Dr. Arnold Kim and Yash Girshbhai Amethiya
Northern Ontario School of Medicine York University, Health Policy and Management
September 30,2026
Abstract.
Healthcare has a rich and precise ontological concept space covering terminologies, treatments, investigations, medications and more. Over-reliance on unstructured text as a carrier of synthesis of clinical impressions and planning forces ad hoc approaches to infer synthesis within a fragmented data environment. This creates significant impedance mismatch in cognition, data and process. At a higher level, what our academic message is saying is that healthcare should join the myriad of "design disciplines" (from engineering to the arts) and move to a more synthesis oriented user experience, data representation, multi-disciplinary collaboration and inline computational automations / workflows. Central to this idea is for healthcare to be more formal in its representations (by moving to a formal language representation such as a Domain Specific Language). Today, formal design is facilitated by recurring motifs within a myriad of tools that are part of CAD. Typically, the front-facing design experience includes a highly reactive and intelligent editing environment, visualizations, a mix of deterministic and GenAI assistance (like "vibe coding"). Other recurring motifs include, "zoom" capability (from higher to lower level concepts), higher level composability with analytical and informative information coming from decompositional algorithms and other supporting inline computations and automations.
Central to the themes of this conference, when health care becomes formal in its representations, this opens up rich neurosymbolic AI possibilities. This mixes higher dimensional modeling (which has "black box" limitations), with human understandable deterministic lower dimensional modeling (as represented by symbols or code). In the case of healthcare, now ontological entities can be brought to formal synthesis to represent clinical impressions and planning.
Our early but growing number of academic collaborators are building a proof of concept CAD tool code named: Aurora Constellations. This will be showcased to interested attendees. The purpose of the prototype is to reify many useful CAD concepts into healthcare to give proof of concept evidence that healthcare can be more formal with reduced cognitive and time burdens, while reaping the benefits of joining the design disciplines. To summarize, our academic message is that healthcare must move away from ad hoc synthesis via unstructured text, and strive to become ontological modelers and designers. The change will not be easy, but we argue that it is necessary.
1. Extended Abstract
Fragmentation continues to be a central problem plaguing healthcare. This creates significant impedance mismatch in cognition, data, process and interoperability. In contrast modern CAD tools invariably strive for a synthesis oriented approach. CAD first debuted in the 1950s to 1960s shortly after formal languages came on scene. Formal languages were central to the first AI attempts using symbolic logic.
In a formal synthesis, relationships between concepts are of paramount importance to ease cognitive burden, allow for deterministic validation and inferences, better data analytics and many opportunities for inline and downstream automation and interoperability. The fundamental thesis of this initiative is essentially driven by seeking useful structural similarities between healthcare and the many differing design disciplines. Examples of design disciplines include architects, software engineers, animation designers, digital audio designers, engineering, etc. Intuitively, all these disciplines are concerned with how their concepts relate to each other. While many of the disciplines focus on spatial concepts and relationships, many are also concerned about more abstract conceptual relationships. Digital audio design is concerned about temporal relationships in addition to a myriad of other conceptual concerns (sound, volume, wave form modifiers etc). Software engineering is another conceptual discipline where integrations are concerned with interfacing with libraries, and compositional deterministic generation from higher level languages to lower level representations (e.g. Java to Bytecode). There are many convergent themes in CAD tools within these disciplines which suggests that there are underlying human preferences when it comes to managing complexity. It is through studying these convergent themes, that our team has prioritized feature development in our research prototype code named: Aurora Constellations.
Currently the main CAD motifs under active exploration are:
Relationships are expressed formally. Relationships are the foundation to the concept of synthesis. As an example, all treatment or investigative elements must have a formal relationship to an element from an ontologically based element in the problem list. “Orphaned” clinical care elements should be regarded as “bad design”.
Organization of conceptual relationships into gradations of higher and lower concepts. Higher level concepts are easily recognizable concepts such as “stable” or “unstable” which aggregate finer detail concepts which justify the classification (e.g. tachycardia, hypoxemia etc). Another example would be a heart failure management policy at the higher level which aggregate elements like furosemide, head elevation and echocardiogram at the lower level.
“Zoom” capability. This allows designers to assimilate and navigate patient care plans from higher level concepts which effectively has been the role of patient summaries and sign over notes, but permit the designer or systems to expand detail to the associated lower level concepts. This allows designers to scrutinize at a higher level of detail (LOD). This eliminates numerous points of duplication (such as patient summaries) which is also a major contributor to fragmentation.
Synthesis preserving composability from higher level concepts. While ordersets are frequently used as a higher-level expression of clinical planning, when patients require multiple ordersets for multiple issues, fragmentation occurs due to lack of tracking of the implied relationships within each orderset. Composability allows for rapid deterministic constructions of complex plans from modular components. To emphasize the importance of this, think about whether most designers prefer designing at the pixel level or from aggregations of pixels under the concept of rectangle, circle or Bezier curve etc.
Decomposition analysis. This allows designers and systems to examine complex compositions of patient care planning and determine which higherlevel conceptualizations contributed to which elements of the patient care plan. In other words this allows analysis of compositionality. This also allows rapid determination of which elements were added de novo by the designer’s choice.
Parametric modeling. This allows designers to mutate planning elements by calculating clinical scoring inline from the formal planning representation and computationally derive alternative pathing or modifying the scale of a planning element. An example would be modifying medication dosing based on age and renal function.
Visualizations are a recurring motif seen in CAD to assist designers to perceive Gestalt level of understanding of complex synthesis and invariably assist designers to navigate the formal representation through “zoom” capabilities. These are once again inline derivations from the formal representation.
Expressiveness. Formal expression emphasize formal concept and relationships. Concepts and expressions that resist discretization (which is common in healthcare) should be expressible through classical narrative as adjunctions to the formal concept model.
With the above ideas and further expansion through academic vision, healthcare could realistically express our concept space quickly, with reduced cognitive burden, and duplication. In an era of GenAI “reviewer fatigue” can be reduced due to automated validation. Because data is now formal, it will reduce friction between stakeholders working towards interoperable systems. It will also be easier to design formal symbolic rules of governance and policies that are actively enforced or flag violations from humans and GenAI for further analysis or active intervention (neurosymbolic AI). Additionally, formal or symbolic representations permit deterministic synthesis preserving constructions which follow human understandable rules (i.e. algebraic rules).
This is a vision that likely will require a Convergence Science approach. This is informed by our experience with interacting with healthcare, human computer interface, medical informatics, policy and management, computer science, medical education, and the EHR industry. This is an exploration that requires a significant amount of consensus building and shared conceptual understanding. At this stage, the most evidence of promise exists within design disciplines. As a latest example, the field of mathematics is at an early stage of adoption of CAD motifs through the adoption of Lean 4. The combination of formalization combined with higher dimensional modeling in mathematics has led to explosive progress along with controversy in mathematics. The larger question to academics is whether we in healthcare should join them.
