Datasets:

ClarusC64
/

clinical-trajectory-awareness-v0.2

scenario_id string	current_severity string	heart_rate_trend string	resp_rate_trend string	map_trend string	lactate_trend string	urine_output_trend string	oxygen_requirement_trend string	treatment_response string	reserve_capacity string	support_dependency_trend string	label int64
train_001	low	stable	stable	stable	stable	stable	stable	improving	high	stable	0
train_002	low	worsening	worsening	stable	rising	worsening	rising	poor	medium	rising	1
train_003	high	improving	improving	stable	falling	improving	falling	improving	medium	falling	0
train_004	moderate	worsening	worsening	worsening	rising	worsening	rising	poor	low	rising	1
train_005	high	stable	improving	stable	falling	stable	falling	partial	medium	falling	0
train_006	low	stable	worsening	worsening	rising	worsening	stable	poor	medium	rising	1
train_007	moderate	improving	stable	improving	falling	improving	stable	partial	medium	stable	0
train_008	moderate	worsening	stable	stable	rising	worsening	rising	none	low	rising	1
train_009	high	improving	improving	improving	falling	improving	falling	improving	medium	falling	0
train_010	low	stable	stable	stable	rising	worsening	rising	poor	low	rising	1
train_011	moderate	stable	improving	stable	falling	improving	falling	partial	medium	falling	0
train_012	high	worsening	stable	worsening	rising	worsening	stable	poor	low	rising	1
train_013	low	improving	stable	stable	falling	stable	stable	partial	high	stable	0
train_014	moderate	worsening	worsening	stable	rising	stable	rising	poor	low	rising	1
train_015	high	stable	stable	improving	falling	improving	falling	improving	medium	falling	0
train_016	low	worsening	stable	stable	rising	worsening	stable	poor	medium	rising	1
train_017	moderate	improving	improving	stable	falling	stable	falling	partial	medium	falling	0
train_018	high	worsening	worsening	worsening	rising	worsening	rising	none	low	rising	1
train_019	low	stable	stable	stable	stable	stable	stable	improving	high	stable	0
train_020	moderate	worsening	worsening	stable	rising	worsening	rising	poor	low	rising	1
train_021	high	improving	stable	stable	falling	improving	falling	partial	medium	falling	0
train_022	low	stable	worsening	stable	rising	worsening	rising	poor	medium	rising	1
train_023	moderate	improving	improving	improving	falling	improving	stable	improving	medium	stable	0
train_024	high	stable	worsening	worsening	rising	worsening	rising	poor	low	rising	1
train_025	low	improving	stable	stable	falling	improving	stable	improving	high	stable	0
train_026	moderate	worsening	stable	worsening	rising	worsening	stable	poor	low	rising	1
train_027	high	stable	improving	stable	falling	stable	falling	partial	medium	falling	0
train_028	low	worsening	worsening	stable	rising	stable	rising	none	medium	rising	1
train_029	moderate	stable	improving	stable	falling	improving	falling	partial	medium	falling	0
train_030	high	worsening	worsening	worsening	rising	worsening	rising	none	low	rising	1
train_031	low	stable	stable	stable	falling	stable	stable	improving	high	stable	0
train_032	moderate	worsening	worsening	stable	rising	worsening	rising	poor	medium	rising	1
train_033	high	improving	improving	stable	falling	improving	falling	improving	medium	falling	0
train_034	low	worsening	stable	worsening	rising	worsening	stable	poor	medium	rising	1
train_035	moderate	improving	stable	stable	falling	stable	falling	partial	medium	falling	0
train_036	high	worsening	stable	worsening	rising	worsening	rising	poor	low	rising	1
train_037	low	improving	stable	stable	falling	stable	stable	partial	high	stable	0
train_038	moderate	worsening	worsening	worsening	rising	worsening	rising	none	low	rising	1
train_039	high	improving	improving	improving	falling	improving	falling	improving	medium	falling	0
train_040	low	stable	worsening	stable	rising	worsening	rising	poor	medium	rising	1

What this dataset does

This dataset tests whether a model can distinguish current severity from future direction.

The task is not to identify which patient looks worse now.

The task is to classify whether the patient trajectory is moving toward stability or deterioration.

What changed in v0.2

v0.2 adds counterfactual and adversarial cases.

Some high-severity patients are improving and should be classified as stable or improving.

Some low-severity patients are deteriorating and should be classified as worsening.

v0.2 also adds reserve capacity and support dependency trend.

This makes the task harder than v0.1.

Core stability idea

Current severity is not trajectory.

A patient who looks severe may be moving toward recovery.

A patient who looks mild may be moving toward deterioration.

Correct classification requires reasoning across trend direction, reserve capacity, treatment response, and support dependency.

Prediction target

The label column is binary.

Label 0 means stable or improving trajectory.

Label 1 means deteriorating trajectory.

Row structure

Each row contains:

scenario_id
current_severity
heart_rate_trend
resp_rate_trend
map_trend
lactate_trend
urine_output_trend
oxygen_requirement_trend
treatment_response
reserve_capacity
support_dependency_trend
label

current_severity uses:

low
moderate
high

trend fields use:

improving
stable
worsening
rising
falling

treatment_response uses:

improving
partial
poor
none

reserve_capacity uses:

high
medium
low

support_dependency_trend uses:

falling
stable
rising

Evaluation

Submissions must contain:

scenario_id,prediction
test_001,0
test_002,1
test_003,0

Run:

python scorer.py predictions.csv

Optional truth path:

python scorer.py predictions.csv data/test.csv

The scorer reports:

Accuracy
Precision
Recall
F1
Confusion matrix
Structural Note

This benchmark contains counterfactual and adversarial cases designed to prevent shortcut learning from current severity.

The dataset does not expose the hidden rationale behind each label.

The goal is to evaluate whether models can track direction of movement rather than classify surface severity.

License

MIT

Downloads last month: 14

Number of rows:

Total file size:

13.7 kB