Optimal Dosage Justification for Datopotamab Deruxtecan in HR-Positive/HER2-Negative Breast Cancer Through Model-Informed Drug Development Approaches.

1
Introduction
Optimal dose selection is a critical step in oncology drug development to ensure patients receive a dosage (dose strength and interval) that maximizes therapeutic benefit while minimizing toxicity. Recognizing the challenges arising from the conventional approach for oncology drug selection (i.e., maximum tolerated dose), the US Food and Drug Administration (FDA) Oncology Center of Excellence launched Project Optimus, an initiative aimed at reforming the dose selection paradigm in oncology by emphasizing the need for robust dose optimization studies to evaluate safety, efficacy, and exposure‐response relationships across a range of dose levels [1].
Model‐Informed Drug Development (MIDD) tools are instrumental in advancing the principles of Project Optimus [2, 3, 4, 5]. Capturing the complexities of treatment‐specific factors, such as drug pharmacokinetics (PK) exposure and tumor biology such as tumor growth and resistance to therapies, tumor growth inhibition (TGI) models have enabled recapitulation of changes in tumor burden over time [6, 7, 8]. This modeling framework linking tumor dynamics and treatment outcomes has demonstrated its potential to support critical decision‐making by providing quantitative predictions of treatment outcomes across different dose levels and patient populations, thereby aiding dose optimization in oncology drug development [9, 10].
Datopotamab deruxtecan (Dato‐DXd) is a trophoblast cell surface antigen 2 (TROP2)‐directed antibody‐drug conjugate (ADC), composed of a humanized anti‐TROP2 IgG1 monoclonal antibody covalently linked to a highly potent topoisomerase I inhibitor payload (DXd) via a plasma‐stable, tumor‐selective, cleavable linker [11]. Dato‐DXd has demonstrated encouraging anti‐tumor effects and manageable safety profiles in multiple tumor types, including non‐small cell lung cancer (NSCLC) and HR‐positive/HER2‐negative (HR+/HER2−) breast cancer (BC) [12, 13]. In the Phase I TROPION‐PanTumor01 study, Dato‐DXd was evaluated at dose levels of 0.27, 0.5, 1.0, 2.0, 4.0, 6.0, 8.0, and 10.0 mg/kg Q3W in patients with NSCLC. The maximum tolerated dose (MTD) was determined to be 8 mg/kg Q3W based on the occurrence of dose‐limiting toxicities (DLTs) [14]. In alignment with Project Optimus, optimal dose selection for Dato‐DXd in patients with NSCLC has been guided using the MIDD paradigm, which integrates PK, pharmacodynamics (PD), and clinical outcomes [15]. A totality of evidence in patients with NSCLC supported the advancement of 6 mg/kg to late‐stage clinical trials, including the evaluation in HR+/HER2− BC in TROPION‐Breast01.
TROPION‐Breast01 trial is a pivotal Phase III study evaluating the efficacy and safety of Dato‐DXd at 6 mg/kg in patients with HR+/HER2− BC who have been previously treated with standard of care therapies, including endocrine therapy and chemotherapy [12]. The study demonstrated that Dato‐DXd significantly improved Progression‐Free Survival (PFS) compared to chemotherapy and had a manageable safety profile at 6 mg/kg Q3W, and was subsequently approved in multiple countries. Dose levels other than 6 mg/kg were not clinically evaluated in patients with HR+/HER2− BC.
The optimal dose selection at 6 mg/kg Q3W for patients with HR+/HER2‐ metastatic BC was further validated in the current analysis. Dato‐DXd PK exposure in lung and breast cancer was compared by the updated population PK (PopPK) model. A TGI model and a TGI‐PFS modeling framework were developed to evaluate the relationship between Dato‐DXd exposure, tumor dynamics, and PFS. Using data from patients treated with 6 mg/kg, we simulated virtual trials to predict efficacy outcomes at untested dose levels, such as 4 mg/kg.

2
Materials and Methods
2.1
Patient Population and Study Design
PopPK modeling analysis dataset included the patients who received at least one dose of Dato‐DXd and had at least one PK record after the first dose. Patient data from four studies were pooled for PopPK model development, including TROPION‐PanTumor01 (NCT03401385), TROPION‐Lung01 (NCT04656652), TROPION‐Lung05 (NCT04484142), and TROPION‐Breast01 (NCT05104866). All four studies were conducted in accordance with the Declaration of Helsinki and were consistent with International Conference on Harmonization and Good Clinical Practice guidelines and applicable regulatory requirements. Written informed consent from participants was obtained before performing any protocol‐related procedures. The four studies provided sufficient data to support the updates in the previously developed PopPK model [15, 16, 17, 18], the speculation of the population of interest (i.e., patients with HR+/HER2− BC), and the generation of PK metrics for downstream analysis. Details of the four included studies were described in Table 1.
Patient‐level data of tumor assessment and PFS from patients with HR+/HER2‐ BC in TROPION‐PanTumor01 and TROPION‐Breast01 were used in TGI and TGI‐PFS model development (TROPION‐PanTumor01 data‐cut off (DCO): 22 July 2022; TROPION‐Breast01 DCO: 17 July 2023). Patients who were included in PopPK analysis and had at least one post‐baseline tumor size measurement were included in TGI and TGI‐PFS analysis. The study designs and key results of TROPION‐PanTumor01 and TROPION‐Breast01 had been published previously [12, 14]. In TROPION‐PanTumor01, tumor assessments were performed at screening and every 6 weeks in the first 36 weeks after Day 1 of Cycle 1 and then every 12 weeks during treatment. In TROPION‐Breast01, tumor imaging assessments were conducted per RECIST v1.1 every 6 weeks (±7 days) for 48 weeks and every 9 weeks (±7 days) thereafter until investigator‐assessed progressive disease.

2.2
Population Pharmacokinetics (PopPK) Model Assessment
In the current analysis, the previously developed PopPK model was used as the initial model and updated with TROPION‐Breast01 data [16, 18]. A full covariate modeling approach emphasizing parameter estimation rather than stepwise hypothesis testing was implemented. The intact Dato‐DXd model was evaluated first. From the initial full covariate Dato‐DXd model, backward elimination was used to confirm the significance of the previous covariates and to find a statistically parsimonious model. After backward elimination, the preplanned covariates of interest were tested by including them in a stepwise addition (forward inclusion), until no more relations were statistically significant. The parameters of the final Dato‐DXd full covariate model were fixed during DXd model development, which followed the same approach as Dato‐DXd model development. Mechanistic covariates (i.e., baseline body weight) were not included in the backward elimination. The covariates of interest in Dato‐DXd and DXd model forward selection included subject level formulation, tumor type, subject level treatment emergent antidrug antibody (TEADA) status, and baseline creatinine clearance (CrCL). The details of PopPK model development were defined in Supporting Information. Individual PK profiles were predicted based on the PopPK model Empirical Bayesian Estimate (EBE) method. Exposure metrics were derived based on predicted individual PK profiles.

2.3
Tumor Growth Inhibition (TGI) Model Assessment
Tumor measurements were fitted using the Stochastic Approximation Expectation–Maximization (SAEM) algorithm implemented in Monolix [21]. The details of tumor growth model development were defined in Supporting Information. Briefly, the TGI base model consists of tumor growth function, tumor inhibition function, and treatment effectiveness, followed by the selection of Dato‐DXd PK metrics related to the tumor inhibition function, and covariate model development. Covariates include baseline albumin, baseline tumor size, region, Eastern Cooperative Oncology Group (ECOG) baseline, smoking status, race, prior line therapy, age, presence of prior line cyclin‐dependent kinase (CDK)4/6 inhibitor, presence of CNS metastases in history, presence of liver metastases, and TROP‐2 status; these were evaluated for tumor growth function and tumor inhibition function.
Diagnostic plots and relative standard error (RSE) of parameter estimates were examined when evaluating the goodness of fit. The difference in Bayesian information criterion (BIC) between nonhierarchical models and the difference in objective function values (ΔOFV) between the reference and nested test models were used for model selection [22]. A drop in BIC of 2 between two models was considered relevant [23]. A ΔOFV ≥ 10.83 (p ≤ 0.001, df = 1) between two nested models was considered statistically significant (approximation of a χ2 distribution).
Post hoc empirical Bayesian estimation was used to obtain individual TGI metrics, which were further evaluated in TGI‐PFS model development:
Individual parameters from the final TGI model: estimated tumor baseline (TS0), tumor inhibition rate (kkill), tumor growth rate (kge), and tumor resistance appearance rate (lambda).

Time to tumor growth (TTG), which represents the predicted time from treatment start to tumor size nadir.

Tumor size ratio at week 6 (TSR6), which denotes the ratio of tumor size at week 6 compared to the initial tumor size (TS0).

2.4
Joint TGI‐PFS Model Assessment
PFS data were firstly explored using Kaplan–Meier (KM) and Cox regression analyses (R, version 4.1.0). Significant variables (p < 0.05) were further evaluated in parametric PFS analysis.
The joint TGI–PFS model was developed sequentially. First, a TGI model was built as described in the previous section, and post hoc TGI metrics were then derived and used as inputs for the PFS time‐to‐event analysis. The PFS profile was characterized with parametric survival regression. The probability density function that best describes the observed PFS was selected among exponential, Weibull, Gompertz, log‐logistic, uniform, Gamma, and lognormal by AIC (Akaike Information Criterion) evaluations and goodness of fit plots. Based on the selected baseline hazard function, TGI metrics were considered for inclusion in the model. Only one TGI metric with the best improvement in OFV (p < 0.001) was retained for further covariate model selection. Other covariates that were statistically significant in CPH (Cox Proportional‐Hazards) univariate analysis were evaluated in the TGI–PFS model based on improvement in OFV (p < 0.001). The TGI–PFS model was developed using Monolix 2021R1.

2.5
Trial Simulation
Each patient's PopPK parameters (CLlinDatoDXd; Q, V
CDatoDXd, V
pDatoDXd, and V
max), TGI parameters (TS0, kge, kkill, and lambda), and baseline covariates (body weight and tumor size) were obtained. Km was not included because no inter‐individual variability was included for Km in the PopPK model. Using the population mean and the covariance matrix of the pooled variables from 390 patients, a total of 1,000,000 virtual patient parameter sets were generated from a multivariate normal distribution (R 4.1.3). After excluding parameters with extreme values (i.e., those less than the minimum or greater than the maximum individual parameter estimates in TGI model analysis population), virtual patient parameters were randomly resampled 500 times to create 500 simulated clinical trials, each consisting of 400 virtual patients. A total of 20,000 virtual patients were generated (400 patients × 500 trials), and each patient was assigned to receive Dato‐DXd at 4, 6, or 8 mg/kg Q3W for up to 25 cycles. This resulted in 60,000 records (20,000 patients × three dose groups). The same set of virtual patients was used for each dose group, meaning patients were not resampled or changed between different Dato‐DXd dose levels. Dato‐DXd PK, tumor size, and PFS were simulated and evaluated at each dose level.

3
Results
3.1
Dato‐DXd PopPK Analysis
A total of 1081 patients, including 644 patients with lung cancer and 437 patients with breast cancer, were included in the analysis. Patients with lung cancer received Dato‐DXd at dose levels ranging from 0.27 to 10 mg/kg. More than 99% of patients with breast cancer received Dato‐DXd at 6 mg/kg (Table 2). Baseline characteristics of patients included in the PopPK analysis were described in Table 2. Dato‐DXd and DXd concentration versus time profiles appear to be comparable between lung and breast patients at 6 mg/kg (Figure 1A).
Dato‐DXd PopPK structural model was described previously (Figure 1B) [16, 18]. Briefly, Dato‐DXd PK was characterized by a two‐compartmental distribution model with parallel Dato‐DXd linear and nonlinear Michaelis–Menten clearance from the central compartment. Nonlinear clearance reflects concentration‐dependent elimination at lower dose levels, driven by target binding and internalization, and is approximated using Michaelis–Menten kinetics [18]. DXd PK was described by a one‐compartment model with first‐order elimination from the central compartment. The release of DXd from the intact Dato‐DXd was equal to the total (linear + nonlinear) elimination rate of Dato‐DXd. The payload: intact ratio decreased with time both within the dosing cycle and between cycles (cycle 1 vs. later cycles).
The current PopPK analysis corroborates the previously published structural model. All preexisting covariates were retained, and no additional significant covariates were identified. Notably, tumor type (lung cancer vs. breast cancer) was not a statistically significant covariate. The parameter estimates for the final Dato‐DXd model and DXd model are reported in Table S1 and Table S2. CLlinDatoDXd and V
CDatoDXd were estimated as 0.416 L/day and 3.02 L, respectively, which are within < 10% of prior estimates (0.386 L/day and 3.06 L) [18]. For DXd, CLDXd and V
CDXd were estimated as 2.68 L/h and 26.22 L, closely aligning with previous values (2.66 L/h and 25.1 L) [18]. Covariate effects were consistent with prior findings. Effects of baseline body weight on CLlinDatoDXd, V
CDatoDXd, V
pDatoDXd, CLDXd, and V
cDXd were characterized in the model. In addition to baseline body weight, baseline albumin, age, region, and sex were included as covariates on CLlinDatoDXd; sex was included for V
CDatoDXd; baseline albumin, aspartate transaminase (AST), baseline total bilirubin (TBIL), and region were included for CLDXd; and sex was included for V
cDXd. Baseline tumor size was identified as a covariate for V
max. Covariate correlation is minimal and unlikely to impact the individual post hoc estimates of Dato‐DXd/DXd exposure. The relationships between the covariates and the model parameters are described in the following equations:

Good agreement between the PopPK model prediction and the PK observations is demonstrated in Figure S1A. The conditional weighted residuals (CWRES) appear evenly distributed around zero over time, indicating no systematic bias in the structural model for both Dato‐DXd and DXd (Figure S1B). Goodness of fit (GOF) was comparable between lung and breast cancer and acceptable in individual fittings. Baseline body weight, baseline albumin, age, sex, region, and baseline tumor had mild impacts (generally within 80%–125% range as compared to reference group) on Dato‐DXd PK as suggested by univariate analysis (Figure 1C). Baseline body weight, baseline albumin, baseline AST, TBIL, region, and sex were included as covariates to describe DXd PK. At the 95th percentile of observed baseline body weight (96.96 kg) in the analysis population, DXd PK changes greater than 30% (+38.8% for DXd first‐cycle area under the concentration‐time curve [AUC], respectively). The impacts of other covariates on DXd PK are within 80%–125% range.
Multivariate analysis was performed to further investigate the impacts of additional covariates of clinical interest, such as tumor type, on Dato‐DXd and DXd PK. At 6 mg/kg, the differences in first‐cycle AUC between lung cancer and breast cancer patients were minimal (2.82% for Dato‐DXd and 0.933% for DXd, respectively) (Figure S1). The differences were not considered clinically meaningful. None of the other covariates included in the post hoc analysis, including formulation, TEADA, region, hepatic impairment, renal impairment, sex, race, age regulatory categories, ECOG, prior IO therapy, and smoke status, demonstrated a clinically significant impact on Dato‐DXd and DXd PK (data not shown). The impacts of TROP2 expression and prior lines of therapy on PK were evaluated in TROPION‐Breast01 only and were not clinically significant (data not shown). PK metrics were derived for further TGI‐PFS model development.

3.2
Joint Modeling of Tumor Growth Inhibition and PFS
Baseline characteristics of the patients in TGI‐PFS analysis dataset were described in Table 2. A total of 390 patients with HR+/HER2− BC, including 39 patients from TROPION‐PanTumor01 and 351 from TROPION‐Breast01, were included in TGI modeling analysis dataset, resulting in a total of 2230 tumor size observations with median (95% confidence interval) follow‐up time of 265 (255–294) days. All the 390 patients received Dato‐DXd at 6 mg/kg. Analysis population baseline characteristics were generally comparable between studies TROPION‐PanTumor01 and TROPION‐Breast01. No correlations were observed between continuous baseline characteristics in the TGI‐PFS analysis population (i.e., age, baseline body weight, baseline albumin, and baseline tumor size) (Figure S2A). Dato‐DXd first‐cycle AUC had strong correlations with other derived Dato‐DXd PK metrics (Figure S2B).
Majority of patients included in the TGI analysis had tumor shrinkage (Figure S2C,D). TGI model selection is described in Tables S3–S5. The final TGI model includes the functions of exponential tumor growth, linear tumor growth inhibition rate correlated with cycle‐specific Dato‐DXd average concentration (Cavg), and resistant appearance as a function of treatment time (Figure 2A). A correlation between the random effects for tumor growth rate (kge) and baseline tumor size (TS0) was incorporated into the TGI model. No covariate‐parameter relationships were identified. The parameter estimations of the final TGI model are denoted in Table S6. The final TGI model recapitulated the observed tumor size data (Figure 2B, Figure S2E).
Model‐predicted tumor response metrics were in good agreement with observed values. The median predicted TSR6 of the 390 patients suggested a 13.5% decrease in tumor size, similar to observations (10% decrease at the first tumor scanning post start of the treatment, respectively). The median values of the TTG were comparable between observation and prediction (90 vs. 116 days, both approximately between the second and third scheduled tumor scanning at week 12 and 18). The median value of predicted initial tumor size was comparable to the observed tumor baseline value (56 mm vs. 55 mm, respectively).
Several tumor response metrics, including TTG, TSR6, initial tumor size, tumor growth rate, and resistance appearance rate, were identified to be significantly correlated with PFS in CPH and KM analyses (Table S7). TTG and TSR6 are correlated with Dato‐DXd PK exposure. Patients with lower Dato‐DXd PK exposure (median first‐cycle Dato‐DXd AUC: 573.4 vs. 738.9 μg mL/day, respectively) exhibited a shorter TTG (107.5 days vs. 134 days, respectively) and higher TSR6 (88% vs. 85%, respectively) (Figure S3A). The differences were statistically significant (p < 0.05).
Significant separations in PFS KM curves were observed across time to tumor growth quartiles. Patients in the higher TTG subgroup (median TTG 225.2 days) had a 209‐day longer median PFS compared with patients in the lower TTG subgroup (median TTG 61.5 days) (Figure S3B). The final joint TGI‐PFS survival model was best described by a Weibull distribution, with TTG as a significant covariate on the scale parameter, without other statistically significant covariates (Tables S8 and S9). A visual predictive check demonstrated that the joint TGI‐PFS model recapitulated the observed PFS data (Figure 2C).

3.3
Progression Free Survival Simulations
A total of 500 virtual trials, each including 400 virtual patients, were simulated as illustrated in Figure 3A. Dato‐DXd dose levels that were not clinically evaluated (e.g., 4 mg/kg) were compared to 6 mg/kg in the 500 virtual trials by simulating the Dato‐DXd PK, patient tumor size, and PFS profiles at different Dato‐DXd dose levels.
Simulated Dato‐DXd concentration‐time profiles at 6 mg/kg closely matched the observed pharmacokinetic data (Figure S4A). The first‐cycle AUC values for Dato‐DXd at 4, 6, and 8 mg/kg in the simulations showed good agreement with observed values in actual patients (Figure S4B). Virtual patients receiving 6 mg/kg demonstrated higher PK exposure compared to those receiving 4 mg/kg. Overlaps in Dato‐DXd PK between 4 and 6 mg/kg were observed in the simulated PK profiles as expected (Figure 3B).
Tumor size time‐course data were simulated for 200,000 virtual patients receiving Dato‐DXd at doses of 4, 6, or 8 mg/kg. The simulated tumor size profiles well‐captured the observed data in patients with HR+/HER2‐ BC treated with Dato‐DXd at 6 mg/kg (Figure S4C). Tumor response metrics were consistent between the virtual patient simulations and the analysis population (Figure S4D, Table S10). The median simulated TTG was 130 days, aligning with the predicted value of 116 days. Similarly, the TSR6 was comparable between the simulation and prediction (83.5% vs. 86.5%, respectively). At 25–30 weeks post‐first treatment, the simulated geometric mean tumor shrinkage at 6 mg/kg was 49%, which was comparable to the observed geometric mean value of 40.2% (based on a patient sample size of N = 268). Simulations suggested that greater tumor shrinkage was achieved at higher dose levels. As shown in Figure 3C, at 180 days post‐first treatment, the geometric mean tumor shrinkage from baseline was 30% at 4 mg/kg, 49% at 6 mg/kg, and 63% at 8 mg/kg, accounting for virtual patient discontinuation.
The comparison of KM curves and survival rates between the 500 simulated trials at 6 mg/kg and the observations suggested acceptable prediction by the parametric PFS model (Figure S4E). Simulated median PFS (95% CI) at 6 mg/kg is consistent with the observation (195.5 [176, 217] vs. 210 [172, 244] days, respectively). Simulations suggested a 34.5‐day increase in median PFS at 6 mg/kg compared to 4 mg/kg (195.5 days vs. 161 days, respectively; Figure S4F). The increase in median PFS from 6 to 8 mg/kg was smaller than that observed from 4 to 6 mg/kg (+21 days vs. +34.5 days, respectively; Figure S4F and Figure 3D). Higher doses resulted in increased survival rates across different tumor scanning intervals. For patients receiving Dato‐DXd at 4 mg/kg who experienced disease progression before the first tumor scan (i.e., within the first 6 weeks of treatment), increasing the dose to 6 mg/kg was predicted to extend median PFS by approximately 100 days (data not shown).

4
Discussion
Semi‐mechanistic models that characterize relationships between PK exposure and efficacy outcomes have emerged as powerful tools for optimizing dosing strategies in oncology drug development [4, 5]. In the current analysis, a joint TGI‐PFS model was developed using data from a single clinically evaluated dose level of 6 mg/kg Q3W in patients with HR+/HER2‐ BC. Virtual trials were simulated at untested Dato‐DXd dose levels (e.g., 4 mg/kg) to further investigate the selection of 6 mg/kg Qdose as the optimal dose for patients with HR+/HER2− BC in the context of Project Optimus.
In the current model, the tumor‐killing effect was driven by Dato‐DXd PK exposure, consistent with the mechanism of action for ADCs [11, 24, 25, 26, 27, 28]. Overlap in Dato‐DXd PK exposure was expected between the 4 and 6 mg/kg dose levels due to inter‐individual variability. Approximately 89% of virtual patients receiving 4 mg/kg had first‐cycle Dato‐DXd AUC values that overlapped with those receiving 6 mg/kg (5th–95th percentiles: 183–819 μg day/mL at 4 mg/kg vs. 306–1300 μg day/mL at 6 mg/kg, respectively). Although the TGI model was developed using data from a single clinically evaluated dose level of 6 mg/kg, the inclusion of Dato‐DXd PK exposure at 4 mg/kg enabled reasonable predictions of tumor size changes at this lower dose. Simulations using the current TGI model indicated that Dato‐DXd 6 mg/kg resulted in greater tumor shrinkage compared to 4 mg/kg in patients with HR+/HER2‐ BC. These results are consistent with findings from a previous TGI model, which was developed across a broader Dato‐DXd dose range (0.27–10 mg/kg) in NSCLC patients from the TROPION‐PanTumor01 study (N = 187) [17]. Simulated maximum tumor size reductions in patients with NSCLC were greater at higher Dato‐DXd dose levels, with predicted maximum tumor shrinkage from baseline in patients with NSCLC 30% at 4 mg/kg, 37% at 6 mg/kg, and 41% at 8 mg/kg [17]. These results highlight the dose‐dependent relationship between Dato‐DXd exposure and tumor shrinkage across multiple indications.
Time to tumor growth has been evaluated as a biomarker to efficacy outcomes [10, 29, 30, 31]. In the current analysis, TTG is defined as the theoretical time to individual nadir that is predicted by the TGI model and based on the sum of the longest diameters of target lesions per RECIST v1.1 [31, 32]. Each patient has a time‐independent, unique estimate of TTG at a certain dose level. Patients with shorter TTG were characterized by significantly greater initial tumor size (62.8 vs. 47.1 mm), faster tumor growth (0.00183 vs. 0.00037 day−1), faster resistance appearance (0.0165 vs. 0.00997 day−1), and lower tumor inhibition rate (0.000137 vs. 0.00022 day−1). Notably, patients with lower Dato‐DXd PK exposure exhibited significantly shorter TTG compared to those with higher PK exposure (Figure S3A), indicating that Dato‐DXd PK exposure is a key determinant of TTG. Associations between shorter TTG and limited tumor eradication capability (correlation coefficient = −0.814) and faster resistance appearance rate (correlation coefficient = −0.768) were observed. Other patient baseline characteristics were comparable between TTG lower and upper halves (Table S11).
Trial simulations were cautiously designed to ensure good predictive ability. Virtual patients were generated using a multivariate normal distribution derived from the existing TGI‐PFS analysis population, ensuring physiologically plausible individual parameters. In the current analysis, virtual patients were not resampled across Dato‐DXd dose levels to reduce the potential impact of patient variability and to better represent hypothetical clinical scenarios. Resampling virtual patients across Dato‐DXd dose levels yielded similar conclusions (data not shown). For virtual patients with predicted TTG values exceeding the maximum observed value in the analysis population (TTG = 482.5 days), a random time from treatment initiation to the analysis cutoff was assigned based on the observed distribution. These patients represented only a small fraction of the total virtual population (1.6%).
Simulated tumor dynamics and PFS profiles indicated that Dato‐DXd at 6 mg/kg provides greater tumor inhibition and longer PFS compared to 4 mg/kg in patients with HR+/HER2− BC. Primary PFS by blinded independent central review (BICR) at Biologics License Application (BLA) was 6.9 months for Dato‐DXd 6 mg/kg versus 4.9 months for investigator's choice of chemotherapy (ICC) [12]; therefore, simulated PFS at 4 mg/kg is closer to ICC arm than Dato‐DXd 6 mg/kg arm, further suggesting the efficacy benefit of Dato‐DXd 6 mg/kg Q3W regimen for patients with HR+/HER2− BC.
On January 17 2025, Dato‐DXd was approved by the US FDA for adult patients with unresectable or metastatic HR+/HER2‐ BC (IHC 0, IHC1+ or IHC2+/ISH−) who have received prior endocrine‐based therapy and chemotherapy for unresectable or metastatic disease without post‐market commitment (PMC) or post‐market requirement (PMR) for dose optimization [33]. The current study leveraged the TGI‐PFS model to simulate clinical outcomes for untested Dato‐DXd dose levels in the target indication, addressing critical gaps in data and supporting dose optimization efforts. We demonstrated that the TGI‐PFS model provides valuable insights into the relationship between dosing, PK exposure, tumor dynamics, and PFS. While the results support the current dose selection of Dato‐DXd in patients with HR+/HER2− BC, further exploration and validation of the modeling framework are warranted in diverse patient populations and across other tumor types. This work underscores the importance of integrating advanced modeling techniques into the dose optimization paradigm.

Author Contributions
Z.T. and D.Z. wrote the manuscript, designed the research, and performed the research; all authors analyzed the data.

Funding
The study was sponsored by AstraZeneca.

Conflicts of Interest
Y. Jiang, M. Munegowda, S. Khan, L. Xu, S. Ren, P. Vajjah, and D. Zhou are current employees of and shareholders in AstraZeneca. Z. Tang, K. Lim, D. Dai, L. Shi, and M. Gibbs were previously employed by AstraZeneca, and the work presented in this article was conducted during their tenure at the company. Y. Pan, Y. Hong are current employees of Daiichi Sankyo and own stock in Daiichi Sankyo.

Supporting information

Data S1: cts70493‐sup‐0001‐DataS1.docx.