The question presupposes a misconception that the vector form of a state $|\psi\rangle$ exists independently of its density operator form $|\psi\rangle\langle\psi|$, which is often described as secondary. In reality, the density operator of a state is all that truly exists --- and even then, it only exists as statistical information. In fact, you can construct quantum mechanics using density operators $\rho$ as the primary model of states. The formalism for this is similar to the Heisenberg picture of quantum mechanics, in which operators evolve over time. In a Heisenberg-type picture, you don't strictly need Hilbert spaces at all at first, you can define everything with abstract "operators". You can introduce a Hilbert space $\mathcal{H}$ later as an algebra representation of all of the operators. You can also introduce vector states $|\psi\rangle$ later as a simplified formalism for operator states that happen to have the form $\rho = |\psi\rangle\langle\psi|$.