AnestheSIM

An open, extensible patient simulator for closed-loop anesthesia research and teaching, driven by VitalDB records (MATLAB/Simulink).

AnestheSIM lets real intraoperative records, or designed dosing schedules, drive interchangeable pharmacokinetic/pharmacodynamic (PK/PD) models of propofol, remifentanil, norepinephrine and inhaled agents, and compares the simulated BIS, MAP and HR with what was measured. Clinical-data acquisition is a native layer of the tool. Signals are requested by canonical name and device, and a thin mapping translates them to repository channels. The same simulation core therefore runs unchanged on VitalDB, on synthetic patients, or on any other database once its loader exists.

• Database abstraction. TrackName, TrackSource, VitalSignal and DatabaseType separate physiological quantities from device- and repository-specific channel names.
• VitalDB acquisition. Cases stream from the VitalDB API with local caching. A cohort can be searched from a signal specification.
• Drugs. Propofol, remifentanil and norepinephrine enter in pump units (mL/h). Sevoflurane and desflurane enter as end-tidal vol%, with automatic agent resolution from the Primus and Solar8000 gas analysers.
• PK. Schnider (propofol), Minto (remifentanil), Joachim covariate model (norepinephrine), and first-order volatile effect sites with Mapleson age adjustment.
• PD. Additive Minto, Bouillon and reduced Greco BIS surfaces run in parallel. A reduced Greco model covers intravenous–inhalational interaction. A turnover MAP/HR model and a unified propofol–remifentanil–norepinephrine MAP model complete the set.
• Validation on real data. One script re-derives a VitalDB cohort from its signal specification, replays every case through the simulator and reports Varvel performance metrics for BIS, MAP and HR.
• Teaching. Synthetic patients and designed dosing schedules run without internet access; every model is a short, documented MATLAB function next to its Simulink block.
• Speed. A 3-hour case simulates in under a second on a laptop.
• Alarms. Excursions beyond configurable safety limits are detected.

Requirements

• MATLAB and Simulink, tested with R2025b. On other releases setup_anesthesim regenerates the Simulink model from simulator/build_anesthesim_model.m.
• Statistics and Machine Learning Toolbox, only for the statistical tests of analyze_validation
• Internet access for the first download of each VitalDB case (cached in ./cache afterwards)

Quick start

setup_anesthesim      % paths; regenerates the Simulink model if needed
main                  % simulate VitalDB case 2809 end to end
run_tests             % unit tests, no internet needed

Simulating a case in a few lines:

signals = vitaldbDefaultSignals('balanced');           % BIS, MAP, HR, propofol, remifentanil, gases
[patient, data] = loadPatientCase(DatabaseType.VitalDB, 2809, signals);
simIn = buildSimulationInputs(data);                    % common 1 s clock
res = runAnesthesim(patient, simIn, 'Data', data);      % Simulink patient model
plotSimulation(res, simIn);

res contains BIS.minto, BIS.bouillon, BIS.greco (intravenous agents only), BIS.volatile, MAP.separable, MAP.unified, HR and all model concentrations, on the time grid res.t.

To test a controller, pass its commanded pump rates as data.Propofol.infRate (and so on) on a synthetic patient. See examples/example_synthetic_patient.m.

Examples

Script	Content
main.m	One VitalDB case: acquisition, simulation, metrics, alarms
example_cohort_search.m	Cohorts defined by a signal specification
example_bis_models.m	Minto / Bouillon / reduced Greco BIS on recorded infusions
example_balanced_case.m	Balanced anesthesia: BIS with and without the inhaled agents
example_synthetic_patient.m	Synthetic patient and designed dosing, no internet needed
study_validation.m	Replays a VitalDB cohort through the simulator (resumable)
analyze_validation.m	Validation tables, statistics and figures
study_agent_resolution.m	Blind check of the sevoflurane/desflurane resolution
reproduce_paper_figures.m	All results of the accompanying paper, written to ./results

Architecture

 Database abstraction        TrackName · TrackSource · VitalSignal · DatabaseType      (database aware)
          │ canonical signal requests
 Acquisition & preprocessing loadPatientCase → vitaldbLoadCase → patientData
          │ Patient + canonical patientData        buildSimulationInputs (common clock)
 ─────────┼──────────────────────────── database boundary ─────────────────────────────────
 Patient & models            Patient · SchniderModel · MintoModel · JoachimModel · VolatileModel
          │ state-space PK, PD parameter structures (bisParameters, mapUnifiedParameters, ...)
 Simulation & output         simulator/anesthesim.slx · runAnesthesim · detectAlarms · plotting

src/
  database/        TrackName, TrackSource, VitalSignal, DatabaseType
    vitaldb/       VitalDB loader, cached downloads, volatile-agent resolution, searchPatients
    population/    synthetic patients (data/population)
  acquisition/     loadPatientCase, canonical patientData
  patient/         Patient (immutable demographics, BMI, James lean body mass)
  models/pk/       Schnider, Minto, Joachim, volatile effect sites
  models/pd/       BIS interaction models and parameter sets
  models/hemodynamics/  turnover MAP/HR, unified MAP, norepinephrine effect
  simulation/      model instantiation, inputs on the simulation clock, runAnesthesim
  validation/      BIS cleaning, Varvel performance metrics
  monitoring/      alarm limits and detection
  plotting/
simulator/         build_anesthesim_model.m and the generated anesthesim.slx
examples/   tests/   data/

The PD equations live in one place, src/models: the Simulink MATLAB Function blocks call the same functions that are available at the command line, and simulateLTI integrates the same state-space matrices as the Simulink State-Space blocks. The test suite checks that both agree.

Pharmacological models

Block	Model	Input → output
Propofol PK	Schnider 1998, ke0 = 1.21 min⁻¹ (Orchestra/Marsh)	mL/h → mg/s → Cp, Ce [µg/mL]
Remifentanil PK	Minto 1997, age-dependent ke0	mL/h → µg/s → Cp, Ce [ng/mL]
Norepinephrine PK/PD	Joachim 2024 covariate model, transport lag, Emax or Windkessel	mL/h → nmol/s → C [nmol/L]
Volatile effect site	first order, ke0 = 0.47 min⁻¹; C50 = C50,40 · 10^(−0.00269 (age−40))	Et [vol%] → Ce [vol%]
BIS (IV)	additive Minto, Bouillon 2004, reduced Greco (in parallel)	Ce → BIS
BIS (IV + volatiles)	U = (U_prop + U_sevo + U_des)(1 + U_remi), BIS = E0 − Emax·U^γ/(1+U^γ)	Ce → BIS
MAP / HR, separable	Garraza-Obaldia 2024 turnover model + Joachim norepinephrine pressor effect	Cp, C_NE → MAP, HR
MAP, unified	C50P,eff = C50P (1 − kR·CeR/(CeR + C50R)); MAP = MAP0 − E_P + E_NE	Ce, C_NE → MAP

In every BIS model Emax is the maximal BIS decrease, so BIS tends to E0 − Emax. All models use published parameters, except the unified MAP structure, whose propofol/remifentanil values are illustrative and meant to be replaced by the user's own.

Extending AnestheSIM

• Add a database. Write a loader that returns a Patient and the canonical structure of emptyPatientData, add a member to DatabaseType, and add a case to loadPatientCase. Nothing in the models or the Simulink core changes.
• Add a drug or signal. Add a TrackName (and a TrackSource for a new device), extend VitalSignal.channelStrings, and map the channel in vitaldbToCanonical. Then add its PK/PD block in build_anesthesim_model. The volatile-agent channels are a worked example.
• Add a PK/PD model. Provide a state-space constructor (PK) or a function plus parameter structure (PD), wire it in build_anesthesim_model, and pass its parameters in runAnesthesim.

Data

AnestheSIM uses only the public, de-identified records of the VitalDB open dataset. Records are downloaded on demand and are not redistributed with this repository. Use of VitalDB data is governed by the VitalDB data-use terms. Please cite: Lee H.-C. et al., VitalDB, a high-fidelity multi-parameter vital signs database in surgical patients, Scientific Data 9, 279 (2022).

data/population contains synthetic demographic records only.

Citation

If you use AnestheSIM, please cite the software:

Birs, I. (2026). AnestheSIM: an open, extensible patient simulator for closed-loop anesthesia driven by VitalDB records. Zenodo. https://doi.org/10.5281/zenodo.23338918

This DOI always resolves to the latest release. The DOI of each release is listed on Zenodo, and the metadata are in CITATION.cff.

License

MIT © 2026 Isabela Birs