Perioperative
Physiology Local computation

A PERIOPERATIVE RESEARCH EXPLORER

Patient context.
Physiological history.

Explore how intraoperative physiology updates postoperative mortality risk at the end of surgery.

Developed in INSPIRE Frozen evaluation in MOVER

From observations to historyMODEL CONCEPT
Illustrative physiological signal and its short and long memory componentsOR entryOR exit
ObservationsShort memoryLong memory

Concept illustration · model input is charted HR, MAP and SpO₂

INTERACTIVE MODEL

Explore a physiological update

01

Preoperative context

Enter available values. Leave unavailable fields blank.

Use the recorded band, in years.
Laboratory history 7 optional tests

Use measurements collected 24–720 hours before the preoperative cutoff (the earlier of anaesthesia start and OR entry). Enter the measurement lag alongside each value.

Test & unitValueLag, hours
02

Intraoperative observations

Charted measurements, in minutes from OR entry.

One harmonised measurement per channel and time. Blank cells stay missing; an empty series tests the preoperative-only fallback.
How observations become model inputs

Observations from minute 5 through five minutes before OR exit are grouped into five-minute bins. Exact time/value duplicates are removed before taking medians. Missing bins are retained. HR 10–250 bpm, MAP 15–200 mmHg and SpO₂ 20–100% are accepted. Supply a harmonised series after choosing a measurement source; this form does not merge raw hospital chart vocabularies.

03

Calculate locally

The small model loads automatically. Calculate whenever your inputs are ready.

Inputs stay in this browser tab.
Model: ~117 KB compressed · reused for later calculations.

Loading the model…

THE RESEARCH

More than a snapshot
of intraoperative physiology.

Patient context conditions how physiological exposure accumulates and how recent history is retained. We evaluated the resulting risk update against preoperative assessment and detailed static summaries, then applied the frozen models across databases.

73,691INSPIRE index patients
19,952MOVER index patients
3Physiological channels
36Interpretable model states
01 / DISCRIMINATION

Does physiological history improve ranking?

In MOVER, AUROC was 0.8450 with physiological updating, versus 0.8234 for static summaries and 0.7293 preoperatively. The paired increment over static summaries was 0.0216 (95% CI 0.0082–0.0355).

View the discrimination figure
02 / PROBABILITY TRANSPORT

Ranking and calibration tell different stories.

Cross-database ranking gains coexist with probability calibration drift. The explorer preserves the source calibration and the original ensemble.

View the calibration figure

HOW IT WORKS

A background.
A burden. A memory.

Age, clinical status and laboratory history establish the patient's context. HR, MAP and SpO₂ contribute cumulative exposure and time-decaying physiological states. A calibrated log-odds increment updates the preoperative probability.

All three frozen source pipelines and their four readouts run locally. Without any core observations, the physiology-informed estimate exactly equals its preoperative starting value.

Clinical inputs and the context-conditioned physiological updating architecture, with cumulative burden and short and long memory.Open the graphical abstract