Does the modified intention-to-treat reporting affect the estimate of the treatment effect in meta-analyses? A meta-epidemiological study

Article type
Authors
Abraha I1, Montedori A1, Cozzolino F1, Orso M1, Luchetta L2, De Florio L3, Schünemann HJ4, Germani A5
1Regional Health Authority of Umbria, Perugia, Italy
2Local Health Unit 2, Foligno, Italy
3Local Health Unit 1, Perugia, Italy
4McMaster University, Canada
5Azienda Ospedaliera di Perugia, Perugia, Italy
Abstract
Background: The intention to treat (ITT) principle helps to preserve the balance of prognostic factors among allocated groups generated by randomized trials (RCTs). The prevalence of trials using modified intention to treat (mITT) may influence estimates of treatment effects.

Objectives: To investigate for any association between estimates of treatment effects among trials with respect to the approach used for analysis (ITT versus mITT).

Methods: We performed a computerized search of Medline and selected a random sample of systematic reviews published between 2006 and 2011 with the following inclusion criteria: (1) dichotomous data, and (2) presence of meta-analysis of at least two randomised trials with at least one reporting mITT. Within each systematic review, trials were classified according to the type of intention-to-treat analyses used as follows: (1) ITT, trials reporting the use of standard ITT analyses; (2) mITT, trials reporting the use of ‘modified intention-to-treat’ analyses; or (3) ‘no ITT’ trials not reporting the use of any intention-to-treat analyses. Risk of bias criteria [sequence generation, allocation concealment and blinding, the type of the center (single vs. multicenter trial), the presence of post-randomization exclusions] and funding source were evaluated for each trial. A meta meta-analysis was performed and ratio of odds ratios (ROR) were calculated comparing ‘mITT trials’ against ‘ITT trials’ (or ‘no ITT trials’). A ROR less than one implying that trials reporting mITT exaggerate intervention effect estimates.

Results: After screening 2268 abstracts, 377 full-text publications were analyzed and 201 meta-analyses remained for inclusion. Of these 56 had at least one RCT with a mITT reporting and overall included 804 trials.

Conclusions: Previous work reported that the proportion of statistical significance within RCTs reporting mITT was high. This study determines whether mITT represents a source of bias. Final results will be presented at the Colloquium.