Of black swans, dragon kings, strange attractors and seahorses

Two rather random events have made a strong impact on me during the twenty years or so I have spent working on clinical information systems.

The first event occurred nearly 20 years ago, while rolling out one of the first ever whole hospital electronic medical records systems. It was a greenfield, new-build hospital site (a 450 bed tertiary care referral center, at the time named Health Care International - HCI), in Glasgow Scotland and we were in the process of implementing Cerner's complete suite for full HIS, Results Reporting, CPOE, MD & RN Documentation...with no plans for any paper storage at all. That paperless EMR implementation was a roller coaster ride and perhaps the topic for a future post.

Here's what happened. I was walking through the Intensive Care Unit six months or so after the hospital had opened its doors and we probably had about eight patients on the unit at the time. An anesthesiologist was staring intently at the results reporting screen for a specific patient of interest, and in particular at a feature that Cerner had made available in the then version of Powerchart, called the "Navigator View" (similar to thumbnail views in Photoshop or other image manipulation apps that allow the user to quickly pan around in a zoomed view). This view didn't actually show discrete result values itself but did show a summary view of all discrete results as well as a colored indication if a result was abnormally high (red) or low (blue). I remember wondering why he didn't just click on the small window to navigate to the result of interest and I asked him that directly.  His answer was both curious and delightful. It stopped me in my tracks and still does whenever I think about it.

What he said to me was this... "You see that pattern of blue and red....well, it kind of looks like a seahorse doesn't it? I've seen that kind of seahorse pattern before and I'm pretty sure that it means this patient is beginning to go into renal failure." OK, so, get this...a doctor who had not looked at any actual test result was reasonably confident in diagnosing incipient renal failure.  He used a tool that was never intended to display "seahorses" and did what humans do best, he pattern matched and suddenly here was an emergent property of a system that had never been anticipated and certainly not designed for.

He had been able to predict what would happen next with this patient, based on pattern recognition. I was blown away and still think about what insights we might be able to gain if we intentionally designed systems to create visual patterns that could be used to make synoptic sense of an ever increasing deluge of patient data.  Essentially tools designed specifically to reduce signal to noise ratio and find the needle in the proverbial haystack of patient data.

The second event occurred two or three years ago when I happened to be on a conference call with Dr. Timothy Buchman of Ochsner Health System, in which I mentioned in passing that I was interested in the potential for predictive modeling in healthcare. After a somewhat slow start to the call, Dr Buchman suddenly became rather animated. It turned out that what I was thinking about and what he was thinking about were two radically different ends of the predictive modeling spectrum.

My interest was focused on exploring and understanding ways that large pools of individualized patient data (claims, EMR, social media etc.) could be leveraged to better predict likelihood of future events such as hospital admissions and ED visits, as well as to anticipate patient safety issues, such as the likelihood of post-operative infection, falls and other potentially avoidable complications, before they occurred.

Dr. Buchman was thinking about something rather different entirely; in fact at the other somewhat murky end of the predictive analytics spectrum that deals with Schrödinger's equations, entropy pumps and low energy "stable attractor" states.  He has published extensively on the topic of predicting the likely evolution of physiological states in critical care environments and the figure below depicts some of the ways that he proposes that physiological states appear to have a propensity to "jump" from one stable attractor state to another.

An external file that holds a picture, illustration, etc.Object name is cc8868-1.jpg Object name is cc8868-1.jpg

As Dr. Buchman might say, not all risk is linear, and we need to understand those wild card situations that may at first appear totally unpredictable. Some have attached wonderfully descriptive labels to these types of catastrophic and tough to predict events - such as black swans and dragon kings.

Whether we are dealing with simple forecasting, complex predictions, seahorses, black swans or dragon kings, one thing does feel certain - healthcare providers are starting to move beyond retrospective analysis and are doing more with predictive analysis.

I'm interested in hearing what healthcare providers around the US (and globally) are doing to look into the future; either a few minutes or hours for the patient in the ICU or months and years for the broader population that are increasingly likely to be managed under some form of accountable care delivery model.

tags: Black Swan, Dragon King, pattern recognition, Predictive Model

Post a Comment

Your email is never published nor shared. Required fields are marked *


You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <p> <pre lang="" line="" escaped=""> <q cite=""> <strike> <strong>

  • About this blog

    Welcome to the SAS Health and Life Sciences blog. We explore how the health care ecosystem – providers, payers, pharmaceutical firms, regulators and consumers – can collaboratively use information and analytics to transform health quality, cost and outcomes.
  • Subscribe to this blog

    Enter your email address:

    Other subscription options

  • Archives