Impact of Dose‐Escalation Design on the Safety and Development of Anticancer Drugs in Clinical Trials

Abstract

ABSTRACT

The operational characteristics of dose‐escalation design in phase I studies have been studied using simulations; however, there is limited analysis regarding its effects on the results of clinical trials. We collected the data of 394 clinical trials involving dose‐escalation studies for anticancer drugs submitted to the Pharmaceuticals and Medical Devices Agency between 2013 and 2022. We used the internal data of the PMDA and published papers and analyzed outcomes such as enrollment and drug development. We identified model‐based designs and rule‐based designs as the two primary designs. The median number of dose‐limiting toxicity (DLT)‐evaluated patients was higher for model‐based designs than for rule‐based designs. The proportion of rule‐based designs was higher in Japanese trials and that of model‐based designs was higher in multiregional clinical trials (MRCTs). The determined recommended phase II dose (RP2D) was consistent with the approved dose in all trials (13/13) involving model‐based designs and in 84.0% (21/25) of trials involving rule‐based designs, although it was not statistically significant. The proportion of progression to the next study phase was 50.0% (61/122) for rule‐based designs and 56.3% (36/64) for model‐based designs. Similar trends in these outcomes were observed when MRCTs and Japanese trials were examined separately. Model‐based designs might require more DLT‐evaluated patients; however, they might have different operational capabilities compared with rule‐based designs, such as selecting an RP2D consistent with the approved dose. The results might help in selecting the optimal dose‐escalation methods in future phase I trials.

---

Study Highlights

• What is the current knowledge on the topic?

  • ○

    Dose-escalation studies of anticancer drugs are important because they determine the recommended phase II dose. Model-based designs have been reported to be superior to rule-based designs in terms of accuracy in defining the maximum tolerated dose in analyses involving computer simulations; however, their effects on the results of clinical trials, such as enrollment number, selection of maximum tolerated dose, and drug development are unclear.

• What question did this study address?

  • ○

    The analyses addressed the impact of model-based designs and rule-based designs on outcomes, such as enrollment number, safety, and anticancer drug development by analyzing data of 394 trials submitted to the Pharmaceuticals and Medical Devices Agency (Japanese regulatory authority) between 2013 and 2022 and published papers.

• What does this study add to our knowledge?

  • ○

    The study revealed that rule-based designs may require fewer dose-limiting toxicity-evaluated patients to determine the maximum tolerated dose than model-based designs. It also revealed that model-based designs might have advantages in operational capability, such as selecting a recommended phase II dose consistent with the approved dose when compared with rule-based designs.

• How might this change clinical pharmacology or translational science?

  • ○

    The results might help determine which dose-escalation method to select in a future phase I clinical trial, even though this is a preliminary result and further studies are necessary to draw a definite conclusion.

---

Introduction

The goal of phase I trials in anticancer drug development is to establish the recommended phase II dose (RP2D) [1]. Traditionally, the recommended drug dose has been based on its toxicity, defined as the maximum tolerated dose (MTD), which is the highest dose considered to be tolerable based on prespecified dose-limiting toxicity (DLT) criteria [2]. This dose determination lies under the paradigm that a higher dose is associated with a higher effect and higher toxicity, especially for cytotoxic drugs [3]. Hansen et al. reported that among 161 trials conducted between 2001 and 2015, the RP2D was determined using toxicity alone in 52 trials and using multiple data, such as toxicity, pharmacodynamics, and efficacy, in 87 trials, suggesting that toxicity is frequently used in the determination of the RP2D in phase I trials [4]. Recently, the concept of optimal biological dose (OBD) was introduced to account for efficacy in addition to toxicity to determine the RP2D [5, 6], and there has been increasing use of pharmacodynamics and efficacy data in phase I trials, especially for molecular-targeted agents and antibody drugs.

Classically, dose-escalation methods roughly fall into two categories: rule-based designs and model-based designs [7]. Rule-based designs use prespecified rules, and the traditional 3 + 3 design is most frequently adopted [1]. Model-based designs use Bayesian statistical models that utilize accumulated toxicity data from all patients treated with previous doses to determine the next dose [1, 2], and the commonly used model-based designs are the continual reassessment method and Bayesian logistic regression model. Recently, model-assisted designs, which are hybrid methods combining the characteristics of rule-based designs and model-based designs, have been developed [2].

Model-based designs have been reported to be superior to rule-based designs in terms of accuracy in defining the MTD in analyses involving computer simulations, and rule-based designs have been reported to have a higher chance of choosing lower doses [2, 8]. However, model-based designs had not been as frequent as rule-based designs in practice until recently. This was mainly because of the need for biostatisticians with expertise and the requirement of computer programs for the planning and execution of trials with model-based designs, which are more complex and require more consideration [9]. The operational characteristics of these dose-escalation designs have been mainly studied with computer simulations [9, 10], and few studies were based on data obtained from clinical trials [11].

Thus, this study aimed to investigate the effects of dose-escalation design on outcomes, such as enrollment number, safety, and drug development, using data from clinical trials.

Methods

In Japan, for the initial submission of drugs with new active ingredients, drugs with new administration routes, and drugs with new medical combinations, the clinical trial sponsor or the person conducting the trial must notify the Pharmaceuticals and Medical Devices Agency (PMDA). To assess the impact of dose-escalation design on the safety and development of anticancer drugs in clinical trials, we analyzed 394 clinical trials involving dose-escalation studies that evaluated tolerability as monotherapy (rule-based designs (e.g., classical 3 + 3 and modified 3 + 3) [n = 195]; model-based designs (e.g., continual reassessment method and Bayesian logistic regression model) [n = 114]; model-assisted designs (e.g., modified toxicity probability interval and Bayesian optimal interval design) [n = 85]) and were submitted to the PMDA between 2013 and 2022 (Figure 1). As background data, the following information was collected using clinical trial notifications: fiscal year when the clinical trial notification was submitted, first-in-human (FIH) trial or first-in-Japanese trial, region where the trial was conducted (Japanese trial or multiregional clinical trial [MRCT]), type of disease (solid tumor, hematological malignancy, or both), mechanism of action (molecular-targeted agent, cytotoxic drug, or monoclonal antibody and antibody–drug conjugate [ADC]), and dose-escalation method (rule-based, model-based, or model-assisted design).

[IMAGE OMITTED. SEE PDF]

Considering the interval from the initiation of phase 1 trials to regulatory approval, the analysis of outcome data, including regulatory approvals, was limited to trials for which submissions were made between 2013 and 2019. The following information was collected using clinical trial consultation documents, common technical documents submitted in new drug applications, notifications of clinical trials of subsequent studies, “” and published papers, and was tracked until April 30, 2024: continuous variable data, including study period (first enrollment to last visit), enrollment number (number of patients enrolled in the dose-escalation part), observed DLT, DLT-evaluated patients and patients administered a dose over the MTD; and nominal variable data, including whether the MTD was determined, whether the MTD was equal to the RP2D, whether the RP2D was equal to the approved dose, proportion of progression to the next study phase, proportion of progression beyond a phase II trial and proportion of regulatory approval. These data were not always presented in the available data source, and each outcome was analyzed by using available data.

The outcome data were presented as medians and interquartile ranges for continuous variables and as proportions and relative risks (RRs) for nominal variables. As the number of clinical trials with model-assisted designs submitted from 2013 to 2019 was small (n = 27), we focused on rule-based and model-based designs, and statistical analyses were performed only for these designs. Specifically, all outcomes, except the number of patients enrolled and study period, were statistically analyzed using the Wilcoxon test for continuous variables and Fisher's exact test for nominal variables. The significance level was set at p < 0.05. In the case of the Wilcoxon test for the number of observed DLTs and the number of patients administered a dose over the MTD, the data were adjusted with division by the number of DLT-evaluated patients.

Results

Table 1 shows the background data of all trials. Among the 394 trials investigated, 195 used rule-based designs, 114 used model-based designs, and 85 used model-assisted designs. The number of trials with model-assisted designs increased between 2013 and 2022, whereas the number of trials with rule-based designs and model-based designs did not change markedly. Domestic trials accounted for 68.2% of rule-based designs, 21.1% of model-based designs, and 25.9% of model-assisted designs. FIH trials accounted for 29.2% of rule-based designs, 73.7% of model-based designs, and 38.8% of model-assisted designs. The distribution of drug mechanisms was similar among the dose-escalation designs.

TABLE 1 Background data of all clinical trials submitted from 2013 to 2022.

	Rule-based	Model-based	Model-assisted
n = 195	n = 114	n = 85
Fiscal year
2013	23 (11.8)	10 (8.8)	0 (0.0%)
2014	20 (10.3)	6 (5.3)	2 (2.4%)
2015	13 (6.7)	10 (8.8)	0 (0.0%)
2016	21 (10.8)	12 (10.5)	4 (4.7%)
2017	22 (11.3)	15 (13.2)	4 (4.7%)
2018	24 (12.3)	14 (12.3)	9 (10.6%)
2019	20 (10.3)	17 (14.9)	8 (9.4%)
2020	19 (9.7)	4 (3.5)	22 (25.9%)
2021	15 (7.7)	13 (11.4)	14 (16.5%)
2022	18 (9.2)	13 (11.4)	22 (25.9%)
Region
Japan	133 (68.2)	24 (21.1)	22 (25.9)
MRCT	62 (31.8)	90 (78.9)	63 (74.1)
FIH
Yes	57 (29.2)	84 (73.7)	52 (38.8)
No	138 (70.8)	30 (26.3)	33 (61.2)
Disease
Solid tumor	151 (77.4)	89 (78.1)	64 (75.3)
Hematological malignancy	39 (20.0)	19 (16.7)	18 (21.2)
Both	5 (2.6)	6 (5.3)	3 (3.5)
Mechanism of action
Molecular targeted drug	118 (60.5)	66 (57.9)	47 (55.3)
Cytotoxic drug	5 (2.6)	1 (0.9)	2 (2.4)
Monoclonal antibody	62 (31.8)	38 (33.3)	29 (34.1)
Antibody-drug conjugate	10 (5.1)	9 (7.9)	7 (8.2)

Our preliminary analysis revealed that the interval from the initiation of phase 1 trials to regulatory approval was approximately 5 years; thus, the analysis of outcome data, including regulatory approval, was limited to the trials that were submitted between 2013 and 2019 to ensure data availability. Table 2 shows the outcomes of continuous variables. The median numbers of observed DLTs, DLT-evaluated patients, and patients administered a dose over the MTD were higher for model-based designs than for rule-based designs (p ≤ 0.01).

TABLE 2 Outcome data of all clinical trials submitted from 2013 to 2019 (1).

Clinical trials submitted from 2013 to 2019	Rule-based	Model-based	p	Model-assisted
n = 143	n = 84	n = 27
Enrollment number
Evaluated trials	60	50		11
Median (IQR)	17 (12–24.5)	34.5 (17.8–75.5)		31 (12–44)
Study period
Evaluated trials	42	28		4
Median (IQR)	974 (553.5–1469)	1141.5 (639–1622.5)		611.5 (410.5–719.5)
DLT number
Evaluated trials	59	51		11
Median (IQR)	0 (0–2)	1 (0–5)	< 0.01a	0 (0–1)
Over MTD
Evaluated trials	17	13		No data
Median (IQR)	1 (0–5.5)	6 (3–11.5)	0.01a
DLT evaluation
Evaluated trials	59	51		11
Median (IQR)	14 (10–21)	29 (15–68)	< 0.01	22 (6–44)

Table 3 shows the outcomes of nominal variables. The proportion of MTD determination (reaching the MTD) and that of the MTD being equal to the RP2D were not different between the two designs. Notably, the determined RP2D was equal to the approved dose in all 13 investigational drugs for model-based designs and in 21 out of 25 drugs for rule-based designs; however, it was not statistically significant. (RR [95% CI]: 1.19 [1.00, 1.41], p = 0.17). Next, the effects of dose-escalation design on drug development were assessed. The proportion of progression to the next study phase was 50.0% (61/122) for rule-based designs and 56.3% (36/64) for model-based designs (RR [95% CI]: 1.13 [0.85, 1.49], p = 0.26). There were no differences between the two designs in the proportion of progression beyond the second phase (RR [95% CI]: 1.00 [0.67, 1.48], p = 0.57) and the proportion of regulatory approval (RR [95% CI]: 1.01 [0.57, 1.79], p = 0.56).

TABLE 3 Outcome data of all clinical trials submitted from 2013 to 2019 (2).

	Rule-based	Model-based	Relative risk [95% CI]	p	Model-assisted
(control)		n = 27
n = 143	n = 84
Yes/N (%)	Yes/N (%)	Yes/N (%)
MTD determination	19/62 (30.6%)	18/54 (33.3%)	1.09 [0.64, 1.85]	0.46	1/10 (10.0%)
MTD = RP2D	11/12 (91.7%)	8/10 (80.0%)	0.87 [0.61, 1.24]	0.92	No data
RP2D = approved dose	21/25 (84.0%)	13/13 (100.0%)	1.19 [1.00, 1.41]	0.17	No approved drug
Progression to the next study phasea	61/122 (50.0%)	36/64 (56.3%)	1.13 [0.85, 1.49]	0.26	10/22 (45.5%)
Progression beyond a phase II triala	43/110 (39.1%)	23/59 (39.0%)	1.00 [0.67, 1.48]	0.57	5/19 (26.3%)
Approvala	25/95 (26.3%)	13/49 (26.5%)	1.01 [0.57, 1.79]	0.56	0/15 (0.0%)

Regarding background data (Table 1), the proportions of MRCTs and Japanese trials differed between rule-based designs and model-based designs. Thus, we examined the effects of dose-escalation design on outcomes separately in MRCTs and domestic trials. Regarding continuous variables, the median numbers of observed DLTs, DLT-evaluated patients, and patients administered a dose over the MTD were higher for model-based designs than for rule-based designs in both MRCTs and Japanese trials (Tables S1 and S2). Furthermore, regarding nominal variables, the determined RP2D was equal to the approved dose in all investigational drugs for model-based designs (MRCTs: 5, Japanese trials: 8) but not for rule-based designs (MRCTs: 5/6, Japanese trials: 16/19) (RR [95% CI]: 1.20 [0.84, 1.72] for MRCTs, 1.19 [0.98, 1.44] for Japanese trials) (Tables S3 and S4). The proportion of progression to the next study phase was higher for model-based designs than for rule-based designs in both MRCTs (21/42, 50.0% vs. 15/35, 42.9%) and Japanese trials (15/22, 68.2% vs. 46/87, 52.9%) (RR [95% CI]: 1.17 [0.72, 1.90] for MRCTs, 1.29 [0.91, 1.83] for Japanese trials) (Tables S3 and S4). These results show that the outcomes of the subgroup analysis (MRCTs vs. Japanese trials) were similar to those of all trials, suggesting that there is no obvious difference between the characteristics of the subgroups.

Discussion

Research regarding clinical trials using public data could have publication bias, as failed studies tend to remain unpublished [12]. A strength of this study is that it used clinical trial notifications in addition to published papers and thus could collect all background data and track the development of all investigational drugs during the research period. This allowed for a broader analysis of the effects of dose-escalation design, especially on anticancer drug development.

Some Japanese trials, which followed FIH trials conducted overseas with the RP2D determined, were performed on a small scale to confirm the tolerability to doses near the RP2D according to requests from the PMDA. In these cases, rule-based designs were often used for a small number of patients [13], and indeed, 68.2% of rule-based designs were used in Japanese trials and 70.8% were used for first-in-Japanese (non-FIH) drugs (Table 1).

Regarding continuous variables, the median numbers of observed DLTs, patients administered a dose over the MTD, and DLT-evaluated patients were higher for model-based designs than for rule-based designs (p ≤ 0.01) (Table 2). Rule-based designs have a higher chance of selecting lower doses as the MTD and could cause early termination of the dose-escalation phase, and they have lower accuracy in defining the MTD, as the dose escalation is based only on the toxicity data of the current dose [2]. Conversely, model-based designs enroll patients up to a targeted sample size. Moreover, they have a higher accuracy in MTD identification by utilizing the accumulated toxicity data from all patients treated at previous doses and by flexibly increasing the enrollment at a dose close to the MTD [2]. Model-based designs require a higher number of DLT-evaluated patients, and they had more observed DLTs and patients administered a dose over the MTD, which were considered to reflect the operational characteristics of the designs. Considering the tendency for patients to receive doses over the MTD, careful observation of each patient might be more important in clinical trials with model-based designs.

Regarding nominal variables, the RP2D doses were consistent with the approved doses for all (13/13) model-based designs and 84.0% (21/25) of rule-based designs (p = 0.17). However, the difference was not statistically significant; this could be due to the limited number of clinical trials analyzed. It might be debatable whether the approved dose is ideal; however, at least within the current drug development paradigm and according to scientific standards, the approved dose is supposed to be effective and safe based on clinical trial data.

Considering these aspects and our findings, model-based designs might require more patients to identify the MTD; however, they might have a different capability for determining the RP2D that is equal to the approved dose. In four trials where the RP2D was not equal to the approved dose with rule-based designs, the drug dose was reduced after determining the RP2D owing to toxicity in subsequent trials. These results suggest that although rule-based designs are considered relatively safe [2], caution might be necessary as a high dose could be administered at times, although this has not been reported, and dose reduction might be deemed appropriate in later trials.

The proportion of progression to the next study phase was higher for model-based designs than for rule-based designs (Table 3). Rule-based designs have a higher chance of selecting lower doses as the MTD, whereas model-based designs have a lower chance. If a lower dose is chosen as the MTD, the development might be terminated due to very low efficacy at the lower dose, and the different operational characteristics of the two designs might be a reason for the potentially different outcomes of drug development in this study. Conversely, the proportion of progression beyond phase II and that of approval had little difference between the two designs. In these later developmental phases, not only toxicity but also efficacy are important, and especially in phase III trials, factors such as selection of the drug and dose in the control arm can significantly affect the results [14]. Therefore, the limited impact of the difference in dose-escalation design on our results is reasonable.

This study has some limitations. We collected the data of five background items and 11 outcome items; however, some potentially informative outcomes, including the number of dose levels examined in dose-escalation studies, were not collected. We analyzed clinical trials using not only published papers but also submitted data from the PMDA. However, some outcome data were missing mainly for the following reasons: (1) The results of some trials had not been submitted to the PMDA yet, especially when they failed; (2) Some data were not described in the publications; (3) Some trials are still ongoing, and some data (e.g., study period) are not fixed; and (4) Some outcomes were not applicable to studies (e.g., when the MTD was not reached, MTD-related outcomes were not applicable). Nevertheless, we collected a lot of background and outcome data, especially data on drug development, using our approach. Therefore, the strength and originality of this study lie in its broader analysis of the outcomes of the clinical trials according to dose-escalation design, using data submitted to the regulatory authority.

The Food and Drug Administration launched Project Optimus in 2021 and issued a guidance in 2024 [15]. This project aims to assist the sponsors in dose selection for cancer treatments and emphasizes the importance of selecting doses that are optimized in terms of efficacy, tolerability, and safety. Although the concept of the OBD is attractive and could be a future direction, some issues need to be resolved. The selection and measurement of adequate biological effects required for OBD identification are not easy [16], and there is no consensus on the efficacy endpoint to be accounted for in the OBD or on the most appropriate dose-escalation strategy to apply when assessing the OBD [5].

In conclusion, our study revealed that model-based designs might require more DLT-evaluated patients to identify the MTD. However, model-based designs have some different operational characteristics compared with rule-based designs, such as a potentially different capability to identify the RP2D consistent with the approved dose. However, at this point, there is insufficient data to draw a definitive conclusion. Even if the paradigm of drug development shifts to the OBD being the optimal dose in the future, a dose-escalation study will still be necessary to determine the optimal dose, and indeed, some researchers have proposed the use of model-based designs [17, 18]. In any case, it is important to understand the operating characteristics of existing dose-escalation methods accurately through both simulations [19] and clinical trial data, and apply the findings to dose-identification studies. The demonstrated operating characteristics of model-based designs may be exerted in both the classical drug discovery paradigm utilizing the MTD and the future drug discovery paradigm utilizing the OBD. This will be revealed by the accumulation of data in future phase I studies utilizing model-based designs.

Author Contributions

A.N. wrote the manuscript; A.N., K.M., K.F., R.K., H.T., and Y.E. designed the research; A.N., K.M., and Y.E. performed the research; A.N. analyzed the data.

Acknowledgments

The authors would like to thank Dr. Mitsuru Arima, Dr. Kayo Shinohara, and Dr. Ryu Matsuo for manuscript preparation. The views expressed in this paper are those of the authors and do not necessarily reflect the official views of the PMDA.

Ethics Statement

This study was conducted using aggregated data derived from anonymized processed information of clinical trials, without any personally identifiable information. Therefore, ethics approval and informed consent were waived in accordance with Japanese ethical guidelines.

Conflicts of Interest

The authors declare no conflicts of interest.

References

J. Shen, B. Swift, R. Mamelok, S. Pine, J. Sinclair, and M. Attar, “Design and Conduct Considerations for First‐in‐Human Trials,” Clinical and Translational Science 12 (2019): 6–19, https://doi.org/10.1111/cts.12582.

D. V. Araujo, M. Oliva, K. Li, R. Fazelzad, Z. A. Liu, and L. L. Siu, “Contemporary Dose‐Escalation Methods for Early Phase Studies in the Immunotherapeutics Era,” European Journal of Cancer 158 (2021): 85–98, https://doi.org/10.1016/j.ejca.2021.09.016.

M. Ochoa de Olza, M. Oliva, C. Hierro, I. Matos, J. Martin‐Liberal, and E. Garralda, “Early‐Drug Development in the Era of Immuno‐Oncology: Are We Ready to Face the Challenges?,” Annals of Oncology 29 (2018): 1727–1740, https://doi.org/10.1093/annonc/mdy225.

A. R. Hansen, N. Cook, E. Amir, L. L. Siu, and A. R. Abdul Razak, “Determinants of the Recommended Phase 2 Dose of Molecular Targeted Agents,” Cancer 123 (2017): 1409–1415, https://doi.org/10.1002/cncr.30579.

R. Ananthakrishnan and S. Menon, “Design of Oncology Clinical Trials: A Review,” Critical Reviews in Oncology/Hematology 88 (2013): 144–153, https://doi.org/10.1016/j.critrevonc.2013.03.007.

P. F. Thall and S.‐J. Lee, “Practical Model‐Based Dose‐Finding in Phase I Clinical Trials: Methods Based on Toxicity,” International Journal of Gynecological Cancer 13 (2003): 251–261, https://doi.org/10.1046/j.1525‐1438.2003.13202.x.

Y. Ji, P. Liu, Y. Li, and B. N. Bekele, “A Modified Toxicity Probability Interval Method for Dose‐Finding Trials,” Clinical Trials 7 (2010): 653–663.

F. Yan, S. J. Mandrekar, and Y. Yuan, “Keyboard: A Novel Bayesian Toxicity Probability Interval Design for Phase I Clinical Trials,” Clinical Cancer Research 23 (2017): 3994–4003, https://doi.org/10.1158/1078‐0432.ccr‐17‐0220.

R. Kurzrock, C.‐C. Lin, T.‐C. Wu, B. P. Hobbs, R. C. Pestana, and D. S. Hong, “Moving Beyond 3+3: The Future of Clinical Trial Design,” American Society of Clinical Oncology Educational Book 41 (2021): e133–e144, https://doi.org/10.1200/edbk_319783.

P. Corbaux, M. El‐Madani, M. Tod, et al., “Clinical Efficacy of the Optimal Biological Dose in Early‐Phase Trials of Anti‐Cancer Targeted Therapies,” European Journal of Cancer 120 (2019): 40–46, https://doi.org/10.1016/j.ejca.2019.08.002.

A. Maki and M. Narukawa, “Factors Associated With Inclusion of Japan in Phase I Multiregional Clinical Trials in Oncology,” Therapeutic Innovation & Regulatory Science 58 (2024): 766–772, https://doi.org/10.1007/s43441‐024‐00655‐0.

P. F. Thall, “A Review of Phase 2–3 Clinical Trial Designs,” Lifetime Data Analysis 14 (2008): 37–53, https://doi.org/10.1007/s10985‐007‐9049‐x.

J. Fraisse, D. Dinart, D. Tosi, C. Bellera, and C. Mollevi, “Optimal Biological Dose: A Systematic Review in Cancer Phase I Clinical Trials,” BMC Cancer 21 (2021): 60, https://doi.org/10.1186/s12885‐021‐07782‐z.

E. L. Korn, J. A. Moscow, and B. Freidlin, “Dose Optimization During Drug Development: Whether and When to Optimize,” Journal of the National Cancer Institute 115 (2023): 492–497, https://doi.org/10.1093/jnci/djac232.

U.S. Department of Health and Human Services Food and Drug Administration, Oncology Center of Excellence (OCE), Center for Drug Evaluation and Research (CDER), Center for Biologics Evaluation and Research (CBER), “Optimizing the Dosage of Human Prescription Drugs and Biological Products for the Treatment of Oncologic Diseases. Guidance for Industry,” (2024), https://www.fda.gov/regulatory‐information/search‐fda‐guidance‐documents/optimizing‐dosage‐human‐prescription‐drugs‐and‐biological‐products‐treatment‐oncologic‐diseases.

J. R. Sachs, K. Mayawala, S. Gadamsetty, S. P. Kang, and D. P. de Alwis, “Optimal Dosing for Targeted Therapies in Oncology: Drug Development Cases Leading by Example,” Clinical Cancer Research 22 (2016): 1318–1324, https://doi.org/10.1158/1078‐0432.ccr‐15‐1295.

T. M. Braun, “The Bivariate Continual Reassessment Method. Extending the CRM to Phase I Trials of Two Competing Outcomes,” Controlled Clinical Trials 23 (2002): 240–256.

B. Nebiyou Bekele and Y. Shen, “A Bayesian Approach to Jointly Modeling Toxicity and Biomarker Expression in a Phase I/II Dose‐Finding Trial,” Biometrics 61 (2005): 343–354, https://doi.org/10.1111/j.1541‐0420.2005.00314.x.

Center for Product Evaluation, Pharmaceuticals and Medical Devices Agency (PMDA), “Statistical Considerations When Planning Phase I Clinical Trials in Oncology From the Safety Perspective. Early Consideration,” (2024), https://www.pmda.go.jp/files/000272425.pdf.