Building Better Clinical Models

It’s Time to Make Your New Year Resolutions

At the start of a new year many people make resolutions to change their lives. We all have our faults, and every year we hope to turn over a new leaf and do some things better. If you build clinical models, then here are six resolutions that can help you build better models.

  I will be able to clearly state my goals.

Before you start building models you need to have a clear idea of what your goals are. This sounds obvious but many people starting a project cannot precisely state just what they are trying to achieve. This causes confusion later on when you need to test performance. Try to write down all of your goals then share the list with your team. This exercise can help to increase focus.

  I will get the training that I need.

Many programmers think that they do not need to take training when starting to model. They already know how to program, so they just want to get started. They may know how to program, but they may not know all of the tricks or nuances needed to use BPM+ standards efficiently. Books can be a start, but an actual training course is better.

  I will get control of data and terminology.

If you do not get control of your data and terminology, then your project will fail. This rule is even truer if you are working as a team. A data object often can be represented in different ways (female, Female, F, woman, Woman, etc), which we tend to overlook. But to the computer each is different and will need to be handled separately, resulting in ambiguity and inefficiency. Data is expensive and you should strive for the minimal viable set. In the same way a term may have different meanings to different people. If you do not explicitly state what a term means, then you may find different usages over time.

  I will secure my foundations before building.

Everyone is eager to get started when they begin a project. It is much more fun to start programming rather than to plan the details in advance and to progress methodically. When a team charges ahead there is often little coordination. As time goes on this can result in the need for significant rework, which is another term for wasted effort. BPM+ models are flexible and forgiving early on, but once you are dealing with platform dependent models there is a lot baked in and rework is no longer simple. Better to plan and coordinate as much as you can before you start building.

  I will choose what to automate carefully.

When building out a complex process (like a clinical practice guideline) you will often have dozens of BPMN and DMN models. It is tempting to automate them all. However, experience teaches that some models are worth automating while others are not.

What is worth automating? Something that has a high ROI, something that users want, or something is very constrained. On the other hand, you should not automate something that is costly to do or that has no demand. Actual clinical practice can disrupt the best of plans, and overambitious models will fail. Automating the right models pays off handsomely while automating the wrong models can be costly.

  I will test and document any changes that I make.

One of the selling points for BPM+ models in healthcare is the ability to quickly make changes to models as conditions change. It is easier to do it yourself rather than trying to work through the EHR bureaucracy. However, this can introduce problems if you do not test and document these changes. A competent clinician can still make mistakes (God forbid), especially when there is a complex model with multiple dependencies.

Conclusion:

I am sure that you can think of other resolutions that you could make. All too often resolutions are quickly left by the wayside. Keep these resolutions and your models will thank you.