Statisics vs. n-of-1
Oncologists are good at giving prognostic life expectancy statistics for a population of patients with a specific cancer
(with associated grade, stage, biomarkers, etc).
However, we are not good (usually) at predicting the life expectancy of an individual patient. This is improving both in the the untreated patient (prognostic markers, etc) and after treatment but is still imperfect.