An alternative method for estimating efficacy of the AN1792 vaccine for Alzheimer disease

In a randomized trial of AN1792 vaccine against A beta in Alzheimer disease (AD), only 20% of vaccine recipients had an anti-AN1792 antibody response. The trialists sought to estimate the efficacy of the vaccine among antibody responders by comparing outcomes among antibody responders in the vaccine group with outcomes among all placebo recipients.

We describe why the method used may be biased. An alternative approach to estimating efficacy is described that compares outcomes between responders in the vaccine group and potential responders in the placebo group. Although potential responders cannot be identified individually, the distribution of outcomes among them can be inferred indirectly, under certain assumptions. Three methods for assessing vaccine effects are compared using data on the ventricular volume boundary shift integral (BSI) from the AN1792 trial and in simulations.

Mean (+/- standard error) increase in BSI relative to controls was 0.16 (+/-0.065) by intent-to-treat, 0.61 (+/-0.116) in the published comparison, and 0.81 (+/-0.320) in the proposed approach. Simulations show that the published method can often yield biased estimates, while the proposed method does not.

Published results from the AN1792 trial may have underestimated the effect of vaccine on progression of cerebral atrophy among patients with an antibody response to the vaccine. For this and future similar trials, we suggest that intent-to-treat results always be reported, and that efficacy estimates be based on the proposed potential-outcomes method.