Introduction: In most Phase I and Phase I/II studies in patients, dose-finding and pharmacokinetics/pharmacodynamics (PK/PD) are still analyzed separately [1]. Various dose-finding methods using PK data have been proposed in recent literature based on different approaches to integrate PK data in the toxicity estimation. Objectives: Our aim was to evaluate the performance of existing prospective Bayesian dose-finding methods incorporating PK in terms of their ability to estimate toxicity and their robustness to model misspecification. Methods: We performed a narrative literature review to identify dose-finding methods including PK models. PKLOGIT [2] opts for a Bayesian hierarchical model with a normal approximation of the log AUC and a logistic regression for the exposure-toxicity relationship. The ED-EWOC [3] method uses a population PK model that consists in estimating the PK parameters for a one-compartment model to then compute the AUC of each patient before obtaining a probability of toxicity for each dose using a logistic regression. We considered two versions, with and without an EWOC decision rule (ED). The TITE-PK [4] method uses latent PK profiles to measure drug exposure and a survival model to compute the probability of DLT for each dose. Since it depends on fixed parameters, we considered two versions: informed TITE-PK where the true values of PK parameters are used, and naive TITE-PK with noisy values for PK parameters. For comparison, BLRM [5] is implemented as benchmark. All methods were evaluated in a simulation study of 1000 clinical trials, assuming four doses, a maximum of 30 patients and cohorts of size 2. Based on the development of the TGF- inhibitor LY2157299 [6] in patients with glioma, PK data was simulated based on a one-compartment PK model with first-order absorption. Toxicity was simulated when the value for the patient-specific linear function of the exposure (AUC or Cmax depending on the scenario) exceeded a specific threshold, thus allowing for within-subject exposure variability. The dose assigned to the next cohort was recommended based on the dose allocation rule of each method using data from previously included cohorts. Several scenarios were implemented reflecting different assumptions about the position of the MTD and misspecification in PK measures of exposure or model. Dose-finding methods were compared using the probability of selecting the correct MTD, the percentage of dose allocation, and the number of dose-limiting toxicities. Results: All methods showed comparable performances in scenarios where the MTD is the lowest dose, except for PKLOGIT and TITE-PK methods. TITE-PK methods often incorrectly chose a higher dose as the MTD. PKLOGIT tended to stop trials prematurely due to an overestimation of toxicity. The issues with PKLOGIT and TITE-PK were less pronounced but still present when the MTD was at the second dose. In scenarios with a high-dose MTD, the three PK dose-finding methods, as well as ED-EWOC and ED, displayed good performance in terms of correct MTD selection. ED-EWOC was effective at avoiding overdose in the first two scenarios, while the ED worked better for high-dose MTDs. Therefore, ED-EWOC has conservative properties regarding dose recommendation for low-dose MTDs but may hinder escalation to higher doses. When all doses were overly toxic and trial should be stopped early, TITE-PK incorrectly recommended the first dose as the MTD despite high toxicity levels, and the BLRM method struggled to accurately assess toxicity probabilities for later doses. ED-EWOC systematically outperformed TITE-PK in terms of correct MTD selection except when the latter was informed with strong prior knowledge or for high-dose MTDs. When toxicity was simulated using Cmax instead of the AUC and PK data computed using a one-compartment model, [1] TITE-PK outshone ED-EWOC and ED except when the first dose was the MTD. Conclusions: Modelling dose-exposure-toxicity is preferred for safer, accurate MTD identification in trials. PKLOGIT struggles with early-dose MTD estimation probably due to poor exposure approximation. TITE-PK consistently performs well, barring low-dose MTDs and misspecification scenarios. ED-EWOC shows potential, especially under misspecification, but is generally inferior to TITE-PK. Integrating toxicity as time-to-first DLT event rather than as a binary endpoint could enhance PK model efficacy. References: [1] Comets, E. & Zohar, S. (2009), A survey of the way pharmacokinetics are reported in published phase i clinical trials, with an emphasis on oncology, Clinical pharmacokinetics 48, 387–395. [2] Ursino, M., Zohar, S., Lentz, F., Alberti, C., Friede, T., Stallard, N. & Comets, E. (2017), Dose-finding methods for phase I clinical trials using pharmacokinetics in small populations, Biometrical Journal 59(4), 804–825. [3] Micallef, S., Sostelly, A., Zhu, J., Baverel, P. G. & Mercier, F. (2022), Exposure driven dose escalation design with overdose control: Concept and first real-life experience in an oncology phase i trial, Contemporary Clinical Trials Communications 26, 100901. [4] Günhan, B. K., Weber, S. & Friede, T. (2020), A Bayesian time-to-event pharmacokinetic model for phase I dose-escalation trials with multiple schedules, Statistics in Medicine 39(27), 3986–4000. [5] Neuenschwander, B., Matano, A., Tang, Z., Roychoudhury, S., Wandel, S. & Bailey, S. (2015), ‘A Bayesian industry approach to phase I combination trials in oncology’, Statistical methods in drug combination studies 2015, 95–135. [6] Gueorguieva, I., Cleverly, A. L., Stauber, A., Pillay, N. S., Rodon, J. A., Miles, C. P., Yingling, J. M., and Lahn, M. M. (2014). Defining a therapeutic window for the novel TGF-β inhibitor LY2157299 monohydrate based on a pharmacokinetic/pharmacodynamic model: Early oncology development based on a pharmacokinetic/pharmacodynamic model. British Journal of Clinical Pharmacology 77, 796–807. Acknowledgments: This study was supported by a grant from Inserm and the French Ministry of Health (MESSIDORE 2022, reference number Inserm-MESSIDORE N° 94).