We know how to deal with random noise in the data and random omissions in the data. However, if data is completly fabricated with no input from reality, then it

We know how to deal with random noise in the data and random omissions in the data. However, if data is completly fabricated with no input from reality, then it's not possible to tell anything about reality from that data, only parameters of the (fictional) model used to generate that data. We can determine the assumptions about the disease which were encoded in that model, we can't validate whether these assumptions match reality.