sepsisim 0.1.0

Gymnasium environments for reinforcement learning in sepsis management

sepsisim

Gymnasium environments for reinforcement learning in sepsis management

---

SepsiSim provides three Gymnasium-compatible reinforcement learning environments for sepsis management in the intensive care unit. Built on established physiological models (Reynolds et al. 2006 inflammation dynamics, cardiovascular fluid response, lactate kinetics), these environments enable researchers to develop and benchmark RL algorithms for critical care decision-making without risking patient safety.

Installation

pip install sepsisim              # Core (numpy, scipy, gymnasium)
pip install sepsisim[train]       # + SB3, PyTorch for RL training
pip install sepsisim[all]         # Everything

Development install:

git clone https://github.com/HassDhia/sepsisim.git
cd sepsisim
pip install -e ".[all]"

Quick Start

import gymnasium as gym
import sepsisim

env = gym.make("sepsisim/FluidResuscitation-v0")
obs, info = env.reset(seed=42)
for _ in range(100):
    action = env.action_space.sample()
    obs, reward, terminated, truncated, info = env.step(action)
    if terminated or truncated:
        obs, info = env.reset()
env.close()

Environments

Environment	Paradigm	Observation	Action	Key Challenge
FluidResuscitation-v0	Continuous	7D (MAP, lactate, UO, damage, volume, bacteria, SOFA)	1D (fluid mL)	Restore MAP while avoiding fluid overload
VasopressorTitration-v0	Continuous	8D (+ vaso dose, hours on vaso)	1D (dose change)	Maintain MAP without excessive vasoconstriction
SepsisManagement-v0	Multi-action	10D (+ abx status, time)	3D (fluid, vaso, abx)	Balance fluids, vasopressors, and antibiotic timing

Architecture

SepsiSim combines four physiological models into configurable Gymnasium environments:

• Inflammation Model (Reynolds et al. 2006): 4-ODE system tracking bacteria, pro-/anti-inflammatory response, and tissue damage
• Cardiovascular Model: Lumped-parameter hemodynamic response to fluids and vasopressors with Frank-Starling saturation
• Lactate Kinetics: Single-compartment model linking tissue hypoperfusion to lactate production and clearance
• SOFA Scoring: Simplified 3-component organ failure assessment (cardiovascular, renal, metabolic)

Each environment supports three difficulty tiers (easy/medium/hard) via initial bacterial load, enabling curriculum learning and systematic benchmarking.

Paper

The accompanying paper is available at:

• PDF (GitHub)

Citation

If you use sepsisim in your research, please cite:

@software{dhia2026sepsisim,
  author = {Dhia, Hass},
  title = {SepsiSim: Gymnasium Environments for Reinforcement Learning in Sepsis Management},
  year = {2026},
  publisher = {Smart Technology Investments Research Institute},
  url = {https://github.com/HassDhia/sepsisim}
}

License

MIT License. See LICENSE for details.

Contact

Hass Dhia -- Smart Technology Investments Research Institute

• Email: [e-mail]
• Web: smarttechinvest.com/research