An Introduction to Longitudinal Synthetic Cohorts for Studying the Life Course Drivers of Health Outcomes and...

Research on the drivers of health across the life course would ideally be based in diverse longitudinal cohorts that repeatedly collect detailed assessments of risk factors over the full life span. However, few extant data sources in the US possess these ideal features. A "longitudinal synthetic cohort"-a dataset created by stacking or linking multiple individual cohorts spanning different but overlapping periods of the life course-can overcome some of these challenges, leveraging the strengths of each component study. This type of synthetic cohort is especially useful for aging research; it enables description of the long-term natural history of disease and novel investigations of earlier-life factors and mechanisms shaping health outcomes that typically manifest in older age, such as Alzheimer's disease and related dementias (ADRD).

We review current understanding of synthetic cohorts for life course research. We first discuss chief advantages of longitudinal synthetic cohorts, focusing on their utility for aging/ADRD research to concretize the discussion. We then summarize the conditions needed for valid inference in a synthetic cohort, depending on research goals. We end by highlighting key challenges to creating longitudinal synthetic cohorts and conducting life course research within them.

The idea of combining multiple data sources to investigate research questions that are not feasible to answer using a single cohort is gaining popularity in epidemiology. The use of longitudinal synthetic cohorts in applied research-and especially in ADRD research-has been limited, however, likely due to methodologic complexity. In particular, little guidance and few examples exist for the creation of a longitudinal synthetic cohort for causal research goals. While building synthetic cohorts requires much thought and care, it offers tremendous opportunity to address novel and critical scientific questions that could not be examined in a single study.