Sackett proposed that popularity bias is: “The admission of patients to some practices, institutions or procedures (surgery, autopsy) is influenced by the interest stirred up by the presenting condition and its possible causes…” (Sackett 1979)
The original definition from David Sackett allows for some ambiguity. Popularity bias could result from an increased awareness of the condition within the general population, making a condition “fashionable” and more likely to be cited as the cause of perhaps somewhat unspecific symptoms. For example, in the last decade, population awareness of coeliac disease (an autoimmune condition due to gluten intolerance) has led to large numbers of media articles on coeliac disease and related putative gluten intolerance syndromes. Sufferers of abdominal discomfort might be therefore more likely to suspect coeliac disease as the cause, and present to healthcare settings and seek investigations.
Also, a provider of health services might be the instigator of increased interest/popularity in a certain disease or condition. Advertisements by health providers for testing for conditions that otherwise might go undiagnosed, or otherwise might cause no overt health problems, could lead to an increase in “popularity” of presenting with certain symptoms or concern about a condition.
Both instances lead to selection bias. If a study is affected by popularity bias it could influence the results if certain groups of people were admitted to certain interventions or observations and other groups, which could happen over time, and these differences between groups might affect the study’s outcomes.
There is also a tendency for more popular items to be recommended more frequently and less popular ones rarely.
Jade Goody’s decision to go public about her cancer diagnosis in 2009 led to a massive surge in the number of young women undergoing screening, nearly 500,000 extra women turned up for smear tests. However, by 2017, when the public messages had dwindled, the number of screenings has reached a 20-year low. (Guardian 2018)
As far as we aware there have been no formal studies of the impact of popularity bias. The impact of selection bias is discussed in our catalogue entry selection bias.
Randomization with proper allocation concealment aims to reduce selection biases and could help prevent popularity bias. Furthermore, studies with historical controls should be clear about the potential for selection bias and report how they have controlled for its presence. Any analysis of incidence or uptake of services should be acutely aware of the public messages that might influence popularity. They should seek to analyse data over the longer term to smooth out any variability in uptake.