Case study

Using ClinBAY Adaptive Design Decision Tool for a Phase 1 dose finding study in Solid Tumor oncology

Request for proposal

Decision on the design

Our client wanted to implement a Phase 1 dose finding study in Solid Tumor oncology. As in almost all such studies the objective was to find the MTD (Maximum Tolerable Dose), and the recommended dose regiment. The client had employees primarily experienced in implementing classical 3+3 designs, in such framework, but wanted to see if there would be added benefits by implementing more modern designs, like the continual reassessment method (CRM) design.

During a kick-off meeting a presentation on the advantages of CRM, which is a model based design, alongside a comparison to the classical 3+3, were explained. The main benefits of the CRM were the following:

  • MTD target rate: The classic 3+3 design although the most widely used conventional design for oncological phase I trials due to its ease of implementation, has no explicit objective in mind, other than to find a dose that gives an observed DLT rate of no more than 33%, while the target toxicity probability level of the 3 + 3 design is explicitly proven to be between 16% and 27%, on average. In addition, 3+3 design does not provide any confidence in what the actual DLT rate of any of the dose levels might be, and thus little confidence in the selected MTD, making it inferior for MTD determination and for treatment of patients at doses close to the MTD.
    On the other hand, in model-based designs like the CRM, the probabilities of DLT for each of the doses are assumed to be fully explained by a parametric model that is a function of the value of each dose. That offers the advantage to explicitly target a specific toxicity probability level of choice that satisfies the pre planned MTD definition(usually 20-30%) as well as providing a level of confidence around that target. This is critical as when the drug is tested for patients with good prognosis where a lower rate is acceptable, and for patients with poor prognosis higher rates are acceptable.
  • Efficiency: The model-based design generally is expected to require less subjects than the 3+3 due the fact that makes use of the full data for dose escalation decisions and not just data specific to each dose level. The addition of prior pre-clinical toxicity information from animal studies through the use of Bayesian formulation of the CRM, can also further contribute to this objective.
  • Overdosing: The model-based design can be fine-tuned into limiting the number of patients exposed to toxic doses during the study to a level acceptable by the client.
  • Quantification of dose-toxicity relationship: Although in theory this is possible from the 3+3 as well, the precision of the dose-toxicity relationship is much higher due the fact that a parametric model is used to descibe the relationship of dose and probability of DLT.
  • Time needed for the study: The model based designs due the fact that they make use of the full data for dose escalation decision and not just data specific to each dose level, require less patients. In addition, they can accommodate a more flexible recruitment plan of more than 3 patients per cohort. The benefit of this is that quite commonly in dropouts are experienced, this means in a 3+3 setting, if we experience a drop-out, we will need to wait until a 4th patient is enrolled and the patient completes 1st (or multiple) cycle(s) of treatment. Whereas, for a model-based design, the decision can be made even if we plan to enroll 4 patients at the time, and eventually all patients end-up being evaluable.

The main concerns raised by the client were the following:

  • Difficulty in conceptualizing the design by the clinical team during the design Phase.
  • Potential delays in waiting for the statistical analysis after a cohort is completed.

In order to address these concerns, the Adaptive Design Decision Tool was presented to the client. The goal was to use the tool to firstly help the clinical personnel to visualize the design during the design phase and enable their accurate contribution to the setup of the study. Secondly, it was suggested that tool would be kept live and updated (if needed), to enable the clinical personnel to anticipate decisions and to avoid waiting for the models to be run. 

Design stage

After the kick-off meeting and the agreement on using the modified CRM design, the Sponsor provided all necessary information to start the design. In summary the Sponsor provided the following information dose levels, target toxicity risk, fold increase (dose skipping was allowed), and the prior information from competitor and animal studies.

For the first simulation the standard rules used by ClinBAY were proposed:

  • Stop if next Dose selected by the model has already 9 patients enrolled.
  • Stop if there is >80% probability that first dose is toxic.
  • Stop if there is >80% probability that maximum dose is safe.
  • Stop if the MTD was precisely estimated.

In addition, hard safety escalation rules, prohibit any further use of a specific dose and above that, for the rest of the dose escalation, if the following criteria are met for that dose:

  • if at least 2 DLTs out of 3 patients,
  • if at least 3 DLTs out of 6 patients,
  • and if at least 4 DLTs out of 9 patients.

Using these rules for performing the dose escalation, the simulations were executed, and the results uploaded to the Adaptive Design Decision Tool for the Sponsors review (example screenshot below).

Adaptive design decision tool

The sponsor exhibited concern that based on the current set of rules, model could propose an escalation if a DLT was observed. (See screenshot above, when a cohort has received 1.5mg/kg and 1 patient had an event out of 3, model proposes escalation to 2.2mg/kg). It was therefore decided to add the following rule:

  • No escalation is allowed for the next cohort, if 1 patient or more in this cohort had a DLT.

The simulation was subsequently re-run and the tool updated. No further feedback was received on the simulation. Subsequently the required simulation package to accompany the protocol was generated.

It was also decided that to reduce timeliness decided 4 patients at the time would be enrolled per cohort, in order to reach 3 evaluable, rather than recruit 3 and only recruit one more if one was non-evaluable. The simulation was not re-run due to this decision, as the intention was to reach at least 3 evaluable patients, and it was expected than in most cohorts 4 patients would have led to 3 evaluable.

Implementation stage

During the course of the study the sponsor referred to the Adaptive Design Decision Tool, for planning and decisions. Notable exceptions were the following:

  • In at least one cohort cases we ended up with 4 evaluable patients. To avoid delays, as this was anticipated well before the end of the treatment cycle, the simulations were re-run with 4 patients in this cohort and the decision tool updated.
  • In cases where de-escalation was proposed, the intermediate doses between dose recommended by the model and previous dose were also assessed. The reason is that the model makes the recommendation based on the most likely dose to be the MTD, however it may be that a higher dose only be “slightly” less likely that the dose selected from the model. In such cases, especially if the “model proposed” dose was unlikely to be efficacious, the higher dose could be discussed with the clinical personel as options. Should a different dose be selected to the recommended, the simulations would have been re-run and the decision tool updated.
  • Clinicians have the final say about the final dose selected, the model is only making a recommendation. Should the clinicians have selected an alternative dose to the one proposed from the model, this would be acceptable with the only impact the requirement to re-run the simulations and update the decision tool.


Above case study, draws components from multiple studies/projects and not only one. Due to the extensive experience of ClinBAY working with these studies, drawing from a single study would not have presented the total flexibility of these approaches.

Man getting organised over statistics

Discover ClinBAY Adaptive Design Decision Tool !

A visualization tool aimed at helping both to understand the recommendations during design phase and to enable quick decision making during the study execution phase.
Find out more
Request for proposal

We will be pleased to hear about your trial project and provide you with a tailored-made, cost effective biometrics solution.

Request for proposal

Interested in getting regular updates about ClinBAY, our products, achievements and solutions, just subscribe.

Thank you! You have been sucessfully subscribed to our newsletter!
Oops! Something went wrong while submitting the form.

Are you looking for a reliable biometrics CRO partner? Kindly contact us.