“Masking” (or “blinding”) refers to the steps taken to ensure that all persons involved in a trial are unaware of the type of treatment that each participant receives. In general, people related to a trial who might be influenced by knowing what treatment each participant is receiving include patients, investigators, care providers, outcome assessors, data collectors, data analysts, and any other trial staff.
The term “single blind” indicates that only patients or investigators are unaware of the intervention; “double blind” indicates that both patients and investigators are blind to the assignment. Blinding can also be extended to include other personnel and data analysts.
Blinding is usually feasible when interventions are similar or can be made to appear similar (ie, a placebo for drug trials). However, there are situations when blinding is not feasible, and, depending on the intervention and the type of outcome, bias can be introduced. The degree and direction of bias in unblinded trials depend on any prejudices for or against the intervention of the staff conducting the trial. If the staff members are neutral toward the intervention, it is unlikely that lack of masking will bias the trial results. Bias introduced by lack of masking tends to exaggerate the effects of new treatments compared with trials in which masking is applied.
Bias from lack of blinding can be generated at the patient level and the investigator or staff level ( Table ).
|Type of bias||Patient level||Investigator level|
|Bias in study management||✓|
|Biased data management and analysis||✓|
At the patient level:
Postrandomization selection bias. A patient who is unhappy with the assigned treatment might be less cooperative, not follow directions, or drop out of the study. For example, in a trial comparing oral hygiene levels with a standard toothbrush and an electric toothbrush, patients who are enthusiastic about the potential effects of electric toothbrushing but assigned to the standard group might be less cooperative and follow the protocol less diligently.
Observer bias. Patients who know their treatment might respond more or less favorably to the intervention, depending on their predisposition to this treatment. Observer bias is potentially worse when the outcome is subjective, because patients might consciously or unconsciously give better scores to the preferred treatment or modify their behavior (Hawthorne effect) in a way to give the expected answer. For example, reporting of pain levels might be influenced by whether the assigned treatment is favored by the patient receiving it.
At the investigator or staff level:
Selection bias. Knowledge of treatment allocation might allow investigators to guess the next allocation, thus influencing their ability to assign future allocations without bias.
Postrandomization selection bias. Investigators and other personnel might exclude patients from the trial after randomization, applying different standards between the treatment arms.
Bias in study management and concomitant care. Investigators and other personnel might follow more closely or more frequently, or provide better care to the preferred treatment group, thus potentially overestimating the effect of the therapy assigned to that group.
Observer bias. Unblinded investigators, depending on their predisposition to the treatment, might record outcomes in a more optimistic manner for the favored group. Investigators might round outcome values up or down, and they might repeat measurements when unexpected values are recorded. With subjective outcomes, the investigators might coerce patients consciously or unconsciously to give the desired response. Also, recording of side effects might be biased in favor of the preferred treatment.
Biased data management and analysis. Unmasked data analysts might introduce bias on the analytic strategies to be used. Analysts might select favorable time points, outcomes, and subgroups, or they might select biased handling of missing data and analyses that emphasize the desired results.
Some measures can reduce bias when blinding is not feasible. If blinding of the assignments is not possible for the patient and the investigator delivering the treatment, blinding might still be possible for other personnel, such as follow-up care providers, outcome assessors, and data analysts. Important measures to take to reduce bias from lack of blinding include establishment of standardized procedures and training and calibration of the staff associated with the trial.
In orthodontic research, blinding—especially of the investigator delivering the intervention—is frequently not feasible; however, masking can often be applied in other aspects of the trial. For example, if we are comparing treatment effects of different fixed appliances and the outcomes are measured on dental casts, scraping the appliances off the casts permits blinded measurements, especially when an independent assessor other than the person delivering the intervention does the measurements. If cephalometric measurements are made as part of the data collection, some masking of the films can be done. For example, in a study comparing the effects of headgear and functional appliances on Class II correction, the maxillary molar area can be masked so that the outcome assessors will not know who had molar bands and presumably wore headgear, and who used a functional appliance. Alternatively, removal of the bands will also allow unbiased outcome assessment.
A third example might be encountered when studying pain. If pain scores are recorded after taking various medications for orthodontic discomfort, the use of a placebo is an easy way to remedy the problem. If a placebo is not feasible, recording of pain levels by an independent assessor is likely to reduce or remove bias, at least at the outcome assessment level.
The next articles will discuss sample sizes for randomized clinical trials.
Blinding is important for valid results, especially when the outcomes are subjective.
Blinding is not always feasible at the patient or investigator level, but it might be feasible during outcome assessment and data handling and analysis.