Temporally Detailed Hypergraph Neural ODEs for Type 2 Diabetes Progression Modeling

TL;DR

TD-HNODE addresses irregular-time, heterogeneous disease progression by modeling progression trajectories as a temporally detailed hypergraph and learning continuous-time dynamics with neural ODEs. It jointly learns a time-aware hypergraph Laplacian via an adaptive incidence mechanism and a learnable hyperedge weight matrix to capture high-order marker interactions within and across trajectories. Validated on two real-world EHR datasets for diabetes and cardiovascular progression, TD-HNODE outperforms a wide range of baselines on accuracy, recall, and F1, demonstrating improved early detection of complications. The approach supports clinically guided pathways for sub-phenotyping and personalized interventions, with potential extensions to infer unknown trajectories and incorporate causal treatment effects.

Abstract

Disease progression modeling aims to characterize and predict how a patient's disease complications worsen over time based on longitudinal electronic health records (EHRs). Accurate modeling of disease progression, such as type 2 diabetes, can enhance patient sub-phenotyping and inform effective and timely interventions. However, the problem is challenging due to the need to learn continuous-time dynamics of progression patterns based on irregular-time event samples and patient heterogeneity (\eg different progression rates and pathways). Existing mechanistic and data-driven methods either lack adaptability to learn from real-world data or fail to capture complex continuous-time dynamics on progression trajectories. To address these limitations, we propose Temporally Detailed Hypergraph Neural Ordinary Differential Equation (TD-HNODE), which represents disease progression on clinically recognized trajectories as a temporally detailed hypergraph and learns the continuous-time progression dynamics via a neural ODE framework. TD-HNODE contains a learnable TD-Hypergraph Laplacian that captures the interdependency of disease complication markers within both intra- and inter-progression trajectories. Experiments on two real-world clinical datasets demonstrate that TD-HNODE outperforms multiple baselines in modeling the progression of type 2 diabetes and related cardiovascular diseases.

Figures (4)

• Figure 1: (a) Patient $u$'s marker status vector from $t_1$ to $t_5$; (b) The corresponding DAG of trajectories $p_1$ and $p_2$; (c) Patient $u$'s TD-Hypergraph with two temporally detailed hyperedges $e_1^{u}, e_2^{u}$; (d) An example of problem definition showing TD-HNODE's input and output.
• Figure 2: (a) Overview of the TD-HNODE, with an example of patient $u$'s encounter sequence; (b) The TD-Hypergraph Laplacian module combining an Attention-based Incidence Matrix and a Learnable Hyperedge Weight Matrix; (c) An illustration of cross attention within hyperedge $e_j$.
• Figure 3: (a) Recall of TD-HNODE on both datasets with varying embedding dimensions. (b) Recall with varying numbers of ODE steps. (c) t-SNE visualization of 1,690 patients from the University Hospital dataset; C1, C2, and C3 denote Clusters 1, 2, and 3.
• Figure 4: Mean onset time (in months from the first encounter) of each marker across patients in each cluster. Earlier values indicate earlier manifestation or faster progression.