Q3 2017 Newsletter - Leading the Change from Descriptive to Predictive Analytics

By Rahul Gupta, PhD

Disclaimer —the views and opinions expressed in this article are solely my own and do not reflect those of my current and past employers.

If I were asked to summarize in one word what separates the hype from reality when talking about predictive analytics in healthcare, I would say “adoption”. While the reasons are numerous and still being passionately debated, I think what is more important is that the problem is increasingly being recognized and there is interest from all stakeholders to come up with practical solutions.

Before I go on, let me briefly explain what I mean by “predictive analytics”. If you have been living and breathing during the last few years, you have probably heard one or all of the following buzz words — data science, predictive analytics, machine learning and artificial intelligence. More than likely you have heard them in multiple different contexts, and, unfortunately, with multiple different meanings. Rather than tease these terms apart, for our purposes here I will define predictive analytics as anything that allows a business to be proactive rather than reactive. This could include advanced statistical analytics, deep neural networks and everything in between.

It is in this spirit that I started as the first data scientist at a $6B vertically integrated healthcare delivery system, to help lead the transformation from descriptive analytics to predictive. It is important to realize that the change was not just about technology, but also about culture and the way we think about care delivery. I had no map when I started, but a general outline of a vision and a broad mandate and leadership support to build on it. In retrospect, I would divide our growth journey into three (overlapping) phases.

— Start with the customer

At the center of it all was the customer. We started our program by meeting with the leaders of different businesses within the health system to understand their challenges. Our objective was to together discover where and how predictive analytics could help them solve their challenges. On the one hand, we explained how predictive analytics was different from the reports and dashboards they were so used to (and found very valuable). On the other hand, we wanted to control the hype around predictive analytics and be very honest and transparent about what it could and could not do for them. At the end of the day, true value is realized only when the frontline staff who care for the patients day in and day out are able to trust and use (i.e. adopt) the analytic solutions. Our job as informaticists and technology providers is then to create solutions that they can understand and find useful. We realized very early that for this to happen, the customers should be treated as partners in the journey, not a consumer of the finished product.

— Build trust and earn credibility

Next came the all-important task of delivering on what was promised. We focused on projects where the emphasis was to demonstrate how predictive analytics can add tangible business value rather than get caught up in the ever-present challenges around data integration and quality. Our projects tended to be small components of larger strategic initiatives where the business leader was on board with some experimentation and piloting.

— Run with it

We had some great successes in our initial pilots and as the program matured we started paying attention to seamlessly embedding predictive analytics in day-to-day operational decision-making, while still supporting the larger strategic projects. What this meant was that the power of predictive analytics was made available to everyone in the health system, not only the ones who asked for it. As an example of the efforts here, we started to create a standardized process to rigorously evaluate any vendor-provided predictive analytics solution in terms of its impact on patient outcome, patient experience and revenue. The expected result was saving time and dollars in implementing a new process change only to find out later that it did not work.

— The promise of predictive analytics in healthcare

As the hype and promise of predictive analytics has grown, more and more organizations in healthcare (large and small) are in various stages of leveraging such methods to build and / or sustain value in an increasingly challenging and dynamic healthcare environment. No matter where the policy changes may lead us, there seems to be general consensus around one thing — the shift from volume to value in healthcare delivery is here to stay. And, I would argue that predictive analytics, if done right, is going to become an increasingly important tool in enabling this transformation.

The field of predictive analytics is rapidly evolving as well where new methodologies like deep learning are leading us toward “true AI”, to wit autonomous systems capable of making complex decisions without the need of human input. However, along with such advances has come the challenge of enormous complexity of such models to the point that it is often impossible to understand why and how a model arrived at a particular decision (sometimes referred to as a “black box”). This is particularly problematic in healthcare where predictions without accompanying clinical / expert reasoning pose a major challenge for adoption. Such shortcomings are an active topic of research as I write, with interesting advances having already been made.

Finally, the question of whether predictive techniques will ultimately cut down the need of the “middleman” (the physicians, nurses and other clinical staff) is an interesting one. There is some level of automation already available in areas such as interpreting images (e.g. radiological, ECG, eye scans, etc.) and even in the surgical suite (via the use of robots). While it seems plausible that automation may soon also start providing clinical diagnoses and treatment for “mundane” conditions, I believe the question here is as much cultural and philosophical as technological. We are long ways off from being able to trust our lives entirely to machines and until that happens, the goal of technology should be to ease the burden on clinicians and prevent “avoidable mistakes”.

That certainly was the goal of our program. While we no doubt had adoption challenges just like any new technology in healthcare, at the end of two years we had built enough credibility to serve as a great launch pad for our third phase of growth. To paraphrase an old saying, adoption is a journey not a destination — bring your customers along with you as partners on this journey, the ones who stay the longest with you will be your best adopters.

Dr. Rahul Gupta is a Data Science and Analytics leader, most recently with St. Joseph Health in Orange County, California. He was responsible for spearheading data science initiatives in the areas of population health management, clinical quality, hospital operations, patient experience, health equity, palliative care and clinical research. He was the first Data Scientist hired by St. Joe’s to help start a new Data Science program and enable the journey towards predictive analytics. In his prior work life, Dr. Gupta was a research scientist at Medtronic, the nation’s largest medical device company, where he led research and analytics in support of novel brain stimulation technologies.

Return to Q3 2017 Newsletter