Improving statistical modeling with a neuropathology‐based APOE genetic risk score

Abstract Background The apolipoprotein E gene (APOE) is the predominant genetic risk factor for late‐onset Alzheimer disease (AD), with three alleles contributing to disease risk and strongly associated with many AD endophenotypes: ε2 (reduced risk), ε3 (reference), and ε4 (increased risk). Researchers often include APOE ε4 carrier status (ε4+/‐) in analyses to account for the genetic effect; however, this does not adequately account for the protective effects of APOE ε2 or the heterogeneous effect of different APOE genotypes (ε2ε2, ε2ε3, ε2ε4, ε3ε3, ε3ε4, and ε4ε4). We hypothesized that accounting for the effect of the full APOE genotype would better recapitulate the influence of APOE in statistical analyses than APOE ε4 status. Method To generate a weighted risk score for APOE genotype (APOE‐npscore), we log‐transformed the odds ratios from a recent study reporting APOE genetic risk for neuropathology‐confirmed AD. We regressed cerebrospinal fluid (CSF) AD endophenotypes on the APOE‐npscore using data from the Wisconsin Registry for Alzheimer’s Prevention (WRAP: 233 cognitively unimpaired (CU), 3 mild cognitive impairment (MCI‐AD)) and Wisconsin Alzheimer’s Disease Research Center (WADRC: 287 CU, 42 MCI‐AD, 48 dementia‐AD). In preliminary analyses, APOE‐npscores were tested against CSF amyloid‐beta 1‐42 (Aβ42) and phosphorylated tau 181 (pTau181) measured using Elecsys® immunoassays (Roche Diagnostics International Ltd, Rotkreuz, Switzerland). Result After adjusting for sex and age at lumbar puncture, the APOE‐npscore was associated with CSF Aβ42 (β = ‐115.9, p = 5.7e‐17) and pTau181 (β = 2.05, p = 4.7e‐8). APOE ε4 carrier status also associated with CSF Aβ42 (β = ‐238.0, p = 1.9e‐13) and pTau181 (β = 4.25, p = 1.1e‐6); however, models using APOE‐npscore explained 2.4% more variation in CSF Aβ42 and 0.8% more variation in CSF pTau181 than APOE ε4 status. These findings were similar in the subset of cognitively unimpaired participants. Conclusion Our preliminary results demonstrate the benefit of using a pseudo‐continuous variable like the neuropathology‐based APOE risk score instead of binary APOE ε4 status. The APOE‐npscore allows researchers to use a continuous measure to account for APOE genetic risk for AD that does not collapse important genotype categories. This increases statistical power, avoids modeling issues that result from small cell sizes, and allows for a more nuanced interpretation of the genetic effect in analyses.