# Deviations from additivity in APOE4-mediated late-onset Alzheimer’s disease risk across races and ethnicities

## TL;DR

This study shows that the genetic risk of Alzheimer's disease from the APOE4 gene varies significantly across different racial and ethnic groups, and that assuming a simple additive model may not be accurate.

## Contribution

The study introduces explicit adjustments for deviations from additivity in modeling APOE4's effect on Alzheimer's disease risk across racial and ethnic groups.

## Key findings

- East Asian participants had the highest APOE4 odds ratios (5.2), while Black participants had the lowest (2.8).
- Homozygote East Asian APOE4 carriers showed a 57% higher risk than predicted by additive models.
- Adjusting for deviations from additivity improved model performance, especially for Black participants.

## Abstract

APOE’s ε4 haplotype (APOE4) is late onset Alzheimer’s disease’s (LOAD) strongest genetic risk factor. Therefore, accurately modeling APOE4’s effect is critical to understanding LOAD. This is especially important as APOE4 odds ratios (OR) vary across racial and ethnic (R/E) groups. We analyzed the APOE4-LOAD association in 3,196 East Asian, 31,105 non-Hispanic White (White), 1,646 Hispanic and Latino (Hispanic), and 6,068 non-Hispanic Black (Black) participants using three genetic models: additive, genotypic, and a reparametrized model accounting for deviations from additivity (DA). Each model adjusted for age, sex, and genetic ancestry. We first calculated additive APOE4 ORs in each R/E group, finding East Asian participants had the largest APOE4 ORs (ORAPOE4: 5.2, 95%CI: 4.4–6.0) and Black participants among the smallest (ORAPOE4: 2.8, 95%CI: 2.6–3.1). Next, we generated APOE4 ORs for heterozygote and homozygote ε4 carriers. These genotypic ORs were statistically the same as the additive APOE4 estimates for all models except homozygote East Asian participants. The measured homozygote East Asian estimate (ORAPOE4: 41.5, 95%CI: 22.1–88.8) was 57% higher than its predicted counterpart based on additivity, indicating significant non-additive contributions. Finally, we used a reparametrized model that adjusted for DA. Although equivalent to the genotypic model, this adjustment provided insight into DA while significantly improving model performance among Black participants. This report demonstrates that APOE4 estimates differ based on the genetic modeling strategy, introduces explicit DA adjustments in the APOE4-LOAD association, and cautions against blindly assuming additivity across R/E groups.

The online version contains supplementary material available at 10.1007/s00439-025-02810-5.