Morgane Philipp (1,2), Sylvie Retout (2,3), Simon Buatois (3), France Mentré (1)
(1) Université Paris Cité, INSERM, IAME, UMR 1137, Paris, France, (2) Institut Roche, Boulogne-Billancourt, France (3) Roche Pharma Research and Early Development, Pharmaceutical Sciences, Roche Innovation Center Basel, Switzerland
Introduction: Identifying covariate clinical relevance (CCR) is essential for patient dose adjustment and is a key step according to health authorities. Clinicians use forest plots [1] to illustrate parameter changes for covariate values relative to typical ones. CCR is inferred from the position of covariate effect ratios (CER) with their 90% confidence interval (CI90) relative to the reference area/line [2]. Previous work [3] compared the accuracy of CCR assessment of Stepwise Covariate Model (SCM) [4], SCM+ [5] and Full Model (FM) [6]. All approaches provided satisfactory results close to those of the reference model (estimations using the simulated covariate models). Here, two alternative approaches were investigated: least absolute shrinkage and selection operator (LASSO) [7] and a genetic algorithm (GA) [8]. LASSO is a regression approach. GA is an heuristic search approach using principles inspired by natural selection and genetics. Both approaches aim to minimize the objective function value (OFV), LASSO by applying a constraint on the sum of covariate parameter estimates, GA by making “individuals” (covariate models) adapt and reproduce within a “population” (set of covariate models) over “generations” (iterations), passing on “beneficial traits” (best covariates based on OFV).
Objective: To compare the results of CCR determination of LASSO and GA with the ones from SCM+ and the reference model in a proof of concept study.
Methods: Two clinical trial datasets from a pharmacokinetic (PK) study [9] involving 383 patients mixing rich and sparse designs were simulated, using a linear one-compartment model with first-order absorption. One using the “base model” including a body weight (BW) effect on CL/F and V/F, and one using the “covariate model” including 4 additional ones with an AGE and albumin (ALB) effect on CL/F, and an AGE and RACE effect on V/F. Overall, covariates showed low correlations (-0.16 to 0.26); only AGE, BW and patient status (STAT) showed moderate ones (0.41 to 0.57).
The three covariate selection approaches were applied using a predefined set of 18 covariate-parameter relationships, including the simulated ones. Parameter estimation was done using NONMEM version 7.4 with FOCEI. SCM+ and LASSO were run using PsN 5.3.2, GA using R 4.0.5 and Perl 5.30.2.
CER were computed for both continuous (between effects at 10/90th percentiles (P10/P90) and the one at the median of the observed covariate distribution) and categorical covariates (between effects of a category and the one of the reference).
Covariates were relevant (R), non-relevant (NR) or with insufficient information (II) if the CI90 of CER fell completely outside, inside or straddle the reference area [0.8 – 1.2], respectively. They were significant (S) if the CI90 of CER did not cross the reference line at 1, and not-significant (NS) otherwise.
Of all the simulated covariate effects, only BW has a strong effect, under both simulated models, with covariate ratios on CL/F and V/F between 0.37 and 0.47 for P10 and between 1.30 and 1.42 for P90. RACE has a moderate effect with a covariate ratio of 0.79. Other covariates have small effects with covariate ratios between 0.8 and 1.2.
Results: LASSO and GA, selected all simulated covariate-parameter relationships for the 2 simulated datasets, as with SCM+. However, few spurious relationships were retained. Under the base model, GA selected an AGE effect on KA (as SCM+), and LASSO selected an AGE effect on CL/F and a RACE effect on V/F, and CL/F. Under the covariate model, GA selected an AGE and STAT effect on KA. The AGE and STAT selection may be explained by moderate correlation between them and also with BW.
Regarding the CCR assessment, conclusions were consistent across all selection approaches and in line with those obtained with the reference models. For the 2 simulated datasets, BW on CL/F and V/F were R. Other simulated covariate effects were either NRS or IIS. The few spurious relationships selected were also either NRS or IIS, hence not R.
Conclusion: LASSO and GA provided satisfactory CCR determination on the two simulated datasets. Further evaluations are necessary to enhance the robustness of those results by using simulation frameworks with increased complexity in model structures, expanded covariate sets, and a larger number of simulated datasets. This should also be extended to Full Random Effect Modeling (FREM) [10].
References:
[1] Menon-Andersen D, Yu B, Madabushi R, et al (2011) Essential Pharmacokinetic Information for Drug Dosage Decisions: A Concise Visual Presentation in the Drug Label. Clin Pharmacol Ther 90:471–474.
https://doi.org/10.1038/clpt.2011.149
[2] Cuzick J (2005) Forest plots and the interpretation of subgroups. The Lancet 365:1308. https://doi.org/10.1016/S0140-6736(05)61026-4
[3] Philipp M, Buatois S, Retout S, Mentre F (in press) Impact of covariate model building methods on their clinical relevance evaluation in population pharmacokinetic analyses: comparison of the full model, stepwise covariate model (SCM) and SCM+ approaches. J Pharmacokinet Pharmacodyn
Preprint: https://doi.org/10.21203/rs.3.rs-3655068/v1
[4] Ahamadi M, Largajolli A, Diderichsen PM, et al (2019) Operating characteristics of stepwise covariate selection in pharmacometric modeling. J Pharmacokinet Pharmacodyn 46:273–285. https://doi.org/10.1007/s10928-019-09635-6
[5] Svensson RJ, Jonsson EN (2022) Efficient and relevant stepwise covariate model building for pharmacometrics. CPT Pharmacomet Syst Pharmacol 11:1210–1222.
https://doi.org/10.1002/psp4.12838
[6] Xu XS, Yuan M, Zhu H, et al (2018) Full covariate modelling approach in population pharmacokinetics: understanding the underlying hypothesis tests and implications of multiplicity. Br J Clin Pharmacol 84:1525–1534.
https://doi.org/10.1111/bcp.13577
[7] Ribbing J, Nyberg J, Caster O, Jonsson EN (2007) The lasso—a novel method for predictive covariate model building in nonlinear mixed effects models. J Pharmacokinet Pharmacodyn 34:485–517.
https://doi.org/10.1007/s10928-007-9057-1
[8] Ronchi D, Tosca EM, Bartolucci R, Magni P (2023) Go beyond the limits of genetic algorithm in daily covariate selection practice. J Pharmacokinet Pharmacodyn.
https://doi.org/10.1007/s10928-023-09875-7
[9] Retout S, Schmitt C, Petry C, et al (2020) Population Pharmacokinetic Analysis and Exploratory Exposure–Bleeding Rate Relationship of Emicizumab in Adult and Pediatric Persons with Hemophilia A. Clin Pharmacokinet 59:1611–1625.
https://doi.org/10.1007/s40262-020-00904-z
[10] Yngman G, Bjugård Nyberg H, Nyberg J, et al (2022) An introduction to the full random effects model. CPT Pharmacomet Syst Pharmacol 11:149–160.
https://doi.org/10.1002/psp4.12741
Reference: PAGE 32 (2024) Abstr 11031 [www.page-meeting.org/?abstract=11031]
Poster: Methodology - Covariate/Variability Models