New Generative Model Integrates Social Determinants with Healthcare Data for Disease Reasoning
Researchers have proposed a novel generative model that integrates social determinants of health (SDoH) with multi-organ sensor data to enhance disease prediction and clinical decision support. Addressing the limitations of existing models that often overlook SDoH, this approach utilizes ICD-coded proxies within a conditioned latent diffusion framework. The model connects tokenized healthcare events with complex data representations, such as brain network connectivity and tabular data from other organ systems. A key innovation is the introduction of a geometric diffusion model to capture the temporal evolution of these complex structures. Extensive experiments conducted on the UK Biobank dataset, involving nearly 500,000 medical history sequences and imaging traits from over 100,000 participants across brain, heart, liver, and kidney systems, demonstrate the model's efficacy. The results indicate significant improvements over state-of-the-art autoregressive models and imaging baselines. This advancement allows for more personalized disease modeling and simulated interventions, offering a robust tool for reasoning about future disease trajectories by accounting for the multi-factorial nature of human health.
Wire timeline
New Generative Model Integrates Social Determinants with Healthcare Data for Disease Reasoning
Researchers have proposed a novel generative model that integrates social determinants of health (SDoH) with multi-organ sensor data to enhance disease prediction and clinical decision support. Addressing the limitations of existing models that often overlook SDoH, this approach utilizes ICD-coded proxies within a conditioned latent diffusion framework. The model connects tokenized healthcare events with complex data representations, such as brain network connectivity and tabular data from other organ systems. A key innovation is the introduction of a geometric diffusion model to capture the temporal evolution of these complex structures. Extensive experiments conducted on the UK Biobank dataset, involving nearly 500,000 medical history sequences and imaging traits from over 100,000 participants across brain, heart, liver, and kidney systems, demonstrate the model's efficacy. The results indicate significant improvements over state-of-the-art autoregressive models and imaging baselines. This advancement allows for more personalized disease modeling and simulated interventions, offering a robust tool for reasoning about future disease trajectories by accounting for the multi-factorial nature of human health.
cs.AI updates on arXiv.org