Project Spotlight: Predicting Outcomes After Cardiac Arrest

Center for Machine Learning and Health primary investigators Dr. Daniel Nagin and Dr. Jonathan Elmer discuss bridging the gap between institutions, and how that partnership has led to developing technology that will use advanced signal processing and modeling methods to allow accurate neurological prognostication sooner than currently possible.

Please share a little about your background and your research experiences.

Dr. Elmer: I am a physician scientist at the University of Pittsburgh and UPMC.  My clinical work and research both focus on treating patients with severe acquired brain injury, particularly after cardiac arrest.

Dr. Nagin: I am faculty member at Carnegie Mellon University’s Heinz College. For most of my career the focus of my research was crime policy and the developmental origins of crime and violence. As part of this research program, I developed a statistical methodology call group-based trajectory modeling (GBTM) which has come to be applied widely in biomedical research.

What led you to the Pittsburgh Health Data Alliance?

Dr. Daniel NaginDr. Nagin: I have explored the use of GBTM in various biomedical domains that were precursors to the approaches being developed in our current shared research. At the same time, Dr. Elmer had begun to use GBTM in his analyses of brain injury after cardiac arrest.  After independently pursuing related research, we saw the Pittsburgh Health Data Alliance as an opportunity to join forces to explore the commercial potential of our research.

 

Walk us through your project.

Dr. Elmer: A particular challenge faced by physicians treating patients who are comatose after cardiac arrest is distinguishing those who will wake up and make good recoveries from those who will not. Although these patients are intensively monitored in critical care units, doctors rely on only a tiny subset of available data to predict likelihood of recovery. Perhaps for this reason, neurological predictions are both slow and imprecise. Our project aims to use better predictive tools to transform the wealth of available data gathered from these patients into clinically useful prognostic information.

Dr. Nagin: GBTM is designed to identify clusters of patients following similar trajectories of bio-markers of interest, in this case various indicators of neurological activity. Dr. Elmer’s earlier research had demonstrated that trajectory groups measuring neurological activity over time had markedly different probabilities of good recoveries. The predictive tool we have developed is designed to provide physicians with well calibrated, real time predictions of the probability of a good recovery.

How do you and your project partners’ strengths complement each other?

Dr. Elmer: This project brings together experts in cardiac arrest at University of Pittsburgh and UPMC with experts in advanced statistical modeling at Carnegie Mellon University. Unfortunately, too often in modern biomedical research there is a gap between the scientists who have access to and understand clinical data and those who can implement novel analytical methods.  This project bridges that gap.

 

 

When you look at Pittsburgh as a region, what role do you see the Alliance playing? What do you foresee the future of innovation looking like here?

Dr. Elmer: Pittsburgh is unique in its wealth of biomedical data and scientists across academic institutions. Collectively, I believe we are poised to make major breakthroughs that meaningfully advance human health. An effective path to commercialization is key to bring innovative new science from the lab to patients’ bedsides. It is this knowledge translation that I see the Pittsburgh Health Data Alliance facilitating in the future.

 

To access Dr. Elmer and Dr. Nagin’s “Using the beta description in group-based trajectory models” article, click here.