PyHealth Clinical ML Pipelines
Builds and checks PyHealth clinical prediction pipelines on EHR, signal, and medical-code data
- Repository
- K-Dense-AI/scientific-agent-skills
- Stars
- 55,780
- Licence
- MIT
- Files
- 8 files
What it does
Once installed, the agent follows PyHealth's structured workflow (dataset, task, model, trainer, metrics) when building clinical machine-learning pipelines, and checks common pitfalls such as patient-level splitting, leakage, and outcome timing. It also knows how to map medical codes (ICD, ATC, NDC, RxNorm) and includes a starter script that runs a demo on invented data to confirm the setup works.
Great for
- Researchers or data scientists prototyping clinical prediction models on EHR datasets such as MIMIC-III or MIMIC-IV with authorized access
- Developers who need a correct PyHealth 2.0.2 starting point and want the agent to avoid legacy 1.x API patterns
- Analysts who need medical code lookups or mapping between coding systems
Skip it if
- Clinical decision-making or patient care: the skill itself states the demo gives no evidence of clinical performance and the upstream mortality task is next-admission prediction, not early warning
- General tabular machine learning with no PyHealth dataset or task, which the skill says it is optional for
- Working with restricted patient data without authorization, since restricted clinical datasets require authorized local access
Try asking
- “Set up a PyHealth 2.0.2 mortality prediction pipeline on my local MIMIC-III extract with patient-level splits”
- “Run the PyHealth starter demo to check my environment works”
- “Map these ICD-10 diagnosis codes to their CCS or ATC equivalents using PyHealth's medical code tools”
- “Review my PyHealth training script for leakage and monitor-metric mistakes”
Install in 3 steps
- 1
Download the skill
The ZIP contains the 8 original files, the MIT licence and an attribution note.
pyhealth.zip - 2
Add it to your agent
Claude Code (all projects)
mkdir -p ~/.claude/skills && unzip pyhealth.zip -d ~/.claude/skills/Claude Code / Agent SDK (this project only)
mkdir -p .claude/skills && unzip pyhealth.zip -d .claude/skills/Claude apps (web/desktop): Settings → Capabilities → Skills → Upload skill, and choose the ZIP.
- 3
Ask for the task
No special command needed — the agent reads the skill's description and loads it when your request matches. Try one of the prompts above.
Get better results
- Run the built-in demo first (uv run ... assets/starter_pipeline.py --demo --epochs 1) to confirm the install before pointing it at real data.
- Give the agent your cohort definition, prediction time, observation window, and evaluation unit up front, since the skill asks for these before modeling.
- Use a fresh cache directory after changing data, configuration, or custom task code, because cache identity does not track file contents.
What's inside
- assets/starter_pipeline.py
- references/datasets.md
- references/examples.md
- references/installation.md
- references/medcode.md
- references/models.md
- references/tasks.md
- SKILL.md
Redistributed unmodified from K-Dense-AI/scientific-agent-skills under the MIT licence. The assessment was drafted with AI from the skill's files on 2026-10-29. Skills can instruct your agent to run code — read SKILL.md and any scripts before installing.