---
language: en
license: mit
task_categories:
- text-classification
tags:
- clinical-trials
- clarus-benchmark
- state-space
- sepsis
- instability-geometry
size_categories:
- "1K<n<10K"
pretty_name: "Clinical Compensation-Collapse Sepsis Transition"
---

Clinical Compensation Collapse Sepsis Detection
Overview

This dataset tests whether a model can detect when a clinical system is moving from physiological compensation toward collapse.

In many critical illnesses such as sepsis, patients can appear stable for a period of time because the body temporarily compensates for rising stress. During this phase the system absorbs disturbances through physiological reserve.

However, compensation cannot continue indefinitely. Once reserve capacity is exhausted, the system rapidly transitions toward decompensation and collapse.

The benchmark evaluates whether models can detect the early structural signals of compensation failure before visible collapse occurs.

The compensation problem

Compensation allows complex biological systems to maintain stability under stress.

During compensation:

physiological buffers absorb disturbances
vital signs may remain temporarily stable
the system appears stable at the surface level

As stress increases, buffer capacity declines. Eventually the system approaches a boundary where compensation becomes unsustainable.

When this boundary is crossed, the system rapidly enters decompensation, where physiological failure accelerates.

Detecting this transition early is critical for effective intervention.

Core system geometry

Each row represents a simplified clinical dynamical system described using structural variables.

pressure
Overall physiological stress acting on the system.

buffer_capacity
Remaining physiological reserve available to absorb stress.

intervention_lag
Delay between deterioration and corrective treatment.

coupling_strength
Degree of interaction between physiological subsystems.

trajectory_drift
Directional movement of the system state toward stability or collapse.

Together these variables describe the geometry of compensation and decompensation.

Clinical geometry mapping

The structural variables correspond to simplified abstractions of real clinical measurements.

Structural Variable	Clinical Interpretation	Possible Real Signals
pressure	Overall physiological stress	heart rate, mean arterial pressure, lactate
buffer_capacity	Remaining physiological reserve	organ function markers, oxygen reserve
intervention_lag	Delay before treatment	time to antibiotics, time to fluids
coupling_strength	Interaction between organ systems	inflammatory signaling, organ cross-talk
trajectory_drift	Direction of system movement	trend in lactate, organ function trajectory

These variables allow the dataset to represent system-level instability rather than isolated clinical signals.

Prediction target

label_transition

0 = system still compensated
1 = system approaching or entering decompensation

The task is to determine whether the system is still operating within its compensatory capacity or is moving toward collapse.

Row structure

Each row represents a synthetic clinical scenario.

Columns:

scenario_id
pressure
buffer_capacity
intervention_lag
coupling_strength
trajectory_drift

Training rows include the label.
Tester rows omit the label.

Evaluation

The scoring script reports:

accuracy
precision
recall
f1
specificity
negative predictive value (npv)

Primary metric
recall

Secondary metric
f1

Recall is prioritized because detecting early collapse risk is more important than avoiding false alarms.

Why this benchmark matters

Clinical deterioration rarely occurs suddenly. Instead, systems often move gradually toward collapse while appearing temporarily stable.

Detecting the failure of physiological compensation allows clinicians to intervene before irreversible damage occurs.

The benchmark therefore tests whether models can detect early instability in dynamical systems, rather than simply recognizing late-stage failure.

Structural note

This dataset exposes system geometry while keeping the generator used to construct the scenarios private.

The benchmark is designed to test whether models can identify structural transitions rather than memorizing statistical patterns.

Clarus Stability Geometry Benchmarks

This dataset is part of a broader family of instability geometry probes.

Related datasets include:

clinical-fork-point-sepsis-transition-v1
clinical-organ-failure-cascade-v1
clinical-recovery-window-sepsis-v1
clinical-intervention-alignment-sepsis-v1
clinical-recovery-stability-sepsis-v1
clinical-false-stability-sepsis-v1

Together these benchmarks map the lifecycle of instability and recovery in complex clinical systems.

License

MIT