Project Spotlight: Predicting Outcomes After Cardiac Arrest (update)

When we last checked in with Dr. Jonathan Elmer, primary investigator for the Predicting Outcomes After Cardiac Arrest project from the Center for Machine Learning and Health (CMLH)  he discussed how the PHDA has helped to bridge the gap for developing technology used to advance signal processing and modeling methods accelerating the process of accurate neurological prognostication. In this project update, check back in with Dr. Elmer as he outlines how his project has progressed, as well as a roadmap for where it’s heading.

Can you please provide a brief refresher on your project and its goals? 

There are about 600K victims of sudden cardiac arrest a year in the U.S. Among those who are successfully resuscitated and survive to hospital care, most are initially in a coma from brain injuries sustained during their cardiac arrest. There are a good portion of those patients who can wake up from their comas and have favorable outcomes. Unfortunately, there are other patients who will ultimately succumb to the sequelae of that schematic brain injury.

Currently, the tools available to predict recovery potential or to estimate recovery potential early after a cardiac arrest are very limited. The result of this is a lot of agonized waiting for families with a lot of uncertainty, which has a significant emotional burden for understandable reasons. It also means that we provide a lot of ICU care that is ultimately recognized retrospectively to have been futile, because the patient turns out to have had no recovery potential. Perhaps the most troubling consequence is that there are a substantial portion of patients nationally who have withdrawal of life-sustaining therapies and a transition to comfort-oriented care and ultimately die shortly after.

The clinical grounding of our project is this desire to get clinical providers and patient families information about recovery potential that’s more accurate and timely. This allows us to alleviate both the burden of futile care on patients, families, and society, but also reduce mortality by preventing avoidable deaths from withdrawal of life-sustaining care. Ultimately, our goal is to hammer out some of the detailed and rigorous methodologic issues necessary to apply trajectory modeling to this clinical problem where the accuracy in predicting poor outcomes needs to be near perfect.

What was most exciting or unexpected that you learned during the project?

I think our current results demonstrate that accurate, timely, and reproducible outcome prediction is possible in these patients in a way that is both faster and more precise than ever before.

What are the biggest takeaways from your project?

Clinicians, family, and society demand an incredible degree of perfection in this particular clinical problem, because this is truly a life-or-death decision for patients and families. I think our takeaway from this is that our findings are promising, but we are at the beginning of this process, not the end.

What are your project’s next steps? 

Supported in large part by the results in the work that was carried out with funding from the PHDA grant, we are now in the first months of a large R01 NIH grant that is further building on our data set and also on some of the analytical tools, including trajectory modeling.

Combining the data and these tools with AI and machine learning approaches and advancing some other aspects of the work that will improve the interpretability and the acceptability of these algorithmic predictions to both providers and families.

The work we’ve done thus far has been foundational and a really important proof of concept, so we are still working towards more definitive results with our ongoing research. It’s daunting, but really exciting.