José Luis Costa, how realistic does it seem to you? Most of us think the health care system is a difficult place to navigate, whether you’re a patient or provider.
Dr. Costa (Thermo Fisher): It will help in the long term. I would like to connect this with the discussion on turnaround time. There’s an opportunity to reduce turnaround time where the test is being performed. Many radiology treatments are executed at the hospital where the patient is seen. If molecular testing is performed at the same hospital, a decentralized model for molecular testing, that’s where time will be saved. AI will not change how long it takes for the specimen to be transported from the hospital to a centralized lab and for results to come back.
Pierre Del Moral, your comment here?
Dr. Del Moral (Illumina): I’m in a global role, so I see that not every country and health care system is created equal. In addition to AI making decisions that help provide better outcomes for patients, we can help developed countries increase testing volumes or enable more testing closer to the patient. In less developed countries, or even in rural areas of the United States, it’s hard to find the professionals who make these interpretations. Through this increased efficiency, there is opportunity to provide more accessibility, including to NGS, and increase volume.
Ul Balis, you serve many referral clients. How do you see this question of how much we need technology in community settings?
Dr. Balis (Michigan Medicine): There’s always an economic argument about make versus send, if your volumes are sufficiently high to justify bringing an assay in-house. That said, we want to provide all tests locally for several reasons. It is faster to do it locally because you’ve eliminated transport time. In terms of data repatriation, if you do testing locally, especially for NGS, maybe you won’t keep the FASTQ files, but you’ll keep the BAM and VCF files and the interpretive pipeline version that was used, so you can revisit it if needed. Having that primary data is essential to refining models as you learn more about outcomes. The more data you can generate locally, the faster that discovery becomes. Such an approach can be combined with the placement of deidentified outcomes and primary molecular data into monolithic repositories, which ideally would be centrally placed in internationally accessible locations. In this manner, the democratized access and larger data set size afforded by aggregated submissions from the expected many contributors would yield greatly enhanced potential for widespread use and discovery.
The economic realities are the limiting factor—you can select some technologies, but you can’t do everything at once. We’ve been interested in long reads for a while. We brought that technology in, and it has fulfilled a missing link—for example, optical genome mapping. Bringing in new technologies can decrease turnaround time but it also allows for acquiring newer classes of rich molecular data. This, in turn, can be federated with clinical outcomes data, clinical observations, and extant lab data, making for a powerful discovery combination. Your best chance of federating is to generate data locally. I’m a fan of repatriating lab testing, specifically molecular, if the economies of scale allow for it.
Jeremy Segal, your thoughts?
Dr. Segal (University of Chicago): In general, I favor testing locally. In certain institutions, that can be more difficult to do, but the technology is always changing, so these decisions change over time.
One of the main things I like about my job is that we’re here and these are our patients. We do the testing and we participate in their care. When questions or difficulties arise, we get in touch with the oncologists or they with us, and we take care of the patient as a multidisciplinary team. That’s the best way to practice genomic medicine, and it can happen only if the expertise is local and you can talk as a group. But it raises the issue of silos. If everything’s local, then there’s a silo problem, which can potentially result in inefficiency. That means institutions need to work together and share data and expertise. That’s what we’re trying to do with the GOAL [Genomics Organization for Academic Laboratories] consortium. Through that group, we’re now trying to set up a multi-institutional variant and interpretation shared database, so a small center can get exposure to the knowledge bases of many institutions. Then, if you see a strange genomic finding in your lab, you can ask if someone somewhere else has seen it and what they said about it. That helps standardize reporting practices and interpretation language and ensures a patient gets the same result and interpretation regardless of institution. The hospital laboratory molecular community is a community, and we shouldn’t think of one another as being in competition. We should be working together to help patients broadly across all our centers.