Statistical variability of parameter bounds for n-pool unidentifiable mammillary and catenary compartmental models (Q1109701)

Analytic (asymptotic) expressions are derived for estimating the variances of the parameter bounds for unidentifiable n-pool mammillary and catenary compartmental models with input and output probes in pool 1 (central, or first pool) only. These equations provide information about how output data errors propagate into the estimation of parameter bounds determined from interval identifiability analysis of the deterministic model. A key feature of the variance algorithm involves use of the Laplace transform to compute output sensitivity functions with respect to the structural invariants of the model, required for estimating covariances in an intermediate step. The resulting variance algorithm is compared with Monte Carlo simulations for several test cases, using data-error models with constant coefficient of variation (CV). The asymptotic variances compare well with simulated results, tending to slightly underestimate the true variances of the parameter bounds with greatest variability. Overall, the results suggest that the variances of parameter bounds for these two classes of models may be uniformly well behaved, at least for constant-CV data errors. This last result appears to hold for quasiidentifiable models in the presence of noisy data, as well as interval identifiable models that are far from being quasiidentifiable.