Neurology -- Correspondence for Koepsell et al., 69 (19) 1868-1872

Correspondence published:

An alternative method for estimating efficacy of the AN1792 vaccine for Alzheimer disease: Michael Grundman, Sid Gilman, Ann Arbor, MI;Ronald S. Black, Collegeville, PA;Nick C. Fox, London, UK;Martin Koller, San Diego, CA (25 February 2008)

Reply from the authors: Thomas D. Koepsell, Yueh-Yun Chi, Erin M. Ramos, William W. Lee, and Walter A. Kukull (25 February 2008)

An alternative method for estimating efficacy of the AN1792 vaccine for Alzheimer disease 25 February 2008

Michael Grundman ,
Sid Gilman, Ann Arbor, MI;Ronald S. Black, Collegeville, PA;Nick C. Fox, London, UK;Martin Koller, San Diego, CA
Send Correspondence to journal:
Re: An alternative method for estimating efficacy of the AN1792 vaccine for Alzheimer disease

michael.grundman{at}elan.com Michael Grundman, et al.

Koepsell et al. [1] suggest an alternative method for analyzing the phase 2a AN1792 studies. [2,3] The original analysis compared immunized patients who mounted a prospectively defined antibody response (titer greater than or equal to 1:2200) to all placebo-treated patients. The authors propose that these immunized antibody responders be compared to hypothetical responders in the placebo group (patients who would have responded to therapy if treated with AN1792).
In this comparison, the authors assumed that the mean outcome among the hypothetical non-responders in the placebo-treated group is equivalent to the mean outcome in immunized non-responders (immunized patients with antibody titers less than 1:2200). This assumption requires that there were no biological treatment effects in the immunized non-responders on the outcomes being compared.
However, as the authors acknowledge, the assumption of no biological effects in the immunized non-responders is questionable. Based upon preclinical and early clinical data, patients in the original phase 2a study were defined as antibody responders if a titer of at least 1:2200 occurred; but it cannot be assumed that lower titers yielded no biological activity in immunized patients. [2]
We reported significant correlations of neuropsychological tests and measures of brain volume with antibody titers below 1:2200 (dose-response). [2-3] Moreover, some patients with titers below 1:2200 developed side effects that appeared related to immunization. It would be false to assume that patients with titers below 1:2200 had no observed biological effects. Because this critical assumption is not valid, we disagree with the suggestion that the data from the original phase 2a study of AN1792 be re-analyzed in this way.
Additionally, Koepsell et al. suggest that if the assumption of no biological effects is indeed false, then the use of an intention to treat analysis (ITT) would be preferred to assess vaccine effects in all recipients. While an ITT analysis would be appropriate if the objective was to evaluate the effects of AN1792 irrespective of antibody titer, the prospectively defined goal of our analyses was to predict benefits that might accrue in subjects with adequate antibody titers (greater than or equal to 1:2200).
As an ITT analysis would include a large number of immunized patients who did not achieve the pre-specified therapeutic threshold or did not mount an antibody response, this method would limit the capacity of the exploratory phase 2a study to predict the effects of adequate antibody titers.
References
1. Koepsell TD, Chi Y-Y, Zhou X-H, Lee WW, Ramos EM, Kukull WA. An alternative method for estimating efficacy of the AN1792 vaccine for Alzheimer disease. Neurology. 2007;69:1868-1872.
2. Gilman S, Koller M, Black RS, et al. Clinical effects of A-beta immunization (AN1792) in patients with AD in an interrupted trial. Neurology. 2005;64:1553-1562.
3. Fox NC, Black RS, Gilman S, et al. Effects of A-beta immunization (AN1792) on MRI measures of cerebral volume in Alzheimer disease. Neurology. 2005;64:1563-1572.
Disclosures:
Dr. Grundman is employed by Elan Pharmaceuticals.
Dr. Gilman participated in the AN1792-201 trial, sponsored by Elan and Wyeth, as Chair of the Safety Monitoring Committee and currently serves the same role in the Phase I, II and III clinical trials of Bapineuzumab. He is also Chair of the Safety Monitoring Committee for the Phase II trial of ELND005, sponsored by Elan and Transition Therapeutics. He has served a similar role in the Johnson & Johnson trial of Topiramate in pediatric epilepsy, and he has consulted for GlaxoSmithKline, Kyowa Pharmaceuticals, Adamas Pharmaceuticals, and ReNeuron. He is a member of the Board of Directors of Balboa Bioscience.
Dr. Black is employed by Wyeth Research.
Dr. Fox participated in the AN1792-201 trial and in the clinical trials of Bapineuzumab.
Dr. Koller consults for Elan Pharmaceuticals.

Reply from the authors 25 February 2008

Thomas D. Koepsell,
Department of Epidemiology
Box 357236, University of Washington, Seattle, WA 98195-7236,
Yueh-Yun Chi, Erin M. Ramos, William W. Lee, and Walter A. Kukull
Send Correspondence to journal:
Re: Reply from the authors

koepsell{at}u.washington.edu Thomas D. Koepsell, et al.

We agree with Grundman et al. on several points but disagree on the following:
First, the potential responders in the placebo group are not hypothetical. Moreover, they are arguably the ideal control group for assessing vaccine effects among individuals who mount an antibody response. Had treatment assignments been reversed, these potential responders would have been singled out for subgroup analysis as being immunologically responsive to vaccine. Unfortunately, we cannot identify them individually except by administering the vaccine.
Second, each of the analytic methods we compared had strengths and weaknesses. The proposed new method assumes that outcomes do not differ between non-responders in the active-vaccine group and their counterparts in the placebo group (potential non-responders). Currently, there is no direct way to test this assumption. Observed correlations between antibody titers and outcomes do raise suspicions, but these correlations could also reflect innate biological differences between high- and low- responders, independent of whether they actually receive the vaccine.
Imbens and Rubin [4] described yet another analytic approach based on potential outcomes, using computer-intensive Bayesian estimation methods, which could also be applied to estimate efficacy among responders. Their method does not require the assumption that outcomes among non-responders are equal between treatment groups. However, without that assumption, their efficacy estimates may be imprecise, even in large samples.
Third, we must remember the alternative. The original analysis compared a non-random subset of vaccine recipients (antibody responders) with all placebo recipients, 80% of whom would not have mounted a qualifying antibody response if they had received active vaccine. This comparison confounds vaccine effects with biological differences between responders and non-responders. Simulation results suggest that this method can be seriously biased.
Both dilemmas could be avoided if an in vitro test can be developed that pre-identifies which individuals will respond immunologically to the vaccine. Until then, the only practical strategy involves administering the vaccine to a mixed population of responders and non-responders.
If biological effects are expected in all recipients, the intent-to-treat (ITT) comparison seems scientifically relevant, even if it is not the only analytic approach. The articles from the AN1792 trial [2,3] reported ITT results for some outcomes but not for others. We merely advocate that ITT results be reported routinely, not selectively.
References
4. Imbens GW, Rubin DB. Bayesian inference for causal effects in randomized experiments with noncompliance. Ann Stat 1997; 25:305-327.
Disclosure: The authors report no conflicts of itnerest.