ClarusC64/clinical-sepsis-trajectory-instability-v0.1

clinical-sepsis-trajectory-instability-v0.1

What this dataset does

This dataset tests whether a model can classify sepsis trajectory instability from short clinical proxy sequences.

Each row describes a patient-like scenario across three time points.

The task is to predict whether the scenario is moving toward instability or remaining stable.

Core stability idea

Sepsis instability does not depend on one variable alone.

A patient may show an abnormal value and still recover.

Another patient may show only moderate abnormalities but deteriorate when trends combine with weak response and delayed intervention.

The dataset tests interaction reasoning across:

• —temperature trajectory
• —blood pressure trajectory
• —lactate trajectory
• —inflammatory burden
• —respiratory strain
• —fluid response
• —intervention delay
• —renal stress

Prediction target

label = 1 means sepsis trajectory instability.

label = 0 means stable or recovering trajectory.

Row structure

Each row includes:

• —scenario_id
• —temperature values across three time points
• —MAP values across three time points
• —lactate values across three time points
• —WBC values across three time points
• —respiratory-rate values across three time points
• —fluid response
• —antibiotic delay
• —renal marker
• —decoy variables
• —label

Decoy variables:

• —inflammation_noise
• —documentation_noise

These appear meaningful but do not define the target alone.

Evaluation

Predictions must use this format:

scenario_id,prediction
S101,0
S102,1

Run:

python scorer.py --predictions predictions.csv --truth data/test.csv --output metrics.json

The scorer returns:

accuracy
precision
recall
f1
confusion matrix
dataset integrity checks
Structural Note

This dataset reflects latent stability geometry through observable proxies.

The generator and latent rule structure are not included.

This dataset is part of the ClarusC64 stability-reasoning benchmark family. Datasets share a latent stability geometry but expose only observable proxy variables.

Production Deployment

This dataset is intended as a compact benchmark for evaluating reasoning over unstable trajectories.

It is not a clinical decision tool.

Enterprise & Research Collaboration

This dataset is designed to invite collaboration around latent stability reasoning, cross-domain transfer, and interaction-based evaluation.

License

MIT