Parameter non-identifiability of the Gyllenberg-Webb ODE model (Q2436586)

Parameter identifiability problems provide useful insights about the performance of the Gyllenberg-Webb model. The model presented here is able to reproduce any tumor growth model written as a first-order ODE, imposed on the total tumor size. The model allows the choice of parameters corresponding to an association function. By using the explicit form of an equivalence relation, it is shown that even a finite-dimension parameter estimation problem, with simple shapes of parametric functions, can be ill-posed. The particular form of the ODE system permits uniqueness in the solutions, in the case of non-Lipschitz parameters, though admittedly in a limited setting. The generalization allows to transfer all relevant results, but in all cases the increasing total of the tumor size is needed as crucial assumption. The main conclusion of the paper is that the non-identifiability of the passage rates is an intrinsic property in all of these models. Any passage rate in a given equivalence class will describe not only global tumor growth in the same manner, but also both proliferating and quiescent cells. The real potential of the Gyllenberg-Webb model can only be tested when it is confronted with experimental data contrasting untreated and treated animals.