Add/dka task

[jhnwu3]

This pull request introduces a new DKA (Diabetic Ketoacidosis) prediction task for the MIMIC-IV dataset, including its integration into the codebase, documentation, and example usage. It also adds comprehensive synthetic test resources for MIMIC-IV, supporting the new task and facilitating robust testing and benchmarking.

New DKA Prediction Task:

• Added the DKAPredictionMIMIC4 class for DKA prediction on MIMIC-IV, and integrated it into the pyhealth.tasks package.
• Documented the new task in the API docs and included it in the list of available tasks. [1] [2]

Examples and Benchmarks:

• Added a new example script dka_mimic4.py demonstrating how to use StageNet for DKA prediction with MIMIC-IV, including data loading, task application, model training, and evaluation.
• Introduced a benchmarking script benchmark_workers_12.py for evaluating MIMIC-IV mortality prediction performance with various metrics and memory usage tracking.

Test Resources:

• Added realistic synthetic MIMIC-IV demo data files (patients.csv, admissions.csv, diagnoses_icd.csv, labevents.csv, d_labitems.csv) to support testing and development of the new task. [1] [2] [3] [4] [5]

These changes collectively enable DKA prediction research on MIMIC-IV within the pyhealth framework and provide the necessary infrastructure for both development and robust evaluation.

[Copilot]

Pull request overview

This pull request introduces a new DKA (Diabetic Ketoacidosis) prediction task for the MIMIC-IV dataset. The PR adds two task classes (DKAPredictionMIMIC4 for general population and T1DDKAPredictionMIMIC4 for Type 1 Diabetes patients), comprehensive synthetic test data, example scripts, and documentation updates.

Key Changes:

• Added DKA prediction task implementation with temporal data leakage prevention
• Created synthetic MIMIC-IV demo dataset with realistic medical codes and lab values
• Provided example scripts demonstrating StageNet model training for DKA prediction

Critical Issues Identified:
The test file tests/core/test_mimic4_dka.py contains multiple critical bugs where it tests DKAPredictionMIMIC4 but uses attributes, methods, and parameters that only exist in T1DDKAPredictionMIMIC4. This suggests the test was written for the wrong class or copied from T1D tests without proper adaptation.

Reviewed changes

Copilot reviewed 17 out of 18 changed files in this pull request and generated 20 comments.

Show a summary per file

File	Description
pyhealth/tasks/dka.py	Implements DKAPredictionMIMIC4 and T1DDKAPredictionMIMIC4 classes with ICD code classification, lab feature extraction, and temporal filtering to prevent data leakage
pyhealth/tasks/init.py	Exports the new DKA prediction task classes
pyhealth/datasets/configs/mimic4_ehr.yaml	Adds hadm_id field to labevents table configuration for admission-level filtering
tests/core/test_mimic4_dka.py	Test suite for DKA prediction (contains critical bugs - tests wrong class attributes/methods)
tests/core/test_mimic4_los.py	Test suite for length of stay prediction task
test-resources/core/mimic4demo/hosp/*.csv	Synthetic MIMIC-IV demo data files (patients, admissions, diagnoses, procedures, prescriptions, lab events)
examples/clinical_tasks/dka_mimic4.py	Example script for general population DKA prediction using StageNet
examples/clinical_tasks/t1dka_mimic4.py	Example script for T1D-specific DKA prediction using StageNet
examples/benchmark_perf/benchmark_workers_12.py	Benchmarking script for mortality prediction (has num_workers documentation inconsistencies)
docs/api/tasks/pyhealth.tasks.dka.rst	API documentation for DKA task
docs/api/tasks.rst	Adds DKA task to the task list

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.