I agree, it is definitely approachable, I like to think very achievable. But the problem with coming up with a prediction for tumor growth for some patient 'A' is that your algorithm might be structured correctly and work great but the data you've/I/we used is biased and therefore 'A' wouldn't receive an inaccurate prediction as to how much their tumor might grow. Consequentially 'A' could be given treatment that is too heavy and unnecessary or too low and ineffective.