When is a randomized controlled trial not a randomized controlled trial?

The question arises from a presentation by Martin Bland from the University of York on a method of allocation designed to ensure a good balance between treatment and control groups in terms of key variables.

The technique was used in a study involving 13 schools in Northern Ireland that varied in terms of size, geography (urban/rural), religion (Catholic Protestant), educational attainment and proportions of children receiving free school meals.

Allocation by minimization takes these differences into account. Schools are allocated one at a time. The first school is allocated randomly after which the computer program asks which group the next school should go into so as to achieve the best balance (and so on until all schools are allocated). For example, if the treatment group has two Catholic schools and the control group one, it is likely that the next Catholic school will go into the control group. If the allocation would make no difference to the balance, the school is allocated randomly.

In the study described, this method achieved a near perfect balance between the groups. All seemed well, but then the argument broke out. Some delegates claimed that this was not random allocation and so the study was not technically an RCT: the Federal Drugs Agency in the US would not accept it.

Others suggested that it was a price worth paying to achieve pre-intervention equivalence between the groups. Bland maintained that it was a good method in a cluster randomized trial if not in trials involving individual clients.

Having been involved in the service development work leading up to this study I know that it was a necessary step in order to secure buy-in from schools and the local community, also to ensure that the results would be taken seriously by policy makers. Without minimization, there was a strong possibility that, for example, the Catholic schools could all end up in one group. It’s all very well being a purist but where the decision affects whether an experiment happens or not I think it is important to be pragmatic.