COPE: Chronic Observation & Progression Events Ontology

COPE: Chronic Observation & Progression Events Ontology A sar a Se n ar at n e & Zen t L im F l i n d e r s U n i v e r s i t y , A u s t r a l i a . O sh an i Se n ev ir atn e R e n s s e l a e r P o l y t e c h n i c I n s t i t u t e , U S A . L ee lan g a S en e vir at ne U n i v e r s i t y o f M or a t u w a , S r i L a n k a . Background •Chronic diseases such as diabetes, cardiovascular conditions, and respiratory illnesses represent a significant burden on global healthcare systems, accounting for nearly 74% of all deaths worldwide. (Source: WHO) •Early detection and effective intervention are critical to managing these conditions and improving patient outcomes. •There is no semantic framework that cohesively brings together knowledge on patients, disease trajectories, potential interventions over time. 2 Wearables EHR AI Models What if we could trace how an AI model’s prediction aligns with a patient’s real-world disease journey? 3 Onset Intervention Outcome •Today, AI models can predict risks or classify disease states, but we often can’t see why they made a certain prediction or how that aligns with a patient’s lived experience. •We need to capture that entire trajectory, so that we can not only understand what an AI predicts, but when and in what context those predictions make sense. Why Existing Ontologies Fall Short 4 Ontology Strength Limitation SNOMED CT Clinical concepts Atemporal ML -Schema AI workflows Not clinical FHIR Interoperability No rich semantics of AI or temporal concepts What’s missing is a unified, time-aware framework that connects clinical knowledge with AI modeling. Our Vision: COPE Ontology 5 Patient & Disease Layer Temporal Trajectory Layer AI Models/Research Layer Design Objectives •Classes and properties that formalize relationships between patient profiles, observed symptoms, clinical interventions, and the dynamic progression of chronic illnesses. •Semantically links these entities to AI/ML models (such as classification, regression, and clustering), data types (such as structured, temporal, and multimodal), sources (such as EHRs, ECG, and wearable sensors), and scholarly outputs (such as datasets, publications, and venues). •The Observation-Event class models timestamped clinical observations and intervention records, enabling representation of time-dependent health trajectories for longitudinal queries such as; •What symptoms emerged after a specific intervention? or •How did the patient’s risk profile evolve across disease stages and episodes? •COPE is designed to: •Support health informaticians, behavioral scientists, and data scientists, and •Predict suitable interventions based on disease and patient profiles. 6 Conceptual Overview of COPE 7 Health Trajectory 8 •A collection of linked Observation Events that tell the story of a patient’s journey. •Each trajectory is specific to an entity that could be an individual patient, or even a larger community trend. •When a patient’s physiological or behavioral data starts to drift — maybe blood pressure rises or sleep patterns deteriorate — that deviation moves them into an anomalous space. •Over time, a person may experience multiple trajectories — some continuing, others resolving or terminating after an intervention. •The concept of participation rank reflects which trajectory the person is currently in. For instance, rank 0 is their active trajectory. Ontology Design •We developed COPE following a structured evidence-driven design approach: (1) Created a taxonomy informed by requirements elicitation and literature analysis. (2) To gather requirements related to personal characteristics and health trajectories, we consulted a behavioural scientist, who is also a co-author of this paper. (3) We also reviewed ~25 research papers focused on the application of AI in chronic disease prediction and management. •We formulated competency questions to define the scope of the ontology, drawing on both our domain expertise and insights from the literature. 9 THANK YOU Oshani Seneviratne [email protected]