I have data on a few millions of patients, about 1000 of them are cases, that is, they were diagnosed having a certain disease at some timepoint in their life.
I think I basically have two options: 1) match to each case a number of controls (e.g. based on age at diagnosis, gender, etc.) 2) Use all the non-cases as a control cohort.
However, someone came with a different proposal: Take all the individuals that can be matched to each case, and pool them together. Since many of the non-cases can be matched to more than one case patient, this will result with a large data set that will containn much more observations than the original data set (in my case, about 4 times larger). This is beecause many of the non-cases are included in the final data set more than once. More over, since one of the proposed matching variables was age at diagnosis, it was argued that a patient who was diagnosed at age 26, for example, can serve as a control for a patient who was diagnosed at age 25.
I cannot find any justification to this approach, but on the other hand I was not able to provide convincing arguments (not that I do not have such arguments, only that the person with that suggestion was not convinced).
My question is if there is any good justification for this inflation of the control cohort.