Negative control outcomes
But there’s a cheaper solution, that can be applied to a study based on electronic medical records: negative control outcomes. Essentially, this means you pick a different outcome: one which requires an equal level of subjectivity or difficulty in diagnosis, but which you are very sure is not at all in any way caused by your exposure of interest. Then, you redo your analysis with this negative control outcome. If the results suggest a relationship between the exposure and the negative control outcome, then your study definitely has detection bias.