Real-world registries ALZ-NET (US) and AD-DMT (Japan) support safety surveillance of anti-amyloid antibodies. Conventional power calculations—dividing required patients by mean per-site caseload—can underestimate the number of centers needed because of patient counts variability.

To develop and evaluate a simulation-based method for site-level sample size planning that incorporates inter-site variability.

We developed a simulation using a zero-truncated negative binomial model to reflect caseload heterogeneity. We estimated the required sites (k) to achieve a target precision (95% confidence interval [CI] width) for ARIA incidence under random and volume-weighted sampling, based on data from published trials. The required number of sites was determined as the point where the CI width met a prespecified precision target (< 0.1).

Simulated ALZ-NET and AD-DMT registry settings using prevalence and ARIA frequencies from published lecanemab and donanemab trials.

Precision (95% CI width) for estimating ARIA incidence in APOE-ε4 homozygotes; comparison of required site counts as estimated by the three methods.

Under random sampling, our method’s site requirement (∼320 sites) was consistent with the ICC-adjusted method, whereas the conventional method underestimated the need (∼220 sites). Critically, our framework showed that strategic volume-weighted sampling could reduce the requirement to as few as 110 sites, surpassing the efficiency of the static analytical methods.

Conventional methods risk underestimating site requirements by ignoring caseload heterogeneity. Our simulation framework provides more realistic estimates and, crucially, quantifies the substantial efficiency gains from strategic recruitment, serving as a flexible tool to optimize registry design.