<img alt="" src="https://secure.lote1otto.com/219869.png" style="display:none;">
Skip to content

Bayesian Dose-Finding Designs – An Overview


imageedit_1_8026331443Cytel recently conducted a webinar on Bayesian Dose-finding Designs for Modern Drug Development, presented by Dr. Yuan Ji.

Dr. Ji is a Professor of Biostatistics at The University of Chicago and a well-known name in the industry. In his presentation, he introduces representative Bayesian designs for dose-finding trials. The webinar offers insights on topics including classical DLT-based dose-finding designs, designs with delayed toxicity using time-to-event endpoints, and designs for combination dose-finding trial. Watch the on demand webinar to see the illustration of Bayesian modeling and inference for dose-finding designs that utilize the concept of probability intervals and related methods for clinical development and decision making.

There are many dose-finding designs that have been developed over the past 30 years and several more are anticipated. Around 2007, a new philosophy was proposed, where the idea was to use the probability of toxicity intervals for the purpose of statistical inference. Following this, the mTPI, BOIN, mTPI-2 Keybrd and i3+3 designs were introduced, and these fell under the interval design framework. Along the way, we witnessed the development of the Bayesian Logistic Regression Model (BLRM) which is a full model-based design; interestingly, the interval idea is also embedded in BLRM where the posterior probabilities of various dosing toxicity intervals were used in the decision making for testing doses.

From 1989 to 2017, the numerous rule-based and statistical models that were the main tools for such inference, came with slight differences in terms of making assumptions or constructing dose-response curves. Last year, a new rule-based design was proposed, namely the i3+3 design which abandoned modeling altogether.

This webinar attempts to resolve the dilemma of which design the sponsors should use, when choosing from the various design options available today. Below, is a brief review of some of these designs as described by Dr. Ji in the webinar :

The 3+3 Design (1989)

The 3+3 design is purely a rule-based design and has no statistical modeling required for its use. Its operationalization can be summarized by specific rules which tell you exactly what to do, and these rules are easy to understand and transparent. It has been used for over 30 years and is still a very popular design in the clinical community. However, it is considered to be a naïve and rigid design, with some limitations. For example, it doesn’t tell you what to do when there are more than six patients per dose. From a statistical point of view, six is not a large number and it cannot capture many variabilities, which do exist in the early phase of trials. There is also no clear definition of Maximum Tolerated Dose (MTD); in Phase 1 trials one of the main goals is to find the MTD and the 3+3 design does not provide that specific measurement very easily, or at least not with precision, although it takes clinicians towards a ballpark.

The CRM designs (1990-2007)

The CRM designs were the first model-based designs that were published. It aims to find the maximum MTD of a new therapy. Both CRM and BLRM designs define MTD in a very scientifically rigorous manner. However, the CRM designs too have some practical issues, for example, there is a need for a statistical expert for inference and decision making, as it can be too complex for the clinical team. There is also the challenge of convincing the non-statisticians to use it.

The Interval-Based Designs (2007-now)

The interval-based designs when compared to CRM, do not need dose response curves and over-dose control. They use simplified models to account for variabilities in the data. They are very flexible and powerful in terms of simulation performance. These designs use Simple Bayesian models and can produce decision tables which can help with real-world clinical trials.
The Interval-Based Designs helped in bridging simplicity and model-based inference for the first time. They effectively challenged the 3+3 design as the only clinically popular method. These designs are widely used in practical trials, and designs like CCD, BOIN and i3+3 further simplified the approaches.

To get an in-depth understanding of each of these designs, watch the on demand webinar.



contact iconSubscribe back to top