CMH SCIENTIFIC · DEFERRA · ANAESTHESIOLOGY

Deferra

Decides when an AI system should assist, hold back, or say nothing during a procedure — and when the measurement is inadequate, tells the clinician what to correct rather than simply going quiet.
RETROSPECTIVE VALIDATION · REAL DATA
What should happen now?
DEFER
Signal is adequate. The course and response relation are both coherent, so there is no reason to interrupt.
Can the system trust the signal?YES
Has the course moved?NO
Has the response changed?NO
Enough evidence?NO · 0/4
Interactive physiologic stream
Accumulated observations
Default state is silence when no supported change requires attention.
Shared information state
MeasurementSUFFICIENTCourseCOHERENTResponse relationCOHERENTStructural memoryUNCHALLENGEDAuthorityCLINICIANAutonomous controlNOT AUTHORIZEDDosing recommendationNOT AUTHORIZED
What changed
CMH Scientific · Deferra anaesthesiology demonstrationIllustrative browser stream · real-data retrospective validation · no dosing or treatment recommendation

Validation & data provenance

The anaesthesiology interaction is illustrative; the intervention-aware routing and physiologic-departure evidence come from real critical-care data.

MIMIC-IV norepinephrine / MAP routing4,473 analyzed routing states · 224 higher-departure and 4,249 stable/lower-departure states · clustered comparison across 337 ICU stays
PhysioNet Sepsis90,626 acute physiologic states · 27,984 high-surprise states in the departure analysis
Result boundaryIn MIMIC-IV departure states, the augmented signal reached 46.9% directional accuracy vs 20.5% clinician and 19.6% standard model; retrospective only
Boundary: no dose is recommended and no autonomous control is performed. The demo shows when the current relation is supported, departed, structurally changed, or requires more measurement.
Dataset / validation references
  1. Johnson AEW, Bulgarelli L, Shen L, et al. MIMIC-IV, a freely accessible electronic health record dataset. Sci Data. 2023;10:1. doi:10.1038/s41597-022-01899-x.
  2. Reyna MA, Josef CS, Jeter R, et al. Early Prediction of Sepsis From Clinical Data: The PhysioNet/Computing in Cardiology Challenge 2019. Crit Care Med. 2020;48(2):210–217. doi:10.1097/CCM.0000000000004145; dataset doi:10.13026/v64v-d857.
  3. Ewing MA. Intelligence Preserves the System It Assists. Research Square. 2026. doi:10.21203/rs.3.rs-10831345/v1. Direct intervention-aware routing validation, including MIMIC-IV MAP/norepinephrine control.
  4. Ewing MA. Intelligence Breaks the System It Controls. Research Square. 2026. doi:10.21203/rs.3.rs-10283184/v2. Precursor experiment motivating Deferra through vasopressor-associated trajectory deformation.