Joe Corkery, Google Cloud product management director, explains how the new API will help developers scale healthcare solutions.
Dan Patterson, senior producer for CNET and CBS News, spoke with Joe Corkery, director of product management, healthcare and life science, Google Cloud, about the use of machine learning in healthcare applications. The following is an edited transcript of their conversation.
Joe Corkery: The Google Cloud Healthcare API is an application, or basically an application layer that we built to enable healthcare data interoperability, to enable healthcare organizations, healthcare application developers, to share a wide variety of different types of healthcare data types. In particular, it’s focused on medical record and medical imaging data, supporting DICOM (Digital Imaging and Communications in Medicine) data for medical imaging, as well as HL7v2 (Health Level Seven International, version 2) messages as well, and the FHIR (Fast Healthcare Interoperability Resources) records for clinical data. It helps with the ingestion, storage and serving of that data to enable organizations to do analytics on that data, to train machine learning models, to build applications on top of that.
SEE: Hiring Kit: Computer Research Scientist (TechRepublic Premium)
I think one of the first things to understand is there are a small handful of industry-standard formats for representing healthcare data. Those are the ones I mentioned earlier, HL7v2, FHIR, DICOM. And we’ve invested heavily in making sure that we are building our applications to meet those open standards. And then, building out tools that allow our customers and partners to take the different flavors of those standards that they might have, or their own data, it comes in other formats. And, basically, build ingestion pipelines where they can do the data transformation that’s required to convert that data into the open standards, if they’re not already in that. If the data is already in an open standard it’s very easy to ingest it through the existing APIs. But if not, we’ve done a lot of work with building out applications to do that harmonization, as well as working with partners that can help customers do that.
We’re really, in many ways, in the early stages of applying machine learning to healthcare, but we’re seeing enormous potential in some of the work that’s been done actually in other parts of Google around research. Particularly at applying machine learning to medical imaging, looking at better diagnostics around diabetic retinopathy, for example. But there have also been great demonstrations of use case looking at predictions based on medical records.
One of the things that we’ve seen some of our customers and users do is taking the data in, and then using it to predict the incidence of disease. We’re seeing a lot of interest in can you use machine learning to predict whether a patient has sepsis, and predict that earlier than you would normally see? We’ve also seen some hospital systems where they’re looking at, can they predict the recurrence of breast cancer, through a combination of their medical records and their medical imaging? And, really, a lot of that is around can they make those predictions earlier than they would have previously, so that way they can intervene more quickly?
We’ve been working on the healthcare API for a couple years because we saw this need in the healthcare industry to be able to break out of the data silos. When you look at different healthcare organizations, one of the common refrains they had for us was that, “We have all this data, we know we can learn from this data, we know we can apply this data to better help our patients. We want to understand our population at large. We want to understand how can we, more quickly, intervene, or do better triage as people come into the ER.” But the thing that they really came to us was that they have lots of data, it’s highly siloed, and a lot of it is stuck in those silos. And particularly, when you’re looking at a large healthcare organization that spans multiple sites. So, you have, potentially, some with two hospitals, some with hundreds of hospitals. How do they bring all that data together, especially when their patients move from one hospital to another?
One of the things that we’re really trying to do right now is help organizations build this data platform, on top of which they can bring together the data all around their different patient populations, so they can have this longitudinal view of the patients in their population. And then, I think part of the future is we expect to see continued improvements to making it easier to do that. But also enabling healthcare organizations, as well as application developers to build on top of that platform. By leveraging open standards, like FHIR, we expect that it’ll be easy for healthcare organizations to build their own applications, as well as third parties to be able to build and easily deploy applications in that environment.
We really expect to see a real growth in the amount of healthcare technology applications that can be built and deployed in these environments. We’re really trying to make it easy for organizations, developers to have a platformThat’s where I really see a rich ecosystem in the future. But I think part of that too, is giving healthcare organizations and researchers, in particular, the ability to take the data, de-identify the data, so that we’ve made a lot of investments in de-identification of healthcare data, so that they can better learn from the data at scale, and use that to build models that can make better predictions that can be applied in a future-looking fashion.