When approaching a Phase 3 clinical trial, the need to ‘de-risk’ the massive investment often leads sponsors on a quest for the perfect risk mitigating adaptation. While a strategically planned clinical trial design can be an important step in giving a new medicine its best possible chance of success, there are a number of other ways that a trial sponsor can minimize study risk.
One strategy, so obvious that it is often overlooked, involves determining novel ways to optimize Phase 2 data so as to minimize risks in Phase 3. High quality early phase data should already be available to sponsors before a Phase 3 trial begins. Strategies that fail to make the most of the information available from this early phase data perform a great disservice to the later stages of a trial.
Consider the following scenarios, all too familiar to those struggling to complete Phase 3 trials:
1: You tested four candidate doses during a Phase 2 trial. While you think the second highest dose might be the right dose going forward, you think a higher dose might prove even more effective. However, the highest dose you tested is clearly toxic. Should you go ahead with the second highest dose? Is there a slightly higher dose that it would be better to test given your resource constraints? Which dose?
2: During an interim look of a Phase 3 trial you discover that one of the doses you are testing seems to have a higher rate of adverse side-effects than your data initially suggested. However, the dose right below it does not seem very favorable. Ideally, you want a dose in between but you did not test any such dose. Which dose should you use going forward?
3: Your interim Phase 3 results appear unfavorable and you would like to alter your dose-selection. Would increasing dosage be more or less risky to the trial outcome?
4: During an interim look it is determined that your trial ought to stop for futility. What next?
All of these scenarios demonstrate the need for having an accurate dose-ranging model. A good statistical model can offer insights into your new medicine that could easily go unnoticed. Since you are likely to appeal to these models for dose-selection, as well as for critical points in your Phase 3 trial, it is important to know that your model is trustworthy. An inaccurate model, providing faulty information during these risky situations, can be costly and expensive.
One method for modeling that has been gaining popularity in recent years is the multiple comparisons procedures modeling approach (MCPMod Approach) developed by Pinheiro, Bornkamp and Bretz .
According to the European Medicines Agency Qualification, “The MCPMod method is efficient in the sense that it uses the available data better than the commonly applied pairwise comparison.” 
MCPMod typically involves a combined proof-of-concept and dose-ranging trial, but results in a statistical model of the dose-range. Instead of conducting independent proof-of-concept trials and then combining data from various trials to determine a dose-range, the MCPMod method allows one trial to determine both of these objectives.
The success of MCPMod depends on the method’s flexibility in choosing from a variety of candidate statistical models to fit the dose-range, and from the approach’s ability to guard against model misspecification using multiple comparison procedures.
If you are interested in learning how you can access MCPMod methods through your validated production environment, (i.e. in compliance with Standard Operating Procedures and Good Clinical Practice,) click here.
Related Items of Interest
 Pinheiro, José, Björn Bornkamp, and Frank Bretz. "Design and analysis of dose-finding studies combining multiple comparisons and modeling procedures." Journal of biopharmaceutical statistics 16.5 (2006): 639-656.
 Bornkamp, Björn, et al. "Response-adaptive dose-finding under model uncertainty." The Annals of Applied Statistics (2011): 1611-1631.
 European Medicines Agency Qualification Opinion of as an efficient statistical methodology for model-based design and analysis of Phase II dose finding studies under model uncertainty