Simpson’s Paradox, Amyloid Paradox, and Racial Differences

Importance: The underlying mechanisms for the paradox of higher prevalence of dementia but lower prevalence of amyloid positivity in self-identified Black US residents than White US residents remain unknown and are important for designing and analyzing prevention and treatment trials of Alzheimer disease. Objective: To assess whether the age of mean amyloid onset (AMAO) explains the paradox of the reported racial differences in dementia and amyloid positivity. Design, Setting, and Participants: This was a retrospective, multicenter, longitudinal cohort study with harmonized amyloid positron emission tomography (PET) and clinical data from self-identified Black and White individuals who were middle-aged or older across 5 academic institutes in the US conducting studies of memory and aging. The study was conducted from June 2020 to March 2026. Exposures: Self-identified race, vascular comorbidities, sex, age, educational attainment, and apolipoprotein E ε4 carrier status. Main Outcomes and Measures: Amyloid PET centiloid and cognition measured as a global cognitive composite constructed by averaging z scores across at least 6 cognitive tests in self-identified Black and White participants. Both cross-sectional and longitudinal analyses were conducted to compare racial groups in amyloid burden and cognition. Data were also analyzed using propensity score-matching of 2:1 (White:Black) by main covariates. Results: Of 2653 participants (mean [SD] age, 69.0 [8.8] years; 1574 [59.3%] female), 406 (15.3%) were Black and 2247 (84.7%) were White. Black participants had a delayed AMAO compared with White participants in the unmatched (83.86 [95% CI, 76.55-91.35] years vs 71.38 [95% CI, 67.74-73.17] years; difference, 12.48 [95% CI, 5.03-22.28 years) and covariate-matched (85.17 [95% CI, 80.16-91.38] years vs 72.61 [95% CI, 70.79-74.55] years; difference, 12.57 [95% CI, 7.25-19.54] years) analyses. Although Black participants had a lower overall prevalence of amyloid positivity than White participants (24.1% vs 31.7%, P = .002), they had a higher prevalence of amyloid positivity than White participants in almost every interval of expected years to amyloid onset (EYAO, defined as participants' simple chronological age minus the group-specific AMAO), hence Simpson's paradox, (eg, -15 to -10 years for Black minus White prevalence in the unmatched cohort, 15.30 percentage points [95% Wilson CI, 0.06-0.26]). No significant racial differences were found in the longitudinal rate of amyloid accumulation after adjusting for the covariates either before the AMAO (eg, unmatched cohort Black minus White difference in slope per year, -0.28 [95% CI, -0.64 to 0.08]; P = .13) or after the AMAO (eg, unmatched cohort Black minus White difference in slope, -1.23 [95% CI, -3.05 to 0.60]; P = .17). Black participants had a higher prevalence of vascular comorbidities than White participants (56.9% vs 39.3%; P < .001), and Black but not White participants showed significant cognitive decline before the AMAO (Black minus White difference in slope per year, -0.016 (95% CI, -0.028 to -0.004)]; P = .009). Conclusions and Relevance: This cohort study found that Black participants had delayed AMAO compared with White participants, partially explaining the paradox of the reported racial differences in prevalence of amyloid positivity and dementia. This finding suggests that EYAO may be an important confounder to analyze racial differences and may serve as a better (than age) inclusion or exclusion criterion to improve the representation of Black US residents in clinical trials of Alzheimer disease.