Healthcare data is often accessed as individual elements, but rich clinical meaning lives in the connections between them. A clinically-grounded knowledge graph makes those connections explicit, breaking down silos and turning disparate data into a complete, credible representation of clinical reality.
As artificial intelligence (AI) and large language models (LLMs) play a growing role in healthcare, the context provided by integrated data becomes increasingly important. Relationships between data points or clinical concepts – architected and governed by clinical experts – are an indispensable foundation for individuals and AI agents alike, grounding more explainable and trustworthy decision-making, no matter the use case. IMO Health’s Knowledge Graph was built to do exactly that, by leveraging the organization’s unique history and capabilities.
- Decades of experience: We’ve been building and structuring our proprietary terminology, ontology, and code mappings for over 30 years. The Knowledge Graph’s 26 million nodes and 1 billion edges now bring it all to life in exciting new ways.
- Ongoing, expert governance: Our team of clinical informaticists, terminologists, data scientists, coders, and mappers collaborate on the continuous maintenance and hydration of the Knowledge Graph, keeping it current, accurate and reliable.
- Unparalleled coverage: From our clinically curated terminology to multi-domain coverage that spans medications, labs, diagnostics, conditions, and more, IMO Health’s Knowledge Graph can go deeper and broader than our competitors, fueling countless AI applications.
But reading about the Knowledge Graph isn’t the same as seeing it in action. With that in mind, we’ve created a contained “sandbox” where we’ll explore two scenarios based on a diagnosis of “Atrial Fibrillation”: 1) getting to greater diagnostic specificity and 2) connecting data across domains. Let’s get started.
Getting to greater specificity
For this first scenario we’ll keep the example simple, demonstrating the depth of our Knowledge Graph by traversing it to find a more specific diagnosis and connecting that diagnosis to the appropriate ICD-10-CM code.
The initial screen under the Diagnostic Specificity tab highlights “Atrial Fibrillation” or AFib.
Clicking on AFib produces a pop-up of the five available relationships. The same list can also be found in the left navigation:
- Narrower Concepts
- Allowed Refinements
- Broader Concepts
- Synonyms
- Code Mappings
By selecting All, the relationships fans out from the center, showing the different directions the Knowledge Graph can take. Do note, however, that for the purposes of this AFib sandbox, we have limited the number of relationship categories. A personalized demo shows the full depth and breadth of what is possible.
Narrower Concepts takes us deeper and reveals five options for drilling down to a more detailed diagnosis.
Clicking on any of those concepts shows their corresponding code mappings. The image below takes us deeper into “Longstanding Persistent Atrial Fibrillation,” which is mapped to a unique ICD-9-CM, ICD-10-CM, and SNOMED® code, respectively.
Why it matters:
Specificity is essential beginning at the point of care but maintaining that level of detail and the clinical intent behind it have ripple effects from the billing and coding departments within health systems to the health tech companies building revenue cycle solutions to support them. For example, in the scenario above, selecting “Atrial Fibrillation, Unspecified” (I48.90) as opposed to “Longstanding Persistent Atrial Fibrillation” (I48.11) has clear revenue cycle implications.
- Risk Adjustment for HCCs/VBC: Codes that lack specificity can negatively impact risk-adjustment factor (RAF) scores and underrepresent patient population severity, impacting reimbursement for value-based care contracts.
- Claims processing: Choosing an unspecified code when the clinical record indicates a chronic or more advanced condition can trigger requests for further documentation and cause administrative delays.
- MS-DRG impact: Vague medical coding masks patient complexity, causing complications and comorbidities (CCs) to be missed, which can lead to lower payments based on the assigned tier. The same lack of specificity can result in payer rejections and expensive rework.
Conneting data across domains
In this scenario we switch tabs to Cross Domain to explore the breadth of the Knowledge Graph. Selecting Atrial Fibrillation here brings up a new menu of relationships, including:
- Treatments
- Supportive Treatments
- Diagnostics
- Associated Problems
- Caused Problems
- Code Mappings
Viewing all of these relationships together reveals the top-level nodes (data points) and edges (relationships) available across domains.
Once again, it is important to note that in the full Knowledge Graph – not the narrow AFib sandbox shown in this blog – any of these nodes can serve as a starting point. If a diagnostic like a Holter monitor is the logical start to a query, it will connect the user to AFib, among other options. If Carvedilol is being used as a treatment, that can also be the jumping off point for the user.
Let’s stay with Carvedilol. From AFib, select Treatments – either from the relationship pop-up menu or the left navigation – to explore the 15 treatment options, including Carvedilol.
Why it matters:
The relationships between nodes on the Knowledge Graph can reveal a great deal about patients and populations. For example, if one knows a patient is taking a specific medication, that information could be leveraged to ground agents with essential clinical context such as the likely underlying condition, related laboratory findings, associated comorbidities, disease severity, and other clinically relevant concepts. In the case of Carvedilol, that context could be used to:
- Support a differential diagnosis and facilitate a course of treatment
- Identify that appropriate medication treatment is occurring
- Help identify that a conservative treatment is not working, and the patient now needs a procedure
- Support the identification of patients who have AFib and are taking Carvedilol as requirements for a clinical trial
IMO Health’s Knowledge Graph: AFib sandbox
Knowledge Graph next steps
Greater diagnostic specificity and cross-domain exploration are just two examples of what IMO Health’s Knowledge Graph can do. But with the ability to bring millions of clinical concepts and the relationships between them together in a fully integrated, clinically grounded framework, the possibilities are endless.
Across clinical AI, analytics, revenue cycle, patient identification, clinical research, and other applications, the same foundation can help turn fragmented healthcare data into information that is more connected, explainable, and actionable.
