Lattice QCD simulations tend to become stuck in a single topological sector at fine lattice spacing or when using chirally symmetric overlap quarks. In such cases physical observables differ from their full QCD counterparts by finite volume corrections. These systematic errors need to be understood on a quantitative level and possibly be removed. I discuss an extension of an existing relation from the literature between two-point correlation functions at fixed and the corresponding hadron masses at unfixed topology, which includes all terms proportional to 1/V2 and 1/V3, where V is the spacetime volume. I also apply the resulting relation to simple models, quantum mechanics on a circle, the Schwinger model and SU(2) Yang-Mills theory.