Joint modeling of time-varying exposure history and subsequent health outcomes: identification of critical windows
share
Long-term behavioral and health risk factors constitute a primary focus of research on the etiology of chronic diseases. Yet, identifying critical time-windows during which risk factors have the strongest impact on disease risk is challenging. Metrics based on a weighted cumulative exposure index (WCIE) have been proposed, with weights reflecting the relative importance of exposures at different times. However, they are usually restricted to a complete observed error-free exposure. When handling an exposure measured with error, the evaluation of exposure and outcome are concomitant while in some contexts, the time-window of exposure history should precede the outcome window. We thus propose a general joint modelling approach to estimate the dynamics of association between the history of a time-varying exposure intermittently measured with error and a non-concomitant outcome using WCIE methodology. Inference is obtained in two stages: (1) the individual complete exposure history is estimated using a flexible mixed model; (2) the trajectory of association (approximated by cubic splines) between the predicted exposure history and the outcome is then estimated. The methodology is validated in simulations and illustrated in the Nurses' Health Study to investigate the association between body mass index history starting in midlife and subsequent cognitive decline after age 70. We show that while higher mid-life BMI is associated with worse cognitive trajectories, the association is inverted at the approach of cognitive assessment, likely reflecting reverse causation. This approach, applicable with any type of outcome, provides a flexible tool for studying complex dynamic relationships in life-course epidemiology.
READ FULL TEXT