of open-source libraries and code
- Overview of current deployment landscape, including discussion of EHR vendor support and a tour of iOS functionality in production
- Overview of standards accelerator programs (Argonaut, CARIN) and how these groups dovetail with advances in US health IT regulations and policy development
- Open discussion
Welcome Reception & Registration
Sunday, May 23 — Day 2
Breakfast & Late Registration
Invited Talk 1 – Can We Teach AI How to Use Clinicians?
If we are ever to see the full promise of AI in the clinic, then there must be an awareness on the part of creators of AI systems of how clinicians think and work. In particular, AI systems must learn to render their findings in terms of arguments that clinicians understand and expect. This talk explores the experience of the Hugh Kaul Precision Medicine Institute at the University of Alabama at Birmingham as it has integrated AI into clinical practice for the treatment of individual patients with rare and/or challenging conditions. In particular, this talk reveals the remarkable utility of classical logic-based reasoners in both discovering possible novel treatments for an individual patient and in presenting findings in terms of what clinicians are ready to accept and implement.
Panel 1 – Helpful or Hypeful?
How much of the buzz around AI in healthcare is just that—buzz? Is it still just excitement around the as-yet-unrealized potential that has been anticipated for decades now? Or is AI making an actual difference in clinical practice and patient care, improving and/or extending people’s lives? This panel will go beyond the buzz to examine real world evidence of whether, where, and how AI is making providers’ jobs easier and patients’ lives better.
Break
Spotlight Talk 1 – Laura Early Warning System (LEWS)
Hugo Morales
Laura Early Warning System (LEWS) – An Artificial Intelligence (AI) Platform for the Management of Clinical Deterioration on the Wards – Preliminary Results from Brazilian Hospitals
Spotlight Talk 2 – System for High Intensity EvaLuation During Radiation Therapy (SHIELD-RT)
Julian Hong
System for High Intensity EvaLuation During Radiation Therapy (SHIELD-RT): A prospective randomized study of machine learning-directed clinical evaluations during outpatient cancer radiation and chemoradiation
Spotlight Talk 3 – Patient Contrastive Learning
Nathaniel Diamant
Patient Contrastive Learning: a New Approach for Constructing ECG Representations for Predictive Models
Spotlight Talk 4 – Deep Learning to Guide Point of Care Cardiac Ultrasound
Baljash Cheema
Deep Learning to Guide Point of Care Cardiac Ultrasound: Initial Validation and Use in the COVID-19 ICU
Spotlight Talk 5 – Real-time Risk Protection
Colin Walsh
Implementing Real-Time Risk Prediction for Clinical Decision Support to Prevent Suicide
Predicting Severe Adverse Events in Chimeric Antigen Receptor T-Cell Trials
Vibhu Agarwal
Predicting severe adverse events in chimeric antigen receptor T-Cell trials
Lunch
Keynote: Leveraging Autonomous AI to Solve Disparities in Healthcare
Autonomous AI in healthcare has huge potential to improve patient outcomes, lower healthcare costs, and increase access. Join Michael Abramoff, MD, PhD as he expounds on the need for an ethical foundation when developing and implementing autonomous AI and the struggles and triumphs of creating a new industry in healthcare. Applying autonomous AI technology in the medical diagnosis and treatment process makes it possible to transform the accessibility, affordability, equity, and quality of global healthcare to solve healthcare disparities.
Invited Talk 2 – Towards Trustworthy ML for Health
We begin by discussing the foundations and progress on Sepsis Watch, a system for detecting sepsis infections in Emergency Departments (ED), using Machine Learning (ML) methodology for infection prediction. We look at the subsequent ethnographic and transparency investigation of the effects of this system’s deployment within an ED environment.
Using this investigation as inspiration, we develop a framework for improved ethical evaluations and transparency. This framework is comprised of three parts: 1) Qualitiative Evaluation, 2) Demographic Slicing, and 3) Distribution Shift and Causal Evaluations. First, I present “Healthsheets”, a transparency artifact for health datasets, in the spirit of “Datasheets”, as work towards improve qualitative evaluation. Secondly, I will make a case for the importance of demographic slicing and the importance of uncertainty in evaluation. Thirdly, we discuss work on identifying models that are not likely to transfer well, wholesale or amongst particular demographics, because they do not encode the causal structure we believe they do. Lastly, I present mobile work that aims to empower chronic disease patients previous to, or together with, in-clinic care. We explore developed data collection mechanisms, and look at the use of Performance Outcome Measures, and demographic data as a novel modality for the prediction of Multiple Sclerosis (MS) progression.
Panel 2 – Bias
When does bias become a social problem rather than a technical challenge? Our panel of experts in machine learning, bias and the intended and unintended consequences of AI applications implemented without careful attention, will address this question. Early guidance on how to avoid the most significant harms will also be reviewed.
Break
Fireside Chat: AI in Medical Publishing
Get cozy with the editors of The New England Journal of Medicine, Nature Medicine and The Lancet Digital Health. At this fireside chat the topic will be what it takes to get artificial intelligence and other computational-driven papers published in top-tier journals. What are the experiments that editors are looking for? What are the innovations that get them excited? Ask your questions, get some answers.
Dinner Banquet
Tuesday, May 24 – Day 3
Invited Talk 3 – How Personal Mobile Devices and Machine Learning Can Provide Early Warning Signals to Potential Health Issues
Large-scale healthcare innovation is happening right on the devices we use every day. Revolutionary sensors in iPhone and Apple Watch can provide health and fitness metrics. Intuitive apps help users understand changes in their health while protecting user data. Virtual, large-scale medical studies bring together academic researchers, medical institutions, and healthcare organizations to accelerate innovation. Leveraging these tools and advancements in machine learning, we’re creating opportunities to detect potential health issues early on and empower users to better manage their health.
Panel 3: Tales from the Trenches
This panel will explore and discuss the challenges and opportunities of deploying artificial intelligence (AI) software and models in the clinic. Topics will include strategies for vetting and validating AI software, building trust among clinicians, bias and fairness, clinical decision support, and integration into clinical workflows.
Break
Invited Talk 4 — Using Predictive Modeling to Drive Proactive Delivery of Mental Health Care: Lessons Learned in Suicide and Overdose Prevention in the Veterans Health Administration
The national opioid overdose and suicide epidemics present health care systems with the challenge of preventing low-frequency, but often fatal events. While effective interventions for reducing risk exist, patients needing help often do not seek health care. Therefore, getting interventions to the right patients, at the right time is a challenge.
VHA developed and implemented multiple predictive models to target preventative interventions to patient populations at high risk of suicide or overdose across their 141 health care systems. We will discuss key lessons learned during implementation and evaluation of these predictive analytics driven programs, highlighting general factors for successful implementation through these specific examples. We will focus on common failure points and how these were avoided or navigated on these successful projects.
Closing Keynote: Regulating AI in Healthcare
The use of ML and AI across health care is rapidly increasing, as all stakeholders adjust to new norms accelerated by the COVID-19 pandemic, and as the evidence generation landscape expands and evolves. This talk will explore the regulatory perspective, looking at the FDA’s AI/ML-Based Software as a Medical Device Action Plan, the Data and Technology Modernization Action Plans, and other initiatives that are shaping agency expectations and capabilities, now and into the future. It will also include lessons learned on how researchers, technologists, and study sponsors can best engage with regulators in order to advance novel and breakthrough technologies that can improve patient care.