Article Text

THU0416 Attrition Bias in Rheumatoid Arthritis Randomised Trials with Different Modified Intention-To-Treat Approaches: A Meta-Epidemiological Study
  1. A. Dossing1,2,
  2. S. Tarp1,
  3. D.E. Furst3,
  4. C. Gluud4,
  5. J. Beyene5,
  6. B.B. Hansen1,
  7. H. Bliddal1,
  8. R. Christensen1
  1. 1Musculoskeletal Statistics Unit, The Parker Institute, Department of Rheumatology, Copenhagen University Hospital, Bispebjerg and Frederiksberg
  2. 2Department of Clinical Medicine, Faculty of Medical and Health Sciences, University of Copenhagen, Copenhagen, Denmark
  3. 3David Geffen School of Medicine, Los Angeles, United States
  4. 4Copenhagen Trial Unit, Centre for Clinical Intervention Research, Rigshospitalet, Copenhagen University Hospital, Copenhagen, Denmark
  5. 5Department of Clinical Epidemiology & Biostatistics, Faculty of Health Sciences, McMaster University, Hamilton, Ontario, Canada


Background The intention-to-treat (ITT) principle preserves randomisation and prevents attrition bias [1]. Frequently data are missing [2] and the ITT analysis is executed by means of non-responder imputation (NRI) [3]. Deviations from ITT, referred to as modified ITT (mITT), excludes a restricted patient-group post-randomisation and is commonly used [4].

Objectives 1) To examine if mITT is associated with a different effect size compared to ITT as evaluated by the composite index ACR20. 2) To examine how imputation and different ITT deviations affect effect size.

Methods A systematic search in MEDLINE, EMBASE, and CENTRAL was performed. Inclusion criteria: Blinded RCTs investigating approved biological treatments for RA; reporting ACR20 as an outcome. Exclusion criteria: Open-label RCTs. A meta-epidemiological random-effects analysis was used to calculate the OR of pre-specified trial characteristics. Formal tests of interaction were performed between intervention groups (active or control) and study characteristics. Statistical analyses were based on GLIMMIX procedure in SAS. Study protocol [5].

Results Among 5,237 references, 87 RCTs were included and analysed (36,924 randomised patients and 36,256 analysed patients), with 185 active intervention groups and 87 control groups.

As illustrated in Table 1, empirical evidence suggests that the choice of the analysis population affects the net benefit of an intervention (p=0.0024); although the OR for the mITT and ITT populations were comparable. The net benefit is further significantly affected by the withdrawal population considered to be non-responders (p=0.0029). The significant differences reflects an over- or underestimation of treatment, with higher OR favouring the intervention.

Table 1

Conclusions The size of treatment effect may be biased based on the population analysed, and the population to which NRI is applied. Our results suggest, however, that the recommended statistical ITT approach is comparable to mITT.


  1. Higgins J. P., et al. BMJ 343 (2011):d5928.

  2. Boers M. Arthritis Rheum. 59.1 (2008): 2.

  3. EMA. Guideline on Missing Data in Confirmatory Clinical Trials. London: European Medicines Agency (EMA), 2010.

  4. Montedori A., et al. Trials 12 (2011): 58.

  5. Dossing A, et al. PROSPERO CRD42013006702, 2013.

Disclosure of Interest : None declared

DOI 10.1136/annrheumdis-2014-eular.2210

Statistics from

Request permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.