Phase I and II trials may have different overall goals (i.e., demonstrating safety vs. efficacy), but the two both struggle with a major challenge in oncology study design: finding the right dose. Luckily, decades of data and innovations have given researchers the tools necessary to plan a successful dose-finding trial. Read on for a look at six of the most common dose-finding trial designs in oncology today.

Rules or Models?

When it comes to Phase I dose-finding trial designs, there are two major types that encompass more specific strategies: rule-based designs and model-based designs. Both of these types work from the fundamental assumption of monotonically increasing dose-toxicity and dose-activity relationships. As their name suggests, rule-based designs use dose escalation based on predetermined rules and algorithms. Rule-based designs include traditional 3+3 design, accelerated titration design and rolling six. By contrast, model-based designs incorporate dose escalation that is guided by a mathematical model. Commonly used model-based designs include the continual reassessment method (CRM), modified CRM and the Bayesian logistic regression model.

1. Traditional 3+3 Design
The traditional 3+3 dose-escalation design starts off by treating three patients at a conservative low dose. If no dose-limiting toxicities (DLTs) are observed, the trial proceeds to the next higher dose with three new patients. If one DLT is observed among the previous three patients, three more patients are treated at this dose. If no DLTs are seen among these three patients, the study moves on to the next dose. If two or more DLTs are found among the six patients, then the maximum tolerated dose (MTD) is defined as the previous dose. Likewise, if two or three out of three patients experience DLTs at any level, the previous dose becomes the MTD. This continues until an MTD is defined.

The advantages of this traditional design are that it is simple and easy to implement, relatively safe and clinical friendly. However, there are many drawbacks to this tried-and-true simple design method. First, there is a general lack of any statistical foundation, which results in lower precision. The 3+3 is also memoryless, as it is based on only the last six patients enrolled. This design also requires many escalation steps with doses that may be too low to be effective, leading to suboptimal treatment for a large number of patients. Finally, this design might not be appropriate for molecularly targeted agents (MTAs) with low general toxicity that may not follow monotonically increasing dose toxicity and dose activity (i.e., higher doses aren’t necessarily more effective and/or toxic).

2. Continuous Reassessment Method
CRM is one of the most popular examples of adaptive design in oncology today. Adaptive design takes advantage of accumulating information while data is still being collected, allowing investigators to make predetermined changes to the trial as necessary. This design strategy helps cut down on costs and has the potential to improve study patient outcomes.

CRM is an adaptive dose-response flexible plan design. The target DLT rate is fixed at the beginning of the study, which then treats patients with predetermined dose levels. One popular example of CRM is the Bayesian logistic regression model with overdose control.

A CRM study is made up of three basic components:

A mathematical model relating the probability of DLT to the dose levels
Dose-escalation method
Stopping rules

The major advantages of CRM are a more precise MTD estimate and that more patients are treated at a therapeutic dose. The only specific disadvantage of CRM is that it uses a higher level of statistical complexity requiring a biostatistician.

3. Other Statistical Designs
There are many different statistical designs for dose-finding studies that have been proposed or are currently in use. Unfortunately, their use is far from universal. For example, one review found that only 1.6 percent of dose-escalation cancer trials follow a Bayesian adaptive design. Another paper looking at first-in-human Phase I trials reported that 90.5 percent utilized rules-based design, while just 7.1 percent followed a model-based design.

Other specific types of statistical design include:

Slope-sign design: Dose finding is directed based on the sine of the estimated slope of the dose-efficacy curve.
Logistic design: Uses a dose-efficacy curve that follows a Bayesian logistic model.
Isotonic design: A non-parametric approach using isotonic regression.
Local logistic design: Wherein the model for the dose-efficacy curve uses two adjacent doses.

4. Designs Using Optimal Biological Dose
Unfortunately, studies that rely on MTD are less effective for dose finding when it comes to targeted therapies, which have moved much of the focus in clinical away from standard cytotoxic chemotherapy. A review of Phase I cancer trials found that MTD was successfully determined in 99 percent of the studies surveyed involving, whereas just 64 percent of targeted agent studies could determine the drug’s MTD. The bottom line? As new modalities in cancer treatment, MTAs require new designs and endpoints.

As an alternative to MTD, the goal of many MTA trials is to instead find the optimal biological dose (OBD), which is the lowest dose that has the highest efficacy rate while still safeguarding patients. Recommended OBD isn’t based on toxicity but on a more rational, scientifically derived set of endpoints. Compared with MTD-based designs, OBD-based designs address the problem of the monotonic assumptions not holding for low-toxicity MTAs.

5. Multi-Agent Dose-Finding Methods
Phase I trials of combination therapies are becoming more common, due in large part to the fact that no single pathway has yet to be identified that uniquely drives the malignant process. With multiple agents being tested, the complexity and need for precision of dose-finding studies both grow tremendously. In particular, special designs are needed for studies involving:

A novel MTA combined with standard chemotherapy
A combination of two MTAs

6. Gehan’s Design
Like the 3+3, Gehan’s Design has been used in clinical oncology trials for decades. It is a Phase II, two-stage design with no control that rejects ineffective treatments early. In the context of an early dose-finding study, Gehan’s strategy can be used to determine how many patients to enroll. The two stages are:

A small number of patients is enrolled. If no successes are observed, the trial ends.
The trial goes forward with a larger sample size, able to estimate the true proportion of success with a given precision.

If the trial ends in the first stage, researchers may choose to explore other indications and dosing strategies.