GraphCare: Enhancing Healthcare Predictions with Personalized Knowledge Graphs

Authors: Pengcheng Jiang, Cao Xiao, Adam Richard Cross, Jimeng Sun

ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental On two public datasets, MIMIC-III and MIMIC-IV, GRAPHCARE surpasses baselines in four vital healthcare prediction tasks: mortality, readmission, length of stay (LOS), and drug recommendation. We evaluated the effectiveness of GRAPHCARE using two widely-used EHR datasets, MIMIC-III (Johnson et al., 2016) and MIMIC-IV (Johnson et al., 2020). Through extensive experimentation, we found that GRAPHCARE outperforms several baselines...
Researcher Affiliation Collaboration Pengcheng Jiang Cao Xiao Adam Cross Jimeng Sun University of Illinois Urbana-Champaign GE Health Care OSF Health Care
Pseudocode Yes Algorithm 1 Subgraph Sampling
Open Source Code Yes Our code is available at https://github.com/pat-jj/Graph Care.
Open Datasets Yes For the EHR data, we use the publicly available MIMIC-III (Johnson et al., 2016) and MIMIC-IV (Johnson et al., 2020) datasets. Both datasets are under Physio Net Credentialed Health Data License 1.5.0
Dataset Splits Yes We split the dataset by 8:1:1 for training/validation/testing data, and we use Adam (Kingma & Ba, 2014) as the optimizer.
Hardware Specification Yes Hardware. All experiments are conducted on a machine equipped with two AMD EPYC 7513 32-Core Processors, 528GB RAM, eight NVIDIA RTX A6000 GPUs, and CUDA 11.7.
Software Dependencies Yes We implement GRAPHCARE using Python 3.8.13, Py Torch 1.12.0 (Paszke et al., 2019), and Py Torch Geometric 2.3.0 (Fey & Lenssen, 2019).
Experiment Setup Yes Based on our hyperparameter study in Appendix G.2, we set learning rate 1e-5, weight decay 1e-5, batch size 4, and hidden dimension 128. All models are trained via 50 epochs over all patient samples, and the early stopping strategy monitored by AUROC with 10 epochs is applied.