Table 2

Potential sources of bias in clinical trials

Bias typeDefinition and examplesIdentifiableQuantifiable
Academic biasThe investigators leading study are advocates for the intervention.YY
Ascertainment biasUn-blinded study design in which the outcome evaluations are susceptible to unmasked observer detection bias. Open-label studies, such as imaging and device trials (without a sham) are susceptible to ascertainment biasYY/N
Comparison group biasIf incorrect control/sham group is chosen, the intervention may appear to be more, or less, effectiveYN
Fraud biasIntentional fraud (rare)YY
Funding availability biasFocus of studies on questions more readily funded (commercial interest)Y/NY/N
Hidden agenda biasStudy designed to demonstrate a prerequired answer.Y/NY/N
Intervention biasEffects of a learning curve when investigating a new technologyY/NY/N
Measurement biasMeasurement influences the respondent’s behaviour and responses, reflecting ‘response shift’ and relatedly a Hawthorne effect. This becomes relevant if there is an interaction between the intervention and the measurement tool (eg, a training effect)Y/NY/N
Observer biasPatients allocated to treatment arm followed more intensely/more favourablyY/NY/N
Publication biasPositive results are more likely to get publishedYY/N
Regulation biasOverly restrictive or permissive review boards confounding the path to first-patient inY/NY/N
Sample choice biasExclusion of minority groups (recruitment bias), older groups (age bias) and women (sex bias)YY/N
Selection biasExclusion of potentially eligible patientsYY/N
Selective reporting biasSelective reporting of positive resultsYY
Withdrawal biasHandling missing data: Are the number of withdrawals and their reasons stated in the report? Are the number of withdrawals similar in each of the groups, or not? Is the overall number of withdrawals comparable to the number of patients that contribute to a difference in the primary outcome?YY/
Wrong design biasIncorrect study design to answer a question (eg, a randomised study rather than post-approval outcome research)YY/N