Khalid Iqbal (1), Karthik Venkatakrishnan (1), Akash Khandelwal (1,2)
(1) Quantitative Pharmacology Merck Healthcare KGaA, Darmstadt, Germany (2) Current affiliation: UCB Biosciences GmbH, Monnheim, Germany
Objectives: Population pharmacokinetic (popPK) and exposure-response (ER) modelling are the critical components of United States Food and Drug Administration (FDA) submissions dossiers to inform dosing and labelling decisions. The aim of the review was to summarize and critically review the key aspects of popPK and ER modelling aspects of FDA submissions in oncology.
Methods: A thorough search of the online FDA/ CDER database [1] was performed where the clinical pharmacology package submitted for new drug applications (NDA)/ Biologics License application (BLA) for oncology indications from March 2015-February 2022 were considered in the evaluation. The relevant sections (popPK and ER modelling, and immunogenicity) of clinical pharmacology dossiers were thoroughly analyzed.
Results:
In total 82 dossiers were found and included in the evaluation. Majority of the submissions were small molecules (65%) while monoclonal antibodies (mAb’s) and antibody drug conjugates (ADC’s) were 23% and 11%, respectively. A two compartmental model was frequently (63%) used to describe the popPK. A high number of covariates were evaluated (median 11, range 3-28) in covariate analysis but only a lower number were retained in the final models (median 3, range 1-15). The stepwise covariate model selection was frequently (28%) used for covariate selection. However, no uniform criteria for statistical significance (p-value) was adopted; for forward selection, two levels (0.01 and 0.005) were used, while for backward selection, multiple levels (0.00005, 0.001, 0.01, and 0.005) were reported. Moreover, time-varying covariates were included in only 5% of the popPK final models while it was not/poorly reported in the rest of the submissions. In 44% submissions the covariate method was not explicitly stated.
In ER analysis, logistic regression was commonly applied (61% and 59% in exposure-efficacy (EE) and exposure-safety (ES) models, respectively). Other methods used were nonparametric Kaplan-Meier analysis (29% in EE models) and graphical methods (17% in ES models). Area under the concentration time curve was the commonly used exposure-metric in ER analysis (51% and 43% of EE and ES, respectively). Interestingly and of concern, for monoclonal antibodies, for the EE analysis, the area under the concentration time curve from first dosing cycle was used as exposure matric in only in 28% of the submissions. The EE analyses were from studies performed with a single dose level in most of the submissions (61%) while on the contrary the ES modelling considered data from across multiple dose levels (74%). Covariate analyses were less frequently (43% in EE and 21% in ES models) reported.
Conclusions: A two-compartment popPK model was the most commonly identified final model. The significance level in covariate selection was not always reported. In the context of exposure-efficacy analysis, data from a single dose level were used which limits the interpretation and applicability of the model for dose selection/justification.
References:
[1] https://www.accessdata.fda.gov/scripts/cder/daf/index.cfm
Reference: PAGE 32 (2024) Abstr 11193 [www.page-meeting.org/?abstract=11193]
Poster: Drug/Disease Modelling - Oncology