Aging
Estimated effects of diabetes on neuroimaging biomarkers: Extending estimates from the MEMENTO cohort to the French population of older adults Yingyan Wu* Yingyan Wu Wu Wu Wu Wu Wu Wu Wu Wu UCLA Fielding School of Public Health; Boston University School of Public Health
Neuroimaging studies show that people with type 2 diabetes have worse neuroimaging biomarkers of Alzheimer’s disease and related dementias. However, neuroimaging samples are often highly selected, raising concerns about generalizability to broader populations. We aimed to extend estimated effects of type 2 diabetes on neuroimaging biomarkers from a clinic-based French neuroimaging sample to the French population of older adults. We used data from the French MEMENTO neuroimaging cohort (n=2,057), which recruited participants aged 60-90 years from memory clinics who had subjective cognitive complaints or mild cognitive impairment, and the survey-weighted French component of the Survey of Health, Ageing, and Retirement in Europe (SHARE-France; n=3,459) to represent 13.8 million French adults aged 60-90 years. After harmonizing and pooling datasets, we compared characteristics between MEMENTO and survey-weighted SHARE-France. We used a doubly robust transportability estimator to extend estimates from MEMENTO to the target population. We estimated inverse odds of selection weights and inverse probability of treatment weights using age, sex, education, self-rated health, and hypertension. We fit outcome models in the MEMENTO sample and combined with weighting to obtain transported doubly robust estimates. Compared with the French population represented by SHARE-France, MEMENTO participants were healthier and more highly educated. Despite these differences, associations between type 2 diabetes and neuroimaging biomarkers were qualitatively similar in the original MEMENTO sample and the transported estimates (e.g., mean difference in brain parenchymal fraction by diabetes status: MEMENTO b=−1.98% [95% CI −2.95%, −1.03%] vs. transported b=−2.68% [−6.21%, −0.36%]) (Figure 1). Despite substantial selection into the neuroimaging cohort, associations between type 2 diabetes and neurodegeneration were robust when extended from a clinic-based sample to the French population.
