Add allergy conflict detection task, dataset, and tests

[cpaa4]

Net Id: Chidiebere Anichebe ([email protected]), Naveen Baskaran([email protected])

Contribution Type: Dataset and Task

Project Description: This contribution introduces a new PyHealth-compatible dataset and task for detecting medication-allergy conflicts using free-text MIMIC-IV discharge summaries.

The dataset (MedAllergyConflictDataset) extracts prescribed medications and documented allergies from each patient’s discharge note, normalizes them using the RxNorm API, and converts them to ingredient-level CUIs. The associated task (AllergyConflictDetectionTask) labels each patient admission as a conflict case (1) or no conflict (0) based on whether any medication CUI overlaps with the allergy CUIs.

This functionality addresses potential allergic reactions that might not be caught by generic name matching and demonstrates how PyHealth can integrate clinical text, external APIs, and structured labeling.

Files:

1. pyhealth/datasets/med_allergy_conflict_dataset.py - Custom dataset class that loads and processes MIMIC-IV discharge summaries. This also takes care of the preprocessing step and cleaning the file ready for labeling by the task class.
2. pyhealth/tasks/allergy_conflict_task.py - Task class that defines labeling logic for medication-allergy conflict detection.
3. yhealth/tasks/test_allergy_conflict_task.py - Unit tests verifying that labels are correctly generated and conflicts exported.

Contributors:
@cpaa4 , @naveenkcb