2022 - Ljubljana - Slovenia

PAGE 2022: Methodology - Covariate/Variability Models
Niclas Jonsson

Efficient and relevant stepwise covariate model building for pharmacometrics

Robin J. Svensson(1,2) and E. Niclas Jonsson (2)

(1) Presently at Medical Products Agency, Sweden (2) Pharmetheus AB, Sweden

Background: Covariate modeling is an important opportunity for pharmacometrics to influence decision-making in drug development. The stepwise covariate model (SCM)[1] building procedure is the most common method for covariate model development[2]. Despite its advantages, the traditional SCM method is known to have long runtimes and sub-optimal ability to select relevant covariates, especially in more complex phase 3 settings.

Objectives: To evaluate two alternative stepwise covariate modelling approaches and compare them to the legacy SCM approach.

Methods: The two alternative stepwise modelling approaches are SCM+[3] and SCM+ combined with stage-wise filtering. SCM+ builds on the legacy SCM algorithm but introduces adaptive scope reduction (ASR) and optimized estimation settings. ASR reduces the defined search scope adaptively during the forward search based on the performance of each covariate-parameter relation in the previous steps. While the first forward step is identical to the SCM, when moving on to the next forward step the number of parameter-covariate relations to be tested is reduced. Once the full forward covariate model is established, backwards elimination proceeds as in the legacy SCM approach. The efficiency of SCM+ is further increased by changing the default NONMEM termination criterion to CTYPE=4 and by dynamically restrict the maximum functiontion evaluations allowed (MAXEVAL in NONMEM) based on the number of function evaluations used by the base model.

With stage-wise filtering, covariates are categorized into 3 groups: mechanistic, structural, and exploratory. Mechanistic covariates are those with a known impact on one or more parameters of the model and are included in the base model without testing. Structural covariates are those that have a strong rationale to impact one or more model parameters; often related to the study design, e.g., formulation or diet status. Exploratory covariates are those that are not mechanistic or structural and are explored for hypothesis-generating reasons. Stage-wise filtering proceeds in 3 phases: (i) addition of mechanistic covariates to the base model; (ii) stepwise inclusion of structural covariates;  and (iii) stepwise inclusion of exploratory covariates. The 3 covariate categories are hierarchical, meaning that a structural covariate cannot replace a mechanistic covariate, and an exploratory covariate cannot replace a structural or mechanistic covariate.

The simulated PK data used in the investigation was setup to mimic an orally administered drug in a phase 3 setting, with two phase 1 studies and one phase 3 study. The covariate characteristics were generated using conditional distribution modelling[4]. Altogether, 100 datasets were simulated and each of them was subject to SCM, SCM+ and SCM+ with stage-wise filtering and the results of the three methods were compared in terms of efficiency and relevance.

Results: 

The two SCM+ methods were considerably more efficient than the traditional SCM: the number of function evaluations was reduced by 70% for SCM+ and by 76% for SCM+ with stage-wise filtering compared to SCM; the corresponding number of executed models was reduced by 44% for SCM+ and 70% for SCM+ with stage-wise filtering. In addition, among the three methods, SCM+ with stage-wise filtering selected the highest number of relevant covariates (on average, 8.0 true covariate-parameter relationships were selected compared to 5.1 selected with SCM and with SCM+).

Conclusions: 

SCM+ and SCM+ with stage-wise filtering is much more efficient than the legacy SCM approach while resulting in similar and/or more relevant covariate models. Given the improved efficiency and ability to select relevant covariates shown in this work, the use of SCM+ and stage-wise filtering can greatly decrease the time needed to establish a covariate model and will therefore increase the time available for additional analyses and simulations that increase the value of the modelling analysis for drug development decision making.



References:
[1] Jonsson, E. N. & Karlsson, M. O. Automated covariate model building within NONMEM. 15, 1463–1468 (1998).
[2] Hutmacher, M. M. & Kowalski, K. G. Covariate selection in pharmacometric analyses: a review of methods. Br. J. Clin. Pharmacol. 79, 132–147 (2015).
[3] Jonsson, E. N., Harling, K, PAGE 27 (2018) Abstr 8429 [www.page-meeting.org/?abstract=8429]
[4] Smania, G. & Jonsson, E. N. Conditional distribution modeling as an alternative method for covariates simulation: Comparison with joint multivariate normal and bootstrap techniques. CPT Pharmacomet. Syst. Pharmacol. 10, 330–339 (2021).


Reference: PAGE 30 (2022) Abstr 10076 [www.page-meeting.org/?abstract=10076]
Poster: Methodology - Covariate/Variability Models
Click to open PDF poster/presentation (click to open)
Top