Performance Bias

Performance bias refers to systematic differences in the care provided to participants in a study, apart from the intervention being evaluated, that arise because participants, clinicians, or other providers know which treatment group a participant belongs to. It is a form of bias that occurs during the conduct of a trial when knowledge of group assignment influences behavior.

Performance bias threatens the internal validity of clinical trials by introducing confounding factors that are not part of the intended intervention. When clinicians know which patients are receiving the experimental treatment, they may unconsciously provide more attentive care, additional monitoring, or supplementary treatments to those patients, inflating the apparent effectiveness of the intervention. Conversely, they may provide more supportive care to patients in the control group out of concern that those patients are receiving a less effective treatment. Patients who know their group assignment may change their behavior or seek additional treatments outside the study protocol. (If they also report their symptoms differently, that is detection bias, because for self reported outcomes the patient is the outcome assessor.) These behavioral changes contaminate the comparison between groups, making it impossible to attribute observed differences solely to the intervention under study.

Performance bias arises through several mechanisms. Clinician behavior is a primary pathway: a surgeon who knows that a patient received a new implant may provide more intensive rehabilitation instructions; a nurse who knows that a patient is on placebo may be more attentive to complaints; a therapist delivering a psychosocial intervention cannot help but know which treatment they are providing and may invest more enthusiasm or effort in the intervention they believe is superior.

Patient behavior is another pathway. Patients who know they are receiving a new experimental treatment may be more motivated, more adherent, and more optimistic about their prognosis (a component of the placebo effect). Patients who know they are receiving the control treatment may be disappointed, less adherent, or more likely to seek alternative treatments outside the trial (co intervention), contaminating the control group.

The primary safeguard against performance bias is blinding (masking). In a double blind trial, neither participants nor clinicians know the treatment assignment, closing both the patient and the clinician pathways of performance bias for as long as the blinding holds. In a single blind trial (typically patient blinded), the patient pathway is addressed but the clinician pathway remains open. In an open label trial, neither pathway is controlled.

When blinding is not feasible (as in trials comparing surgery to medical therapy, or in behavioral interventions), several strategies can mitigate performance bias. Standardizing care protocols for all groups reduces variation in co interventions. Using objective outcome measures and blinded outcome assessors reduces detection bias, a separate problem, but does not stop the groups from receiving different care. Monitoring protocol adherence and documenting co interventions allows post hoc assessment of the extent of performance bias. The original Cochrane Risk of Bias tool assessed performance bias in a domain on blinding of participants and personnel; its successor, RoB 2, covers it in the domain on bias due to deviations from intended interventions.

Assess and address the risk of performance bias in the design of every clinical trial. Implement blinding whenever feasible. When blinding is not possible, document the strategies used to minimize performance bias and discuss the potential impact on results. When reading or appraising published trials, evaluate the risk of performance bias as part of your quality assessment, noting whether blinding was implemented and what additional safeguards were described.

A common mistake is assuming that randomization alone protects against performance bias. Randomization, together with allocation concealment, addresses selection bias (systematic differences in group composition at baseline) but does nothing to prevent differences in post randomization care; blinding, where feasible, is the main safeguard against those. Another error is claiming that a trial is double blind when the nature of the intervention makes effective blinding implausible (for example, comparing acupuncture to pharmacotherapy without a sham procedure). Failing to assess and report the risk of performance bias is a further oversight, particularly in systematic reviews that evaluate the quality of included trials. Finally, dismissing the importance of performance bias for "hard" outcomes such as mortality overlooks the fact that even mortality can be affected by differential care, co interventions, and treatment seeking behavior.

A randomized trial compares a new physiotherapy protocol to standard care for recovery after knee replacement. Because participants and physiotherapists cannot be blinded to the treatment assignment, the trial is open label. To mitigate performance bias, the protocol specifies that all participants receive identical postoperative medical care, that the number of physiotherapy sessions is standardized, and that outcome assessors are blinded. Nevertheless, a monitoring log reveals that physiotherapists delivering the new protocol spend an average of 8 minutes more per session than those delivering standard care, and patients in the new protocol group report higher satisfaction with their physiotherapists' engagement. The researchers note this as a limitation, acknowledging that the performance difference may contribute to the observed outcome advantage of the new protocol.