A New Way to Look at Healthcare Data Models
We frequently use technical explanations developed by engineers to explain the healthcare data model driving our late binding data warehouse. But the tech talk isn’t always what our audience wants to hear, at least not right up front. That’s why we developed a more relatable grocery list analogy and a video to illustrate how one of these data models might work (or not).
Why Your Healthcare Data Model Matters
In healthcare, having an adaptive data model allows you to remain flexible while still being structured and efficient. Changes resulting from regulation, scientific advancement, patient populations and other sources can be accommodated with minimal development effort with an adaptive model. Just as importantly, data is prepared and delivered to users efficiently.
It’s easier to see how the adaptive model works when you compare it to two other types of data models: dimensional and enterprise. Still, the concept of data modeling is pretty dry and technical. It’s easier to digest when you see how these models would play out with of a more common task. That’s where our grocery list comes in.
Let’s say you’re baking cookies. Your recipe indicates that you need four eggs, two cups of brown sugar and four cups of flour, among other things. So you head to the store with your recipe in hand and get the ingredients you need for that one recipe. It’s very efficient for you (although not so much for the store, as you’ll see in our video). You come home with exactly what you need. No extras, no waste.
But suppose you learn later that day that you also need to bake a cake? You’ll need to the new recipe and go back to the store – measuring cups in hand. Our video shows you how this might play out.
This is how the dimensional data model starts. You have a one-hit-wonder – a single-purpose “point” solution that addresses a relatively specific analytic need. It sounds great at the start: it’s easy and efficient to build. But as the demand for analytics grows – and it always does – your organization makes repeated trips back to the same data and you end up with a mass of redundant data feeds from the same source systems. It’s like the shopper making redundant trips to the store. What originally felt efficient is now wasteful and very inefficient. It just doesn’t scale.
Enterprise Data Model in Healthcare
With an enterprise model, you start with a different type of list, predefined with all of the categories you think you might ever need. As you start making your list you quickly discover you have to compromise because the predefined list doesn’t support everything you need. Say you need cream cheese, tomato soup and butter. Your list doesn’t account for all of these ingredients nor for this level of specificity. Instead, you’ll have to settle for plain old cheese and soup – even if they’re not quite right for the recipe.
Using this approach to analytics can work well in industries like retail and banking, where the information you need to capture – units of a specific tennis shoe, for example – is relatively standardized and stable over time. You can outline the type of information you’ll be tracking up front and build your data model to find and accommodate that information.
But healthcare is a completely different beast. Here the enterprise data model gets messy. Medical knowledge is forever expanding and changing, making it impossible to anticipate what new data or new requirements will look like and how they can fit into a model. If you take the enterprise approach in healthcare analytics, you may be able to call up the right ingredients for certain tasks – sometimes. But when regulations, requirements or even patients change, your model needs more flexibility to change with them.
Adaptive Data Model in Healthcare
To us, the best solution is the adaptive model. It’s customizable but with enough structure to provide the organization you and your system needs. For shopping, the list may start out looking something like this.
Making the list one store at a time keeps us organized, and we can always add new lists for other stores to keep up new requests and ingredients. The blank lines let us jot down the items we need and arrange them in ways that make sense to us. Plus, when a store starts carrying two brands of organic pasta or soup in different size cans, we can pencil in exactly the one we want without overhauling the list or starting over from scratch.
This adaptive approach is very similar to Health Catalyst’s adaptive data model and key to our Late-Binding™ Data Warehouse. It’s structured but customizable. “Stores” represent source systems that are filled with “ingredients” or data that are ultimately brought into the data warehouse. It’s up to you, with the adaptive data warehouse, to decide which ingredients you need and where they’ll come from, and exactly which components are to be used for any specific analytics recipe. The good thing is, however, that you’ll be well stocked.
With the adaptive model, you have access to all of the data you need, and you don’t have restrictions associated with a pre-defined list. You have the flexibility to include new data points or make other changes whenever the need arises.
To us, it’s this adaptive approach that allows you to get exactly what you need without compromise or redundancy.
Tell us: how would you benefit from an adaptive approach?