/wp-content/uploads/Blog-John-Deterministic-Modeling-Thumbs.JPG.jpg Our Experts' Blog /blog/ Determinedly Deterministic: The Benefits of Deterministic Models in Healthcare Using AI in healthcare puts clinicians at malpractice risk. Learn how determinsitic modeling to provide reliable evidence-based care and ensure patient safety.
Dr. John Svirbely's blog post - Determinedly Deterministic: The Benefits of Deterministic Models in Healthcare
Dr. John Svirbely, MD
Blog

Determinedly Deterministic: The Benefits of Deterministic Models in Healthcare

By Dr. John Svirbely, MD

Read Time: 2 Minutes

What is Deterministic Modeling?

BPM+ offers a family of visual modelling languages to capture how things get done. A great benefit of BPM+ models is that they are deterministic. A deterministic model is one for which a given set of data inputs always gives exactly the same output.

Deterministic models are attractive for several reasons:

  • The models can be tested, and their performance analyzed and compared.
  • The models can be reviewed by experts.
  • The models are appropriate when regulatory compliance is an issue.
  • The models can be efficient to use.

"Winging It" with AI

Most of today’s AI is based on Large Language Models (LLMs). LLMs are non-deterministic by design, which may feel to some people like “winging it”.

How does a deterministic approach compare with "winging it" when using a large language model? When winging it, you may be using the LLM on a topic that you are unfamiliar with and unable to judge for accuracy. You may be pressed for time, with little time to document. Maybe you haven't tested the output or only done so cursorily. Large language models can generate huge amounts of material that can take a long time to review. Are you sure that the references given offer the best evidence, or that they were even used to generate the output? Every time you run the script or make a minor change maybe the results are different. Busy people may choose to take it all on faith and get on with the next problem.

When a critical system such as healthcare is involved, can you afford to rely on something coming out of black box? If something goes wrong and your actions are under the microscope, then maybe you will be viewed as negligent.

And what about costs? Calling a LLM several times a day can become expensive in tokens, while verifying the result can consume large amounts of time. What works for a single episode may become prohibitive when used more often.

Does this mean that you give up AI? No - AI is simply too valuable. But relying on luck may not be the best strategy.

AI and Creation of Deterministic Models

Creating deterministic BPM+ models does require some effort up front. In the past you might have to develop new skills or hire people to create them. Now you can use LLMs to help with the heavy lifting.

Creating deterministic models entails several tasks:

1

Gathering the evidence

2

Summarizing the evidence

3

Designing the appropriate models

4

Generating the applicable DMN and BPMN models

5

Preparing a test plan and data

6

Preparing essential documentation

7

Integrating it with data sources

Each of these tasks can be partially performed by LLMs. An expert can make sure that what is generated is relevant and applicable. Can you do everything by AI? Maybe in the future but not now. The best approach is to use it to make your knowledge workers more efficient.

Once you create the BPM+ model then it can be executed cheaply. And it can be infinitely reused, distributing costs over a large number of calls.

Medical knowledge is becoming a commodity (although experience will continue to have value). Medical knowledge may have an expiration date after which it is not as reliable. BPM+ models implementing medical knowledge need to be curated and updated or retired as needed based on the best available evidence. AI can help with these tasks which otherwise might be neglected.

Future Vision

No one knows how many decisions or processes are being used in healthcare around the world. My opinion is that there are at least half a million and probably more. Trying to create models for so much content is an enormous task that might not be possible if done by historical methods.

With AI the situation has changed. With enough resources it is now possible to envision a semi-automated factory capable of generating BPM+ models. Claude currently does a very credible job creating DMN models. Using AI to implement this content could create a world class resource capable of providing evidence-based care and tackling cost and quality issues plaguing healthcare.

Take Home Message

Can you just "wing it" when using AI? Sure, but you may not be providing the best care. If you want to excel, then you can use deterministic models to provide reliable evidence-based care and ensure patient safety.

Blog Articles

John Svirbely

View all

    All Blog Articles

    Read our experts' blog

    View all

      Learn how it works

      Request Demo

      Confirm your budget

      Request Pricing

      Discuss your project

      Request Meeting
      Graph