Bias
Bias is a systematic deviation of a study result from the effect that would be obtained under the conditions defined by the relevant study design and target estimand. In clinical research, bias can arise from the selection of participants, conduct of the study, measurement or classification of variables and outcomes, handling of missing data, or selection of results for reporting.
Bias is a methodological concept rather than a GCP or regulatory term with a single generally applicable definition. The terminology and classification of bias vary according to study design and the risk-of-bias assessment method used.
Types of bias
Common categories of bias in clinical research include:
Selection bias
Selection bias arises from systematic differences associated with how participants enter a study, are allocated to intervention groups, remain under observation, or are included in an analysis. In randomized trials, inadequate concealment of the allocation sequence can result in differences between intervention groups at baseline and is classified as selection bias in the Cochrane framework.
Information bias
Information bias arises from systematic differences in the collection, measurement, classification, or recording of information between study groups or participants. It may affect exposure, intervention, outcome, or covariate measurements.
Measurement bias is commonly used as a related term when the source of systematic error is the measurement process.
Observer bias
Observer bias occurs when the expectations, knowledge, or other characteristics of an observer systematically influence the assessment or recording of study data or outcomes. Blinding of outcome assessors can reduce some forms of observer bias.
Detection bias
Detection bias refers to systematic differences in how outcomes are ascertained or assessed between study groups. In randomized-trial methodology, detection bias is associated particularly with differences in outcome assessment and lack of blinding of outcome assessors. Cochrane includes detection bias among the principal sources of bias assessed in trials.
Performance bias
Performance bias refers to systematic differences in care or behavior between intervention groups after allocation, other than differences attributable to the interventions being compared. In randomized trials, knowledge of intervention assignment by participants or study personnel can contribute to performance bias. Cochrane includes performance bias among the principal sources of bias in trials.
Attrition bias
Attrition bias arises when missing outcome data or losses to follow-up differ systematically between study groups or are otherwise related to the outcome of interest. The resulting analysis may differ from the effect that would have been estimated if the relevant outcome data had been available.
Reporting bias
Reporting bias occurs when the availability or characteristics of reported results depend systematically on the findings or other characteristics of the results. Selective reporting of outcomes, analyses, or time points can produce reporting bias.
Publication bias is a related form of bias in which the likelihood that a study is published depends on characteristics of the study or its results. Publication bias is generally considered at the level of the body of evidence rather than as a bias arising solely from the conduct of an individual study.
Confounding
Confounding occurs when an extraneous variable is associated with both an exposure or intervention and an outcome and distorts the estimated association or effect.
Confounding is generally distinguished from bias in methodological terminology. Confounding is a form of distortion caused by the relationship between variables, whereas bias refers more broadly to systematic deviation arising from flaws or systematic processes in study design, conduct, analysis, or reporting. Risk-of-bias frameworks for non-randomized studies nevertheless commonly assess confounding as a specific domain because uncontrolled confounding can produce systematic distortion of an estimated effect.
Risk of bias
Risk of bias is the likelihood that a study result is systematically distorted because of problems in the design, conduct, analysis, or reporting of the study. Risk-of-bias assessment evaluates specific domains rather than assigning a single generic label to a study without considering the mechanisms that could produce distortion.
The domains used for assessment depend on the study design and the assessment tool. For example, the Cochrane approach to randomized trials identifies selection, performance, detection, attrition, and reporting bias among its principal domains. RoB 2 is designed for randomized trials, while ROBINS-I is designed for non-randomized studies of interventions and assesses domains including confounding, selection into the study, classification of interventions, missing data, measurement of outcomes, and selection of the reported result.
References
Cochrane. 2011. Cochrane Handbook for Systematic Reviews of Interventions, Chapter 8: Assessing risk of bias in included studies. Available at: https://training.cochrane.org/handbook/current/chapter-08
Sterne JAC, Hernán MA, Reeves BC, et al. 2016. ROBINS-I: a tool for assessing risk of bias in non-randomized studies of interventions. BMJ 355:i4919. Available at: https://www.riskofbias.info/welcome/home/original-2016-version-of-robins-i