Background: Intention to treat (ITT) is an important approach used for the analysis of randomised controlled trials. It implies that subjects are included in a trial and analysed regardless of whether they satisfied the entry criteria, the treatment to which they were originally allocated and subsequent withdrawal or deviation from protocol. Excluding patients from analysis may introduce bias and lead to misleading results. In recently published trials the "modified intention to treat" (mITT) approach has repeatedly appeared; its meaning is unclear. No study has been interested in the meaning of mITT.
Objectives: The purpose of this study is to evaluate the frequency, definition and implication of the mITT approach in randomised controlled trials. It will also assess if true ITT analysis was used in the same trials and if it was used in the primary analysis, and what makes mITT different from true ITT analysis.
Methods: To retrieve a sufficient number of pertinent articles we shifted the search from Medline to Highwire and ScienceDirect which are databases that provide full-text search options. We used the following terms 'modified intention to treat', 'modified intent to treat', and 'modified ITT'. Randomised controlled trials published from 1990 to 2006 were considered for evaluation. The description of mITT has been examined and classified according to defined criteria.
Results: Out of 573 records 337 randomised controlled trials were retrieved. All trials cited the modified intention to treat approach. Only 11 papers appeared in 2000, while almost 50% of the reports were published between 2005 and 2006. The issues of interest in the trials were infectious disease in 23%, neurological or psychiatric disease in 14% and pain management in 10%. The most frequent interventions investigated were antibiotics (20%), followed by coxibs (7.2%), antifungal (6.6%) and antiviral drugs (3.1%), and monoclonal antibodies (2.8%). To define the modified intention to treat authors fixed one condition in 125 (37.1%) reports, two in 137 (40.7%), three in 44 reports (13.1%); no condition was given in 27 reports (8%). Receiving one or more doses of the drug was the most required condition to be considered in mITT analysis: it represented 42% in trials with only one condition, 66% in trials with two conditions and 75% in trials with three conditions. In at least 21% of the reports specified, features such as clinical, microbiological, radiological or laboratory data had to be positive for inclusion. In addition baseline or post-baseline assessment, follow-up completeness, compliance, protocol violation, missing data were some of the many other descriptions used for consideration in the mITT analysis. Eighty-eight trials mentioned the intention to treat analysis too. In such trials ITT was used in primary analysis in no more than 16, whereas it was incorrectly defined in 34. Moreover, in 19 trials ITT though cited was not used in any analysis.
Conclusions: Trials testing new drugs seem likely to use modified intention to treat. The definition given to mITT in randomised controlled trials was irregular and arbitrary. In most trials authors did not provide adequate or plausible explanations for using mITT instead of ITT. MITT analysis allows a subjective approach in entry criteria, which is highly hazardous to manipulation and consequently to bias.