Intention to treat analysis

At the Centre for Evidence in Transplantation (CET) we have been rating the methodological quality of randomised controlled trials (RCTs) published since 2004. The quality score is a composite score assessing allocation concealment, intention to treat (ITT) analysis and the Jadad scale which comprises items on the adequate description of randomisation, double blinding and withdrawals. The quality scores are available through the Transplant Library.

We have applied the strict definition of ITT, namely that analysis should be based on all patients randomised. Recently, this definition of ITT has stirred a discussion among some of the authors whose trials we have assessed and among ourselves on how this item should be interpreted when assessing trials for their quality. 

Randomisation of patients at the start of a trial ensures that known and unknown prognostic factors are balanced between groups. This is the strength of an RCT. An ITT analysis maintains this balance as all randomised patients are included when analysing the results of a trial. The ITT also ensures that the power of the study is preserved.

However, an ITT analysis gives rise to concern when patients drop out or there are deviations from the study protocol. In the case of patient dropout and thus missing data an ITT analysis can only be carried out with a complete data set. This can be achieved by imputing data using techniques such as last observation carried forward or replacing missing values with mean change scores. But these imputations are subject to certain assumptions and may introduce a bias. In the case of deviations from the study protocol participants are often excluded from the analysis, for example participants who did not receive a transplant or participants who did not take (a minimal amount of) the study drug. However, following the ITT these patients should still be included because the balance between randomised groups is no longer maintained when patients are excluded. Excluding patients also affects the power of the study.

The 2010 CONSORT statement (http://www.consort-statement.org/) acknowledges these difficulties when deciding how to handle the ITT and therefore an ITT analysis is no longer requested when reporting a trial. Instead it is now required that authors  provide a detailed description of whether the analysis was by original assigned group and which patients were included in each of the analyses (this information is also clearly identified in a flow chart).

The CET will follow the approach outlined in the new CONSORT statement. Instead of assessing whether a trial was based on the ITT we will rate the statistical analysis of trials as one of four different analysis strategies:

1. Strict intention to treat:
   All patients are included in the analysis. Missing data are imputed in case of dropouts.
2. Available case analysis:
   Data are analysed according to the assigned intervention for every participant for whom the outcome was obtained. Missing data are not imputed.
3. Modified ITT:
   Data are analysed according to the assigned intervention. The analysis excludes participants who did not adhere to the protocol. For example, patients not transplanted or patients not receiving the study drug.
4. Per protocol analysis:
   Only patients who sufficiently complied with the protocol are included in the analysis. 

These ratings will be available in the Transplant Library from September 2011.