Jeremy Segal, how does federated information—well integrated, easily accessible—help solve turnaround time problems?
Dr. Segal (University of Chicago): It may not help that much with the actual test being performed, but it can help accelerate care decisions on the back end. Decisions about what to do with a patient can sometimes be difficult and may take integrating a lot of information. Frequently, decisions require discussions with oncologists, radiologists, pathologists, and others, which requires assembling and looking at a large amount of data. If that data were more easily assembled, it may be an opportunity for AI-based systems to precondense and summarize the data to help the clinical team manage the patient more effectively. Clearly, you don’t want to eliminate the multidisciplinary nature of care, but it would be valuable to free up physician time and accelerate decision processes.
Ul Balis, we will have predictive diagnostics, in which we’ll be able to say on the basis of everything we know now, we’re going to do this or we know this will happen. Tumor boards will turn more into explaining—Mike Becich of UPMC coined the term “explainable AI” a few years ago. Do you agree?
Dr. Balis (Michigan Medicine): I do. Right now it’s an operational issue. Getting to the point where a number of domain experts can discuss the data will accelerate the process, but only if all the data are available. For example, most labs across the country are seeing an explosion in serum protein electrophoresis testing volumes. To carry out SPEP analysis properly, you need to constantly switch workflow between the EHR and LIS. It’s time-consuming. Using AI tools and machine learning along with federated architecture, instead of pathologists having to hunt for key clinical data elements, a “gopher” type application could instead automatically mine the chart for medications, radiographic contrast agents—which can contribute to MGUS [monoclonal gammopathy of undetermined significance]—and procedure notes. This simple shifting of effort to a “computational assistant” holds the potential to condense a nine- to 10-minute data extraction exercise per case to fewer than 45 seconds, on average. This approach can also be applied to molecular interpretation, where you have a multiplexed assay based on germline and somatic mutations as well as clinical observations.
The constant workflow breakage caused by context switching between the EHR and LIS is artificial and unnecessary. That could go away by having agentic tools that don’t replace the laboratorian or pathologist, but rather augment the quality of the primary data that could be made visible in one place, which is very akin to realizing a sign-out cockpit. Having such constructs will facilitate subject matter experts getting together and not spending time trying to sequester information. Rather, using the multidisciplinary tumor board as an exemplar, providers would readily have all the information needed to make an immediate and well-informed decision on a case. Decisions are explainable because members of the team would have, in the ready, all the information that’s needed and not be spending the 10 minutes per case that would otherwise be needed. If you’re on the fence as a clinical pathologist as to whether or not some clinical data element needs to be reviewed, you might not look in the clinical chart for each and every case. But if you could pull all the same information for every case—this being standard work—you would elevate your level of practice to a consistently higher level, with more granular and precise diagnoses being the result. This is what we’re attempting to accomplish for serum protein electrophoresis workflow at Michigan, as an example.
My colleague Dr. David Keren wrote the book on SPEP interpretation. In watching him sign out cases, I know there’s a difference between an expert and your average pathologist signing out SPEPs. Having the primary information available changes the quality of diagnosis, and you can elevate every case to that same expert level of review when you have tools that reverse federate clinical data back to the laboratory—for every case.