For some time now, Oscar Marroquin, M.D., a practicing cardiologist and epidemiologist, has been helping to lead a team of clinical data experts at the vast, 40-hospital UPMC health system in Pittsburgh. Dr. Marroquin and his colleagues have been busy harnessing the power of creating and nurturing purpose-specific teams focused intensively on the management of data to power performance improvement and clinical transformation.
Recently, Dr. Marroquin shared his thoughts on the subject of artificial intelligence (AI), which has received enormous attention this year in the U.S. healthcare industry, with Healthcare Innovation Editor-in-Chief Mark Hagland. Below are excerpts from that interview.
When you look at the potential of artificial intelligence in healthcare, what are you seeing right now?
I feel that we are certainly at an inflection point as an industry in starting to make meaningful use of artificial intelligence in the healthcare space. I would say there is a lot of work that is still being done that is translational work, that has gone from an analysis of data sets, to beginning to explore whether one can use AI for diagnostic purposes. AI is in fact already being leveraged, or at least beginning to be leveraged, for diagnostic purposes in radiology, in dermatology, in ophthalmology, cardiology, and in pathology. Those are all areas where AI is being explored as an enabler or used as an aid for physicians to make diagnoses from images.
There are also AI models being built in the healthcare management and risk assessment prediction areas. We’re probably a little bit further along in the advancement of the utilization of different data sources than most places, mostly because we are doing it not only in the academic arena, but also to help us operationally; at UPMC, we now have five models that are in production that are AI-based. We’ve derived the models and trained them and are now using them. That spans from the operational—understanding our capacity for the volume of patient visits in our offices, to predicting the patients who are at the highest risk of rehospitalization, to understanding different levels of risk in patients with chronic diseases. We’re calculating the risk of re-hospitalization to all patients admitted to the hospital. We went live in all of hospitals in January of 2019. We had done pilot projects all of 2018, but in January of 2019, we populated it into our apps that unit directors can access, and also in the EMR for clinicians to use within their workflow. And two more models are being rolled out to our clinics, to help us identify patients at high risk of progression of kidney disease, high risk of complications of diabetes, and those at high risk of having complications after surgeries.
When will those be live?
We piloted those three last ones in the last two quarters of 2019 and we will be rolling them out for operational (“live”) use in the first quarter of 2020; our goal is to alert physicians as they take care of the patients of the different levels of risk that we predict in these patients.
What have been the couple of biggest learnings from the inpatient experience so far?
Probably the biggest learning is that—as clinicians, we’ve traditionally thought that a patient’s rehospitalization results from physiological issues seen in the original hospitalization. But the reality is that often, patients are rehospitalized because of social issues or because of de-conditioning. With the model, it has been interesting to see that maybe we haven’t had as good a holistic view of predicting rehospitalization as we might have had when we only relied on physiologic parameters. Second, it becomes clear that once one has the ability to predict levels of risk, you can actually deploy resources more efficiently. So now we’re more efficient at having our social workers and case managers better spend time with patients prior to discharge.
What will happen in the next five years?
Just in the same way that AI has transformed industries like banking, investing, and retail, I think that we’re going to see a similar revolution in healthcare. Again, specifically talking about how we use our own data to deliver care more efficiently. We are in an environment where we actually have to come up with ways to deliver as good or better care at lower cost, because we are all facing those pressures. I’m convinced that the only way to achieve that is by making more intelligent use of our data, and artificial intelligence affords us the opportunity to use the data in a more efficient way, so that we can derive insights and take action sooner rather than later, per the very long cycles we’ve been used to in the past.
What would your advice be for CMIOs, CIOs, and other healthcare IT leaders?
I would hope that most would agree that having a team dedicated to do analytics, and as part of that, focus on advanced uses like machine learning and AI, will be a necessity. In the future, it will become not a ‘nice-to-have,’ but essential to deploy teams to help enable the work that our clinicians are doing. Artificial intelligence is not going to replace what we do, but will be an enabler for us.