AI-based System Reduces Denied Claims For Community Medical Centers Of Fresno
The provider saw a 22% decrease in one type of denial and an 18% decrease in another. Those improvements have resulted in more than 30 hours per week in eliminated work for follow-up staff. A new AI-based system has been implemented by Community Medical Centers of Fresno to reduce denials. The system uses claims and payer remittance data to predict denials before claims are filed and also score incoming denials with a “probability of recovery” score. The tool was able to stop claims on the front end, allowing follow-up staff to identify areas where they are likely to be paid quickly after an initial appeal. Despite these improvements, the team continued to search for new tools to improve their efficiency. The new system was developed with Experian Health and used the same direct connection to our EHR as our clearinghouse claims operations/denials.
Published : 2 months ago by Liam in Tech
For Community Medical Centers of Fresno, like many other healthcare facilities, denials are a constant source of pain, and managing them is an ongoing battle.
Over roughly the past decade, payers have increased their delay tactics — one method is to do so with denials, says Eric Eckhart, director of patient financial services at Community Medical Centers of Fresno.
“This increased volume translates into a higher workload for my team,” he explains. “We knew that many of our historical workflows were not efficient enough to handle this new level of denial volumes. This situation led us to look for new systems that we could implement internally or systems that involved a vendor-based system used to be.
“Over the past five years we have been on a journey to put the team in the best possible position to succeed, by streamlining staff reporting structures, improving denial reporting, updating EHR workflows and eliminating paper processes,” he continued. “In addition, more attention has been paid to preventing refusals upstream.”
Despite all the positive changes that were made, Eckhart knew that this was not the end of the improvement journey. The team had to keep looking for new tools to add to the toolbox.
Eckhart was approached by its clearinghouse vendor in late 2022 to be a beta tester for a new AI-based system that would predict denials before claims were filed and would also score incoming denials with a “probability of recovery” score.
Both parts of this new product used Community Medical Centers of Fresno’s claims and payer remittance data to learn how payers denied and paid claims.
“Because the vendor was our clearinghouse, the data was readily available and no additional assistance was required on our part,” Eckhart noted. “The denial prediction tool used the same direct connection to our EHR as our clearinghouse claims operations/denials, so no additional programming was required. A similar trajectory was also used for the ‘probability of recovery’ score.
“Having this tool allowed us to stop claims on the front end, giving us one last chance to resolve outstanding items that could lead to a denial,” he continued. “This gave our billing team a tool to highlight areas that needed a second look.”
And if a denial did occur, follow-up staff had a scoring tool to focus their workflow on areas where they were likely to be paid quickly after an initial appeal.
“The overall intent of these tools was not to eliminate all denials, but to provide my staff with another tool to help prevent denials and also to guide their appeals efforts after the denials are received,” Eckhart said. “The main reason I chose to move forward and partner with Experian Health was because this system was not a new workflow for staff.
“There’s nothing worse than trying to get your staff to disrupt the workflow they know and log into a different system to use a resource,” he added. “This system also made it possible to customize which operations and scores my employees could see in the EHR.”
Community Medical Centers of Fresno began the denial prediction portion of the tool in early 2023. The initial rollout of the predictions was slow and very intentional to ensure leadership had buy-in from the billers who would be working on these operations.
“Over several weeks I reviewed the forecast data by IP/OP, payer, CARC code, etc.,” he recalls. “The tool is very good at predicting future denials, but not all of them can be prevented; therefore, only a select set of predictions are relevant to the team.”
The team decided to implement two CARC code predictions:
• 109 – Claim/service not covered by this payer/contractor. The claim/service must be sent to the correct payer/contractor. (Medicaid payers only.)
“The 197 prediction allowed us to ensure that an authorization process was followed upstream and it allowed us to ensure that the auth number made it onto the claim – a technical issue at the time caused this issue,” Eckhart said . “The prediction of 109 was a second check managed Medicaid enrollment issues that occurred upstream and were necessary before we could implement a coverage automation system on the front end.
“The ‘likelihood of recovery’ scoring component was implemented later in the year with our commercial follow-up team,” he continued. “There were some initial challenges in ensuring this score was easily accessible from our current work queues. But once these were resolved, we were able to integrate this into the staff’s daily workflow and it also provided a method to ensure we get the easy cash to the door as quickly as possible.”
Fresno Community Medical Centers saw significant results almost immediately. In the first six months of implementation, it saw a 22% drop in 197 denials and an 18% drop in 109 denials.
“Both metric improvements have resulted in more than 30 hours per week of additional work eliminated from follow-on staff workloads,” Eckhart reported. “This one tool has been able to free up staff time and allow for additional appeals work in the future.”
The most important advice Eckhart gives to other organizations is to make sure they get staff buy-in and start slowly.
“AI tools are just that: tools,” he noted. “Maybe we’ll get to the day when they do everything, but we’re not there yet. Human intervention and guidance are the key to a successful outcome. Also make sure that the AI is trained on data that is relevant to your organization. If the model is not trained on relevant data, it undermines the entire purpose of AI. You might as well go back to your analyst team with some spreadsheets.
“Choose the right tool for your situation and needs,” he concluded. “As with other technologies, some AI tools make sense for some and not for others. I know many of us like to be at the forefront of technology, and AI is that buzzword that we all feel we should be a part of .’ Don’t fall into this trap: find the AI technology that helps and isn’t just the next vendor offering.”
Follow Bill’s HIT coverage on LinkedIn: Bill Siwicki
Email him: [email protected]
Healthcare IT News is a HIMSS Media publication.
Topics: AI