The objective of phase 1 dose-finding studies is to determine the maximum tolerated dose (MTD), defined as the highest dose at which a pre-specified proportion of subjects experience a dose limiting toxicity (DLT). Ideally, the trial design would not only be easy to implement but also efficient and ethical. In recent years, several statistical models for adaptive trial designs have emerged.
Even with numerous publications detailing novel designs, the classical rule-based 3+3 design remains a popular phase I trial design. In this trial design, a cohort of patients (n=3) is treated and decisions regarding escalations to the next dose level are determined based on the number of DLTs observed. Decisions are made on cohort sizes of 3 or 6. There are no statistical models, and therefore, it is easy to understand and implement. However, performance of this trial design depends on the number of dose levels planned for the study. The greater the number of dose levels to be tested, the higher the potential for exposing more subjects to subtherapeutic doses in the search for the MTD.
There are several alternative design options to the 3+3 design to consider when designing a trial. One such option is the Bayesian Optimal Interval (BOIN) design. For this design, decisions to escalate or de-escalate simply involve comparing the observed DLT rate at a given dose with prespecified escalation and de-escalation boundaries. In fact, the 3+3 design is nested within the BOIN design when the target DLT rate is 30% and the cohort size is 3. However, the 3+3 design requires that the number of subjects at a dose level does not exceed 6, whereas the BOIN design does not impose this requirement. In addition, newer methodologies exist that allow late onset toxicities to be considered. For example, immune-related adverse events may occur outside the DLT evaluation period, especially as subjects continue to receive additional doses. One example of a newer methodology is the time-to-event BOIN(TITE-BOIN), which allows real-time dose assignment for new subjects while toxicity data is still pending for some currently enrolled subjects.
A newer rule-based design is the i3+3 design. This design is similar to BOIN in that upper and lower bounds for the target toxicity level are chosen. The difference between the designs is what happens when the DLT rate is above the upper bound. When this happens, testing whether removing just one toxicity would cause a big jump from above the upper bound to below the lower bound is warranted. Should a large difference occur, this would imply that there is not enough information to tell if the boundary is tipped, and the decision would be to retain the current dose rather than de-escalate in an effort to accrue additional data.
Model-based designs provide an alternative to the rule-based designs and can offer some advantages. For example, much like the BOIN design, which utilizes an isotonic regression method to pool information across doses, model-based designs allow for the use of all available toxicity data from all dose levels. In addition, they can be cost effective and improve efficiency through fewer subjects receiving sub-optimal doses and fewer subjects receiving potentially toxic doses. Furthermore, they can identify the true MTD a higher proportion of time compared to the rule-based 3+3 design. However, there are also disadvantages of model-based designs. Specifically, they can be statistically and computationally complex, they often require simulations before the trial starts to calibrate design parameters and confirm acceptable operating characteristics, and they often require a procedure that allows study staff to obtain dose assignments in real time when a subject needs to be treated. Additionally, time needed to run ‘models’ for model-based designs may delay safety review meetings.
References
- Ji, Yuan, et al. “A Modified Toxicity Probability Interval Method for Dose-Finding Trials.” Clinical Trials: Journal of the Society for Clinical Trials, vol. 7, no. 6, 2010, pp. 653–663., doi:10.1177/1740774510382799.
- Pallmann P, et al. “Adaptive designs in clinical trials: why use them, and how to run and report them.” BMC Med. 2018 Feb 28;16(1):29. doi: 10.1186/s12916-018-1017-7.
- Yuan Y, et al. “Bayesian Optimal Interval Design: A Simple and Well-Performing Design for Phase I Oncology Trials”. Clin Cancer Res. 2016 Sep 1;22(17):4291-301. doi: 10.1158/1078-0432.CCR-16-0592. Epub 2016 Jul 12.
- Wheeler GM, Mander AP, Bedding A, Brock K, Cornelius V, Grieve AP, Jaki T, Love SB, Odondi L, Weir CJ, Yap C, Bond SJ. How to design a dose-finding study using the continual reassessment method. BMC Med Res Methodol. 2019 Jan 18;19(1):18. doi: 10.1186/s12874-018-0638-z.
- James GD, Symeonides SN, Marshall J, Young J, Clack G. Continual reassessment method for dose escalation clinical trials in oncology: a comparison of prior skeleton approaches using AZD3514 data. BMC Cancer. 2016 Aug 31;16(1):703. doi: 10.1186/s12885-016-2702-6.
- trialdesign.org
- Wages, Nolan A et al. “Design considerations for early-phase clinical trials of immune-oncology agents.” Journal for immunotherapy of cancer vol. 6,1 81. 22 Aug. 2018, doi:10.1186/s40425-018-0389-8.
- Emens LA, Bruno R, Rubin EH, Jaffee EM, McKee AE. Report on the Third FDA-AACR Oncology Dose-Finding Workshop. Cancer Immunol Res. 2017 Dec;5(12):1058-1061. doi: 10.1158/2326-6066.CIR-17-0590.
- Liu M, Wang SJ, Ji Y. The i3+3 design for phase I clinical trials. J Biopharm Stat. 2020 Mar;30(2):294-304. doi: 10.1080/10543406.2019.1636811. Epub 2019 Jul 15.
- Detangle modern dose-finding designs: A tutorial https://www.youtube.com/watch?v=49WgGtAEJ8g