Skip to main content

Evidence-Based Medicine: Bias

Bias (Systematic Error)

Any process or effect at any stage of a study from its design to its execution to the application of information from the study, that produces results or conclusions that differ systematically from the truth. Bias can be reduced only by proper study design and execution and not by increasing sample size (which only increases precision by reducing the opportunity for random chance deviation from the truth). Almost all studies have bias, but to varying degrees. The critical question is whether or not the results could be due in large part to bias, thus making the conclusions invalid. Observational study designs are inherently more susceptible to bias than are experimental study designs.

Randomization in Clinical Trials

Types of Bias

Confounding is the distortion of the effect of one risk factor by the presence of another. Confounding occurs when another risk factor for a disease is also associated with the risk factor being studied but acts separately. Age, breed, gender and production levels are often confounding risk factors because animals with different values of these are often at different risk of disease. As a result of the association between the study and confounding risk factor, the confounder is not distributed randomly between the group with the study risk factor and the control group without the study factor. Confounding can be controlled by restriction, by matching on the confounding variable or by including it in the statistical analysis.

Confounding bias is a systematic error due to the failure to account for the effect of one or more variables that are related to both the causal factor being studied and the outcome and are not distributed the same between the groups being studied. The different distribution of these "lurking" variables between groups alters the apparent relationship between the factor of interest and the outcome. Confounding can be accounted for if the confounding variables are measured and are included in the statistical models of the cause-effect relationships.

Systematic error that occurs when an association observed between variables representing group averages is mistakenly taken to represent the actual association that exists between these variables for individuals. This bias occurs when the nature of the association at the individual level is different from the association observed at the group level. Data aggregated from individuals (e.g. census averages for a region) or proxy data from other sources (e.g., the amount of alcohol distributed in a region is a proxy for the amount of alcohol by individuals in that region) are often easier and less expensive to acquire than are data directly from individuals.

Systematic error that occurs when, because of the lack of blinding or related reasons such as diagnostic suspicion, the measurement methods (instrument, or observer of instrument) are consistently different between groups in the study.

  • Screening Bias: The bias that occurs when the presence of a disease is detected earlier during its latent period by screening tests but the course of the disease is not be changed by earlier intervention. Because the survival after screening detection is longer than survival after detection of clinical signs, ineffective interventions appear to be effective unless they are compared appropriately in clinical trials.

Systematic errors of interpretation made during inference by the user or reader of clinical information (papers, test results, ...). Such biases are due to clinical experience, tradition, credentials, prejudice and human nature. The human tendency is to accept information that supports pre-conceived opinions and to reject or trivialize that which does not support preconceived opinions or that which one does not understand. (JAMA 247:2533)

Systematic error that occurs when, because of design and execution errors in sampling, selection, or allocation methods, the study comparisons are between groups that differ with respect to the outcome of interest for reasons other than those under study.

The bias that occurs in a prospective study when individuals are found and enrolled in such a fashion that unintended systematic differences occur between groups at the beginning of the study (stage of disease, confounder distribution). Cohort studies are susceptible to zero-time bias if the cohort is not assembled properly.

  • Non-differential Bias: Opportunities for bias are equivalent in all study groups, which biases the outcome measure of the study toward the null of no difference between the groups.
  • Differential Bias: Opportunities for bias are different in different study groups, which biases the outcome measure of the study in unknown ways. Case-control studies are highly susceptible to this form of bias between the case and control groups.

Blinding (Masking)

Blinding is those methods to reduce bias by preventing observers and/or experimental subjects involved in an analytic study from knowing the hypothesis being investigated, the case-control classification, the assignment of individuals or groups, or the different treatments being provided. Blinding reduces bias by preserving symmetry in the observers’ measurements and assessments. This bias is usually not due to deliberate deception but is due to human nature and prior held beliefs about the area of study.


A placebo is the shame treatment used in a control group in place of the actual treatment. If a drug is being evaluated, the inactive vehicle or carrier is used alone so it is as similar as possible in appearance and in administration to the active drug. Placebos are used to blind observers and, for human trials, the patients to which group the patient is allocated.

Source Information