Incidence of randomised trials with modified intention-to-treat analysis in the medical literature: a systematic review update

Article type
Authors
Orso M1, Abraha I1, Cozzolino F1, Cherubini A2, Rimland J2, Lombardo G3, Eusebi P1, Marchesi M4, Germani A4, Cirocchi R5, Montedori A1
1Regional Health Authority of Umbria, Italy
2Geriatrics and Geriatric Emergency Care, Italian National Research Center on Aging, Ancona, Italy
3Clinica Odontoiatrica, Università di Perugia, Italy
4Medicina Trasfusionale, Azienda Ospedaliera di Perugia, Italy
5Chirurgia Generale e Oncologica, Università di Perugia, Italy
Abstract
Background: It has been reported that trials using the modified intention-to-treat (mITT) analysis are increasingly being published in the medical literature. In 2010, CONSORT recommended abandoning any mention of intention-to-treat analysis in favour of a clear description of who was included in each analysis. Recently a meta-epidemiological study showed that trials deviating from intention-to-treat overestimated the treatment effect compared to trials using standard intention-to-treat (Abraha, BMJ, 2015). Consequently, we were interested whether mITT is still being used in randomised trials.
Objectives: To determine the incidence of mITT trials published between 2007 and 2014.
Methods: PubMed, Embase, the Cochrane Central Register of Controlled Trials (CENTRAL), ISI Web of Knowledge, Ovid, HighWire Press, Science-Direct, Ingenta, Medscape, BioMed Central, Springer, and Wiley, were searched to identify randomized trials that used the modified intention-to-treat as the primary analysis. The search strategy used the following terms: 'modified intention-to-treat', 'modified intent-to-treat' and 'modified ITT'.
Results: Using the reported search strategy, a total of 1951 records were retrieved; 250 records were randomly selected from these and the respective full-texts were obtained for evaluation. Trials that used only a modified intention-to-treat for a primary analysis amounted to 222 (88.8%). The remaining trials also used an intention-to-treat analysis (n = 24) or a per protocol analysis (n = 4). Overall, the number of mITT trials increased 3.4 times from 2007 to 2014. The incidence of mITT trials among all the published randomised trials increased from 0.5% in 2007 to 2.5% in 2014 (P value for trend = 0.010).
Conclusions: Trials that report a mITT are increasingly being published in the medical literature. The incidence is underestimated since the search was limited to trials that used the word 'modified' in the intention-to-treat description. The use of mITT should be abandoned in favour of a clear description of who was included in final analysis.