On the validity of drift-reduced fluid models for tokamak plasma simulation