Aziz Ouerdani 1, Estelle Chasseloup 1, Alberto Russu 1, Vincent Duval 1, Nele Goeyvaerts 1
1 Johnson & Johnson Innovation Medicine (Beerse, Belgium)
Introduction:
Forest plots (FPs) originate from the field of statistical meta-analysis to visualize individual study effects and overall pooled effects for multiple clinical studies [1]. The concept was extended to represent the effect of intrinsic and extrinsic factors on exposure metrics [2]. Since then, FPs are increasingly used in pharmacometrics (PMx) to visualize and interpret the effect of covariates on population PK (PopPK) and PK/PD model parameters and/or derived individual exposure metrics such as AUC and Cmax. FPs are used to support drug development decisions such as dose adjustments and to communicate with regulatory authorities [3]. In an FP, covariate effects are generally presented by point estimates with horizontal lines representing confidence intervals (CI), relative to a reference line, indicating the reference subject, and a reference interval.
Objectives:
Different types of FPs are used in the field of PMx, and the interpretation of covariate effects depends on the type used. The aim of this work is to propose a general classification along with a clinical data application, complementing published tutorials on FPs in PMx [4, 5].
Methods:
The proposed classification of FPs in PMx was mainly based on in-house practice and two tutorials, one from Marier et al. [4] introducing workflows and the coveffectsplot R package and one from Jonsson and Nyberg [5] providing guidance on the presentation and interpretation of FPs. For each type of FP, the general objective, example drug development questions, and key advantages and limitations, were identified. In addition, a computational workflow was established: integrating covariate and exposure data sources, derivation of PK parameters and exposure metrics, and calculation of FP statistics (point estimates and intervals). To illustrate the difference in the interpretation of covariate effects, different types of FPs were applied to niraparib, using a PopPK model based on pooled phase 1-3 PK data from subjects with metastatic castration-resistant prostate cancer [6]. FPs were created in R version 4.4.1.
Results:
Four types of FPs in PMx were identified: EBE/NCA, marginal, empirical and prediction FPs:
• EBE/NCA FPs aim to explore marginal or joint effects of predefined factors on exposure metrics derived from Empirical Bayes estimates (EBE) or non-compartmental analysis (NCA). Because covariate effects are estimated as geometric mean ratios relative to a reference category via linear regression modeling, EBE/NCA FPs are straightforward to generate, but they do not provide a direct link to PopPK parameter estimates and their associated uncertainty.
• Marginal FPs aim to explore marginal effects of PopPK model covariates on exposure, changing one covariate at a time, keeping the others fixed at the reference values. While marginal FPs enable exploring continuous covariate effects at extreme values, their applicability is limited to those covariates included in the PopPK model.
• Empirical and prediction FPs both aim to explore PopPK model-based joint effects of predefined factors on exposure within a given population. Empirical FPs are based on actual observed clinical data, while prediction FPs are based on a virtual population dataset and include between-subject variability to support extrapolation to a future population of interest (e.g. paediatrics). Similar to EBE/NCA FPs, empirical and prediction FPs support evaluation of covariates not included in the PopPK model, provided these covariates are correlated with PopPK model covariates.
The CI in FPs are based on the uncertainty of the PopPK parameter estimates (e.g. based on re-sampling from the variance-covariance matrix, or other techniques [5]), except for the EBE/NCA FP in which it is derived from the linear regression model.
For niraparib, results were comparable between the EBE and marginal FPs and supported that no dose adjustment in specific patient subgroups was required.
Conclusions:
FPs are frequently used in PMx to visualize the impact of covariates on exposure and therefore it is critical to understand what the FPs are presenting. The proposed classification of FPs in PMx will facilitate clear communication as the FP name itself reflects key assumptions and informs interpretation of covariate effects, including dose adjustment considerations. This work stimulates discussion on which FP type should be included in reports and summaries to better align with regulatory authorities’ expectations.
References:
[1] Chang Y, Phillips MR, Guymer RH, et al. The 5 min meta-analysis: understanding how to read and interpret a forest plot. Eye (Lond). 2022 Apr;36(4):673-675. Erratum in: Eye (Lond). 2023 Dec;37(17):3704.
[2] Menon-Andersen D, Yu B, Madabushi R, et al. Essential pharmacokinetic information for drug dosage decisions: a concise visual presentation in the drug label. Clin Pharmacol Ther. 2011 Sep;90(3):471-4.
[3] U.S. Food and Drug Administration. Population pharmacokinetics: Guidance for industry. Silver Spring, MD: FDA; 2022. Available from: https://www.fda.gov/media/128793/download
[4] Marier JF, Teuscher N, Mouksassi MS. Evaluation of covariate effects using forest plots and introduction to the coveffectsplot R package. CPT Pharmacometrics Syst Pharmacol. 2022 Oct;11(10):1283-1293.
[5] Jonsson EN, Nyberg J. Using forest plots to interpret covariate effects in pharmacometric models. CPT Pharmacometrics Syst Pharmacol. 2024 May;13(5):743-758.
[6] Russu A, Hazra A, Tian H, et al. Population Pharmacokinetics of Niraparib/Abiraterone Acetate Administered as Single-Agent Combination and Dual-Acting Tablets Plus Prednisone for Metastatic Castration-Resistant Prostate Cancer. Adv Ther. 2025 Apr;42(4):1860-1880.
Reference: PAGE 34 (2026) Abstr 11960 [www.page-meeting.org/?abstract=11960]
Poster: Methodology - Covariate/Variability Models