8.3.4  Collecting information for assessments of risk of bias

Despite the limitations of reports, information about the design and conduct of studies will often be obtained from published reports, including journal papers, book chapters, dissertations, conference abstracts and web sites (including trials registries). Published protocols are a particularly valuable source of information when they are available. The extraction of information from such reports is discussed in Chapter 7. Data collection forms should include space to extract sufficient details to allow implementation of the Collaboration’s ‘Risk of bias’ tool (Section 8.5). When extracting this information, it is particularly desirable to record the source of each piece of information (including the precise location within a document). It is helpful to test data collection forms and assessments of risk of bias within a review team on a pilot sample of articles to ensure that criteria are applied consistently, and that consensus can be reached. Three to six papers that, if possible, span a range from low to high risk of bias might provide a suitable sample for this.

 

Authors must also decide whether those assessing risk of bias will be blinded to the names of the authors, institutions, journal and results of a study when they assess its methods. One study suggested that blind assessment of reports might produce lower and more consistent ratings than open assessments (Jadad 1996), whereas other studies suggested little benefit from blind assessments (Berlin 1997, Kjaergard 2001). Blinded assessments are very time consuming, they may not be possible when the studies are well known to the review authors, and not all domains of bias can be assessed independently of the outcome data. Furthermore, knowledge of who undertook a study can sometimes allow reasonable assumptions to be made about how the study was conducted (although such assumptions must be reported by the review author). Authors must weigh the potential benefits against the costs involved when deciding whether or not to blind assessment of certain information in study reports.

 

Review authors with different levels of methodological training and experience may identify different sources of evidence and reach different judgements about risk of bias. Although experts in content areas may have pre-formed opinions that can influence their assessments (Oxman 1993), they may nonetheless give more consistent assessments of the validity of studies than people without content expertise (Jadad 1996). Content experts may have valuable insights into the magnitudes of biases, and experienced methodologists may have valuable insights into potential biases that are not at first apparent. It is desirable that review authors should include both content experts and methodologists and ensure that all have an adequate understanding of the relevant methodological issues.

 

Attempts to assess risk of bias are often hampered by incomplete reporting of what happened during the conduct of the study. One option for collecting missing information is to contact the study investigators. Unfortunately, contacting authors of trial reports may lead to overly positive answers. In a survey of 104 trialists, using direct questions about blinding with named categories of trial personnel, 43% responded that the data analysts in their double-blind trials were blinded, and 19% responded that the manuscript writers were blinded (Haahr 2006). This is unlikely to be true, given that such procedures were reported in only 3% and 0% of the corresponding published articles, and that they are very rarely described in other trial reports.

 

To reduce the risk of overly positive answers, review authors should use open-ended questions when asking trial authors for information about study design and conduct. For example, to obtain information about blinding, a request of the following form might be appropriate: “Please describe all measures used, if any, to ensure blinding of trial participants and key trial personnel from knowledge of which intervention a participant had received.” To obtain information about the randomization process, a request of the following form might be appropriate: “How did you decide which treatment the next patient should get?” More focused questions can then be asked to clarify remaining uncertainties.