In part 2, Lisa LaVange, Ph.D., Director of CDER’s Office of Biostatistics (OB) shares some of the activities that CDER statisticians take part in along with their routine work to help further the field of biostatistics and the challenges they are facing.
Activities outside of applications
In addition to our routine work of reviewing sponsor submissions, OB’s statisticians participate in activities that have the potential to positively influence future sponsor submissions. We advocate for industry-wide use of data standards such as those developed by the Clinical Data Interchange Standards Consortium (CDISC) to improve the quality of data and clinical studies as a whole, and also to facilitate efficiency in the review process and streamline the integration of data across multiple submissions when we investigate safety signals. We are very involved in the review of drug development tools (DDT) qualification submissions, particularly in characterizing the relationship between the DDT and a clinical outcome, and understanding the strengths and uncertainties of that relationship.
Our statistical research programs are aimed at developing better study designs and innovative methods to analyze data. We present our research at public meetings and in peer-reviewed publications, and we also collaborate with industry and academia to find novel solutions to challenges statisticians face.
Challenges CDER statisticians are facing
There has been a lot of attention in recent years about how to account for missing data, so much so that the FDA commissioned the National Academies of Science to provide feedback on how to deal with missing data in regulatory trials. Since their report published in 2010, OB has actively been working on ways to put their recommendations into practice. The urgency of this problem is reflected in the fact that the amount of missing data and its impact on the analysis is often a key consideration in approval decisions and frequently discussed during advisory committee meetings.
Many outside groups have performed research to develop or refine statistical techniques that accommodate for missing data, and most of these methods involve imputing values to replace those missing. But, as regulators, we have to think about why the data are missing - are patients not able to take the drug, or are they not staying on the drug because of side effects, or because they perceive the drug is not helping them? We need to know the reasons behind the missing data, how those reasons may affect what we want to estimate, and what analysis methods should be used for the estimation itself.
In the past year, our missing data working group has given a series of talks and workshops to provide more clarity to CDER’s expectations for handling missing data, and we are actively participating in an international working group to update the International Conference on Harmonisation’s Statistical Principles for Clinical Trials (ICH E9) guidelines to include missing data methods. We are also working to develop an internal policy for responding to missing data in applications to ensure a consistent approach across all of our statistical review divisions and consequently across all therapeutic areas. I'm optimistic that by providing greater detail on this issue, sponsors will know how to plan ahead to minimize missing data in the trial itself and make their primary analysis strategy sturdy enough to accommodate any unavoidable missing data.
Another challenging area for OB statisticians is the analysis of subgroups of patients, such as men and women or older versus younger patients. Subgroup analysis may sound simple on the surface – a sponsor shows that the drug works in the overall study population, and then checks to see how well the drug works in particular age, gender and racial ethnicity subgroups. Most would expect that the drug will look a little bit more efficacious in some subgroups compared with other groups just by chance. Well, at what point, or at what level of difference between subgroups, does a sponsor or statistician start thinking that a drug is really only working in one particular subgroup or not working in another? And when is the number of patients in a subgroup so small that the evidence is too unreliable to make a decision? That is just a small slice of the subgroup problem, and there are many more layers of complexity and variations on this theme. A cross-agency working group of statisticians is developing a white paper to help clarify FDA’s thinking on the issue of subgroups.
Exploring novel statistical solutions for a development challenges both old and new
The missing data and subgroup problems are just two of many areas where OB is actively seeking solutions. Our statisticians are also working on novel statistical solutions for evaluating treatments for diseases that affect small populations. For example, drug development in rare diseases and for pediatric patients needs statistical methods that work in small populations, and we are investigating methods based on Bayesian statistics that leverage information from other sources to help address this problem. Post-marketing trials to evaluate cardiovascular risk of diabetes drugs are big, lengthy and expensive. OB’s statisticians are exploring approaches using Bayesian and other methods that may be able to give us an answer sooner based on the accumulating data in the trial.
The past few years have seen significant advances in targeted drug development for cancer, and we have been strong advocates for the use of adaptive designs and master protocols to develop these drugs safely and efficiently. For the past two years, OB’s anti-bacterial team has explored a number of similar innovative statistical ideas to facilitate development in this critical area of unmet medical need, and our anti-viral team is now part of a larger group of statisticians working on trial designs for therapies to counteract the growing Ebola crisis. Our work in all of these areas illustrates our ability to think outside the box as well as our desire to help get good drugs to market and keep bad drugs off.
Looking to the future
Clarity, consistency, and transparency are three critical attributes OB statisticians strive to achieve as we work to support FDA’s mission to improve public health. We have to provide sponsors with a clear understanding of our expectations or they may waste resources on a poorly designed study that misses the mark. We have to be consistent in our recommendations across therapeutic areas, when it is reasonable to do so, and across sponsors in the same therapeutic area. We have to be transparent in our evaluation of a sponsor’s analyses, particularly when our analyses give different results that may impact the drug’s review. We have to show our work -- our processes, analyses and reviews -- so that sponsors understand what we're doing and what we're looking for, and perhaps statisticians will have to throw less cold water in the future as a result.
In Part 1, Dr. LaVange shared her perspective on a statistician’s role in CDER and on some of the work her more than 175 staff members take on every day.
Dr. LaVange joined FDA in September of 2011 as director, Office of Biostatistics in the Office of Translational Sciences. She received a B.A. in Mathematics from the University of North Carolina at Chapel Hill (UNC), a M.A. in Mathematics from the University of Massachusetts Amherst, and a Ph.D. in Biostatistics from UNC. Dr. LaVange’s early career was spent doing research at the non-profit RTI International, followed by ten years in the pharmaceutical industry. Immediately before joining FDA, she was a professor and Director of the Collaborative Studies Coordinating Center in the Department of Biostatistics at UNC.
- Clinical Data Interchange Standards Consortium
- ICH E9 Statistical Principles for Clinical Trials
- Article by Dr. LaVange in TIRS