Response to letter by Berger: The success of masking should be tested routinely and correctly

Abstract

Bello, Moustgaard, and Hrobjartsson [[1]Bello S. Moustgaard H. Hrobjartsson A. The risk of unblinding was infrequently and incompletely reported in 300 randomized clinical trial publications.J Clin Epidemiol. 2014; 67: 1059-1069Abstract Full Text Full Text PDF PubMed Scopus (26) Google Scholar] pointed out that unmasking can lead to a host of biases, including cointervention bias, attrition bias, response bias, and observer bias. Crucially, unmasking can also lead to yet another bias, specifically selection bias. The ability of perfect masking to preclude the possibility of selection bias means that any test for selection bias also doubles as an objective test for the success of masking [[2]Berger V.W. On the insufficiency of reporting masking.J Rehabil Med. 2013; 45: 221Crossref PubMed Google Scholar]. The Berger–Exner test of selection bias [[3]Berger V.W. Exner D.V. Detecting selection bias in randomized clinical trials.Control Clin Trials. 1999; 20: 319-327Abstract Full Text Full Text PDF PubMed Scopus (120) Google Scholar] relies on neither perfect memory nor perfect honesty (and note that if investigators were able to decipher the allocations, then they may have no incentive to implicate themselves by admitting this, so asking investigators to guess which patients received which treatments is highly unreliable). Unmasking “allows for” selection bias but does not guarantee it, so one limitation is that the Berger–Exner test can demonstrate unmasking but cannot by itself prove that masking was perfect. Yet, given the absence of any better analysis, the Berger–Exner test should be used routinely any time the success of the masking is in question [4Mickenautsch S. Fu B. Gudehithlu S. Berger V.W. Accuracy of the Berger-Exner test for detecting third-order selection bias in randomized controlled trials: a simulation-based investigation.BMC Med Res Methodol. 2014; 14: 114-123Crossref PubMed Scopus (14) Google Scholar, 5Berger V.W. Selection bias and covariate imbalances in randomized clinical trials. John Wiley & Sons, Chichester2005Crossref Scopus (190) Google Scholar]. And when would this be the case? The better question would be when would this “not” be the case? Bello, Moustgaard, and Hrobjartsson [[1]Bello S. Moustgaard H. Hrobjartsson A. The risk of unblinding was infrequently and incompletely reported in 300 randomized clinical trial publications.J Clin Epidemiol. 2014; 67: 1059-1069Abstract Full Text Full Text PDF PubMed Scopus (26) Google Scholar] noted that often it is just assumed that masking is perfect unless it is somehow known (in the absence, of course, of any effort at ascertainment) to be otherwise. An unsubstantiated claim of masking should never be accepted as convincing [[2]Berger V.W. On the insufficiency of reporting masking.J Rehabil Med. 2013; 45: 221Crossref PubMed Google Scholar], especially given the ready availability of an objective analysis [3Berger V.W. Exner D.V. Detecting selection bias in randomized clinical trials.Control Clin Trials. 1999; 20: 319-327Abstract Full Text Full Text PDF PubMed Scopus (120) Google Scholar, 4Mickenautsch S. Fu B. Gudehithlu S. Berger V.W. Accuracy of the Berger-Exner test for detecting third-order selection bias in randomized controlled trials: a simulation-based investigation.BMC Med Res Methodol. 2014; 14: 114-123Crossref PubMed Scopus (14) Google Scholar] to support (albeit without being able to actually prove) this claim, which may be used alone or in conjunction with other measures of the success of masking. As an aside, it is also worth noting that if it is suspected that perfect masking may be unattainable, then it might be a good idea to employ de facto masking [[6]Berger V.W. De facto masking and other measures to prevent contamination.J Clin Epidemiol. 2012; 65: 1236Abstract Full Text Full Text PDF PubMed Scopus (10) Google Scholar] to minimize the impact of unmasking. This would involve concealing which treatment is the active one and which is the control, for example, by augmenting the control group with special information or a special diet or special fitness training that may be considered to be an active treatment in its own right, although not affecting the primary endpoints of the trial. The risk of unblinding was infrequently and incompletely reported in 300 randomized clinical trial publicationsJournal of Clinical EpidemiologyVol. 67Issue 10PreviewTo assess the proportion of clinical trials explicitly reporting the risk of unblinding, to evaluate the completeness of reporting on unblinding risk, and to describe the reported procedures involved in assessing unblinding. Full-Text PDF Re: “A note on unmasking” (Letter commenting on: J Clin Epidemiol. 2014;67:1059-69)Journal of Clinical EpidemiologyVol. 69PreviewDr. Berger states that “unmasking” may lead to allocation bias (“selection bias” is his term) [1]. In our article on the reporting of risk of unblinding in trial publications [2], we followed a convention of distinguishing between “blinding” and “allocation concealment.” Allocation concealment is usually understood as the procedures to mask the person responsible for patient inclusion to the allocation sequence (ie, before and during patient inclusion). Blinding is usually understood as the procedures to mask investigators and patients to the allocated interventions (ie, after patient inclusion). Full-Text PDF