BIAS (DEFINITION):
A systematic deviation of observations, results, inferences, or conclusions from the truth1. Bias is considered to be an unfavourable element of medical science.
The term ‘Bias’ is used in many disciplines, including statistics, where is often used to imply a deviation from the truth. However, the term ‘Bias’ in the context of medical science, implies this deviation from the truth is systematic/predetermined/follows a non-random structure. Thus, ‘Bias’ is often used as a synonym for systematic error. To clarify these differences across disciplines, sometimes the term ‘systematic bias’ is found in the literature, but not generally preferred in medical science.
In the context of medical science, Bias should not to be confused with random error; which is a non-systematic deviation from the truth. Bias is sometimes thought of as an ‘unfairness’ in the balance of information across, or within study groups. Bias arises due to how participants/data are selected into studies, how researchers measure participants/data, and how researchers interpret information in research studies. Thus, Bias is due to problems, or mistakes, in the process of research1. Bias may arise to due overt, covert, conscious, or unconscious errors in study processes.
It has been proposed2 that all types of Biases can be classified into two high-level categories: 1) Information Biases, and 2) Selection Biases; depending on whether the errors are related to how participants/data were selected or how participants/data were measured/interpreted.
Theoretically, Bias can be prevented or mitigated during study conduct, but arguably cannot be truly corrected/removed using statistical techniques; as many causes of Bias cannot be measured precisely. Further, some argue that Bias is only obvious in retrospect3. Thus, it is assumed that all medical research studies are Biased to some degree; so it is the job of the scientist to; 1) minimize the threat of Bias in a research study as much as possible, 2) evaluate the degree of Bias in the study after mitigation, and 3) convince the users of the research, that the degree of Bias present is small enough to not affect the conclusions drawn, or disclose that the degree of Bias is so large that it is difficult to draw any meaningful conclusions from the study. Also see: Systematic Bias, Debiasing, Design Bias, Selection Bias, Information Bias, and Confounding Bias.
References:
1. Porta M, ed. A Dictionary of Epidemiology. Sixth ed. New York, NY: Oxford University Press 2014. (Link to Reference)
2. Celentano DD, Szklo M. Gordis Epidemiology. Sixth ed. Elsevier 2018. (Link to Reference)
3. Kaptchuk TJ. Effect of interpretive bias on research evidence. BMJ. 2003;326(7404):1453-5. (Link to Reference)