SummaryThe ultimate goal of this methodological project is to establish a lifespan AI method that enables the forecast of individual-level health trajectories over extended time spans by combining data from multiple cohorts. We will combine and harmonize data of multiple cohorts and aim to identify key requirements with regard the design features of multi-cohort studies that need to be fulfilled for predictions and causal discovery across the lifespan. We will leverage so-called mixed-effects machine learning (ME-ML) approaches that combine the structure of generalized linear mixed models with advanced modelling capabilities of machine learning to enhance lifespan predictions based on pooled cohort data. We further aim to answer the key question on which time span can be validly predicted based on the newly developed method. Finally, we will evaluate the predictive performance of the newly devised method by application to empirical and simulated data as well as in different epidemiological settings and compare it to conventional methods. This work is part of the DFG-funded research group “Lifespan AI – From longitudinal data to lifespan inference in health” (https://lifespanai.de/) which has the overall aim to advance AI methods and tools to model, predict, and explain health outcomes from multi-dimensional data that span the entire life.

Keywords lifespan mixed-effects-machine-learning multi-cohort-data neural-network prediction random-forest

InstitutionsLeibniz-Institut für Präventionsforschung und Epidemiologie - BIPS, Leibniz Institut für Präventionsforschung und Epidemiologie - BIPS, Universität Bremen