As providers look for more ways to add efficiencies, produce better patient outcomes and reduce costs, many are turning to partnerships with large technology companies.
On July 20, University of Pittsburgh Medical Center announced that the health system would enter a five-year partnership with Microsoft to better utilize data the provider collects throughout its 40 hospitals.
UPMC’s clinical teams will have access to Microsoft’s cloud computing, artificial intelligence and machine-learning tools to improve patient care. The two companies will work together to mine more than 13 petabytes of clinical data and 18 petabytes of imaging data with the goal of creating actionable insights for care teams.
Previously, UPMC has leveraged data to identify higher-risk patients pulling information from more than 1 million surgical procedures. Now, upon scheduling a surgery, the model runs in real-time and allows providers to better care for the patient’s post-surgery.
Dr. Oscar Marroquin, the head of UPCMC’s healthcare data and analytics team and Christian Carmody, UPMC’s chief technology officer, spoke with Digital Health Business & Technology about the processes and institutional framework needed to create meaningful insights for care teams. The interview has been edited for length and clarity.
Download Modern Healthcare’s app to stay informed when industry news breaks.
What were you looking for in a data partner? How is the data being used?
Marroquin: I think it's no secret that healthcare as an industry has been a little bit late to the game in terms of using analytics and more advanced aspects of how we consume data with artificial intelligence, machine learning and compared to other industries. But the reality is with all the digitization of our medical records, with the advancements in computational power and technologies, we have started to take advantage of all the data that we have accumulated over the years. It produces insights that we can use to deliver better care for our patients.
Why use Microsoft and not another provider?
Carmody: The platform that Microsoft offers is text-based analytics where you basically run through those clinical documents and extract out meaningful data. That was one of the key sources of data as we went along. I think at one point we were up to 28 million documents that were being processed through in a very real time perspective to enable more information to come in. We’ve leveraged Microsoft technologies for a time and put them through the ringer. We've tested and done a tremendous number of proof of concepts with their technology platforms in this case with Azure.
What does implementing such a process look like?
Carmody: At UPMC, we’ve been digitizing data for 25 years, but there was always that great opportunity to get more out of it, to derive meaning and value from the data, especially when you look at it from an aggregate perspective. About 10 years ago, UPMC made an investment to create an analytics platform. The data that we're creating continues to grow. This whole arrangement is all about re-platforming this analytics solution [and allowing] us to scale and to grow.
Do you have an example of a process that is enhanced with the new platform?
Carmody: One of the challenges that everyone in healthcare experiences is that we deal with a lot of legacy applications and systems. Whether it's local electronic health records or different ancillary systems, but there are some meaningful datasets there that organizations want to ingest. Our older, legacy platform used to take 240 minutes to extract a daily load, pulling it out of the existing system and transforming it. We've got that down to about 14 minutes right now. These are large data sets, so we can push those insights out at the point of care close to the real time.
What are some of the biggest challenges in implementation?
Carmody: The inconsistency and the data and the data types. If everyone could have agreed to a standard 30 years ago when the first electronic health record came out, we'd be in a different place right now. That’s what differentiates us from other industries from an analytics perspective, much of what they leverage is based upon structured data. Approximately 80% of our data is unstructured data. It's in clinical notes, it's in paragraphs and sentences that aren't always alike. So, a lot of our challenges have been around that data acquisition just to bring it in to one place where it can be consumed and used.
How far is UPMC away from implementing these insights into clinical care? Are there things clinicians can do to aid the process?
Marroquin: Ten years ago, that was our aspirational goal, I can say that now we are at a place where that's what we are doing. There is not a simple plug and play solution. In my opinion, the only way we've been able to get to where we are is because we have made the data and the analytics and the clinical enterprise work together in the process. UPMC made a deliberate decision to build the team on those processes. How we [as clinicians] went about doing our job, whether it is writing notes or whether it is choosing diagnosis codes. You can't get to have trust from the clinicians unless they have been part of that whole process.
