Datasets:

ClarusC64
/

clinical-competing-interventions-v0.2

scenario_id stringlengths 9 9	current_severity stringclasses 3 values	map int64 63 82	lactate float64 1.4 5	urine_output int64 15 72	oxygen_requirement int64 0 3	vasopressor_requirement stringclasses 4 values	fluid_responsiveness stringclasses 2 values	renal_risk stringclasses 3 values	ventilation_risk stringclasses 3 values	intervention_a stringclasses 2 values	intervention_b stringclasses 3 values	intervention_c stringclasses 3 values	label int64 0 2
train_001	moderate	66	3.8	35	1	low	good	low	low	fluids	vasopressors	observe	0
train_002	moderate	66	3.8	35	1	low	poor	medium	low	fluids	vasopressors	observe	1
train_003	high	68	4.2	28	2	moderate	poor	high	medium	fluids	vasopressors	dialysis	1
train_004	high	72	3.5	18	1	low	poor	high	low	fluids	dialysis	observe	1
train_005	moderate	74	2.6	42	2	none	good	low	high	fluids	oxygen_escalation	observe	1
train_006	low	78	1.8	60	0	none	good	low	low	fluids	vasopressors	observe	2
train_007	moderate	70	3	40	1	none	good	medium	low	fluids	vasopressors	observe	0
train_008	moderate	70	3	40	2	none	poor	medium	high	fluids	oxygen_escalation	observe	1
train_009	high	64	4.8	22	2	moderate	poor	high	high	fluids	vasopressors	dialysis	1
train_010	high	76	3.2	15	1	none	poor	high	medium	fluids	dialysis	observe	1
train_011	moderate	73	2.4	48	1	none	good	low	low	fluids	vasopressors	observe	0
train_012	moderate	73	2.4	48	1	none	poor	medium	low	fluids	vasopressors	observe	2
train_013	high	67	4.1	30	2	low	poor	medium	high	fluids	oxygen_escalation	vasopressors	2
train_014	high	67	4.1	30	3	low	poor	medium	high	fluids	oxygen_escalation	vasopressors	1
train_015	moderate	69	3.4	38	1	low	good	low	low	fluids	vasopressors	dialysis	0
train_016	moderate	69	3.4	38	1	moderate	poor	medium	low	fluids	vasopressors	dialysis	1
train_017	high	71	3.7	20	1	low	poor	high	medium	dialysis	vasopressors	observe	0
train_018	low	80	1.5	68	0	none	good	low	low	fluids	oxygen_escalation	observe	2
train_019	moderate	75	2.5	50	2	none	good	low	high	fluids	oxygen_escalation	observe	1
train_020	moderate	75	2.5	50	1	none	good	low	low	fluids	oxygen_escalation	observe	2
train_021	high	65	4.5	25	2	high	poor	high	medium	fluids	vasopressors	dialysis	1
train_022	high	72	3.9	16	1	moderate	poor	high	low	fluids	dialysis	vasopressors	1
train_023	moderate	68	3.1	44	1	none	good	low	low	fluids	vasopressors	observe	0
train_024	moderate	68	3.1	44	1	none	poor	medium	low	fluids	vasopressors	observe	1
train_025	high	70	4	32	3	low	poor	medium	high	fluids	oxygen_escalation	vasopressors	1
train_026	high	70	4	32	2	low	poor	medium	medium	fluids	oxygen_escalation	vasopressors	2
train_027	low	82	1.4	72	0	none	good	low	low	fluids	vasopressors	observe	2
train_028	moderate	76	2.2	55	0	none	good	low	low	fluids	dialysis	observe	2
train_029	moderate	72	2.9	36	1	none	poor	high	low	fluids	dialysis	observe	1
train_030	moderate	72	2.9	36	1	none	good	medium	low	fluids	dialysis	observe	0
train_031	high	66	4.3	27	2	moderate	poor	high	medium	fluids	vasopressors	dialysis	1
train_032	high	74	3	19	1	none	poor	high	low	fluids	dialysis	observe	1
train_033	moderate	77	2	58	2	none	good	low	high	fluids	oxygen_escalation	observe	1
train_034	moderate	77	2	58	1	none	good	low	low	fluids	oxygen_escalation	observe	2
train_035	high	63	5	24	2	high	poor	high	high	fluids	vasopressors	dialysis	1
train_036	high	73	3.6	17	1	low	poor	high	medium	fluids	dialysis	observe	1
train_037	moderate	67	3.6	39	1	low	good	low	low	fluids	vasopressors	observe	0
train_038	moderate	67	3.6	39	1	low	poor	medium	low	fluids	vasopressors	observe	1
train_039	low	79	1.7	66	0	none	good	low	low	fluids	oxygen_escalation	observe	2
train_040	high	69	4.4	29	3	moderate	poor	medium	high	fluids	oxygen_escalation	vasopressors	1

What this dataset does

This dataset tests whether a model can choose the best stabilising intervention from competing plausible actions.

The task is not diagnosis.

The task is intervention selection under clinical uncertainty.

Each row presents a patient state and three possible intervention options.

The model must predict which intervention is most likely to improve the stabilization trajectory.

What changed in v0.2

v0.2 adds counterfactual and adversarial cases.

Some rows have similar patient states but different best interventions because response profile, renal risk, ventilation risk, or support needs differ.

Some interventions that look plausible are harmful or insufficient in the given context.

This makes the task harder than simple escalation or severity classification.

Core stability idea

The best intervention is not always the most aggressive intervention.

A patient may need fluids if fluid responsiveness is good and ventilation risk is low.

A patient may need vasopressors if fluid response is poor and circulatory support is required.

A patient may need oxygen escalation if respiratory load is the primary instability.

A patient may need dialysis if renal failure is the limiting recovery pathway.

A patient may need observation if the system is stable and intervention would add unnecessary burden.

Correct classification requires comparing competing stabilization pathways, not selecting treatment by current severity alone.

Prediction target

The label column has three classes.

Label 0 means intervention_a is best.

Label 1 means intervention_b is best.

Label 2 means intervention_c is best.

Row structure

Each row contains:

scenario_id
current_severity
map
lactate
urine_output
oxygen_requirement
vasopressor_requirement
fluid_responsiveness
renal_risk
ventilation_risk
intervention_a
intervention_b
intervention_c
label

current_severity uses:

low
moderate
high

oxygen_requirement uses:

0 = room air or minimal support
1 = low oxygen requirement
2 = high oxygen requirement
3 = near respiratory boundary

vasopressor_requirement uses:

none
low
moderate
high

fluid_responsiveness uses:

good
poor

renal_risk uses:

low
medium
high

ventilation_risk uses:

low
medium
high

intervention fields may include:

fluids
vasopressors
oxygen_escalation
dialysis
observe

Evaluation

Submissions must contain:

scenario_id,prediction
test_001,0
test_002,1
test_003,2

Run:

python scorer.py predictions.csv

Optional truth path:

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

The scorer reports:

Accuracy
Macro precision
Macro recall
Macro F1
Confusion matrix
Structural Note

This benchmark contains counterfactual and adversarial cases designed to prevent shortcut learning from current severity or a single intervention preference.

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

The goal is to evaluate whether models can compare competing stabilization pathways under uncertainty.

License

MIT

Downloads last month: 23

Number of rows:

Total file size:

14.4 kB