# Continual Learning of Longitudinal Health Records

[![arXiv](https://img.shields.io/badge/arXiv-2112.11944-b31b1b.svg)](https://arxiv.org/abs/2112.11944) [![PyTorch](https://img.shields.io/badge/​-PyTorch-%23EE4C2C.svg?style=flat&logo=PyTorch)](https://pytorch.org/) [![License: GPLv3](https://img.shields.io/badge/license-GPLv3-green.svg)](https://opensource.org/licenses/gpl-3-0) [![Python](https://badges.aleen42.com/src/python.svg)](https://www.python.org/) ![uv](https://img.shields.io/badge/%E2%80%8B-uv-%23A100FF.svg?style=flat&logo=uv&logoColor=A100FF) 

Repo for reproducing the experiments in [*Continual Learning of Longitudinal Health Records*](https://arxiv.org/abs/2112.11944) (2021). Release [v0.1](releases/v0.1) of the project corresponds to published results.

Experiments evaluate various continual learning strategies on standard ICU predictive tasks exhibiting covariate shift. Task outcomes are binary, and input data are multi-modal time-series from patient ICU admissions.

## Setup

1. Clone this repo locally.

2. Request access to [MIMIC-III](https://www.physionet.org/content/mimiciii/1.4/) and [eICU-CRD](https://www.physionet.org/content/eicu-crd/2.0/).<sup>1</sup>

3. Download the [preprocessed datasets](https://physionet.org/files/mimic-eicu-fiddle-feature/1.0.0/) to the `/data` directory.

## Results

To reproduce main results:

```zsh

uv run main.py --train

```

Figures will be saved to `/results/figs`. Instructions to reproduce supplementary experiments can be found [here](/results/README.md). Bespoke experiments can be specified with appropriate flags e.g:

```posh

uv run main.py --domain_shift "hospital" --outcome "mortality_48h" --models "CNN" --strategies "EWC" "Replay" --validate --train

```

A complete list of available options can be found [here](/config/README.md) or with `uv run main.py --help`.

## Citation

If you use any of this code in your work, please reference us:

```latex

@misc{armstrong2021continual,

      title={Continual learning of longitudinal health records}, 

      author={J. Armstrong and D. Clifton},

      year={2021},

      eprint={2112.11944},

      archivePrefix={arXiv},

      primaryClass={cs.LG}

}

```

---

### Stack

For standardisation of ICU predictive task definitions, feature pre-processing, and Continual Learning method implementations, we use the following tools:

| Tool                        | Source               |

|-----------------------------|----------------------|

|ICU Data                     | [MIMIC-III](https://www.physionet.org/content/mimiciii/1.4/)<br> [eICU-CRD](https://www.physionet.org/content/eicu-crd/2.0/) |

|Data preprocessing / task definition | [FIDDLE](https://www.physionet.org/content/mimic-eicu-fiddle-feature/1.0.0/) |

|Continual Learning strategies| [Avalanche](https://avalanche.continualai.org/)

> [!NOTE]

> Temporal Domain Incremental learning experiments require linkage with original MIMIC-III dataset. Requires downloading `ADMISSIONS.csv` from [MIMIC-III](https://physionet.org/content/mimiciii/1.4/) to the `/data/mimic3/` folder.