Artificial Intelligence in Healthcare - A Medical Coder's Perspective

As the “work done by machines” increasingly contributes to the provision and intensity of medical services, the taxonomy established and governed by the American Medical Association will assist coders in accurate reporting, coverage, and reimbursement throughout its evolution. Appendix S of the Current CPT code set describes the various types of AI and how it is organized within the CPT code set. This will help facilitate patient access to the technology and accurate payment to physicians. 

It seems like everywhere we turn artificial intelligence (AI) is a topic of discussion in healthcare. I’m sure coders have long wondered if it was possible for AI to accurately perform the tasks required of coders. For many years I used computer-assisted coding (CAC) as part of my job duties and was told it was just to enhance the number of encounters I could send out the door in a day, but was not reliable enough to simply turn on and not be monitored by a coder. Certain types of services were able to be turned on at almost 100%, maintaining just a little oversight, but only because the documentation was consistent and so were the reportable codes. Today, AI has continued to advance and it seems like it is everywhere and so are terms like autonomous or augmentative AI, leaving some to wonder who is in charge of this AI machine and how big is it going to get.

This same question was floating around among providers and healthcare technology arenas, and questions about how to organize or categorize AI in the healthcare industry to ensure accurate documentation, oversight, and reporting. In 2022, the AMA developed and published a new CPT appendix, “Appendix S "Artificial Intelligence Taxonomy for Medical Services and Procedures”, which categorizes three types of AI, which supports and assists human activities, with the ultimate decision-making authority remaining with humans. 

Assistive AI: As the name describes, this type of AI assists the provider in their tasks. An example of this that everyone might understand is the use of a virtual assistant, like Siri or Alexa. The assistive AI can respond to user queries and perform tasks based on very specific instructions but not actually make decisions independently. In healthcare, a similar form of virtual assistant is done in image analysis. The AI system analyzes the medical images to identify potential abnormalities, lesions, or patterns that may indicate a specific condition or disease (e.g., mammography with computer-assisted device (CAD). This type of AI highlights anomalies or areas of interest on the image that could be problematic so a radiologist can easily identify and investigate them. This automation, performed in the initial analysis, is assistive because it helps to focus the radiologist’s attention to areas of concern faster and more efficiently. It also is like a second set of eyes providing insights and helping to detect any subtle or early-stage abnormalities that might otherwise be challenging to identify through a manual review alone. And finally, it helps with workload distribution, by automating routine tasks so the healthcare provider can concentrate on the more complex aspects of patient care.  
Assistive, augmentative, and autonomous AI in healthcare is designed to complement human expertise, not replace it. The final diagnosis and treatment decisions still rest with the healthcare professionals, with AI serving as a valuable tool to enhance their capabilities.

Augmentative AI: As the name describes, this form of AI also enhances human capabilities by working collaboratively with individuals, combining human expertise with AI capabilities to improve overall performance. In medical diagnostics, there are AI systems that augment the healthcare provider’s ability to identify patterns or anomalies, leading to more accurate diagnoses and more, such as: 

• Remote Patient Monitoring: Using wearable technology to monitor and analyze in real-time a patient’s health data and provide continuous monitoring of patients with chronic conditions or high-risk conditions to enable earlier intervention. 
• Predictive Analytics for Patient Outcomes: This can be used in patient risk stratification to predict a likely patient outcome or it could be used to analyze provider coding data to determine if a provider or payer is high risk for inaccurate coding and reimbursement and therefore may be added to a kind of watchlist for monitoring. 
• Natural Language Processing (NLP) for Healthcare Records: This is what many coders experience when they use forms of AI to help with coding. We’ve all seen those insanely long encounters, where it seems like every single sore throat and toothache has been documented when the patient was just in for a fractured arm. Who wants to read the entire encounter hoping to find the key diagnoses? Well, AI is fantastic in this capacity, augmenting the work of a medical coder but leaving the ultimate decisions about which codes are relevant and which are not to the coder.
• Healthcare Fraud Detection: Along the same lines as NLP for healthcare records, payers like Medicare use AI for healthcare fraud detection. AI algorithms analyze vast amounts of healthcare data to identify irregularities and patterns indicative of fraudulent activities and highlight those anomalies for review. 

Autonomous AI: This form of AI can operate independently of human intervention. Not quite like the movie I-Robot, but more along the lines of the autonomous automobile (well, ok a little more like I-Robot). The ability to assess data and make decisions followed by taking action without oversight. It learns from experience and adapts to changing circumstances. Once the autonomous AI is programmed, receives instructions or rules, it can make decisions without human control but it does not remove the responsibility of the human to intercede and even, at times, provide new rules or retraining. A couple examples of autonomous AI include: 

• Robotic surgery systems: These help perform minimally invasive surgical procedures where the decisions made without direct intervention by the surgeon are less common due to the less critical nature of the procedure. There are many types of robotic systems used in surgery, which are equipped with AI algorithms and assist the surgeon in performing very precise and controlled movements during the surgery. Some have adaptive capabilities, allowing them to respond to the surgeon’s movements in real-time and even adjust their actions accordingly.
• Data Analysis: The AI component can analyze data input from sensors and provide insights to the surgical team about the patient's condition throughout the procedure.
• Autonomous medical coding: This form of AI learns from the coding habits of the entity using the AI and over time retrains itself to know which codes are sequenced first, or require an additional code. It codes the report or an entire outpatient encounter (e.g., history and physical, images, operative report) and provides all the correct codes, with correct coding edits and sequencing for the coder to review and approve or deny. 

While these systems have autonomous features, they operate under the control and supervision of skilled providers, who often limit their use to specific aspects of the procedure, and the ultimate responsibility for decision-making rests with the human.

With regard to the AI systems used in healthcare, the AMA stated, 

As we move into the future with AI, we are sure to see some incredible advances in assistive, augmentative, and autonomous AI. With the 2024 CPT code updates, we saw the following newly added CPT code that contains language categorizing the service as augmentive AI:  

The AMA states they do not have a specific, formal definition of AI, as the field is dynamic and rapidly evolving and any definitions would need to be updated frequently to ensure new perspectives and insights are provided as the definition evolves. Personally I think that is an open-minded approach to AI, to not try and constrain it but allow it to develop and grow while managed by highly trained and skilled healthcare personnel.

We should expect to see continued advances in AI use in the provision of medical services and as such, should take the necessary time to study these services and code descriptions to ensure documentation meets the code descriptions for reporting purposes and reimbursement. Staying up to date with the new AI coding information will help you remain relevant in your career as a medical coder and provide a source of confidence in the ability to ensure proper code assignment and provider education where needed. 

###

Questions, comments?

If you have questions or comments about this article please contact us.  Comments that provide additional related information may be added here by our Editors.

Latest articles:  (any category)