You're forgetting a much simpler scenario that could wildly throw off survival statistics: early detected cancer might have a chance of never actually killing the patient even without treatment. Therefore early detection increases survivability in the statistics because it's counting more "duds".