A test or treatment for a disease may perform differently according to some characteristic of the study participant, which itself may influence the likelihood of disease detection or the effectiveness of the treatment. Detection bias can occur in trials when groups differ in the way outcome information is collected or the way outcomes are verified.
Larger men have bigger prostates, which makes diagnosing prostate cancer via biopsy more difficult (it is harder to hit the target). Therefore, men with larger prostates are less likely to be accurately diagnosed with prostate cancer. Thus, a real association between obesity and prostate cancer risk may be underestimated.
This was shown by researchers who found that obesity increased prostate cancer risk. Without accounting for the size of the prostate, however, the relationship between obesity and prostate cancer was under-estimated.
Underestimation of the association between obesity and overall Prostate Cancer risk may exist in the literature. Rundle A et al 2017.
Detection bias can either cause an overestimate or underestimate of the size of the effect. For example, a recent systematic review showed on average non-blinded outcome assessors in randomised trials exaggerated odds ratios by 36%.
Other studies have suggested that rates of second breast cancers may be higher among women taking statins and lower among women taking antibiotics than a comparison group of women not taking these treatments. In a cohort study of breast cancer survivors, there were systematic differences in how much screening the women received, depending on which medicines they were taking. This meant that any associations observed might be affected by detection bias.
A cohort study investigating the relationship between smoking and risks of basal cell or squamous cell cancer found that current smokers had significantly lower risks of basal cell carcinoma, but higher risks of squamous cell carcinoma. Former smokers had similar risks for each cancer as did never smokers. However, when they looked more closely, the researchers discovered that current smokers had had fewer skin examinations and procedures than never smokers.
Intervention studies should be designed to ensure that all groups have an equivalent chance of being affected by known factors that influence detection. The use of randomisation in intervention studies also aims to generate groups equivalent in unknown factors. In observational studies, potential sources of detection bias should be sought out, and if identified, adjusted for or stratified by to clarify the observed associations of interest.
Statistical adjustment for perceived differences can be used to help reduce the problem, for example adjusting for age to avoid discrepancies in age leading to a spurious observation of association/effect. Detection bias can also be due to the knowledge of the allocated interventions by outcome assessors; therefore outcome assessors should be blinded to the intervention.