What is autonomous coding, and why is it so important to get it right?

Medical coding automation utilizing artificial intelligence (AI) has been happening for more than 10 years. When the industry was getting ready for the immense change that was moving from ICD-9 to ICD-10, health information management (HIM) departments had to embrace automation to help with the transition without damaging their bottom lines.

Throughout the years, as AI grew, deep learning and now generative AI models were developed, we are now in the next wave of massive technology adoption to further and fully automate coding workflows. This time, the focus is on combatting the coder shortage crisis and stiff budget challenges that have escalated over the last five years. Enter autonomous coding.

The quest to define autonomous coding for HIM

After attending last year’s AHIMA 23 annual conference in Baltimore, my biggest takeaway was the increased attention to AI in healthcare and medical coding automation. However, it seemed like there wasn’t a clear definition of autonomous coding. It had been talked about for the past two years or so, and as the industry was trying to embrace it, defining it was the first step.

A recent KLAS report pointed out that the definition of autonomous medical coding varies among healthcare organizations and technology providers. So much so that organizations sometimes consider any type of coding automation as autonomous coding as long as it removes or decreases the need for human review. The industry also agreed that for it to be effective, it needed to provide 95% accuracy – the current industry standard for human coders.

An organization presenting at the AHIMA23 conference pointed out how they trained their coders on a specific coding area within the emergency department to not review the codes generated by the technology if it met certain criteria. However, the coders were still touching each record by simply clicking “accept.” They called it autonomous coding. But was this autonomous? Maybe the human coders were, but not the technology.

So, what really is autonomous coding?

Solventum defines autonomous medical coding as the ability to process electronic patient data (e.g., a chart or documentation from an encounter) and generate a confident final code set that is ready for the next step in the billing process without human interaction.

Autonomous coding solutions typically leverage AI and/or logic to generate codes automatically and qualify the encounter as “final coded ready” without any human intervention. Thus, humans are eliminated from interacting with the chart or documentation from the encounter that successfully went through this process.

The case for a complete, compliant coding automation environment

AI modeling alone can only get you so far. Medical coding is complex and requires a fully integrated process where changes in documentation or new documents get updated and entered into the autonomous coding process. But that’s just the start. A well thought out program needs to consider coding guidelines, local facility or department criteria, state specific payer rules, etc.

And what about quality assurance? Accuracy is paramount to avoid denials after a bill has been submitted. Dealing with denials can be costly and takes up precious human time. Having control over what goes out the door automatically (again, without coder interaction) is a tall order and makes many HIM leaders understandably nervous over the idea of letting visits “go out the door” without a human reviewer. So, it is imperative that an autonomous coding system include capabilities for administrators to:

• Decide what specific chart categories or specialties go through the process without human intervention
• Decide what percentage of visits qualified for automation are stopped for quality review
• Create enterprise specific criteria to further qualify visits for automation by flagging a visit for coder review, if needed
• Easily review what the system coded and what changes, if any, were provided by coders
• Track and compare any changes a human coder makes against the system-generated codes
• Pull reports to analyze what was coded and drill down to the code or encounter level

A comprehensive autonomous solution must seamlessly handle all patient visits, whether they are qualified for automation or not. Human coding or validation of auto-suggested codes still need to happen for those patient visits that don’t qualify for automation.

Yet, how can these cases also be expedited to improve productivity? A semi-autonomous integrated workflow that provides automation at the coding level can enhance the current workflow of coders, who will likely see greater productivity. The good news is that this is already possible with enhanced computer-assisted coding (CAC) that uses expert-guided AI for confidence level assessment to automatically move codes closer to the end of the process, all within one workflow.

The bottom line – a complete, compliant coding automation environment must provide transparency and control over fully autonomous coding workflows that have no-coder intervention; a comprehensive, integrated solution that supports a semi-autonomous workflow for those non-qualified charts or encounter documentation; and last but not least, thorough, engaging reporting capabilities with the ability to drill down to the necessary detail to give HIM professionals the power to make staffing decisions and process improvements.

Giannina Rachetta, licensed scrum product owner (LSPO), is a senior product marketing manager at Solventum.