HiDiNet: High-Dimensional Interpretive Network forModeling Aging Health and Survival

Aging is a high-dimensional and stochastic process through various pathways in which healthy functioning can change withtime. While many studies focused on prediction of aging cross-sectionally, few methodologies have been developed to modellongitudinal aging process. Modeling longitudinal data with a Stochastic Differential Equation (SDE) is an emerging area ofaging research. There are frameworks that have been proposed, but have only implemented for few health variables or limitedto binary values. We propose HiDiNet (High-Dimensional Interpretive Network), a framework for predicting individual healthtrajectories and survival over continuous time. Unlike traditional sequential models that operate on fixed, discrete timesteps,HiDiNet uses stochastic differential equations (SDEs) to represent health evolution and integrates Multi-Visit Attention tocapture long-range temporal dependencies among irregular clinical visits. A three-dimensional interaction network supportsinterpretability by visualizing cross-variable effects and pairwise correlations. Evaluated on the English Longitudinal Study ofAging (ELSA; 10 waves, 1998–2019), HiDiNet outperforms Recurrent Neural Network (RNN) and Elastic-Net models (Brierscore 0.33 vs 0.42 vs 0.70; C-index 0.968 vs 0.951 vs 0.700) and achieves more reliable calibration than a Transformer-onlymodel (D-calibration p = 0.932 vs 0.280), while maintaining comparable discrimination (C-index 0.968 vs 0.978). We alsocompare HiDiNet to latent space models with varying dimensions to demonstrate that HiDiNet is comparable to other high-dimensional models in prediction. Finally, we demonstrate HiDiNet’s interpretability through a visualized pairwise correlationnetwork of the various health variables. HiDiNet is the first three-dimensional interaction network to uncover high dimensionalinteractions among health variables during the aging process while capturing its stochasticity in longitudinal data. It canbe applied to a wide range of high-dimensional health data and ultimately improve our understanding of aging process and transform health policy approaches for aging populations.