AI Talk: American Medical Informatics Association – Annual Symposium 2021 (part two)

Nov. 19, 2021 / By V. “Juggy” Jagannathan, PhD

Continuing the discussion of my observations from last week, here are a few highlights from various sessions conducted during the 2021 AMIA National Symposium. The sessions can be grouped under three broad themes: pandemic response; clinical decision support and noteworthy advances.

Pandemic response

The first theme was examining the impact of the pandemic. There was a panel devoted to lessons learned on this topic. Dr. Topol referred to an MIT Technology article about the hundreds of AI tools developed in response to the pandemic, none of them being very good. The fundamental problem here is that all these researchers were developing solutions based on small, siloed datasets.

To counter this, the NIH launched a big effort: The National COVID Cohort Collaborative (N3C). This program supports the pooling of resources, harmonizing the data format and providing controlled data access to researchers with strict use agreements in a GovCloud. The researchers would not be allowed to download the dataset. Research findings are now beginning to emerge from this democratized clinical data infrastructure. Currently, the N3C dataset has 9.8 billion rows of data, 8.6 million patients covering more than 3 million COVID-19 cases.

Clinical Decision Support

The second area of focus was on clinical decision support (CDS). This particular area has been around forever, so what is new now? I saw a demo of a new system that had an interesting approach. CDS is about providing diagnostic and therapeutic guidance to physicians. The demo focused on how to design and evaluate CDS solutions. The goal is to impact some outcome and to impact the outcome, some process must change. For the process to change, timely intervention needs to happen (nudge). An analytic framework for the design of CDS should incorporate the above three elements – systematically tracking how effective nudges are, how often the desired process change happens and whether it has an impact on the outcome. 

Another effort which caught my attention in the CDS space was work that tracks new research and identifies what clinical guidelines are being invalidated by new research. Primarily, underscoring the reality that CDS and clinical practice guidelines have to continuously be updated to factor in new research. We all know this to be true in the COVID-19 era!

Noteworthy advances

Lastly, there were a number of talks that pointed to the fact that CDS needs to be personalized—the recurring theme of precision (or rather accurate) medicine.

The third area I want to highlight is everything else! I was disappointed that only a portion of the conference was available in the virtual format. Nonetheless, one of my favorite sessions was available to the virtual audience: “Year in review.” This particular session has been around for more than a decade, where a leading researcher—in this case Dr. James Cimino—summarizes noteworthy research published in the past year. AMIA has a dozen or more working groups on every imaginable topic. The working groups submit recommendations and the moderator selects a few from each group to highlight. This gives a waterfall view of all that is happening in terms of research. A few that got my attention:

  • ”Swarm learning” or federated learning approach where data remains at a specific institution while the models are centrally aggregated.
  • Precision medicine was another area of emphasis.
  • N3C work discussed above made the list.
  • A model dubbed UmlsBERT – specifically trained to deal with clinical domain concepts and create better embeddings of clinical concepts.

Of course, I am not doing justice to the hundreds of papers and posters presented, but overall the event was very enlightening.

I am always looking for feedback and if you would like me to cover a story, please let me know! Leave me a comment below or ask a question on my blogger profile page.

V. “Juggy” Jagannathan, PhD, is Director of Research for 3M M*Modal and is an AI Evangelist with four decades of experience in AI and Computer Science research.