scenario_id int64 | infection_pressure float64 | physiological_buffer float64 | intervention_delay float64 | organ_coupling float64 | microcirculatory_instability float64 | label_sepsis_cascade int64 |
|---|---|---|---|---|---|---|
1 | 0.85 | 0.27 | 0.67 | 0.72 | 0.76 | 1 |
2 | 0.24 | 0.86 | 0.1 | 0.21 | 0.18 | 0 |
3 | 0.73 | 0.4 | 0.59 | 0.64 | 0.68 | 1 |
4 | 0.31 | 0.79 | 0.16 | 0.27 | 0.24 | 0 |
5 | 0.81 | 0.33 | 0.63 | 0.69 | 0.72 | 1 |
6 | 0.2 | 0.88 | 0.09 | 0.18 | 0.16 | 0 |
7 | 0.7 | 0.45 | 0.56 | 0.61 | 0.65 | 1 |
8 | 0.27 | 0.82 | 0.13 | 0.24 | 0.21 | 0 |
9 | 0.77 | 0.36 | 0.6 | 0.66 | 0.7 | 1 |
10 | 0.22 | 0.85 | 0.1 | 0.2 | 0.17 | 0 |
What this repo does
This dataset models the transition from controlled infection to septic cascade using a five-variable interaction structure.
The goal is to detect when a patient is drifting toward septic boundary failure before overt systemic collapse is fully established.
Sepsis often unfolds as a cascade: infection pressure rises, physiological reserve narrows, intervention delays reduce reversibility, organ interactions amplify instability, and microcirculatory dysfunction reduces tissue-level recoverability.
The five-node structure extends the core Clarus clinical logic beyond a quad and captures a higher-order cascade surface.
Core five-node structure
infection_pressure
physiological_buffer
intervention_delay
organ_coupling
microcirculatory_instability
infection_pressure reflects pathogen burden and inflammatory stress.
physiological_buffer reflects reserve capacity and resilience against septic deterioration.
intervention_delay captures lag in antibiotics, fluids, source control, imaging, or escalation.
organ_coupling reflects how infection-driven dysfunction interacts with circulatory, renal, respiratory, and metabolic instability.
microcirculatory_instability reflects tissue perfusion disruption and failure of capillary-level compensation.
Clinical Variable Mapping
| Node | Clinical Measurements | Typical Risk Signals |
|---|---|---|
| infection_pressure | lactate, procalcitonin, WBC, temperature, CRP | rising lactate, high procalcitonin, fever, inflammatory burden |
| physiological_buffer | age, albumin, baseline reserve, frailty, comorbidity burden | low albumin, frailty, limited reserve |
| intervention_delay | delay in antibiotics, fluids, source control, escalation | antibiotics delayed, delayed source control |
| organ_coupling | SOFA interaction, urine output, oxygen requirement, hypotension | infection spilling into multi-organ instability |
| microcirculatory_instability | capillary refill, MAP, mottling, peripheral perfusion, lactate clearance | delayed refill, mottling, poor lactate clearance, unstable perfusion |
Prediction target
label_sepsis_cascade
Binary classification.
0 = infectious state remains stable or recoverable
1 = sepsis cascade boundary is approaching
Binary simplification note
The Cascade Transition framework supports a full five-stage trajectory:
0 stable regime
1 deterioration drift
2 near cascade boundary
3 active cascade propagation
4 recovery trajectory
This v0.1 dataset intentionally uses a binary formulation.
Binary classification is easier to validate clinically and aligns with how sepsis monitoring and escalation systems are used in practice. Operational systems usually require a clear alert condition rather than a multi-stage taxonomy.
The full five-stage structure remains part of the broader framework and may appear in future dataset versions.
Why five nodes here
Most of the clinical suite uses quad structure because quad coupling is easier to validate, explain, and deploy.
This repo is a deliberate flagship extension.
The additional fifth node captures a clinically important layer that is often decisive in sepsis progression: tissue-level perfusion breakdown. That makes this dataset suitable as an advanced boundary set inside the wider clinical suite.
Row structure
Each row represents a simulated infectious patient state.
Columns:
scenario_id
infection_pressure
physiological_buffer
intervention_delay
organ_coupling
microcirculatory_instability
label_sepsis_cascade
Values are normalized between 0 and 1 for training simplicity.
Higher infection_pressure increases risk.
Lower physiological_buffer increases risk.
Higher intervention_delay, higher organ_coupling, and higher microcirculatory_instability increase risk.
Files
data/train.csv
data/tester.csv
scorer.py
train.csv contains labeled rows.
tester.csv contains unlabeled rows with the same schema except for the target label.
scorer.py evaluates binary classification performance.
Evaluation
The scorer computes:
accuracy
precision
recall_cascade_detection
false_safe_rate
f1
confusion matrix
The primary metric is recall_cascade_detection because the main task is to detect approaching septic boundary failure rather than simply optimize overall accuracy.
false_safe_rate captures the proportion of positive danger cases missed by the model.
License
MIT
Structural Note
This dataset is part of the Clarus Cascade Transition Dataset family.
These datasets model how complex systems move from stable regimes into cascading failure states.
In infection medicine this corresponds to the transition from controlled infectious stress into systemic septic deterioration.
This five-node dataset functions as an advanced boundary set within the clinical suite.
Production Deployment
Potential applications include:
sepsis boundary detection
ICU infection monitoring
advanced clinical deterioration modeling
decision support for unstable infectious patients
research prototypes for higher-order cascade detection
Enterprise & Research Collaboration
Clarus datasets explore stability boundaries in complex systems including clinical deterioration, infrastructure failure, financial contagion, and AI system stability.
- Downloads last month
- 80