odyssey

Odyssey

A library for developing foundation models using Electronic Health Records (EHR) data.

Visit our recent EHRMamba paper

Introduction

Odyssey is a comprehensive library designed to facilitate the development, training, and deployment of foundation models for Electronic Health Records (EHR). Recently, we used this toolkit to develop EHRMamba, a cutting-edge EHR foundation model that leverages the Mamba architecture and Multitask Prompted Finetuning (MPF) to overcome the limitations of existing transformer-based models. EHRMamba excels in processing long temporal sequences, simultaneously learning multiple clinical tasks, and performing EHR forecasting, significantly advancing the state of the art in EHR modeling.

Key Features

The toolkit is structured into four main modules to streamline the development process:

1. data:
   • Tokenizes data and creates data splits for model training.
   • Provides a dataset class for model training.
2. models:
   • Implements models including XGBoost, LSTM, CEHR-BERT, BigBird, MultiBird, and EHRMamba.
   • Offers various embedding classes necessary for the models.
3. evals:
   • Includes tools for testing models on clinical prediction tasks and forecasting.
   • Provides evaluation metrics for thorough assessment of model performance.
4. interp:
   • Contains methods for interpreting model decisions.
   • Features interactive visualization of attention matrices for Transformer-based models.
   • Includes novel interpretability techniques for EHRMamba and gradient attribution methods.
     
     

Data Preprocessing

The data extraction and preprocessing pipeline requires running the repository located at MEDS repository. The pipeline extracts and preprocesses the MIMIC-IV dataset to generate a patients’ sequence of events.

Installation

Clone and install the required repository locally:

git clone --branch odyssey https://github.com/VectorInstitute/meds.git
cd meds/MIMIC-IV_Example
pip install .

As mentioned in the MEDS repository two (optional) hydra multirun job launchers for parallelizing extraction and pre-processing pipeline steps: joblib (for local parallelism) and submitit to launch things with slurm for cluster parallelism.

To use either of these, you need to install additional optional dependencies:

1. pip install -e .[local_parallelism] for joblib local parallelism support, or
2. pip install -e .[slurm_parallelism] for submitit cluster parallelism support.

Running the Extract Pipeline

The run_extract.sh script performs the following steps:

• Unzips the MIMIC data files if necessary.
• Batches hospital lab events and chart events into multiple Parquet files to prevent memory issues during processing.
• Runs the pre_MEDS pipeline.
• Executes the extract pipeline, which:
  • Converts raw data.
  • Shards events.
  • Splits subjects into train, test, and holdout sets (note: in our case, we process all data as train and perform the split later in the Odyssey pipeline).
  • Converts data to sharded events.
  • Merges data into a MEDS cohort.

Note that the events that are extracted and included in the MEDS cohort are defined based on the event configs files.

Run the extract pipeline using:

./run_extract.sh path_to_raw_data_dir path_to_preMEDS_dir path_to_MEDS_dir

Options:

• do_unzip=true|false (Optional) Unzip CSV files before processing (default: false).
• batch_files Run batch_files.py before processing (requires extra arguments):
  • --lab_input=<path> (Required if batch_files is set) Path to labevents CSV.
  • --chart_input=<path> (Required if batch_files is set) Path to chartevents CSV.

To use a specific stage runner file (e.g., to set different parallelism options), you can specify it as an additional argument

export N_WORKERS=5
./run_extract.sh path_to_raw_data_dir path_to_preMEDS_dir path_to_Extract_dir \
    stage_runner_fp=slurm_runner.yaml

The N_WORKERS environment variable set before the command controls how many parallel workers should be used at maximum.

Running the Preprocess Pipeline

The run_preprocess script executes the following steps:

• filter_codes: Remove specific codes from patient records if necessary.
• filter_subjects: Exclude patients with fewer than a minimum number of events.
• filter_labs: Remove lab records without numerical values.
• filter_meds: Exclude medications with a specific code (0 in our case).
• update_transfers: Rename certain transfer codes.
• add_age: Compute and add patient age as the first event in records.
• add_cls_token: Add CLS token as the first event.
• quantize_labs: Quantize lab values based on binning strategy.
• add_time_tokens: Apply binning strategy to account for time intervals between events using predefined bins (hour2bin and minute2bin).
• generate_sequence: Concatenate all patient records to form the final sequence.

Run the preprocess pipeline using:

./run_preprocess.sh path_to_Extract_dir path_to_Processed_DIR

Modifying Default Configuration

To customize the default parameters for each pipeline step, modify the following configuration files:

• extract_MIMIC_seq.yaml
• preprocess_MIMIC_seq.yaml

Contributing

We welcome contributions from the community! Please open an issue.

Citation

If you use EHRMamba or Odyssey in your research, please cite our paper:

@misc{fallahpour2024ehrmamba,
      title={EHRMamba: Towards Generalizable and Scalable Foundation Models for Electronic Health Records},
      author={Adibvafa Fallahpour and Mahshid Alinoori and Arash Afkanpour and Amrit Krishnan},
      year={2024},
      eprint={2405.14567},
      archivePrefix={arXiv},
      primaryClass={cs.LG}
}

This site is open source. Improve this page.