Design: Systematic review and meta-analysis of 21 phase 3 randomised trials
Population: 12,610 patients with early Alzheimer's disease on anti-amyloid antibodies
Primary outcome: Incidence, severity and risk factors for ARIA
Effect: ARIA-E 5.4%, ARIA-H 10.3%; APOE4 homozygote OR 5.12 for ARIA-E

A meta-analysis pooled 21 phase 3 trials and 12,610 patients with early Alzheimer's disease treated with anti-amyloid monoclonal antibodies, quantifying amyloid-related imaging abnormalities — the oedema (ARIA-E) and microhaemorrhage (ARIA-H) seen on surveillance MRI.

ARIA-E occurred in 5.4% and ARIA-H in 10.3%, most cases asymptomatic and radiographically mild to moderate. Risk was strongly genotype-dependent: APOE4 homozygotes had about five times the odds of ARIA-E (odds ratio 5.12) and heterozygotes about twice (1.91), with higher antibody dose (2.0) and baseline microhaemorrhages (1.43) also raising risk. Homozygotes were likewise more likely to have symptomatic and radiographically severe ARIA-E.

The practical implication is that APOE genotyping is not optional before these drugs — it stratifies who is most likely to come to harm, informs the consent conversation, and sets how closely to monitor. As anti-amyloid therapy reaches Indian memory clinics, this is the safety framework that has to travel with it.

ARIA-E occurred in about 5% and ARIA-H in about 10% of treated patients, mostly asymptomatic.
APOE4 homozygotes had roughly five times the odds of ARIA-E; heterozygotes about twice.
Higher antibody dose and baseline microhaemorrhages added further risk.
Genotype APOE before starting, and match MRI monitoring intensity to the risk.

Why it matters

It makes genotype a prerequisite, not an afterthought, in selecting and monitoring patients for these drugs.

The statistics, in plain English

Most ARIA is asymptomatic and spotted only on MRI, so the incidence figures describe imaging findings, not symptomatic harm; the odds ratios show how steeply genotype shifts that risk rather than how often patients become unwell.