Was EMILIA Biased?

Dr Kimberly Blackwell

In 2012, I got the opportunity to represent Fight Colorectal Cancer as a patient research advocate at the American Society of Clinical Oncologist (ASCO) meetings.  It was an amazing experience.  At those meetings I had the opportunity to see Dr Kimberly Blackwell present findings from the Phase 3 trial measuring the effectiveness of T-DM1 (Kadcyla) for HER2 metastatic breast cancer patients (EMILIA).

Recently, I was looking at data on EMILIA in ClinicalTrials.gov and noticed that the results of trial may suffer from attrition bias.  Just under 1,000 patients participated in the trial with half in the T-DM1 arm and half receiving the combination of Lapatinib and Capecitabine (X+L).  What I noticed was that 48 patients (about 10%) left the X+L arm of their own volition, while only 28 patients (about 6%) left the T-DM1 arm of their own volition.

EMILIA was an "open-label" study, so the patients and their doctors knew what drug they were taking.  If patients who were more likely to do well on T-DM1 relative to X+L, were more likely to stay in the trial, then the results may suffer from attrition bias.  At the 24-month mark there were 111 patients remaining in the T-DM1 arm and 86 patients remaining in the X+L arm.  Some of the difference is due to the difference in survival between the two arms, but some of the difference may be due to difference in patient/doctor choices.  

To the extent it is the second, then we no longer have random assignment between our two comparison groups.  In particular, the patients left in the T-DM1 arm may include a disproportionate number of patients who will live longer on T-DM1 relative to X+L.

Given these observed differences, you may think attrition bias gets a relatively detailed discussion in the New England Journal of Medicine article on the trial or the FDA's approval.  You would be wrong.  The issue is not raised in either forum.

For more discussion of attrition bias in randomized control trials check out this recent paper at BMJ.com or the 2010 National Academies publication.