Simulating the outcome of a clinical intervention from a data‐driven model of Alzheimer's disease progression

Background The recent outcomes of clinical trials in Alzheimer's Disease (AD) underline the critical importance of timing for drug intervention, and the need to identify optimal intervention windows along the disease history to maximize cognitive benefit. In this work, we propose a data‐driven progression model of AD allowing simulation of the effect of disease modifying drugs along the entire history of the pathology. Method We propose a generative model of disease progression, aimed at identifying the dynamical relationships between multi‐modal clinical and imaging data. Our data consists of AV45‐PET, FDG‐PET, MR images, and clinical scores for 311 ADNI individuals (49 healthy, 113 MCI, 33 MCI converted to AD, and 116 AD patients) for a total of 2188 longitudinal measures. The model is based on a dynamical system relating the multi‐modal information, and is formulated under realistic clinical hypothesis to provide plausible disease progression simulation and predictions. Result Figure 1 shows the modelled disease progression for grey matter atrophy, glucose hypometabolism, amyloid concentration, and cognitive decline. We notice a global increase of amyloid, while grey matter atrophy and glucose hypometabolism map prevalently temporal and parietal regions. Clinical scores exhibit a non‐linear profile, accelerating during the latest disease stages. The individual disease severity associated to such progression shows significant separation across clinical groups (p