Humility and Other Trends in Health Analytics

by Jeff Terry and Andy Day. This essay originally appeared in Analytics Magazine.

Our team at GE Healthcare uses analytics to help caregivers improve outcomes and efficiency. Our work includes retrospective analytics to assess performance and monitor KPIs, simulation analytics to align leaders with a strategy by testing alternative futures, and actionable analytics to support real-time decisions with predictive and prescriptive information. Here are some recent trends we observe in health analytics:

1. Humility. We notice growing humility among healthcare leaders regarding the use of data and artificial intelligence (AI) to support real-time decisions. It’s quite striking. A few years ago, the CIO, CTO or CMIO of many major institutions would hold forth about their big data strategy and growing army of data scientists. Vendors spoke of data lakes, AI platforms and digital envelopes; it was an international broken record of buzzwords and futures. If only we had all the world’s data, then we’d find something to do with it.

Fast forward to present day, and something is different. Much of that old drumbeat continues of course, but a contrarian cadence has risen to meet it. Leaders talk about moving from “PowerPoint to proof points.” Consider recent headlines:

• “How to Cut Through the Artificial Intelligence Hype” – TechEmergence
• “Fake Artificial Intelligence vs. the Real Thing” – Enterprise Tech
• “Artificial Intelligence in Business: Separating the Real from the Hype.” – McKinsey & Company

You see it up close with healthcare leaders too. One of the authors recently had the honor to facilitate a panel with three CIOs from major health systems in New York City, each a great leader from a great institution. What struck us most was their humility. They see the potential of predictive analytics and AI, but they were also upfront about the relatively small steps their teams have made so far. “Very basic” use of machine learning to route ambulances. Tracking refrigerator temperatures. Recognition that even though all sites may be on a single EMR instance, it still takes rework to implement an analytic at two sites in the same system. And so on.

We see this humility all over the world. A new appreciation that while health data is “small” compared to telecom and retail, its complexity dwarfs all other domains combined. In other words, a new appreciation of the chasm from imagining to actually implementing a real-time health analytic. And it’s into that chasm that healthcare analytics leaders must dive.

For us, this is the most positive trend of all. As unrealistic optimism fades, data scientists, data translators, change leaders and clinicians are getting down to the intensely difficult work of solving problems with real-time data in service of caregivers, patients and families. Humility. That’s good news.

2. Scenario planning with simulation analytics. We notice accelerating adoption of digital twins and other simulation analytics to test alternative futures and eventually align leaders with a strategic plan. Using simulation analytics to test alternative surgical block schedules, bed complements, bed algorithms, winter-surge strategies and unit-level staffing plans is not exactly new. Our team has been doing this work for more than 10 years. But acceptance of these methods, and even eagerness to adopt them, has undoubtedly turned a corner. Ten years ago, our initial client presentations had to have an educational flavor. Today, many healthcare leaders understand the concepts and instead ask pointed questions about the scalability and durability of our technology.

Data scientists, change leaders and clinicians are getting down to the difficult work of solving problems with real-time data.

Simulation analytics take many forms. Given the complexity of hospitals and health systems, not to mention the human factor of patient decision-making, it’s hard to argue that discrete-event simulation and agent-based modeling are the most appropriate tools for healthcare strategic planning. Growing awareness of these tools is one factor driving their use.

The other factor is increasing utilization of tertiary and quaternary facilities driven by consolidation, an aging population and the relative infancy of population health efforts. We see this all over the world. Healthcare executives in the United States, Canada, United Kingdom, France, Brazil, Mexico, Germany, Saudi Arabia, Australia and virtually everywhere else that we work face budgets that have grown more slowly than patient demand, not to mention patient expectations.

There are no easy answers, but more and more healthcare leaders are using simulation analytics to test alternative investment strategies and, in the process, align divergent stakeholders to an agreed-upon path forward. The key is that there are many good ideas. The risk is to launch many small projects rather than the few crucial high-impact programs. The challenge is that each project affects the others. Simulation analytics are increasingly valued for their ability to measure the overlap, understand the potential ramifications, and align the team.

3. Analytics career paths. We notice growing appreciation of a simple fact: “data scientist” must be a career and not just a job. To be viable, an analytics capability must offer employees a long-term career path. This seems so obvious, and yet organizations have spent years and large sums to learn it. “We have a team of 20 data scientists,” or “we hired an amazing data scientist with [amazing clinical and data science credentials]” just isn’t enough. Imagining, building, validating, deploying, refining, supporting and expanding real-time decision support analytics is not a year’s work. It’s decades of work.

In recent discussions with CIOs and CMIOs there is a growing appreciation of this reality. In short, an appreciation that the capability they seek requires career paths to attract, retain and grow great employees in stable roles over decades so those employees can invest in their lives and families.

The good news is that the macro-trend toward fewer, larger systems makes such career paths possible in more and more large health systems. Even so, much work remains to craft those career paths, compensation structures, etc. By the same token, we work with leaders of smaller institutions who have spotted this trend and choose to outsource their analytics capability.

4. Problem-back machine learning. We notice that while machine learning (ML) is the most accessible rung on the AI ladder, the technology alone is insufficient. Fully harnessing the power of ML to predict census and bottlenecks requires three competencies: technology, nuanced local knowledge of healthcare delivery, and a data translator to bridge the two. As with humility, there seems to be a growing appreciation that it is the combination of these competencies that is essential to impact care delivery. Together, staff can work “problem-back” to first pinpoint the problem, then the desired response and finally the analytic that will trigger it

5. Early warning explosion. We notice a proliferation of early warning algorithms to predict inpatient deterioration. There’s HEWS, MEWS, CEWS, TEWS, PEWS and 50 more. This trend is related to the democratization of machine learning. It also demonstrates a wonderful aspect of modern patient care: the data exists in the EMR and patient-monitoring equipment to identify with some measure of accuracy those patients who are likely to have problems tomorrow. A remarkable reality.

That’s the good news; it is increasingly possible to anticipate patients who will decline. It’s not yet routine, but no longer science fiction either. The harder part is putting this into practice. Why might that patient deteriorate? Who should see that information? What to do? Answering these questions is the work ahead, and many great caregivers are sorting them out. We’re trying to help through our command center work. The challenge is to incorporate new methods into an acute care setting that is already strained, already delivering great patient care, and (sometimes rightly) resistant to change.

It’s a wonderful time for analytics in healthcare and always an honor to serve caregivers.

Jeff Terry and Andy Day both work at GE Healthcare Partners.