Attrition bias is the potential skewing of data/results that arises due to the attrition, or dropout rate, in a study. A certain amount of people leaving a study is normal. Some move away or move within the area and get lost to follow-up, some may die (depending on the condition, their age, the length of the study, etc.), some have work or family or other circumstances come up that require their time, some cannot manage the side effects of a condition, and some simply choose never to show up again. What should draw your attention is any study where attrition is particularly high (such as more than a third of participants) or where attrition is substantially higher in one arm than in another.
Deeper dive
Attrition bias is just one of many biases that can potentially distort findings, but it’s one of the easier ones for journalists to look for since studies should report the starting and ending number of participants in each study arm. If an intervention group, for example, has much higher attrition than the placebo arm, that may call into question how reliable the statistical analysis is at comparing effects between the two unless the researchers specifically account for those attrition differences, if they’re able to. Either way, it’s worth at least asking about. Sometimes higher attrition in an intervention group occurs for reasons directly related to the intervention, such as low tolerance for serious or bothersome side effects or difficulty adhering to a particular regimen.
While attrition bias is especially important to look at in randomized controlled trials, you should also look at differences in groups in case control and other observational studies. If one group is substantially higher than another, there may be confounding factors that influence the findings and which the researchers should either account for in their statistical analysis, list as a limitation of the study, or at least speculate about in their discussion.
For example, consider a study looking at drinking in pregnancy that includes one group of women who completely abstain, one group of women who have a few drinks during pregnancy, and one group who drink frequently during pregnancy. If attrition is twice as high in the group of abstainers than in the other two groups, it’s important to ask why and how that might affect the results. Is there some characteristic specific to abstainers that influenced their higher rate of attrition? If so, could that characteristic also influence the outcomes of their child? Enough to confound the comparison to the children born to women in the other two groups? Does that characteristic relate to why the women abstain in the first place?
Sometimes it just works out that one group has higher attrition than another, and sometimes attrition has little effect on the results, but more often than not, noticeable differences or large amounts of attrition are likely to introduce bias that the researchers should address in their statistical analysis and/or discussion.