Without accurate and reliable revenue cycle management, it will be very difficult to run a successful hospital or healthcare system. Accurate and reliable medical coding is also a big part of this.
A growing number of healthcare organizations see the potential of artificial intelligence to boost revenue and reduce administrative burdens that extend all the way down to providers. But medical coders worry about AI’s accuracy and the potential for future job losses.
The goal, however, should not be autonomy, but rather augmentation – rather than removing the human from the equation, the human can be empowered with new generative AI-powered assistive software, said Varun Ganapathi, co-founder and chief technology officer at Akasa, a provider of generative AI-based technology for revenue cycle management. Ganapathi holds a PhD in computer science, specializing in artificial intelligence, from Stanford University.
Healthcare IT News spoke with Ganapathi to get his insights on how genAI can be used in revenue cycle areas like medical coding, how the technology can be trained to share justifications for its recommendations, what organizations should consider when exploring generative AI systems and their underlying technology, best practices for genAI implementation specific to coding and revenue cycle processes, and how genAI-driven tools can help solve obstacles in the employee revenue cycle.
Q. How can genAI be used in the revenue cycle, such as medical coding? How would you alert users?
A. GenAI is the future of the revenue cycle. This complex industry relies on highly manual and time-consuming tasks that involve navigating complicated EHRs and payer portals, dealing with endless documentation, and navigating ever-changing regulations and policies—all while prioritizing the patient experience.
GenAI is powered by large language models that deeply understand patient records and clinical data that was previously opaque to computers. With LLMs, all of that is now accessible. Training an LLM to understand the healthcare domain and its data opens up a lot of possibilities, including an easier path to really solving some of the biggest pain points in the revenue cycle.
Historically, coding has been too complex for most legacy technology to adequately solve. This is not the case with genAI, which can be trained specifically on healthcare system data. This allows genAI to surface relevant information from patient data and workflows, as opposed to a massive online database.
From there, genAI can collaborate with existing coding specialists and generate suggested quotes and codes. Because genAI runs on LLMs, it also continues to learn. So if coding rules are updated in a particular state or new patient data is added, genAI can quickly adapt.
Users should be cautious about autonomous coding. While the idea is exciting, autonomous coding carries a number of risks in execution, including inaccurate coding suggestions. That’s why healthcare systems should always use a trusted genAI model that is tailored to their data and involves humans in the process.
Q. You suggest that showing work is key to using genAI in the revenue cycle. Why? And how can the technology be trained to share justifications for its recommendations and insights?
A. Showing your homework is crucial in healthcare, especially when working with genAI. Hallucinations, or genAI output that is not accurately based on actual data, are a real concern. Imagine a tool that suggests false codes for a patient. What started as a routine visit could turn into huge medical bills and a misdiagnosis.
AI has historically been a big part of the technology we use, a black box where we can’t see the technology workings or understand where the results come from. With the right genAI tool, coders can see what coding suggestions are being made and why. Where in the patient records is the information coming from?
By showing its work, genAI allows teams to vet that suggestion before it reaches a payer for accurate reimbursement. It trains on specific data, so it learns how to capture the most successful codes for each individual organization and casemix index.
Q. What should organizations consider when exploring generative AI systems and the underlying technology?
A. Organizations need to do some groundwork before implementing genAI. While genAI can learn on the fly and adapt to different workflows, it still needs some help getting off the ground.
First, healthcare systems need to digitize as much of their information as possible. Again, you want genAI trained on the healthcare system, data, and workflows, and this can only happen if information is digital.
Next it is essential Each genAI tool works with the systems provided. Is it compatible with an organization’s EHR? The provider portals used? Can it scale across service lines, even complex ones?
Last but not least, organizations need to think about security. What is the genAI vendor’s data retention policy? Do they audit and encrypt all data? And the same goes for organizations. Does an organization encrypt data, audit, and only retain what is necessary?
Q. What are some best practices for genAI implementation specific to coding and revenue cycle processes?
A. It’s easy for organizations to get excited about genAI, and even easier to want to streamline everything possible. Instead, look at the low-hanging fruit. What are the problem areas that aren’t overwhelmingly large and complex?
More importantly, which areas have the most data to train genAI? These would be great areas to test the technology and prove results.
For example, with coding, an organization can get more out of its team by using a genAI tool that specializes in generating quotes and coding suggestions. This can even help with looping in doctors for coding suggestions.
Q. And how can genAI tools help solve revenue cycle staffing issues, such as the shortage of medical coders?
A. There is currently a severe shortage of medical coders. Experienced coders are retiring and not enough new people are entering the workforce. Coding teams are having to do more with less. But how? Historically, technology has been the answer.
Staff shortages create time pressure and require programmers to work faster than they otherwise would. This results in a lack of completeness because documents are skipped or small details are missed. These details can lead to missing codes or incorrect codes, which can ultimately have a negative impact on quality metrics.
GenAI can help find codes that humans would otherwise miss. Some genAI models can review clinical records faster than human operators – and dig deeper into the files – resulting in greater accuracy and achieving appropriate revenue at a lower cost.
Some models suggest correct codes, leaving programmers to check or double-check the work. This not only frees up senior programmers from having to spend time on simple tasks, but also allows new hires to work at the speed of a seasoned programmer.
Think of genAI as giving programmers superpowers. It helps them work faster and perform better. Now imagine that kind of potential over an entire revenue cycle.
Follow Bill’s HIT reporting on LinkedIn: Bill Siwicki
Send him an email: bsiwicki@himss.org
Healthcare IT News is a publication of HIMSS Media.