Deriving the Prescriptive Analytics Function

While Gartner’s 2017 Healthcare Provider Hype Cycle still places most prescriptive analytics solutions on the rising-to-peaking portion of the curve, not all “prescriptive” analytics are created equal.  In fact, I’d argue that providers who aren’t engaged in some way in this highest quadrant of the analytics maturity model are leaving significant improvements in quality, efficiency, and velocity of care on the table today.

In advance of my upcoming talk at HIMSS, this post frames a view of prescriptive analytics that’s more function than solution.  It’s a concept that hit me while on a run this weekend past the outlook shown in the image above – specifically:

Prescriptive Capability of an Analytics Solution

Technology + (People * Continuous Improvement Process)

With a nod to the operating model’s “Golden Triangle,” at the heart of realizing the power of any technical analytics solution is a process that requires purposeful intent and deep integration across people in governance, strategy, support, and operations.  HOW to achieve this integration will be the main focus of my platform sessions at both HIMSS and Becker’s Health IT + Clinical Leadership later this spring – here, I’ll argue that varying technical implementations across the analytics maturity model below can still get to prescriptive with the right people and processes behind them.

Image result for analytics maturity model

The promise of prescriptive analytics is a world where our models not only answer the questions “What happened?” and “Why?” but also tell us what will happen next AND what actions to take.  In its most autonomous form, I think of prescriptive analytics like the algorithm-driven trading dominating Wall Street.  Countless investment house algorithms train on mountains of data, look for patterns, test predictions, and find the actions that led to the best historical outcomes.  Once let loose upon the market, these same models then monitor trading in real-time, executing actions in vivo based on the patterns they’ve learned in training while simultaneously digesting each new day’s worth of data to refine and learn new pattern / action combinations.

While most of the clinical delivery of healthcare can’t presently (and arguably will never be) *fully* autonomous of human action, we CAN make much better use of prescriptive algorithms today that optimize real people’s time and energy.  At this most technically “prescriptive” end of the Technology + (People * Continuous Improvement Process) function lives Mission’s first internally-developed machine-learning algorithm – Readmissions Predictor.  Powered by a model developed by Dr. Andrew Johnson and his Data Science team – one whose development we spoke to in detail at HAS17 and will highlight again at #HIMSS18 – the LASSO Readmissions Predictor model was trained on millions of rows of data including hundreds of variables from Mission’s own enterprise data warehouse.  Importantly, the model not only serves a prediction on likelihood of 30-day readmission for every discharged patient on the day immediately post-discharge, but also provides the user with the variable that most strongly drives the model’s prediction for each patient.

Though the model handily beats LACE and maximizes the “Technology” portion of the prescriptive analytics function, without the right workflow to drive the right people’s actions it still would sit like a Ferrari in the garage gathering dust.  Instead, because the model was co-developing with clinical leaders from the outset, version 1.0 is now live in pilot form with transitional care managers actively tweaking the model and its user interface with the technical team each day.  Serving care managers with both better and more actionable data lets them spend more time doing what they’ve trained to do:  using their experience and brainpower to figure out HOW best to support the most concerning patients rather than spending time ferreting out WHICH patients most need their help.

On the less technically-advanced part of the maturity model curve are the merely “predictive” and “descriptive” analytics solutions.  While a report – even when multi-sourced and served in real-time – can never be truly prescriptive on its own, I believe these analytics solutions can get to prescriptive when integrated into a well-oiled team’s continuous improvement platform.

Take two of the other examples we’ll cover at HIMSS – the Care Process Model (CPM) Explorer and Ambulatory CPM Explorer dashboards.  These dashboards drive the work of over 60 CPM teams across Mission Health focused on improving specific clinical issues (like preventative screening) or disease states (like sepsis, cellulitis, and heart failure).  By providing a CPM team – composed of physicians, ACPs, nurses, pharmacists, performance improvement guides, informatics and analytics professionals, clinic or hospital leaders and support staff, and many others – with real-time data at the individual provider and individual patient levels, these dashboards plug directly into a continuous improvement flywheel and serve the team with the information they need to drive subsequent turns.

Want to know whether the most recent revision to the “rib fracture” electronic workflow is working?  The inpatient CPM team can pull up the dashboard and compare outcomes like mortality, readmission, length of stay, and cost per case in the pre/post intervention periods and find out what’s working, why, and where to focus its next PDSA cycle.

Need to have a discussion with a specific trauma surgeon?  The CPM team’s surgeon lead can pull up the dashboard and compare her CPM usage statistics and those of her partner along with their respective patients’ outcomes side-by-side.

The same is true for Ambulatory CPM Explorer, where teams are driving improvements in areas like the management of COPD exacerbation, preventing both ED visits and inpatient hospitalizations while increasing GOLD guideline adherence.

Again, its the “People * Continuous Improvement Process” portion of the CPM equation that drives a predictive function much greater than the Technology component alone could justify.  But without the technology these teams would have a much harder time knowing what happened, why it happened, and where to focus next.

Finally, the CPM process is only one of several continuous improvement examples that get into prescriptive analytics by leveraging more “basic” analytics technology while driving “People * Continuous Improvement Process.”  Teams today are using similarly creative dashboards in real-time or near real-time to drive significant improvements in emergency department throughput and inpatient length of stay, optimize surgical scheduling, and manage through incredible periods of patient surge while achieving better outcomes for patients and the community.  These teams are committed to a common goal – getting better, and better, and better, and better… – and their work is driven by predictive/descriptive analytics that become prescriptive when coupled to purposeful people and powerful processes.

Thus, while HIMSS18 will be filled with the incredible promise of technology – and it is quite amazing – I invite you to pause and remember the power inherent in the people who’ll be using it.  People and our processes make the world go round… equip the right people, using the right process, with powerful technology, and there’s nothing beyond our reach.

About the Author

Dr. Chris DeRienzo is a dedicated husband, a proud father, and a mediocre triathlete. He’s also a doctor dedicated to improving the quality, safety, experience and sustainability of healthcare for all Americans. You can read more about him at or follow him on Twitter

Leave a Reply

%d bloggers like this: