Technology
Generative AI llm

We have written other articles about generative AI like OpenAI’s ChatGPT and whether it has value in revenue cycle management (ChatGPT In RCM) right now, which you can read here. But what happens when you ask AI about AI itself with respect to RCM?

Here is what happened when I prompted a Generative AI application like (in this case OpenAI ChatGPT) to write an article for our blog about generative AI in RCM ( ChatGPT in RCM). Specifically, I prompted it with the following input:

“write a blog post about whether you need generative AI or how it could help in healthcare revenue cycle management”

It created this article:

Do You Need Generative AI in Healthcare Revenue Cycle Management?

Generative AI has been making waves across industries, and healthcare is no exception. From enhancing patient care to streamlining administrative tasks, the potential of AI in healthcare seems limitless. However, when it comes to the specific niche of healthcare revenue cycle management (RCM), the role of generative AI warrants a closer look. Is it a necessity, or a shiny new toy with overestimated benefits?

Understanding Generative AI in RCM

Generative AI refers to systems capable of generating new content, be it text, images, or even predictive analytics, based on the input data they are trained on. In the context of RCM, generative AI could theoretically handle tasks like automating billing processes, generating appeal letters, predicting payment timelines, and analyzing coding errors.

Potential Benefits of Generative AI in RCM

  1. Improved Efficiency: Generative AI can streamline repetitive tasks like generating billing statements, which could potentially reduce the time and errors associated with manual data entry.
  2. Enhanced Accuracy: AI algorithms can analyze large volumes of data to identify coding errors or discrepancies in billing, which might otherwise be missed by human eyes. This leads to more accurate claims and fewer denials.
  3. Predictive Analytics: By analyzing historical data, generative AI can predict payment cycles, flag potential delays, and suggest optimal billing strategies to maximize revenue collection.
  4. Personalized Communication: AI can be used to generate personalized communication for patients regarding their bills, payment options, and financial assistance programs, thus improving patient engagement and satisfaction.

Challenges and Considerations

  1. Data Quality and Availability: The effectiveness of generative AI is heavily reliant on the quality and comprehensiveness of the data it is fed. In healthcare, data can be fragmented and inconsistent, posing a significant challenge.
  2. Integration with Existing Systems: Integrating generative AI solutions into existing healthcare IT systems can be complex and costly. Ensuring compatibility and smooth data flow is crucial.
  3. Regulatory Compliance: Healthcare is a highly regulated industry. Any AI application must comply with data privacy regulations like HIPAA, adding another layer of complexity to implementation.
  4. Cost vs. Benefit: While generative AI can potentially bring about significant efficiency gains, the cost of implementation and maintenance must be weighed against the expected benefits.

Is Generative AI a Necessity in RCM?

The necessity of generative AI in RCM largely depends on the specific needs and capabilities of the healthcare organization. For larger organizations with complex billing processes and a high volume of claims, generative AI can provide significant value by improving efficiency and accuracy. However, for smaller practices or those still struggling with basic technology infrastructure, the immediate benefits might not justify the costs and effort involved.

Conclusion

While generative AI holds promise for revolutionizing healthcare RCM, its implementation should be strategic and aligned with the organization’s readiness and needs. It is essential to focus on building a solid foundation with existing technologies and processes before diving into generative AI. The key lies in understanding where AI can add the most value and ensuring that your organization is well-prepared to harness its potential.

<end GAI article>

ChatGPT In RCM

The Reality of Generative AI in RCM

Great article right? Well written, coherent. The only problem is that Generative AI doesn’t even understand what Generative AI does or even what it is sadly. While this starts to get a little meta, it is actually pretty funny that GAI is so clueless. And there are some learning lessons here I think about the deployment of technology in healthcare and RCM (aka medical billing).

Common Misconceptions about Generative AI in RCM

There are a couple of problems with this. First, it is flat out wrong and doesn’t even make sense in a bunch of its suggestions. Benefit #1 of eliminating repetitive tasks like generating statements is really just automation, which is not technically considered AI by experts and is certainly not generative AI. If it means by generating the content like the words in the patient statement, it doesn’t understand that this is not done manually currently, it is automatically generated by the practice management system PMS already.

Misinterpretation of Predictive Analytics

Suggestion #3 of predictive analytics is again really wrong. This is where it goes completely off the rails. Generative AI, particularly for text, operates by assigning numerical values, known as embeddings, to words based on their relationships within a vast dataset. These embeddings represent how closely words are related in meaning. For example, “leash” and “dog” would have embeddings that place them near each other because they are commonly associated. “House” might also be nearby, as dogs often live in houses. However, “bog,” despite being similar in spelling, would be placed far from “dog” in this numerical space due to its lack of semantic connection. 

This method is powerful for tasks involving language generation or understanding, but it isn’t suitable for tasks like predicting when a claim will be paid. Predictive tasks involving structured data—such as dates, amounts, CPT codes, and payer IDs—are better handled by other AI techniques like machine learning (ML). ML is designed to analyze patterns in this kind of discrete, structured data, making it much more appropriate for predictive analytics in scenarios like claims processing.

AI’s Acknowledgment of Its Own Errors

When I pointed out to the GAI application that I thought it was really wrong about it’s suggestions and didn’t seem to understand AI (I have to say it again – AI doesn’t understand itself yet), it acknowledged I was right, and it screwed up:

“You’re right in highlighting that revenue cycle management (RCM) data is primarily discrete, structured, and numerical. This clarifies that while predictive analytics is a component of broader AI capabilities, it’s generally associated with analytical AI, not generative AI, which is more about creating new content or outcomes based on patterns and learned inputs.”

The High Cost of Generative AI

In the “Challenges” it mentions cost vs. benefit. But even a company like Oracle has said the cost for the hardware alone to do Generative AI would be over $100 million to get started. So the cost is astronomical at this point and basically means it is a non-starter. Could you link together something using a third party app like OpenAI, Gemini, etc. that generates content after viewing a file or some unstructured data? Sure, but that sounds like a giant HIPAA violation at this point. I’m sure it could be overcome, but I’m still not sure I see what benefit it would provide.

Potential Use Case: Appeal Letters

A fun potential use case came up recently, which was generating appeal letters in medical billing using Generative AI. I think this could have long term potential, especially if you combine it with a much larger complex process including automation and some other forms of AI like machine learning / ML. But the secret sauce in that technology stack would be the automation and machine learning, not the GAI as far as I can tell.

If you gave two different billers the same inputs of payer rules, procedure and diagnostic codes, patient and payer data, and other info, combining it with the right logic to persuade the payer, it would be hard to believe it makes a giant different on what words are surrounding the content in the appeal letter. But who knows, that might be interesting to look at if you had a ton of time and money. In the meantime, you could just implement the automation and ML and get a massive bang for your buck and wait for GAI to advance.

I’m an engineer and a technology evangelist. I believe in AI and many other technologies. And market hype often gets out of hand. GAI is not aware enough yet to realize it says really worthless things sometimes. So please don’t just do whatever it says even if it sounds really knowledgeable. Do spend time on advancing your technology, but don’t look too far at the horizon or you’ll miss the steps in front of you that are effective and generate huge ROI. Talk to us about what you should be doing now.

Author

voyant

Leave a comment

Your email address will not be published. Required fields are marked *