Welcome to MedPipe

Configuration-driven, TRIPOD+AI compliant ML pipelines for tabular clinical data.

MedPipe is a Python framework designed to streamline and standardize clinical machine learning workflows on tabular data, for both classification and regression outcomes. It requires Python 3.12 or later and is installed with pip install medpipe. Built around scikit-learn and TRIPOD+AI reporting guidelines, it transforms complex clinical modeling into reproducible, TOML-configured pipelines. By automatically logging execution details and persisting models, environment metadata, and evaluation metrics into a standardized artifact hierarchy, MedPipe ensures full transparency and reproducibility from initial experimentation to publication.

Key features

TRIPOD+AI Guideline Compliant: Designed from the ground up to follow rigorous clinical machine learning reporting standards.
Scikit-Learn Compatible: Integrates seamlessly with standard Scikit-Learn estimators, transformers, and pipelines.
Classification and Regression: MedpipeClassifier models binary outcomes and MedpipeRegressor models continuous outcomes, including full predictive distributions with OrdBoost.
Declarative TOML Configuration: Define entire experiments, from preprocessing to plotting aesthetics, in a single validated TOML configuration file.
Multi-Outcome Pipelines: Model distinct clinical outcomes in a single run with independent hyperparameter and recalibration overrides.
Integrated Recalibration & Fairness: Native probability recalibration and stratified subgroup analysis to evaluate algorithmic fairness across patient cohorts.
Clinical Metrics & Visualizations: Automatic generation of Decision Curve Analysis (DCA), reliability diagrams, ROC/PR curves, and bootstrapped 95% confidence intervals.
Standardized Artifact Tracking: Automatic, structured output management for models, metrics, environment metadata, and execution logs in a unified artifacts/ hierarchy.

Where to go next

Installation: install the package.
Quickstart: run your first classification or regression pipeline.
Examples: scripts for the manual workflow, restoring a run, custom components, and more.
Configuration guide: describe an experiment in TOML, one page per block, and what changes for a regression.
The run folder, confidence intervals and TRIPOD+AI reporting: what a run produces and how it supports reporting.
API reference: documentation of the public classes and functions.

Contributing

We welcome contributions from the community! Whether you are fixing bugs, improving documentation, or proposing new features:

Feel free to open an issue or start a discussion on our GitHub Repository.
Submit Pull Requests targeting the main branch.
Ensure all unit tests pass before submitting (pytest).

License

This project is licensed under the Apache-2.0 License. Developed and maintained by the Surgical Recovery and Safety Lab.

Acknowledgements

This package was developed using Claude Sonnet 5.5. The code was reviewed and edited by humans.