Modern healthcare is increasingly complex, and the need for accurate, well-organized data is crucial to clinical, operational, and financial success. Superior data quality in healthcare is essential to identify patterns, make diagnoses, and tailor treatment plans. Data also supports the integration of advanced technologies like artificial intelligence (AI) and machine learning, which rely on structured data to generate reliable insights. 

Value sets – standardized lists of codes and terms that define clinical concepts – play a pivotal role in this data-driven environment and ensure consistency and accuracy across systems, applications, organizations, and even the departments within them.  

With this backdrop, large language models (LLMs) have the potential to be powerful tools in the creation and maintenance of value sets. They can help clinicians and analysts develop accurate and standardized value sets, reducing manual burdens and human error and making contextually relevant data usable at scale. 

How LLMs contribute to value set management 

As data grows more intricate and dynamic, the role of LLMs in managing value sets becomes increasingly significant. Here’s how LLMs can enhance the creation, standardization, and maintenance of value sets: 

Criteria understanding 

LLMs can interpret textual inclusion and exclusion criteria, extract relevant clinical concepts, and map them to standardized codes. For example, criteria like patient with Type 2 diabetes would be recognized as the concept Type 2 diabetes mellitus, and mapped to ICD-10 code E11.9 for Type 2 diabetes without complications. 

Incorporating inclusion and exclusion rules 

LLMs can also apply inclusion and exclusion rules to ensure only the most relevant codes are selected. This automation helps narrow down the value set to those codes that best match the specified criteria, reducing the risk of including unrelated or unnecessary codes. 

Managing complex clinical relationships

LLMs can manage complex relationships like comorbidities or contraindications. For instance, if the criteria require excluding patients with both diabetes and heart disease, the LLM can ensure that codes related to both conditions are filtered out when generating the value set.  

Contextual understanding 

LLMs bring a deep understanding of the context in which terms are used, like a specific diagnosis or treatment protocol, helping to refine value sets. 

Customization and specificity 

LLMs can adapt the code generation process to suit specific clinical settings, patient populations, or research purposes. This customization ensures that the value sets generated are fine-tuned to meet the unique needs of the organization, whether it’s a small clinical practice or a large research institution. 

Incorporating LLMs into the creation and maintenance of value sets not only streamlines the process but also enhances accuracy, relevancy, and adaptability and lays the foundation for more efficient, scalable, and clinically meaningful value set management. 

Advantages of LLMs for value set management 

LLM-powered value set management can enhance data workflows with unprecedented speed and scalability, improve customization, and maintain accuracy. These advancements are crucial in a field where precision and efficiency are critical to financial success and competitive standing. 

Speed and scalability 

Automation through LLMs dramatically accelerates processes such as concept extraction, terminology mapping, and value set updates. This efficiency reduces the manual effort but also allows organizations to scale their data utilization without a significant increase in resources, freeing up time and enabling teams to focus on strategic priorities such as improving care, advancing research, and optimizing clinical workflows. 

Consistency and error detection 

Manual processes are prone to errors and variability, which can lead to inconsistencies in data and reporting. LLMs, by contrast, reduce the likelihood of such errors and minimize discrepancies through flagging, ensuring value sets are uniform and enhancing overall data reliability and quality. 

Validation and iteration 

LLMs support continuous refinement by allowing for easy updates and adjustments to value sets as new criteria emerge or medical knowledge evolves. This makes it easier to keep value sets accurate and up to date without requiring manual intervention each time clinical practices change. 

Clinical relevance and meaning 

By managing complex relationships and ensuring value sets are aligned with intended uses, LLM-supported value set management can further improve decision support and interoperability. 

Considerations for LLM-powered value sets 

When implementing LLMs for value set management, it’s important to recognize that, while these models offer significant benefits, they also require careful oversight and customization. The balance between automation and oversight is essential to maximize accuracy and trust in the resulting data. 

Training and fine-tuning 

LLM effectiveness is highly dependent on the quality of the training data. To achieve accurate and relevant results, LLMs must be trained on high-quality, domain-specific healthcare data. This involves curating and preprocessing large datasets of medical terminology, code systems, and other relevant information. Proper training ensures the LLM can understand and process complex medical terminology and concepts accurately, leading to more reliable value sets. Organizations must invest in the right training processes to maximize the benefits of LLMs. 

Human oversight 

Despite what LLMs can bring to value set management, human oversight remains essential. Clinical experts play a critical role in validating the value sets generated by LLMs. Their expertise is necessary to ensure that the codes and terminologies proposed by the models align with current clinical practices and standards. Human validation also helps catch any errors or discrepancies that the LLMs might miss and provides an additional layer of assurance that the value sets are accurate and applicable. This collaborative approach combines the efficiency of AI with the nuanced understanding of human experts, leading to optimal outcomes in healthcare data management. 

Moving the value set needle 

The current state of value set management reveals significant challenges with traditional in-house approaches and open-source tools and data, which often struggle with scalability and maintenance.   

In response to demands for streamlined data management and more control over value sets, IMO Health developed IMO Precision Sets, backed by clinical terminology expertise. Delivering a first-of-its-kind editing and maintenance tool, plus highly accurate pre-packaged value sets, IMO Precision Sets goes beyond what is available in the market today and is primed for LLM integration to make value set workflows even faster and more accurate.