Are you asserting that it is overfit? Also, before they tested on the other smaller dataset from a different source, aiui, they also trained only on the earlie

Are you asserting that it is overfit? Also, before they tested on the other smaller dataset from a different source, aiui, they also trained only on the earlier subset of the first source, and used the later portion from the first source (with no overlap in patients) for the testing. (also, I'm not sure that 252 is really all that small?)