(M-055) Real World Data Based Evaluation of a Novel TMDD Approximation Model

Hyeseon Jeon – College of Pharmacy – Chungnam National University, Daejeon, Republic of Korea; Hwi-yeol Yun – College of Pharmacy – Chungnam National University, Daejeon, Republic of Korea; Soyoung Lee – College of Pharmacy – Chungnam National University, Daejeon, Republic of Korea; Jae Kyoung Kim – Biomedical Mathematics Group – Pioneer Research Center for Mathematical and Computational Sciences, Institute for Basic Science, Daejeon, Republic of Korea; Jung-woo CHAE – College of Pharmacy – Chungnam National University, Daejeon, Republic of Korea; Jong Hyuk Byun – Department of Mathematics and Institute of Mathematical Science – Pusan National University, Busan, Republic of Korea

Objectives: Target-Mediated Drug Disposition (TMDD) occurs when a drug binds to its pharmacological target with high affinity, leading to nonlinear pharmacokinetics. Due to the increasing number of parameters in TMDD models, approximation methods such as the Michaelis-Menten (MM) and quasi-steady-state (QSS) models have been proposed to simplify data fitting.

In our previous study (Byun et al., 2024) [1], we introduced the pTMDD model, derived by applying a first-order Taylor expansion to the qTMDD (QSS TMDD) model under the assumption that Rtot (total receptor concentration) was constant. This study showed two main advantages: (1) Broader applicability (both C0+KM≫Rtot and ≪Rtot) compared to mTMDD (only C0+KM≫Rtot) and (2) reduced computation time relative to qTMDD. However, this was restricted to simulated and limited datasets.
This study aimed to evaluate the usability of the pTMDD model proposed by Byun et al.[1] by applying it to five clinical trial datasets and comparing its estimation accuracy, estimation precision and computational efficiency against full TMDD, mTMDD, and qTMDD models.

Methods: The five clinical trial data analyzed were: (1) anakinra, (2) anti-PD-L1 mAb, (3) IL-1Ra Fc fusion protein, (4) anti-CD-47 monoclonal antibody and (5) IL-7 Fc fusion protein. Cases 1–4 corresponded to (C0 + KM)/Rtot > 1; case 5 to ≪ 1.
For estimation accuracy, we defined “estimate accuracy” and “imprecision accuracy” as how close each parameter estimate and relative standard error (%, RSE) were to those of the full TMDD model. Parameter uncertainty was assessed using (1) a parametric approach with variance-covariance matrix from NONMEM .cov files and (2) a nonparametric approach with bootstrap results, following Jonsson (2024) [2].
Estimation precision was assessed by comparing the absolute RSE values across the four models for each parameter using Dixon’s Q test. Computational efficiency was compared by measuring estimation times from NONMEM .lst files. All runs were performed on an IdeaPad 5 Pro with an AMD Ryzen 7 5800H processor and 8 GB RAM.

Results: For estimate accuracy, mTMDD showed deviations in 1/4, 2/12, 1/5, and 4/11 parameters (cases 1–5), while qTMDD deviated in 1/11 parameters (case 5) with the parametric approach. With the nonparametric approach, mTMDD showed deviations in 1/10, 1/5, and 3/9 parameters (cases 3-5), and qTMDD in 2/9 parameters (case 5).In case 5, (C0+KM≪Rtot), pTMDD remained close to full TMDD estimates, while mTMDD clearly deviated. AIC values followed the order mTMDD < pTMDD < qTMDD when averaged across all cases. With respect to imprecision accuracy, no model demonstrated superiority over the others.
For estimation precision, only pTMDD showed no cases where its RSE was significantly larger than those of the other models.For computational efficacy, pTMDD had the shortest elapsed time in all cases except for case 2, with the largestdifference observed in case 4, which had the most complex model.

Conclusions: pTMDD offered a reliable, precise and efficient alternative to existing TMDD approximation methods. Additionally, we developed Q2PCONV, an R Shiny application to convert NONMEM code from qTMDD to pTMDD to make it more accessible to researchers. (https://hyese.shinyapps.io/Q2PCONV/)

Citations: [1] Byun, Jong Hyuk, et al. "Validity conditions of approximations for a target-mediated drug disposition model: A novel first-order approximation and its comparison to other approximations." PLOS Computational Biology 20.4 (2024): e1012066.
[2] Jonsson, E. Niclas, and Joakim Nyberg. "Using forest plots to interpret covariate effects in pharmacometric models." CPT: Pharmacometrics & systems pharmacology 13.5 (2024): 743-758.

Keywords: Pharamacometrics, Pharmacokinetics, Target-Mediated Drug Disposition, first-order approximation of the total QSSA (pTMDD)