Supplementary Material for: HETEROGENEITY ANALYSIS OF ASSOCIATIONS INVOLVING THE LARGE SCALE ONLINE MINDCROWD SURVEY MEMORY TEST

Alzheimer’s disease and related disorders (ADRD), as well as general age-related cognitive decline, are known to be multifactorial with heterogeneous etiologies. Identifying and accommodating heterogeneity in any one data set can be pursued using different analytical techniques, each with different assumptions or purposes. Whereas a great deal of research has explored clustering individuals of variables that exhibit greater similarity in some way, little research has explored evidence for heterogeneity in the relationships between relevant outcomes, such as a performance on a memory test, and risk factors such as environmental exposures, behaviors or genetic factors among individuals. Heterogeneity in relationships between outcomes and risk factors could compromise the development of accurate risk prediction models. We explored evidence of heterogeneity in the relationships between ability on a memory test and multiple social and demographic risk factors using the large MindCrowd database (n>90,000 individuals). We focused on mixtures of regression models, but compared models assuming many interaction effects among independent variables as well as random effects. We ultimately find substantial evidence for heterogeneity and offer an intuitive explanation for this heterogeneity involving motivation for participation. We provide suggestions to validate and understand these claims using companion or ancillary studies, such as those being pursued as part of the Precision Aging Network (PAN) consortium. Our results also motivate caution in the use of large epidemiological study or survey-oriented data sets to build predictive models of clinical or subclinical pathologies without exploring or accommodating heterogeneity.