Studies are assessed as at unclear risk of bias when too few details are available to make a judgement of ‘high’ or ‘low’ risk; when the risk of bias is genuinely unknown despite sufficient information about the conduct; or when an entry is not relevant to a study (for example because the study did not address any of the outcomes in the group of outcomes to which the entry applies). When the first reason dominates, it is reasonable to assume that the average bias in results from such studies will be less than in studies at high risk of bias, because the conduct of some studies assessed as unclear will in fact have avoided bias. Limited evidence from empirical studies that examined the ‘high’ and ‘unclear’ categories separately confirms this: for example, the study of Schulz et al. found that intervention odds ratios were exaggerated by 41% for trials with inadequate concealment (high risk of bias) and by 30% for trials with unclear concealment (unclear risk of bias) (Schulz 1995b). However, most empirical studies have combined the ‘high’ and ‘unclear’ categories, which were then compared with the ‘low’ category.
It is recommended that review authors do not combine studies at ‘low’ and ‘unclear’ risk of bias in analyses, unless they provide specific reasons for believing that these studies are likely to have been conducted in a manner that avoided bias. In the rest of this section, we will assume that studies assessed as at low risk of bias will be treated as a separate category.