pyhealth

Build clinical ML pipelines with datasets, tasks, models, and trainers.

22|4|Updated May 25, 2026

Search Tags:#data-pipelines #ehr #mimic #research-tools #healthcare-ai #clinical-ml #medical-coding

Ready to run

npx skills add https://github.com/crazymsn/academic-skills --skill pyhealth-crazymsn

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: pyhealth
Source: https://github.com/crazymsn/academic-skills/tree/main/academic-skills/pyhealth
Command: npx skills add https://github.com/crazymsn/academic-skills --skill pyhealth-crazymsn

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes assets (resource) and references (resource) components.

What problem does it solve?

PyHealth provides a unified, modular toolkit to build end-to-end clinical deep-learning pipelines, from raw EHR/clinical data to trained models and evaluation metrics.

Core Features & Use Cases

Modular components: dataset/task/model/trainer/metrics unify clinical workflows and reduce boilerplate.
Scenario coverage: supports MIMIC-III/IV, eICU, OMOP, SleepEDF, ChestXray14, and EHRShot for mortality, readmission, LOS, drug recommendation, sleep staging, and medical coding tasks.
Real-world workflow: quickly prototype end-to-end pipelines that map datasets to tasks, train models, and compute clinically relevant metrics.

Quick Start

Run the starter pipeline to see an end-to-end MortalityPredictionMIMIC3 workflow from dataset to evaluation.

Frequently Asked Questions about pyhealth

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I build a clinical ML pipeline for MIMIC-III data?▼

To build a clinical ML pipeline for MIMIC-III, use a standardized 5-stage workflow: Dataset, Task, Model, Trainer, and Metrics. This structure maps raw EHR data to tasks like mortality prediction and trains models while computing clinically relevant evaluation metrics.

What healthcare datasets and tasks can I use for clinical machine learning?▼

Clinical machine learning tasks support datasets including MIMIC-III/IV, eICU, OMOP, SleepEDF, ChestXray14, and EHRShot. You can apply these to mortality, readmission, length of stay, drug recommendation, sleep staging, and ICD/ATC/RxNorm medical coding workflows.

How do I set up a deep learning pipeline for medical coding?▼

Set up a deep learning pipeline for medical coding by defining a Dataset, Task, Model, Trainer, and Metrics. This modular approach processes clinical EHR data and handles ICD, ATC, and RxNorm coding tasks end-to-end without extensive boilerplate.

Does this toolkit support sleep staging with SleepEDF data?▼

Yes, the toolkit supports sleep staging using SleepEDF data. It unifies signal data processing through the same modular pipeline structure, mapping raw inputs to trained models and evaluation metrics for sleep stage classification.

Can I use OMOP CDM data for patient readmission prediction?▼

Yes, you can use OMOP CDM data for patient readmission prediction. The toolkit maps OMOP formatted EHR records to readmission tasks, training deep learning models and evaluating outcomes using standardized clinical metrics.

What is the best way to reduce boilerplate in healthcare ML workflows?▼

The best way to reduce boilerplate in healthcare ML workflows is using a modular component system with unified dataset, task, model, trainer, and metrics stages. This standardizes pipelines across EHR, signal, and imaging data to accelerate end-to-end prototyping.