Modeling Temporal Relationships Between Multivariate Repeated Markers Along With Clinical Endpoints: Application to Alzheimer's Disease and Related Dementias.

Diseases often involve multiple dimensions of interrelated impairments. Although significant advances have been made in joint models to simultaneously assess these processes in relation to clinical endpoints, they often fail to evaluate how these dimensions influence each other. We propose an original joint modeling framework to describe the temporal relationships between the processes involved in Alzheimer's disease and related dementias (ADRD) progression, and assess their association with ADRD diagnosis and death. The longitudinal submodel is a dynamic model that combines a structural multivariate mixed model based on differential equations to explain the instantaneous change over time of each latent dimension according to the others, and observation models that can accommodate ordinal, binary, and continuous (whether Gaussian or non-Gaussian) biomarkers. The association of the biomarkers with ADRD diagnosis and death are described via a shared random-effect joint modeling approach. The estimation procedure, carried out within the maximum likelihood framework is made available in the DynNet R package. The methodology is validated in a simulation study and is applied in a population-based French cohort study to disentangle the temporal relationships between three major drivers of ADRD natural history, depression, cognition, and functional dependency in link with the two major clinical events in ADRD progression: ADRD diagnosis and death. The methodology and application are designed to help understand the complex interplay between biomarkers over time.