Graph based model improves ICU death risk prediction with missing data
GLR-MM: Graph-Based Global-Local Reconstruction for Robust Multimodal Chest X-ray and EHR Representation Learning under Missing Modalities
Machine LearningArtificial Intelligence
Summary
Making predictions about patient outcomes is harder when some medical tests or records are missing. The authors designed a system called GLR-MM to fill in missing information by looking at connections both within a patient’s records and across similar patients. Their method combines images from chest X-rays and electronic health records to better predict risk of death in intensive care. It works especially well when a lot of data is missing, outperforming other methods in that setting.
What this means in practice
- •For icu data scientists: Enable early mortality risk prediction even when chest X-rays or health records are partially unavailable by reconstructing missing data using clinical and patient similarity graphs.
- •For clinical ai developers: Improve robustness of multimodal medical AI systems that need to handle incomplete chest X-ray and electronic health record inputs in hospital environments.
Authors
Surbhi Sharma, Nikhil Manali, Devesh Maheshwari
Abstract
Clinical multimodal models must often predict before all chest X-ray (CXR) and electronic health record (EHR) inputs are available. Existing approaches align observed representations, model missingness, or reconstruct across modalities, but do not jointly exploit within-patient and clinically similar inter-patient evidence. We propose GLR-MM, a Graph-Based Global-Local Reconstruction framework for early ICU mortality prediction. It maps five CXR-EHR modalities to a shared space, reconstructs missing embeddings through complementary local cross-modal and global graph-attention branches, adaptively fuses their estimates, and optimizes class-balanced prediction, reconstruction, and contrastive objectives. On 9,620 MIMIC-derived ICU stays, we evaluate 10%, 30%, and 50% random modality missingness with shared deterministic masks. MUSE performs better under mild and moderate missingness, whereas GLR-MM achieves higher AUROC and AUPRC at 50% by 0.0088 and 0.0249, respectively. These results indicate that graph-guided reconstruction is most useful when inputs are severely incomplete.