Beyond Sequential Patterns: Rethinking Healthcare Predictions with Contextual Insights

About Our Work

Update: 2025/04/22: We have created a repository for the paper titled Beyond Sequential Patterns: Rethinking Healthcare Predictions with Contextual Insights, accepted TOIS2025. In this repository, we offer the original sample datasets, preprocessing scripts, and algorithm files to showcase the reproducibility of our work.

Requirements

• openai==1.3.5
• torch==1.13.1+cu117
• dgl==1.1.2
• pyhealth==1.1.4
• seaborn==0.13.0

Data Sets

Owing to the copyright stipulations associated with the dataset, we are unable to provide direct upload access. However, it can be readily obtained by downloading directly from the official website: MIMIC-III, MIMIC-IV.

The structure of the data set should be like,

data
|_ drugrec-gpt
|  |_ MIII
|  |_ _processed
|  |_ _ _datasets_pre_stand.pkl
|  |_ MIV
|  |_ _ _datasets_pre_stand.pkl
|_ los
|_ mortality-gpt
|_ readmission-gpt
|_ ready-gpt
|_ _MIII
|  |_ _processed
|  |_ triples_id.pkl
|  |_ entity_emb.pkl
|  |_ re_id.pkl

RUN

# process the data
python kg_gen.py
# run the file
python main_los.py

Acknowledge & Contact

You can contact czhaobo@connect.ust.hk for more information and help.