• Center on Health Equity and Access
  • Clinical
  • Health Care Cost
  • Health Care Delivery
  • Insurance
  • Policy
  • Technology
  • Value-Based Care

How Data Extraction Artificial Intelligence Is Helping to Alleviate Health Care Worker Burnout, Staffing Shortages

Commentary
Video

Bevey Miner, executive vice president of health care strategy and policy, Consensus Cloud Solutions, shares how extraction artificial intelligence is helping train machine learning models to recognize and extract information from documents, providing structured data with confidence scores.

Bevey Miner, executive vice president of health care strategy and policy, Consensus Cloud Solutions, shares how extraction artificial intelligence (AI) is helping train machine learning models to recognize and extract information from documents, providing structured data with confidence scores.

This approach to extracting data is presented as a solution to address the challenges in health care and helps organizations comply with evolving standards, even with limited resources, Miner explains.

Transcript

What role do emerging technologies such as AI large language models play in shaping digital health strategy?

When you look at this movement to consume, send, and receive structured data, mostly using HL7 FHIR [Fast Healthcare Interoperability Resources] standard, which is the latest HL7 messaging, or even X12, a method that's been around forever for sending and receiving and sharing data, you can think about unstructured data being a document.

So, if you look at a fax, it's a PDF document. There's so much that's a document, like a fax, scanned images—there's lots of scanned images all over the place in health care—or even TIFF images where pictures are taken, that's all unstructured. So how do you get it to be structured?

The way I'm referring to structured data is that all of the different important fields are consumable as separate fields. Let's just start with the patient's demographics: first name, last name, address, their insurance number, etc. That is not structured data. Therefore, somebody has to data enter that in order for any of those documents to be attached to a patient's chart.

We have a workforce shortage, we have burnout everywhere, and we are even asking nurses to sit and do a lot of data entry before they can start to treat a patient, and that's becoming a huge challenge in health care today. I've heard quotes from CIOs [chief information officers] that have been in this business for 2 decades who’ve said they've never seen it as bad as it is with not having the right kind of staff to treat patients, and a lot of them are having to do data entry.

So where artificial intelligence plays a role—and I just want to take a step back a little bit and talk about the differences of artificial intelligence—there’s a lot of buzz right now around generative AI. What generative AI does is it generates something that didn't exist before using prompt questions. There's a lot of applicability for generative AI, there's a lot of concern in the industry for using generative AI and Chat GPT.

The kind of intelligence I'm talking about is intelligent document extraction. So, that allows AI to look at a document and contextually see that John Smith is a name and that this ICD-10 [International Classification of Diseases, 10th Revision] code is this diagnosis, and that there's handwriting on a prior authorization that has urgent written on it where it can recognize the handwriting. That is machine learning that has been trained to understand forms and can extrapolate those forms into structured data. It's machine learning, not large language models, that go into generative AI.

The other thing that's important to note is that with extraction AI, you always have a source of truth. You can go back to that original document, solutions like ours, our clarity solution [Consensus Clarity] gives you a confidence score that John Smith, 100% we know that's John, we know that’s Smith, and we know that's a name. So, you can look at every single solitary field and you can see the confidence scores.

As a system, you can say, "I want you to extrapolate data that's at a 90% confidence score. If urgent is 85% or 90%, we still want you to pull out urgent and we'll make a decision whether we want a human to intervene, we want a human to look at that more." But you can always go back to a source of truth.

With generative AI, you don't have a source of truth. These large language models are crunching through a whole bunch of content, and there's concern that there's what's called a hallucination effect with generative AI that all that data may make up something that really is not proven with any kind of a source. So, the kind of AI I'm talking about is extraction AI, and that's how AI can make a difference in terms of how we move towards a structured type of format and provide all types of organizations the ability to meet the new FIHR standards that are coming out, even if they don't have any HR [human resources] or the money to invest in expensive technology.

Related Videos
Joseph Zabinski, PhD, MEM, vice president, head of commercial strategy and AI, OM1
Mila Felder, MD, FACEP, emergency physician and vice president for Well-Being for All Teammates, Advocate Health
Ross Margulies, health care attorney and partner at Foley Hoag LLP.
Pat Van Burkleo
dr robert sidbury
dr mitzi joi williams
dr lawrence eichenfield
dr krystyn van vliet
Raajit Rampal, MD, PhD, screenshot
Related Content
© 2024 MJH Life Sciences
AJMC®
All rights reserved.