OneEHR

Run longitudinal EHR experiments from standardized event tables, a TOML config, and one saved run directory.

OneEHR covers preprocessing, model training, testing, analysis, and figures for conventional ML/DL models plus LLM or agent systems. The same config and artifacts are used by the CLI, Python API, and notebooks.

Run the quickstart Prepare data

Python 3.12+ TOML config MIMIC / eICU ICD / CCS / ATC Parquet + JSON

Input 3-table EHR schema dynamic.csv, static.csv, label.csv

Workflow Preprocess to plot One config, one run directory

Models 42 built in ML, DL, multimodal, KG, survival

Outputs Structured artifacts Predictions, metrics, analysis, figures

Start Here ¶

Installation

Set up Python 3.12+, install OneEHR, and verify the CLI.

Quickstart

Run the bundled TJH example from CSV conversion through analysis and figures.

Data Model

Prepare the dynamic, static, and label CSV files used by every workflow.

Standard Workflow ¶

Preprocess

Bin events, encode features, create labels, and save a patient-level split.

Train

Fit every model listed in the TOML config against the saved artifacts.

Test

Evaluate trained models and configured systems on the held-out test split.

Analyze

Write comparison, feature importance, fairness, calibration, statistical test, and missing-data outputs.

oneehr preprocess --config experiment.toml
oneehr train      --config experiment.toml
oneehr test       --config experiment.toml
oneehr analyze    --config experiment.toml
oneehr plot       --config experiment.toml