Table 4, Principle domains of bias in randomised controlled trials - Multimorbidity: Assessment, Prioritisation and Management of Care for People with Commonly Occurring Multimorbidity

Limitation	Explanation
Selection bias (sequence generation and allocation concealment)	If those enrolling patients are aware of the group to which the next enrolled patient will be allocated, either because of a non-random sequence that is predictable, or because a truly random sequence was not concealed from the researcher, this may translate into systematic selection bias. This may occur if the researcher chooses not to recruit a participant into that specific group because of: knowledge of that participant's likely prognostic characteristics, and a desire for one group to do better than the other.
Performance and detection bias (lack of blinding of patients and healthcare professionals)	Patients, caregivers, those adjudicating or recording outcomes, and data analysts should not be aware of the arm to which patients are allocated. Knowledge of the group can influence: the experience of the placebo effect performance in outcome measures the level of care and attention received, and the methods of measurement or analysis all of which can contribute to systematic bias.
Attrition bias	Attrition bias results from an unaccounted for loss of data beyond a certain level (a differential of 10% between groups). Loss of data can occur when participants are compulsorily withdrawn from a group by the researchers (for example, when a per-protocol approach is used) or when participants do not attend assessment sessions. If the missing data are likely to be different from the data of those remaining in the groups, and there is a differential rate of such missing data from groups, systematic attrition bias may result.
Selective outcome reporting	Reporting of some outcomes and not others on the basis of the results can also lead to bias, as this may distort the overall impression of efficacy.
Other limitations	For example: Stopping early for benefit observed in randomised trials, in particular in the absence of adequate stopping rules. Use of unvalidated patient-reported outcome measures. Lack of washout periods to avoid carry-over effects in crossover trials. Recruitment bias in cluster-randomised trials.

Limitation

Explanation

Selection bias (sequence generation and allocation concealment)

If those enrolling patients are aware of the group to which the next enrolled patient will be allocated, either because of a non-random sequence that is predictable, or because a truly random sequence was not concealed from the researcher, this may translate into systematic selection bias. This may occur if the researcher chooses not to recruit a participant into that specific group because of:

knowledge of that participant's likely prognostic characteristics, and
a desire for one group to do better than the other.

Performance and detection bias (lack of blinding of patients and healthcare professionals)

Patients, caregivers, those adjudicating or recording outcomes, and data analysts should not be aware of the arm to which patients are allocated. Knowledge of the group can influence:

the experience of the placebo effect
performance in outcome measures
the level of care and attention received, and
the methods of measurement or analysis

all of which can contribute to systematic bias.

Attrition bias

Attrition bias results from an unaccounted for loss of data beyond a certain level (a differential of 10% between groups). Loss of data can occur when participants are compulsorily withdrawn from a group by the researchers (for example, when a per-protocol approach is used) or when participants do not attend assessment sessions. If the missing data are likely to be different from the data of those remaining in the groups, and there is a differential rate of such missing data from groups, systematic attrition bias may result.

Selective outcome reporting

Reporting of some outcomes and not others on the basis of the results can also lead to bias, as this may distort the overall impression of efficacy.

Other limitations

For example:

Stopping early for benefit observed in randomised trials, in particular in the absence of adequate stopping rules.
Use of unvalidated patient-reported outcome measures.
Lack of washout periods to avoid carry-over effects in crossover trials.
Recruitment bias in cluster-randomised trials.

From: 4, Methods

Multimorbidity: Assessment, Prioritisation and Management of Care for People with Commonly Occurring Multimorbidity.

NICE Guideline, No. 56.

National Guideline Centre (UK).

London: National Institute for Health and Care Excellence (NICE); 2016 Sep.

NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

Table 4Principle domains of bias in randomised controlled trials