Uncovering disease progression trajectories across multiple timescales with hybrid ODE-mixture models

LCPTA hybrid ODE–mixture modeling recovered relevant long-term subgroup structure and trajectory patterns in synthetic heterogeneous hypertension and 30-year insulin sensitivity cohorts, including 75 virtual individuals.

• Why it matters: Existing models often capture only short-term physiology or only long-term progression, and methods that span both scales can struggle with inter-individual variation or need large datasets.
• What they did: The framework combines mechanistic ODEs for short-term dynamics with latent class trajectory modeling for long-term parameter evolution, implemented via TemporalMixtureModels.jl. It was benchmarked against a conditional neural ODE and a conditional neural network for long-term reconstruction and subgroup separation.
• The result: LCPTA recovered subgroup structure in both synthetic studies, while neural methods were less faithful in the insulin sensitivity setting; with sparse, irregularly sampled unseen individuals, LCPTA maintained strong classification performance.

AI summary of the abstract · may contain errors · not medical advice · read the paper.