mip-examples

Example notebooks and scripts for the Medical Informatics Platform (MIP) Jupyter notebooks. They use the mip Python client to reach the platform; they never touch data directly.

A notebook in this repo is the analysis, not a run of it. If you have executed one, clear the results before saving: no cell outputs, no execution counts, no attachments. What a run printed or plotted stays on your machine.

Path	What it is
algorithm_examples.py	Every Pipeline algorithm and preprocessing builder, with # %% cells. Also the default scratch template of jupyter-mcp (examples/algorithm_examples.py).
Diabetes/	Three hospital-diabetes scenarios: HbA1c and readmission, 30-day readmission, medication and glucose.
ECREAM/	Door-to-doctor time, sex and age by cohort, a mortality negative control, and the live demo.
FERES/feres_analysis.ipynb	FERES stroke territory analysis.
IST/	Six International Stroke Trial scenarios, from aspirin to bleed-or-clot.
NHAMCS/	Three emergency-department scenarios: flow, opioids, COVID-19 admission.
NHCS/	Six hospital scenarios, from ED visits through inpatient stay to the ED and the ward side by side.

Using them

This repo is synced read-only into every notebook, at ~/work/examples/ (about a minute after a change lands on master). You cannot edit the files there. To adapt one, open it and use File > Save As into scratch/, then edit your copy. A run in scratch/ can keep its results; a copy proposed back here cannot.

Contributing

master is copied into every notebook in every federation within about a minute, and users run it holding their platform token. Only the repository's writers can change it, and they may push to master directly; there is no required review. So check your own change against the list below before you push.

The check workflow runs on every push and pull request (only .ipynb, .py and .md files; notebooks with cleared outputs; no symlinks, no files over 1 MiB; a grep for direct network, shell or environment access). It reports problems but does not block a push: a failing commit is already in the notebooks, so fix it with the next push. CI is a tripwire, not a sandbox: it only catches the obvious.

Before you push:

• outputs, execution counts and attachments are cleared, including after a local run;
• there is no data and no patient-level printouts;
• the platform is reached only through mip;
• nothing is federation-specific (it runs in every federation);
• no text aimed at AI agents or users that tells them to run, fetch or send anything; read markdown cells, docstrings and comments too.