Reconsidering the 3+3 Dose Escalation in Oncology Studies
When it comes to first-in-human clinical trials in oncology, most clinical researchers opt for a simple, tried-and-true approach to finding a recommended phase 2 dose (RP2D), which in oncology is typically a standard 3+3 dose escalation design. This automatic default to the 3+3 design when developing a phase 1 protocol may be due to a lack of understanding of why we should consider alternative designs, and a lack of knowledge about what other options even exist.
In recent years, we have seen an emergence of molecularly targeted agents (MTAs) and immunotherapies as well as a transition away from cytotoxic agents. As the treatment landscape changes, it is becoming increasingly important for researchers to consider alternative study designs. For instance, patients being treated with MTAs and immunotherapies may not experience a dose limiting toxicity (DLT), and the side effects of these compounds may not be dose-dependent; therefore, dose escalating until the maximum tolerated dose (MTD) is identified may not be relevant.
Aside from the 3+3 dose escalation, there are alternative phase 1 study designs that should be considered which will help achieve RP2D faster, while exposing fewer patients to lower, less effective doses of the study drug. Alternate study designs have proven to be safer and more reliable when compared with the industry-standard 3+3 design, and some can be easily implemented with minimal statistical complexities
Rule-based designs apply simple rules to allow for step-up or step-down dosing in the absence or presence of toxicities seen at each dose level. The most widely used rule-based design in clinical practice is the traditional 3+3 design. Most people who have ever worked on an oncology clinical trial have experience with this design. The 3+3 design enrolls 3 patients at a given dose level. If no patients experience an adverse event that qualifies as a protocol defined DLT, then dosing may proceed to the next dose level. If one patient experiences a DLT within a specified timeframe (typically 1 cycle), then that dose level is expanded to include 3 more patients. If a second patient at that same dose level experiences a DLT, then the MTD is considered to have been exceeded and the next-lower dose level should be expanded to confirm the MTD.
By far the most popular phase 1 study design in oncology, the 3+3 dose escalation has been utilized in more than 95% of published phase 1 trials in the past two decades (Ji & Wang, 2013). One of the benefits of the 3+3 design (and likely a major factor to its popularity) is that it is very simple and can be easily understood by study investigators and clinical researchers. The logistical simplicity of the design together with familiarity with the escalation rules by clinicians and researchers are likely precluding exploration and implementation of novel study designs (Hansen et. al., 2014).
However, our continued reliance on the 3+3 design should be questioned. Statistical simulations have shown that the MTD is identified in as few as 30% of clinical trials that utilized a 3+3 design (Hansen et. al., 2014). Furthermore, the 3+3 design exposes an unnecessary number of patients to subtherapeutic doses. Due to the number of escalations and the number of patients required to be treated at each dose level, a large proportion of patients are treated at low doses that are potentially subtherapeutic, while few patients receive doses at or close to the RP2D (Le Tourneau et. al., 2009; Simon et. al., 1997).
Model-based designs use data from each dose level to model a dose-toxicity curve and provide a confidence interval for the RP2D, once achieved (Le Tourneau et. a., 2009). While these designs do require biostatistics expertise and statistical modeling software, model-based designs can achieve better estimations of the target probability of a DLT at the RP2D while minimizing suboptimal dosing (Hansen et. al., 2014). Particularly for agents with a low expected toxicity profile, it may make sense to consider a model-based design, given that that model-based designs assume a relationship between the study drug dose and the likelihood of occurrence of a DLT. Furthermore, in model-based design, medical decisions are based on statistical inference, which reduces subjectivity in the dose escalation decision-making process (Ji & Wang, 2013).
Traditional model-based designs such as the continual reassessment method (or CRM) were first introduced nearly 30 years ago, and yet they are still scarcely used in practice due to a perception of being too statistically complex (Wheeler et. al., 2019; Yan et. al., 2017). More recently, there has been increased interest in a combination design that incorporates the simplicity of a rule-based design with the better performance of a model-based design; a model is used for decision making but it allows for the decision making rules to be pre-tabulated before the trial begins (Yan et. al., 2017).
One such combination design, the modified Toxicity Probability Interval (mTPI) design, is equally as simple, transparent, and cost less to implement as the 3+3 design (Ji & Wang, 2013). The mTPI design requires a biostatistician to generate a simple decision table to be included in the protocol based on the number of planned dose levels in the study. In the decision table, the dose may be escalated, de-escalated, or eliminated based on the number of subjects treated and the number of DLTs. “Eliminate” means that the current and higher doses will be eliminated from the trial to prevent treating any future subjects at these doses because they are overly toxic. In a simulation of 2,000 trials comparing the operating characteristics of the 3+3 design and the mTPI design, it was concluded that compared with the 3+3 design, the mTPI design is safer, because it treats fewer patients at doses above the MTD and is more likely to identify the true MTD than the 3+3 design (Ji & Yang, 2013).
One of the drawbacks of the mTPI design and model-based designs, in general, is that while they can accelerate dose escalation by treating fewer patients at sub-therapeutic dose levels, the inclusion of one patient per dose level may also deprive the study team of data on interpatient pharmacokinetic variability (Le Tourneau et. al., 2009). However, this limitation can easily be addressed by expanding the cohort size if additional PK data are needed.
Looking Beyond the 3+3
The limitations of the 3+3 study design and potential for alternative designs have been discussed for decades, with little to no increase in the number of phase 1 studies utilizing alternate designs. In 1997, a simulation comparing the 3+3 design with 3 accelerated titration designs was conducted. The results showed that the alternate designs were favorable for a number of reasons: they reduced the duration of trials, reduced the number of patients exposed to subtherapeutic doses, and provided an estimate of the population distribution of the MTD where the 3+3 design did not (Simon et. al., 1997). Nearly 25 years later, these results have had seemingly little to no impact on clinical trial designs as the 3+3 design continues to be commonly utilized.
Further substantiating the 1997 simulation, a recent comparison of 172 rule-based versus model-based oncology trials were conducted. The results showed that rule-based designs took a median of 10 months longer than model-based designs to complete, fewer patients were treated at sub-optimal dose levels in model-based versus rule-based studies, and that despite the savings in time and minimization of suboptimal treatment, safety was preserved in the model-based design (Brummelen et. al., 2016).
While alternative designs have remained more the exception than the rule, FDA has begun encouraging more innovative and adaptive designs in early phase studies; recent guidance calls out a need to consider adaptive trial designs in exploratory and dose-finding studies as a way to ensure optimal dose selection while affording the opportunity to learn more about exposure, pharmacodynamics, and variability in patient response (FDA, 2019). As new drugs are brought to the clinic, it is important to understand that there is no one best escalation scheme that can be applied across all scenarios. While the 3+3 design still may be appropriate in some situations, clinical researchers should take into consideration the mechanism of action of their drug as well as the expected toxicity profile when considering study design and resist opting for the 3+3 design in every instance for its simplicity. Implementing alternative study designs as a means to efficiently achieve a safe and optimal recommended phase 2 dose is a crucial consideration for clinical researchers in 2020 and beyond.
Brummelen, E. M. J. V., Huitema, A. D. R., Werkhoven, E. V., Beijnen, J. H., & Schellens, J. H. M. (2016). The performance of model-based versus rule-based phase I clinical trials in oncology. Journal of Pharmacokinetics and Pharmacodynamics, 43(3), 235–242. doi: 10.1007/s10928-016-9466-0
FDA. Guidance for Industry: Adaptive Design Clinical Trials for Drugs and Biologics. 2019. https://www.fda.gov/downloads/drugs/guidances/ucm201790.pdf. Accessed 18 Feb 2020.
Hansen, A. R., Graham, D. M., Pond, G. R., & Siu, L. L. (2014). Phase 1 trial design: is 3+ 3 the best?. Cancer Control, 21(3), 200-208.
Ji, Y., & Wang, S. J. (2013). Modified toxicity probability interval design: a safer and more reliable method than the 3+ 3 design for practical phase I trials. Journal of Clinical Oncology, 31(14), 1785.
Le Tourneau, C., Lee, J. J., & Siu, L. L. (2009). Dose escalation methods in phase I cancer clinical trials. JNCI: Journal of the National Cancer Institute, 101(10), 708-720.
Simon, R., Rubinstein, L., Arbuck, S. G., Christian, M. C., Freidlin, B., & Collins, J. (1997). Accelerated titration designs for phase I clinical trials in oncology. Journal of the National Cancer Institute, 89(15), 1138-1147.
Wheeler, G. M., Mander, A. P., Bedding, A., Brock, K., Cornelius, V., Grieve, A. P., … & Bond, S. J. (2019). How to design a dose-finding study using the continual reassessment method. BMC medical research methodology, 19(1), 1-15.
Yan, F., Mandrekar, S. J., & Yuan, Y. (2017). Keyboard: a novel Bayesian toxicity probability interval design for phase I clinical trials. Clinical Cancer Research, 23(15), 3994-4003.