David Lareau is CEO of Medicomp systemsmaking medical data relevant, usable and actionable.
The healthcare industry has seen rapid advances in artificial intelligence (AI) and machine learning capabilities in recent times – from virtual health assistants and drug discovery to development and personalized medicine. These advances are transforming healthcare, making it more efficient, personalized and effective.
As these technologies mature, there may be growing interest in using them to support clinical documentation, quality measurement, and medical coding. Although AI shows promise on these fronts, significant challenges still have to be overcome before it might be used effectively and equitably in healthcare.
The role of AI in clinical documentation and coding
One area where AI shows remarkable potential is clinical narrative processing. AI, particularly through natural language processing (NLP) and huge language models (LLMs), can reliably extract clinical concepts from patient encounter notes and conversations, which has previously been difficult attributable to the complexity and variability of natural language in medical documentation.
By training on extensive medical texts, these AI models can accurately discover relevant codes and quality measures, streamline the documentation and billing process, and help ensure compliance with healthcare standards. Not only this technological leap increases efficiency but in addition improves accuracy the management and billing of medical records, which ends up in higher health management and Patient care outcomes.
For example, by recognizing phrases like “nocturnal shortness of breath” in a patient’s description of symptoms, AI could suggest investigating relevant conditions like sleep apnea, heart failure, or COPD that could be eligible for hierarchical condition category (HCC) risk adjustment. This ability to link free text with standardized diagnosis codes and terminology could possibly be a useful support for physicians looking for comprehensive documentation of care. The next step must be to offer tools that allow a clinician to diagnostically filter the codes and terminologies to discover information relevant to every patient condition or diagnosis.
Challenges and limitations in AI-driven coding
However, significant obstacles remain in accurately mapping different patient narratives to coded criteria. A serious limitation today is the consideration of individual demographic aspects that impact health and diagnosis. Most LLMs source training data from broad clinical corpora that reflect the patterns of the bulk population. Therefore, they often overlook the unique needs or clinical characteristics of minority groups. To ensure truly equitable AI support in healthcare, much larger data sets that capture underrepresented perspectives are required.
Data integration and terminology issues also proceed to hinder the usage of AI for quality coding. The multitude of diagnostic code systems equivalent to ICD, CPT, SNOMED and RxNorm have different formats that don’t work together easily. Medical language itself presents one other hurdle, with complex synonymy, polysemy and negation that AI still struggles to process properly. The distinction between expressions equivalent to “feeling of falling” attributable to temperature or illness illustrates this subtlety.
Without the flexibility to create holistic patient profiles from fragmented data sets and validate coded results, AI cannot yet autonomously perform clinical quality measurements (CQM) and HCC identification. Hybrid human-AI approaches are promising here. Using AI to represent clinically relevant information from charts will help human abstractors discover applicable codes and quality measures. But clinical experts still must review the AI suggestions to discover any omissions or inaccuracies before signing off on the ultimate coding.
Realizing AI’s full potential while overcoming its current limitations would require continued progress from healthcare organizations, AI developers, and policymakers alike. Representative data collection, unbiased algorithms and human-AI partnerships will pave the way in which forward. If used equitably and responsibly, AI-powered clinical intelligence could significantly improve quality measurement, risk adjustment, and care outcomes for all patient populations.
Forbes Technology Council is an invitation-only community for top CIOs, CTOs and technology managers. Am I qualified?
