A novel method based on unbiased correlations tests for covariate selection in nonlinear mixed effects models: The COSSAC approach

Abstract

Building a covariate model is a crucial task in population pharmacokinetics and pharmacodynamics in order to understand the determinants of the inter-individual variability. Identifying a good covariate model usually requires many runs. Several procedures have been proposed in the past to automatize this task. The most commonly used is Stepwise Covariate Modeling (SCM). Here, we present a novel stepwise method based on statistical tests between individual parameters sampled from their conditional distribution and the covariates. This strategy, called COSSAC, makes use of the information contained in the current model to choose which parameter-covariate relationship to try next. This strategy greatly reduces the number of covariate models tested, while retaining on its search path the models improving the log-likelihood (LL). In this article we detail the COSSAC method and its implementation in Monolix, and evaluate its performance. The performance was assessed by comparing COSSAC to the traditional SCM method on 17 representative data sets. For the large majority of cases (15 out of 17), the final covariate model is identical (11 cases) or very similar (4 cases with LL differences less than 3.84) with both procedures. Yet, COSSAC requires between 2 to 20 times fewer runs than SCM. This represents a decisive speed up, especially for models which take long to run and would not be tractable using the SCM method.

Joint modelling and simulation of M‐protein dynamics and progression‐free survival for alternative isatuximab dosing with pomalidomide/dexamethasone

“…The and Ri are large (69% and 61.5%), they did not impact the covariate analysis because the sampling of conditional distribution was used instead of conditional mode estimates, allowing avoid the associated bias of large shrinkage to be avoided. 29 The estimated link between serum M-protein slope and PFS is high at 11.9, consistent with IMWG criteria, in which the decrease in serum M-protein in response to treatment is the main component directly impacting PFS. Thus, in case of initial response, the serum Mprotein decrease is associated with a current slope <0 and hence a reduced risk of progression.…”

“…The parameter‐covariate relationship was first explored graphically using individual parameter estimates. The Conditional Sampling for Stepwise Approach based on Correlation tests (COSSAC) covariate selection algorithm was then used for automatic building of the covariate model 29,30 . The best covariate model was selected using the corrected version of Bayesian Information Criteria (BICc) 31 .…”

SAMBA: A novel method for fast automatic model building in nonlinear mixed‐effects models

“…To assess the performances of the SAMBA procedure compared to SCM and COSSAC procedures, we replicate the illustration provided in ref. 4. We applied the three routines to a collection of 10 representative datasets, including PKs, pharmacodynamics, and disease models.…”

“…Each covariate addition or deletion is tested in turn selecting models at each step leading to the best adjustment according to the objective criterion. Approaches such as Wald Approximation Method (WAM) 3 and COnditional Sampling use for Stepwise Approach based on Correlation tests (COSSAC) 4 are less computationally intensive as they use, respectively, a likelihood ratio test and a correlation test to move in the covariates space, which allows the testing of less models. All these methods are nevertheless computationally intensive as they require to re‐estimate the model parameters and the likelihood many times.…”

Welcome to the statistics and pharmacometrics themed issue

“…Understanding the factors driving interindividual variability in pharmacokinetics and pharmacodynamics is a goal of many population analyses. In this issue, Ayral et al 1 present a novel stepwise covariate modeling method that makes use of the information contained in the model at one step to choose which parameter-covariate relationship to fit in the next. Hartung et al 2 derive and evaluate nonparametric goodness-of-fit tests for parametric covariate models, transferring concepts from statistical learning to the pharmacological setting.…”