Application of modified Poisson regression for risk estimation of major neuropathology outcomes: National Alzheimer's coordinating center

Odds ratios (OR) can overestimate risk when the prevalence of outcomes is more than 10%. We compared logistic and modified Poisson models in 5843 National Alzheimer's Coordinating Center decedents to examine associations of apolipoprotein (<i>APOE</i>) ε4, age at death, sex, and education with 7 neuropathologies. OR for neuritic plaques (6.28) and Braak stage (4.90) attenuated to prevalence rate ratios of 1.38 and 1.08; similar attenuation occurred for Lewy body disease, arteriolosclerosis, cerebral amyloid angiopathy, and infarcts. Modified Poisson regression is critical for valid risk estimation, reliable interpretations, and accurate risk stratification in non-rare outcome studies.