intention to treat

A randomised controlled trial earns its strength at the moment of allocation: chance decides who gets what, so the groups differ in the treatment and not in who chose it. The analysis can give that strength away again. ICH E9, the guideline on statistical principles for clinical trials agreed by the regulators of Europe, the United States and Japan, states the intention to treat principle as the rule that the primary analysis should include all randomised subjects, and says that preservation of the initial randomisation in analysis is important in preventing bias and in providing a secure foundation for statistical tests. Complete follow-up of everyone is rarely achieved in practice, so the guideline uses the term full analysis set for the set of participants that comes as close to that ideal as the data allow.

The alternative is the per protocol set, which ICH E9 also calls the valid cases or the evaluable subjects sample: the narrower group who completed the protocol as written, took enough of the treatment and had no major violations. The guideline names its danger plainly. The bias, which may be severe, arises from the fact that adherence to the study protocol may be related to treatment and outcome. People who stay the course tend to differ from the people who leave, and a treatment judged only on those who tolerated it can look better than it is. That is why ICH E9 says the full analysis set tends to avoid over-optimistic estimates of efficacy, since the non-compliers it keeps in will generally diminish the estimated treatment effect.

Reading a paper, the question is which set carries the headline number, and how many people are missing from it. Where a trial reports its main result on those who completed, the guideline asks for the other analysis alongside it: it is advantageous to demonstrate a lack of sensitivity of the principal trial results to alternative choices of the set of subjects analysed, and when the two sets lead to essentially the same conclusions, confidence in the trial results is increased. One caution is attached to that comfort. The need to exclude a substantial proportion of subjects throws some doubt on the overall validity of the trial, however closely the two analyses agree.