Lessons from the Parachute Trials, Losing Control of Control Groups

By | April 28, 2011

The problem we reported during Day 1 of the Trials with a 100% crossover rate from the placebo group to the parachute group may have been funny and a bit hyperbolic, but it outlines a very serious flaw in the design of all studies with crossover arms. Let’s be frank about it, the presence of a crossover arm is a sop to the ethical misgivings of the researcher who designed the study and the institutional review board that approved it. Real crossover comparisons should always be done by a two stage trial wherein all participants crossover at the midpoint.

The integrity of a control group is diminished with every person that crosses over from one group to the other. If the groups were well matched and randomized at the start, the patient motivated, deterministic alteration of these groups introduces confounding influences that cannot really be corrected for. Even if the crossover rate from control to treatment were equal to the rate from treatment to control, this still does not mean that the movement was random or even motivated by similar concerns.

There are only two ethical ways to report data on studies with crossovers. Firstly data can be reported up until the point of the first cross over. Secondly, if the number of crossovers is relatively small, the crossovers from the experimental group to the control group can be scored as treatment failures and all crossovers from the control group to the treatment group can be scored as successes for the control group. This is not often done because with more than a handful of crossovers, statistical significance can be impossible to achieve. Sadly not all researchers adhere to such rigorous standards, which once again brings to mind Mark Twain’s famous quote, “Facts are stubborn, but statistics are more pliable.”

Leave Your Comment