Selection bias

A selection bias is any type of bias that results in a sample population that is not representative of the population at large or the population of interest, or that results in study groups that are systematically different from each other. For example, the type of individual who is willing to volunteer for a study on diet and weight loss might be significantly different from the population of patients struggling to lose weight.

A classic example of this is the use of hormone replacement therapy. Multiple observational trials indicated that hormone replacement therapy reduced the rate of coronary artery disease. However, when studied in a randomized fashioned, HRT actually increased the rate of CAD. The difference was probably explained by women more dedicated to their health self-selecting into the hormone replacement group in observational trials.

There are a number of sub-types of selection bias:

A primary research tool aimed at preventing selection bias is randomization. Other aspects to consider when appraising a trial are the inclusion and exclusion criteria (do these patients looks like the patients I am treating?), how the different groups compare at baseline, evidence of unblinding or improper allocation concealment, and how data from missing patients was handled. An intention to treat analysis is also important, because it prevents selection bias from occurring after randomization.

The existence of research bias does not indicate wrongdoing on behalf of the researchers. For example, a researcher could run a methodologically perfect weight loss trial, but because people that volunteer for such trials are systematically different from the general population, the results could be impacted by selection bias, limiting the generalizability of the results.



Odgaard-Jensen J et al.Randomisation to protect against selection bias in healthcare trials. Cochrane Database Syst Rev 2011 Apr 13; (4): MR000012.

Sackett DL. Bias in analytic research. Journal of chronic diseases. 1979; 32(1-2):51-63. PMID: 447779