Checklist ROAI 2026 County Stage, Grades 11–12 · Task 1
The Good Nurse vs. Andrew Anarchy
Predict 28-day mortality of mechanically ventilated ICU patients when most training labels have been deleted.
The task
A hospital studies the effect of indwelling arterial catheters on 28-day mortality in patients mechanically ventilated on their first ICU day. A saboteur replaced most values of the target day_28_flg in the training set with NaN, leaving only a small labelled subset alongside many unlabelled records.
Subtask 1 asks for the number of missing labels. Subtask 2 asks for the mortality percentage among labelled patients with sofa_first > 10 and with sofa_first ≤ 10. Subtask 3 asks for a probability of death for each test patient, using the labelled and unlabelled training rows.
Abridged by SOTA from the official materials. The official statement has the exact rules, and it wins wherever this summary differs.
At a glance
- You get
train.csv(day-1 demographics, vital signs, laboratory values, comorbidity flags;day_28_flgmostly NaN) andtest.csv.- You submit
- CSV with subtaskID, datapointID and answer: one row for subtask 1, two rows for subtask 2 (datapointID 1:
sofa_first> 10; 2: ≤ 10), one probability per test patient for subtask 3. - Scoring
- Subtask 1 (10): exact count. Subtask 2 (20): both percentages correct. Subtask 3 (70): ROC-AUC; full marks for AUC strictly above 0.91, partial credit scaled towards that threshold.
- Rules
- Only the provided
train.csvmay be used to build the model - On-site at partner county centres, in the prepared local Jupyter environment
- Only the packages listed in the round rules may be used
- Mobile phones and other electronic or information sources are forbidden in the room
- Only the provided
- Format
- ROAI 2026 county stage, grades 11–12, 8 March 2026 (09:10–12:40 UTC on the judge; 3-hour test).