Search papers, labs, and topics across Lattice.
This paper introduces a contrastive representation learning framework that models multivariate disease trajectories as temporal graphs, addressing the complexities of longitudinal clinical data. By employing contrastive graph neural networks, the method captures both temporal continuity and structural similarities among patient observations, leading to the generation of embeddings that maintain crucial contextual information. The approach not only facilitates robust clustering of patients based on disease progression patterns but also uncovers latent structures within the longitudinal data, enhancing our understanding of disease dynamics.
Clustering patients based on disease trajectories reveals hidden patterns that could transform personalized medicine approaches.
Understanding disease trajectories from longitudinal clinical data remains challenging due to complex temporal dynamics and heterogeneous patient cohorts. Here, we present a contrastive representation learning framework that models multivariate disease trajectories as temporal graphs and learns representations using contrastive graph neural networks. Nodes represent patient observations over time, while edges capture temporal continuity and structural similarity between trajectories. Structure-aware random walks guide contrastive learning to generate embeddings that preserve temporal context and trajectory topology. The resulting representations enable robust clustering of patients with similar disease progression patterns and reveal latent structure in longitudinal data.
