PROJECT SUMMARY
Characterizing the biological mechanisms of Alzheimer's disease and related dementias (ADRD) is crucial for
identifying effective strategies to prevent and treat dementia. Careful and substantial investments have
enabled characterization of preclinical Alzheimer's disease (AD) using the AT(N) framework based on
biomarkers, but the limited diversity of biomarker study samples may limit generalizability of findings from
these studies. Unfortunately, early attempts to increase diversity have not solved the problem of non-
generalizable study results due to differential selection processes across sociodemographic groups. Thus,
increasing diversity of study samples is necessary, but not sufficient, for achieving generalizable study results.
Our overall objective is to develop tools and approaches to enhance the generalizability of ADRD biomarker
studies. It is increasingly recognized that evaluating generalizability of ADRD study results is important, but
researchers are stymied by lack of tools for systematically evaluating generalizability of findings. Standard
analytic methods do not take full advantage of currently available data and do not reveal the extent of
uncertainty in estimates when generalizing to the entire population. New epidemiologic and statistical tools,
including weighting and g-computation (“transport tools”), allow for generalizing results from a study to an
external population of interest. We propose to apply transport tools to develop population-level estimates of the
predictive accuracy of ADRD AT(N) biomarkers on cognitive outcomes and prevalence of preclinical AD (Aim
1) and effects of risk factors on ADRD AT(N) biomarkers (Aim 2) using data from Alzheimer's Disease
Neuroimaging Initiative (ADNI), the French MEMENTO cohort, and Kaiser Healthy Aging and Diverse Life
Experiences (KHANDLE) and Life After 90 (LA90), two newly available, exceptionally diverse ADRD biomarker
samples with harmonized measures. Generalizing from a study sample to a population always increases
uncertainty in estimated effects relative to estimates in the sample (e.g. wider confidence intervals and reduced
power), but the magnitude of added uncertainty due to generalizing depends on the composition of the sample.
Currently, there are no tools to evaluate how the composition of a proposed study will impact statistical power
to detect population-representative effects (i.e. draw generalizable inferences from the proposed study). In Aim
3, we will develop a power calculator for population-level estimates based on the sociodemographic
composition of the proposed sample, enabling researchers to evaluate generalizability in the study design
stage. This project advances critical analytic tools to improve generalizability of existing ADRD biomarker
studies, which can be applied to novel biomarkers and help prioritize recruitment goals. Developing transport
tools for ADRD biomarker research will help researchers obtain the best possible evidence on how to prevent
and effectively treat ADRD in the entire population.
