The principle of an ‘intention to treat’ analysis is that the participants in a randomised trial are analysed in the group to which they were randomised, regardless of what treatment they received. So in a hypothetical trial of salbutamol vs. aminophiline infusion for severe asthma, regardless of what the child got, they are placed in their ‘you should have’ group…
The concept comes from the core of RCT philosophy – that chance has settled all prognostic factors evenly between the two* arms – and so the only reasonable way of preserving this is to analyse the outcomes according to this sorting.
What this does is, if some folk in the ‘intervention’ arm don’t get the intervention (e.g. Salbutamol infusion, but their K+ was falling prior to starting) then it reduces the observed effect of the drug. This is then ‘unfair’.
But wait. Pragmatic RCTs, ones of treatments as we use them, test an intervention. They test not ‘salbutamol infusion’ but the intervention – which might be characterised as ‘what if we have an approach that says we should use salbutamol infusions for pts unless its clear they need something different … like PICU .. now … can someone ring 2222 please …’
If there are lots of deviations, crossovers and non-receipts of the allocated intervention, it’s very important to looks why. The way we were proposing to do ‘the intervention’ clearly doesn’t work in practice — so it needs reassessing — not necessarily having the ‘treatment’ element thrown out.
* OK – so it could be three, four etc arms. It’s just that two is easier to think about. And commonerer.