In a recent webinar, Enhancing Medical Coding with AI and LLMs, IMO Health experts Vidhya Sivakumaran, PhD, VP of Clinical Informatics & Terminology Data Engineering, and Jingqi Wang, PhD, SVP of Data Science & Chief AI Architect, shared how LLMs are transforming coding workflows — and where the technology needs improvement.
From foundational challenges to practical applications, here are five key takeaways from the session — including clips and commentary you won’t want to miss.
Pressed for time? Continue scrolling for key insights.
1. LLMs alone aren’t enough for medical coding accuracy
“LLMs are not appropriate for use on medical coding tasks without additional research.” — Wang
General-purpose LLMs like GPT-4 or Llama may seem capable of coding on the surface, but without guardrails, they often return incorrect, vague, or even fabricated codes. As Wang pointed out, some early models generated non-existent ICD-10-CM codes or lacked the specificity needed for reimbursement. That’s a deal-breaker in clinical workflows.
2. The secret sauce: Combining LLMs with curated clinical terminology
“LLMs can achieve this performance — but only when powered by rich semantics, which is powered by IMO Health terminology.” — Sivakumaran
The power of AI lies not just in the model, but in the data it draws from, Sivakumaran emphasized. IMO Health’s curated terminology includes more than one million unique clinical concepts, enabling models to understand how clinicians actually document care. Plus, meticulous mapping to code sets like ICD-10-CM, SNOMED®, CPT®, LOINC® is what enables AI to perform at a clinically acceptable level.
3. AI in medical coding is not about replacing humans — It’s about augmentation
“This evolution isn’t about replacing human coders but rather augmenting their capabilities and allowing them to focus on the most complex cases.” — Sivakumaran
LLMs are powerful tools, but they’re most effective when they work alongside human coders — not instead of them. AI can take the first pass at routine coding tasks, flag inconsistencies, or surface relevant secondary codes. This lets human experts focus on edge cases, compliance nuances, and patient context that models can’t (yet) fully interpret.
4. AI agents improve transparency and trust
“With AI agents, the solution is not a black box anymore. You can literally see how the LLM tries to solve the problem.” — Wang
One of the most impressive parts of IMO Health’s approach is transparency, said Wang. Using AI agents and prompt engineering, the system not only provides code, but it also explains why. Whether it’s synonym recognition, rule-based logic, or multiple query attempts, you can trace how the AI arrives at its answer which is critical for adoption and trust.
5. Better coding = Better risk adjustment and reimbursement
“When we integrate IMO Health’s terminology with LLMs for medical coding, we’re building the strongest possible foundation for accuracy and completeness.” — Sivakumaran
Accurate coding isn’t just about billing — it drives everything from quality reporting to HCC risk scoring. IMO Health’s AI-enhanced system automatically suggests secondary codes, flags modifiers, and ensures that complex conditions are captured with the right specificity. That means fewer denials, faster payment, and better representation of patient complexity, said Sivakumaran.
LLMs hold promise, but they need a strong clinical foundation to reach their full potential.
Out-of-the-box models simply can’t deliver the accuracy or compliance healthcare demands. However, when AI is grounded in proven terminology — and guided by clinicians, coders, and informaticists — it becomes a true partner in care delivery and revenue integrity.
Download our recent guide, Optimizing LLMs for smarter medical coding, to learn more.
SNOMED and SNOMED CT are registered trademarks of SNOMED International.
CPT is a registered trademark of the American Medical Association. All rights reserved.
