One of the most frustrating things for me about starting a new clinical process improvement (PI) project is just how flipping long it takes to get actionable results. Most of us use the tried and true PDCA (Plan, Do, Check, Act) method of W. Edward Deming fame.

Predictive analytics short-circuits this PDCA process by looking at huge chunks of data (from your EHR, lab, pharmacy, etc.) and by simultaneously examining lots of variables associated with the process being improved, in order to show which of those variables have the strongest association with the outcome you are measuring.

Let me elaborate…by using the PDCA approach to PI we hope to improve quality. The first step is to “plan” what you’d like to change – what are the project’s objectives and how will we know it was successful. This is not a trivial step; in fact it’s probably the most important. Just like trying to cook a meal, if you don’t plan exactly how, when and what you’re preparing, the result is usually not all that tasty.

The second step is to “do” the planned change and collect the data elements that you decided would be helpful in seeing if your change worked. Once the data is collected, the third step is to “check”or “study” the results and see whether the change worked the way you thought it was going to (nice job) or didn’t work (back to the drawing table).

The data is key in the third step, without it, your analysis is likely to be highly subjective and make it impossible to figure out where the process worked or didn’t work and what was the most likely cause for failure or success.  The last step, “act”, takes into account the discoveries of the first three steps and uses them to make a change or continue on current path based on the expected results from the “plan” step. The cycle then repeats until the process is as optimal as possible. As you can imagine, this takes a lot of time.

One of the most time-intensive steps, once the planning is done and the intervention has been put in place, is collecting enough data to see if the intervention worked. And, because the processes we work on in medicine are necessarily complicated, we can’t control for all the variables or see all the key influencers in the results we obtained.

Think of it this way; imagine you’re trying to find a new and hopefully better way to cook eggs.  You start with changing the temperature of the stove and then measure (check) if the egg has been cooked better. If yes then you make this change permanent and now plan to change another variable to see if you make the egg even better. If no, then you vary the temperature until you get the best egg possible.

There are several limitations to this process. First, you don’t know if changing the temperature will really make the egg better. Also, if the egg isn’t better by whatever criteria being used, the only way to figure out if temperature has anything to do with making the egg good, is to vary the temperature over and over until either you find the perfect temperature or determine that temperature really has no effect. By that time, you’ve wasted a lot of time and made a lot of bad eggs. Translate that into clinical processes and besides wasting time, you’ve produced a lot of bad clinical outcomes on real patients…

So  imagine instead that you already collected not only temperature data, but also hundreds of other variables related to cooking eggs. Once you’ve built a model that can accurately predict which eggs are most likely to be good, you can very easily see the most important influencers of making good eggs and bad eggs, almost instantaneously!

Wouldn’t speeding up the PDCA cycle improve care and patient outcomes save money and reduce waste?

That’s the value of predictive analytics in healthcare process improvement! Predictive analytics can quickly and easily find relationships and patterns in the big data common in healthcare and identify opportunities to optimize clinical processes.