Sex differences of APOE neuropathology‐based scores in brain aging

Abstract

AbstractBackgroundBy reinterpreting the difference between a model prediction and calendar age, machine learning algorithms allow for estimation of “brain age” using MRI. There is growing evidence for sex differences in the influence of APOE genotype (evaluated as APOE ε4 carrier status) on Alzheimer’s disease (AD) endophenotypes at various stages of disease. The purpose of this study is to examine the influence of an APOE neuropathology‐based score (APOE npscore) on the brain aging estimated from deep learning.MethodsBrain age from 274 participants (Fig. 1) in the Wisconsin‐ADRC was estimated from T1w‐MRI using a publicly available deep learning model called two‐stage‐age‐network (TSAN) that was pre‐trained on over four thousand scans from OASIS, ADNI‐I and PAC‐2019. TSAN uses a novel rank‐based loss along with mean squared loss and makes predictions in two steps which does not require a posteriori bias correction. This is a significant improvement from prior deep learning models in reducing the “regression‐to‐mean bias” typically present in brain age models. APOE genotypes (ε2ε2, ε2ε3, ε3ε3, ε2ε4, ε3ε4, ε4ε4) were weighted by the log(odds‐ratio) from a large study of autopsy‐confirmed AD cases/controls, providing a variable representing the relative amount of risk across APOE genotypes. Mood’s median test was used to examine the sex differences between the excess brain aging (EBA) as a function of the APOE npscore.Results Fig. 2 shows the brain age predictions. Fig. 3 shows that there are significant differences between females and males in the EBA for the APOE npscore <0. Median EBA for both males and females is <0 when APOE npscore <0 suggesting protective effects of ε2 on brain aging. The median EBA >0 for both the sexes when APOE npscore ≥0.ConclusionWe report sex differences in the influence of APOE genotype on brain aging. There is heterogeneity in the differences within the ε4+ (APOE npscore >0) and ε4‐ (APOE npscore ≤0) groups. These sex differences may not be observed when using just ε4 carrier status instead of the APOE npscore. This provides evidence for a shift in paradigm for the way we analyze APOE genotype because of latent nuances of the ε4‐status influence on AD outcomes.