In pharmacometric modelling, it is often important to know whether the data is sufficiently rich to identify the parameters of a proposed model. While it may be possible to assess this based on the results of a model fit, it may be difficult to disentangle identifiability issues from other model fitting and numerical problems. Furthermore, it can be of value to ascertain identifiability beforehand from study design.
The aim of this work was to develop two new methods for identifiability analysis prior to model optimization that can be applied to a wide range of models and detect a variety of potential identifiability issues. Their use is illustrated with an example problem.
An R package implementing these methods is available under the GPL
license:
Download
the package
The methods were applied to a quasi-equilibrium (QE) approximation to a Target Mediated Drug Disposition (TMDD) model describing leukaemia inhibitory factor (LIF) data in sheep, with parameter values, sample times and dose levels obtained from Abraham et al. (Abraham, Krzyzanski, and Mager 2007). The output consisted of the PK concentration at the reported sample times. The methods were presented at the PAGE conference (Noort and DeJongh 2023).
The parameters of a model are identifiable if two different parameter vectors always lead to two different model outputs. In practice one often focuses on local non-identifiability, characterized by a curve in parameter space of constant model output. The tangent to the curve is the unidentifiable direction. If a model is locally unidentifiable then its parameters cannot be determined uniquely from the data. Two methods for local identifiability were developed.
The sensitivity matrix (SM) is the matrix of derivatives of the model outputs with respect to the structural parameters. Unidentifiable directions in parameter space correspond to vectors in the null space of this matrix. As the null space is difficult to determine numerically, three proxy indicators were developed that identify near-singularities of the SM and the corresponding parameter vectors.
1. The Skewing Angle, which measures the
angle between the images in the output space of the parameter vectors. A
small angle indicates non-identifiability;
The Fisher Information Matrix (FIM) determines the approximate shape of the objective function value (OFV) with respect to the parameters. Unidentifiable directions in parameter space correspond to vectors in the null space of this matrix. As for the SMM, proxy indicators are determined.
1. Curvatures of the OFV surface. A low
curvature corresponds to non-identifiability;
2.
The maximum change in parameter values corresponding to
a given OFV change. Large changes indicate non-identifiability;
3. Relative standard errors (RSE), where high
values characterize non-identifiability.
Unlike the SM method, this method can handle random effect
parameters. It assumes Gaussian distributions.
This shows overall identifiability indicators from the two methods. For each one, a high value indicates identifiability. In this way, different scenarios can be compared in one view.
For unidentifiable scenarios, the minimal parameter relation indicates the direction of unidentifiability, that is, the parameters that are badly identifiable. They correspond to high values.
The L-norm of a parameter characterizes the distance of this parameter from all the others ones. If Lnorm is zero, this parameter is unidentifiable. The larger the L-norm, the more identifiable the parameter is.
Low curvatures correspond to a shallow OFV surface as function of the parameters. The number of low or zero curvatures is the number of unidentifiable parameter directions.
The plot shows a red “warning” zone where curvatures are low. In this case it is defined below the minimal Lmin*100, but could be optional for the user to choose.
For each scenario and curvature, the following plot has a column showing the maximum relative parameter changes corresponding to a change in OFV that is not significant (at 95%). Different colors are used for relative changes above and below a threshold. Changes corresponding to non-positive curvatures (if any) are also colored differently. The dot size indicates the size of the change.
Developed by: Richard Hooijmaijers, Rolien Bosch, Martijn van
Noort
2023-07-04
Contact person: Martijn van Noort (m.vannoort@lapp.nl)