In this 2-part series, I will be exploring some high-level concepts of the data analytics continuum and how they apply to the operating room. In my second post, I will consider how data analytics can be applied to help OR Directors improve their case on time start performance. However, in this first post, we’ll start with a simpler example: teaching a child to ride a bike.
What Do We Mean by Data Analytics?
“Data analytics” is a phrase that all too easily fades away into the buzzword background as it gets loosely thrown around by technology companies in just about every industry.
While the phrase “data analytics” may be carelessly tossed into marketing materials, the field of data analytics, or data analysis, is incredibly important. Our societies are currently being reshaped by the intersection of data analytics and machine learning.
Wikipedia has a succinct and useful definition of data analytics, stating that it is, “a process of inspecting, cleansing, transforming, and modeling data with the goal of discovering useful information, informing conclusions, and supporting decision-making.” (reference: https://en.wikipedia.org/wiki/Data_analysis).
Technology companies in multiple sectors – perhaps most notoriously in advertising – are working furiously to collect and analyze large data sets (often referred to as “big data”) to answer key questions that might give them an edge in developing solutions to their industries’ problems.
This is no less true in healthcare, and, specifically, in the operating room. For example, data analytics is a major component of our very own solution, OR Orchestration (link to our solution).
As more and more companies begin to promote data analytics solutions to healthcare professionals, now is a good time for those professionals to learn more about the field of data analytics and consider how it may be leveraged to benefit clinical care.
Throughout this article, we will examine the data analytics continuum and then consider the application of the continuum.
The Data Analytics Continuum
First things first. When it comes to understanding data analytics solutions, it is useful to categorize the levels or types of data analytics performed by a particular solution.
Gartner (https://www.gartner.com) developed a now popular chart to visualize these different types or levels of data analytics seen in the figure below.
As seen in the chart above, data analytics solutions can often be categorized as performing one or a combination of four different types of data analytics: Descriptive, Diagnostic, Predictive, and Prescriptive. What’s especially useful about this chart and its terminology is that it boils down many complex concepts into four simple categories with intuitive names.
Descriptive analytics generally seeks to answer the question: what happened? With the right real-time data and real-time processing, it is worth noting that descriptive analytics may also answer the question: what is happening?
Diagnostic analytics generally seeks to answer the question: why did it happen? Again, it is worth noting that with the right real-time data and real-time processing, diagnostic analytics may also answer the question: why is it happening?
Predictive analytics generally seeks to answer the question: what will happen? Here is where we begin to especially see how machine learning and artificial intelligence (AI) intersect in revolutionary ways by using previous data to predict the future.
Prescriptive analytics generally seeks to answer the question: how can we make it happen? This is where things start to get complicated and “futuristic.” This is where all the other levels of data analytics can be leveraged to prescribe particular actions to trigger a desired outcome.
At each level, more intelligence is required. A deeper understanding of the data is required, and often better data itself is required (more data, higher quality data, more data sources, etc.). Each level delivers a higher value than the last to a problem: clearly, it is more useful to know how to trigger an outcome than it is to simply describe an outcome that has already come to pass. However, each level also comes with more complexity, so it’s important to pick the right tools for the right job.
Rather than dig into all the complexities and nuances of each of these levels of analytics, let’s consider a very simple example: a child learning to ride a bike using training wheels.
Applying Data Analytics: Meet Mikey
Imagine that proud parents of Mikey have just given him a new bike with training wheels. Mikey is excited to learn how to ride his new bike, and his parents are there to enthusiastically guide him to bike-riding proficiency.
After Mikey gets plenty of practice learning how to pedal and stay upright with his training wheels on the ground, his parents decide that the time has come to raise the training wheels a bit.
After raising his wheels two inches from the ground, Mikey’s parents send him back out on his bike. Then, the trouble starts.
Mikey starts to fall, stop and put his foot down, request support from behind, and more. He can barely go a few feet unassisted. As Mikey loses confidence, he slows down, which makes his problems worse: more falls, more stops, more uncertainty.
After seeing his problems not only persist but worsen, Mikey’s parents decide to make a change. They decide to lower the training wheels to just one inch off the ground to see if this improves matters.
At first, Mikey is apprehensive. But, after experiencing a one-inch tip, which is less scary, he starts to gain confidence and starts going a little faster – fast enough to stay upright. Now, he’s able to go further before tipping. Then further. Then further. Before long, Mikey can bike confidently and proficiently with his training wheels one inch off the ground.
Ok, let’s get back to data analytics…
Mikey and Descriptive Analytics
Descriptive analytics seeks to describe what was happening at each step of the way. Descriptive analytics examine the data and present precise metrics.
In the case of Mikey, descriptive analytics would seek to precisely answer questions such as:
- How fast exactly was Mikey going with the training wheels fully on the ground? What was his minimum speed? Maximum speed? Average speed?
- How far exactly did he travel before stopping?
- Does Mikey’s speed correlate or not correlate to his distance traveled? How so and to what extent?
- Was his average speed and average distance traveled increasing with more practice? To what extent?
- Once Mikey’s training wheels were raised two inches off the ground, what were his metrics then, including how many falls, tips, stops, etc.?
- Did the same trends persist when Mikey’s wheels were raised two inches (e.g., any correlation between speed and distance traveled)?
- How about when his training wheels were lowered to one inch off the ground?
Describing these metrics, and better yet, summarizing them into an easy-to-interpret report that calls out important trends could serve useful to bike-riding teachers & experts, such as parents. It would be useful to know whether Mikey’s progress was entirely halted, partially halted, or reversed when the wheels were two inches off the ground.
Mikey’s parents performed descriptive analytics in their heads: Mikey was making visible progress with his training wheels on the ground, but he clearly wasn’t able to make progress with his training wheels two inches off the ground.
Perhaps Mikey’s parents could not answer all the questions asked in this section, but they chose to take the “good enough” approach with a few obvious insights.
Mikey and Diagnostic Analytics
Diagnostic analytics seeks to answer questions about why metrics and trends may be changing.
At first glance, the cause of Mikey’s impeded progress may seem obvious: the training wheels were raised, so Mikey’s progress was impeded. But, why?
Perhaps Mikey lost all confidence when the wheels were raised, which may call for one solution; but perhaps Mikey was never actually going fast enough with the wheels down to stay upright at all, which would likely call for an altogether different solution; or perhaps it was a combination of both.
Also, is there now a headwind that picked up? Or, did Mikey change directions to go against the wind instead of with it?
Were Mikey’s parents saying different things that changed Mikey’s response? Perhaps their overly enthusiastic cheers of “you can do it” were indicating to Mikey that there was a hurdle to overcome.
Since Mikey has biked to a different part of the neighborhood with his wheels down, is the grade now steeper?
So many questions!
Diagnosing what the cause was of Mikey’s impeded progress is critical to identifying solutions that can optimize his journey towards bike riding proficiency. Here again, Mikey’s parents performed diagnostic analytics in their heads: Mikey’s wheels were raised, so his progress was impeded. They concluded that this was the primary cause of his lack of progress, and knowing this was good enough.
Mikey and Predictive Analytics
Predictive analytics seeks to help us think ahead.
If Mikey’s progress was impeded primarily because tipping two inches to one side or another was alarming enough to trigger Mikey to go more slowly, then perhaps we can use the data we have to predict that Mikey’s progress will restart if he doesn’t have to tip quite so far.
Perhaps we might be able to use data (particularly if we have high-quality descriptive analytics) to predict to what extent Mikey’s progress may resume if we lower his training wheels a given amount. Perhaps lowering 0.2 inches would yield 15 more feet of travel before stopping.
Perhaps positioning Mikey to bike slightly downhill would help restart progress since the additional speed will help him stay upright, or perhaps that would erode his confidence by making the whole experience scarier.
High-quality descriptive analytics combined with deep diagnostic insights of what caused Mikey’s progress to halt will help consider (and model) what changes might cause progress to resume.
Predicting how Mikey’s progress might improve or decline given a set of changes is critical to solving his current progress problem. Mikey’s parents performed predictive analytics in their heads: they concluded that if Mikey’s wheels are lowered, he will regain confidence since he won’t tip as far.
Mikey and Prescriptive Analytics
Finally, prescriptive analytics seeks to combine diagnostic insights with predictive insights to prescribe optimal solutions.
To apply prescriptive analytics, we need to have clear goals – desired success outcomes. Let’s imagine that Mikey has a specific goal in mind: he wants to learn to ride his bike proficiently without training wheels “as quickly as possible.” The goal here is to achieve proficiency with as little practice time as possible. We might better define proficiency as having only two falls per mile. It is worth noting that this is a different goal than being able to “ride fast” without training wheels or being able to “ride far” without training wheels.
Once a goal is identified, prescriptive analytics can help Mikey achieve the desired outcome(s) as optimally as possible.
There are likely optimal solutions, and prescriptive analytics helps uncover those. Perhaps Mikey should practice with his training wheels 0.5 inches off the ground for 20 minutes biking down a surface with a 4.6% grade with a 2 mph tailwind and then practice with his training wheels off 1.34 inches off the ground for 10 minutes biking down a surface with a 5% grade with a 1 mph tailwind and so on.
This is certainly overanalyzing the situation for a child learning to ride a bike, but for some of the world’s biggest problems, prescribing solutions like these can yield many benefits.
As has been true for all levels of analytics in this example, Mikey’s parents performed prescriptive analytics in their heads when they decided to raise the training wheels two inches off the ground and then lower them back to one inch off the ground. Their first prescribed solution didn’t quite work, so they prescribed another. This will likely work out well for Mikey as it has for countless children without requiring his parents to overanalyze the situation
The Right Tool for the Right Job
Speaking of overanalyzing… It’s important to have the right tool for the right job.
Mikey’s parents may not realize it, but they went through an exercise of descriptive, diagnostic, predictive, and prescriptive analytics:
- They watched how Mikey progressed in his journey and noted how his progress changed as they raised his training wheels (descriptive analytics).
- They diagnosed the cause of his impeded progress when they raised his training wheels too high (diagnostic analytics).
- They predicted how Mikey’s progress might improve if they lowered his training wheels halfway back toward the ground (predictive analytics).
- They prescribed a path towards Mikey’s biking proficiency that consisted of gradually raising the training wheels (prescriptive analytics).
In this case, Mikey’s parents are using all levels of analytics, but their selected tool is very simple: they’re using their minds. This seems like the right tool for the right job.
But, here’s the trouble: the human mind is simply not as good at perfectly and accurately analyzing data as well-designed software and can even cause more problems than it solves – particularly when datasets are massive, when data sources are different, and when many variables impact a particular situation. For those interested, a great book that contemplates our difficulty identifying actual trends as humans is ‘The Drunkards Walk; by Leonard Mlodinow.
(https://www.amazon.com/Drunkards-Walk-Randomness-Rules-Lives/dp/0307275175)
While there may be diminishing returns for Mikey, there’s little doubt that intuition alone cannot perform the high level of descriptive, diagnostic, predictive, and prescriptive analytics for complex problems.
Coming Next
In my next post, I will dig into one such complex problem: the operating room. In particular, I will explore the example of case on time starts.
Thanks for reading!