Clinical Research
What the data never recorded.
Clinical AI is trained on outcomes. Patients live in decisions. My research is about the layer in between — the state, context, uncertainty and timing behind each clinical decision, none of which survives into the record.
Peer reviewer and published author, Diabetes Care (American Diabetes Association). Living with Type 1 diabetes since 1967.
Peer-reviewed publications
Calendar-based CGM segmentation misaligns with continuous circadian glucose oscillators. Circular time representation and Shannon entropy analysis show midnight segmentation produces a 2.8-fold reduction in continuity correlation and 12.1% information loss.
Simultaneous hospital alarms elevate cortisol and directly counteract insulin effectiveness in DKA. Argues for environmental audit and sensory-aware design as patient-safety interventions.
CGM systems fail to capture the behavioural and cognitive context behind glucose variability — and physiological data alone does not reflect adherence or clinical intent.
Frameworks
ESP-360° Decision-Aware CGM Dataset
A novel annotated dataset pairing continuous glucose monitoring signals with real-time cognitive decision context: 536 records, 341 structured variables and 21 human-state annotation layers.
Decision Shadow / Decision Provenance
Formalises how prior clinical decisions generate persistent interpretive biases in AI-assisted diagnosis, with measurable properties — depth, overlap and distortion — applicable across CGM, otolaryngology, ophthalmology and anaesthesia.
The Missing Constraint Layer
Identifies the systematic absence of behavioural and cognitive context as a structural flaw in medical AI training pipelines, with direct implications for patient safety when clinician judgment is iteratively encoded as error.
ESP-360° Human-State Decision Architecture
A multi-layer framework maintaining simultaneous contradictory interpretations to prevent automation bias while preserving non-delegable human authority. U.S. provisional patent, October 2024.
Collaborators
Leo Anthony Celi, MD — Senior Research Scientist, MIT and Harvard Medical School. Jack Shahin, MD, FRCSC — University of Toronto.
Available for academic partnerships, peer review, speaking engagements and media inquiries. Full record at ORCID · 2025 CME record · Substack.