TL10: Precision Medicine: Statistical issues, trial design and regulatory aspects
*Amir A Handzel, Astrazeneca 

Keywords: Predictive biomarkers, Early clinical trials, Umbrella trials, Trial deisgn, Precision Medicine

The White House has just announced a national Precision Medicine Initiative to improve the ways healthcare is conducted and diseases treated. The initiative includes allocation of funding to the NIH and the FDA for creating special databases and enhanced methods to promote the goal. These would complement tremendous activity by the industry which has been accelerating in recent years. Precision Medicine (PM) initially was applied in Oncology due to the nature of the diseases and thanks to the maturity of the underlying science, but it is now expanding to other therapeutic areas. As PM evolves and its boundaries expand with increasing complexity, new statistical methods need to be developed and challenges addressed. Participants in the proposed session will have the opportunity to discuss and learn about the following topics that we, as a community, are currently facing:

• Design of early PM clinical trials. With the successful registration of an increasing number of drugs within the PM framework, it appears that at least several key issues have been addressed for late stage registrational trials, including sub-group analyses, multiple hypothesis testing and the need for prospective design of trials. Less attention has been paid to the design of early clinical trials where typically less is known about the associated biomarkers and greater uncertainty exists regarding potential efficacy. What statistical methods and designs are optimal for early to mid-stage clinical trials? How should they be considered within the complete drug development process?

• Combinatorial biomarkers. One of the most recent trends to emerge in PM, mostly in Oncology, is the characterization of patients’ status using a combination of biomarkers, not a single biomarker as has become accepted for many drugs. In fact, panels of genetic markers have recently been developed (e.g. by IIlumina and by Foundation Medicine) that are now being implemented in clinical trials. Potential use of these multiplex biomarker panels goes beyond the biomarker triage algorithms that exist in several diseases, such as breast cancer. They further slice the patient population to much smaller and more precise groups thereby presenting difficulties in the development and approval of drugs using them. Statistical power for testing each combination of biomarkers quickly diminishes with the number of biomarker components and the complexity of clinical trial increases. What are suitable clinical trial designs for combinatorial biomarkers? How can statistical power be preserved in this setting? Are there particular issues that the FDA should consider in this context?

• Combination therapies. In several therapeutic areas, most notably Oncology and Infection, effective treatment requires the use of a combination of drugs. Oftentimes, each drug is associated with a specific biomarker and its inclusion in the “cocktail” dependent on biomarker status. In developing combination therapies, especially when two or more of the agents are novel, there is difficulty in attributing relative benefit to each constituent in the “cocktail” or assigning “guilt” for causing adverse events. Other challenges are closely related to the combinatorial problem of multiplex biomarkers. How can combination therapies be tested for efficacy in this combinatorial setting? What are suitable designs for such trials?

• Multidrug (“Umbrella”) and multi-indication (“Basket”) trials. With the increase in resolution of patient sub-populations using biomarkers, the size of each well defined sub-population for whom a targeted therapy is expected to work appears small as percentage of the total population in a given primary disease. This presents both statistical and operational challenges for conducting effective clinical trials. In part as a response, a novel approach has been introduced of combining multiple drugs – often from different companies – in a single primary indication (Umbrella trials) or testing a single drug in multiple disease types sharing similar molecular (genetic) profile. Pioneered with the BATTLE and iSPY trials in cancer these diverse designs provide a new way of organizing clinical trials in much more efficient and effective way. On the statistical side, what are the main questions that we face in planning and conducting such trials? Are there statistical conclusions that can be drawn from the trials that have already been conducted in recent years in order to improve this novel framework?

• PM at the FDA. In addition to all the above topics that the industry has to address in order to develop new PM drugs, the FDA may need additional tools for broader visibility of the field as a whole and for assessment of risk and efficacy. The national PM Initiative specifically mentions the assembly of dedicated databases to support the regulatory work of the FDA. What statistical methods and tools would be most useful for the FDA in fulfilling its task?