Anesthesia Coding Alert

Prevent denials by using AI to analyze past claim denial trends.

After decades of development, artificial intelligence (AI) has gone from being a thing of science fiction to commonplace technology that is becoming increasingly pervasive. While AI is not new to healthcare, its applications are expanding beyond leveraging the technology to deliver better care as those on the business side of healthcare are realizing the extent of its potential.

As AI evolves and becomes more sophisticated, it will inevitably play a more significant role in revenue cycle management (RCM). Stay ahead of the curve by learning about the ways that AI can be used in RCM — including claims processing, denials management, patient billing, and collections — to boost the timeliness, accuracy, and effectiveness of your efforts.

Add AI to Augment Your Revenue Cycle

A 2020 Change Healthcare study found that 98 percent of respondents planned to use AI in RCM by 2023. At the same time, approximately 65 percent of respondents claimed to use AI in RCM at the time of the study.

Two areas where AI would be a natural fit into your revenue cycle include:

1. Predicting Future Claim Denials

One of the biggest challenges that medical practices face is managing denials. AI can help predict claim denials by analyzing past denial trends and alerting HIM professionals of the potential denial in advance of billing.

By providing the AI with a significant amount of data related to your past claim denials, the technology will be able to predict future claim denials. This can be done in real-time or as a final analysis before the claim is submitted electronically. “Denials are rising. Since 2016, the average denial rate was 9 percent and as of the second quarter of 2020, it had risen to 10.8 percent,” said Holly Ridge, BSN, RN, CPC, CPMA, manager of medical necessity and authorization denials for Duke Health in Durham, North Carolina, during AAPC’s HEALTHCON 2022 session, “Medical Necessity Denials — When and How to Appeal.”

By adding AI to your claim submission process, coders and billers can know if a claim may receive a denial before submitting the claim to the payer. Regardless of whether you receive the possible denial warning as you’re filling out the claim or as a final pre-submission check, the AI’s warning will allow you the chance to make adjustments to ensure an accurate claim.

Important: You should know that the technology will only be able to identify denials based on the information it’s provided. This machine learning uses the submitted data to develop a training dataset, which the technology compares to new information when the AI is activated.

2. Filling in During Staffing Shortages

In January 2022, AKASA published findings of a survey that looked at revenue cycle staffing shortages in hospitals and health systems. The survey showed the number of days to fill vacant roles for different levels of revenue cycle talent:

• Entry-level talent: 84 days
• Mid-level talent: 153 days
• Senior-level talent: 207 days

AI can help maintain production in your workflow during these staffing shortages. The technology can handle simple coding cases with minimal human input, which allows human coders to tackle more complex cases. But human coders still need to review the information to ensure the AI-chosen codes are correct.

Additionally, with the number of denials rising, AI could review denied claims and offer suggestions on how to appeal the claim.

The ultimate goal of incorporating AI into medical coding isn’t to replace human interventions, “but to make human interventions smarter,” so the coder can use their time and knowledge elsewhere, explains Devanshu Yadav, vice president of strategy for Raapid in Louisville, Kentucky.

While there is significant promise for AI to improve your revenue cycle, obstacles still stand in the way of the technology’s growth.

Understand the Challenges for AI in Medical Coding

OpenAI released ChatGPT to the public in November 2022. On Feb. 9, 2023, a study was published showing that the technology successfully passed the United States Medical Licensing Exam (USMLE). While this prompt-based technology was able to “[perform] at or near the passing threshold for all three exams” of the USMLE, obstructions still present themselves for using AI with medical coding, billing, and the revenue cycle, the study noted.

One such challenge involves the complexity of the data. “Healthcare data is complex and understanding the clinical context is difficult for AI,” Yadav adds. At the same time, clinical documents feature different styles, abbreviations, and layouts — all of which the AI can learn over time, but these complexities will require humans to teach the technology to get it up to speed.

Another problem involves gathering datasets for the machine learning. Healthcare organizations and hospitals are abundant spaces for data collection, which will be beneficial to training the AI to understand physician’s notes, operational notes, lab results, and imaging reports. For AI to accurately identify billing errors, the technology will need a complete picture of a patient’s record.

The necessary data includes:

• Electronic health record (EHR) data
• Admission notes
• Discharge summaries
• Physician’s notes
• Lab results
• Pharmacy information

Teams can gather this information through Health Level Seven^® (HL7^®) international interfaces or EHR connectors, health information exchanges (HIEs), and data dumps and pickups (via Secure File Transfer Protocol (SFTP)).

Also, for AI systems to code encounters accurately, the technology will need to be up to date on the medical codes, instructional notes, and guidelines. Implementing this information into the AI system will require careful guidance and assistance from human coders to ensure the output is correct. Simultaneously, subjectivity is a skill humans possess that allows them to determine the correct codes for each case.

These challenges don’t mean that AI cannot function in medical coding, but more that a sophisticated team of AI and natural language processing (NLP) specialists will be needed to make adjustments. “The key point is there cannot be one-size-fits-all AI tools, but customization needs to be done on the AI tool every time for the technology to be successful for that particular organization,” Yadav says.

Medical Coders and AI Working Together

Technologies, such as computer-assisted solutions, enable medical coders to automate manual processes and have the technology perform a primary review before the human begins the coding work. The computer-assisted solutions are helpful, but still require a human coder to use their knowledge, subjectivity, and critical thinking skills to code the medical reports.

How would a system with AI implemented change the coder’s role? “The role of the coder with AI is now more of a validator or reviewer. However, coders need to be trained and onboarded comprehensively on how to use AI tools effectively,” Yadav says. With Raapid, humans are involved throughout the coding process to provide feedback and confirm code selections to maintain accuracy.

Bottom line: AI isn’t meant to replace human interventions but to make those interventions smarter, Yadav explains. By using the AI to handle the recurring tasks, the human coder is freed up to tackle more complex coding cases, spend time on other tasks that require greater attention, interpret cases, and review clinical outputs to make determinations. This will “make human experts do what matters — interpretations and ensure excellence of care,” he says.