AI-Based Population-Level Disability Estimation and Trajectory Forecasting

Abstract Estimating disability and forecasting the progression of chronic diseases at the population level present major challenges for healthcare systems. This session explores how routinely collected healthcare data, when combined with advanced analytical techniques such as neural networks, can be used to generate disability-adjusted estimates and simulate disease trajectories over time. Traditional statistical models often face limitations in handling the complexity, heterogeneity, and non-linearity inherent in real-world health data. In contrast, artificial intelligence methods-particularly neural networks-offer greater flexibility in modelling high-dimensional datasets, capturing dynamic relationships, and incorporating time-varying covariates. These approaches enable more accurate and timely predictions, which are critical for health system planning and resource allocation. Applying these models to large, linked datasets supports a deeper understanding of the current burden of disability and enables projections of future trends, even in the context of uncertainty. This is especially relevant for progressive chronic conditions, where early identification and timely intervention can substantially influence outcomes. The session will present examples of how neural networks can be applied to routinely collected data to map levels of disability and model its progression in neurological conditions such as Alzheimer's disease, other dementias, and multiple sclerosis. These applications demonstrate how AI-driven tools can enhance population health monitoring, support data-informed policy, and ultimately contribute to improved care for individuals affected by complex, high-burden conditions.