Search papers, labs, and topics across Lattice.
This paper introduces HealthPoint (HP), a novel clinical point cloud paradigm to handle multi-level incomplete Electronic Health Records (EHRs) for in-hospital mortality prediction. HP represents EHR data as points in a 4D space (content, time, modality, case) and uses a Low-Rank Relational Attention mechanism to model interactions between point pairs, addressing temporal misalignment, modality imbalance, and limited supervision. Experiments on large-scale EHR datasets demonstrate that HP achieves state-of-the-art performance and robustness compared to existing methods, especially under varying degrees of data incompleteness.
A new point cloud representation and attention mechanism unlocks state-of-the-art in-hospital mortality prediction from messy, incomplete real-world EHR data.
Deep learning-based modeling of multimodal Electronic Health Records (EHRs) has become an important approach for clinical diagnosis and risk prediction. However, due to diverse clinical workflows and privacy constraints, raw EHRs are inherently multi-level incomplete, including irregular sampling, missing modalities, and sparse labels. These issues cause temporal misalignment, modality imbalance, and limited supervision. Most existing multimodal methods assume relatively complete data, and even methods designed for incompleteness usually address only one or two of these issues in isolation. As a result, they often rely on rigid temporal/modal alignment or discard incomplete data, which may distort raw clinical semantics. To address this problem, we propose HealthPoint (HP), a unified clinical point cloud paradigm for multi-level incomplete EHRs. HP represents heterogeneous clinical events as points in a continuous 4D space defined by content, time, modality, and case. To model interactions between arbitrary point pairs, we introduce a Low-Rank Relational Attention mechanism that efficiently captures high-order dependencies across these four dimensions. We further develop a hierarchical interaction and sampling strategy to balance fine-grained modeling and computational efficiency. Built on this framework, HP enables flexible event-level interaction and fine-grained self-supervision, supporting robust modality recovery and effective use of unlabeled data. Experiments on large-scale EHR datasets for risk prediction show that HP consistently achieves state-of-the-art performance and strong robustness under varying degrees of incompleteness.
