It Is Never Too Early to Think About Statistical Leadership

Richard C. Zink, Principal Research Fellow, JMP Statistical Discovery

Author’s Note: Meijing Wu kindly asked me to draft this article shortly before her passing; it is dedicated in her memory.

Statisticians and data scientists occupy a unique space on a multidisciplinary team. So much so that it may often feel like a salmon swimming upstream against the current produced by everyone else. Teams want to move fast and meet deadlines, but we know firsthand that every decision—even seemingly innocuous ones—can have major implications and consequences for all downstream activities.  

This is especially true for clinical trials—large, exceptionally expensive experiments involving human beings that often take several months or years to complete and include the following: developing the protocol; designing and testing the database; conducting the study; collecting the data; producing the final analysis; and authoring the final study report. The team does their utmost to produce the best possible product at each stage but, without fail, multiple changes often occur over the course of a study.

While changes can create headaches for individual groups, statisticians need to be aware of the entirety of the process, since changes can—and likely will—affect the final analysis. 

In the worst-case scenario, a modified protocol triggers modifications to an electronic data capture system and the underlying raw data, which triggers modifications to symbol digit modalities test programs, which triggers modifications to analysis data model programs, which triggers modifications to programs for tables, figures, and listings. That is a lot of places in need of revision and plenty of opportunity for things to go wrong.  

Every passing day is one fewer day prior to database lock to get things in their proper place, so there is little time for rethinking, revising, and revalidating the analysis. It is no wonder standard operating procedures and process improvement are so important in medical product development; it is extraordinarily complex, and there is so much at stake. 

I have often read a protocol and thought to myself, “There is no way a statistician has reviewed this”? Even sections such as inclusion and exclusion criteria, study endpoints, and study conduct can be written in a manner so vague or inconsistent that it is unsurprising when protocols are often in need of amendments. When asked to provide the statistics section to protocols, I start at Page 1, add comments throughout, and—more often than not—find major issues that require discussion by the team before I can even begin to write text describing an appropriate statistical methodology and the accompanying summary tables.  

When team members grow frustrated with my changes and suggestions, especially since they believe they are near the end of protocol development, my response is to include me earlier in the process. It does not take long for this to start happening, and it results in fewer hiccups down the road.  

The take-home message is this: Though the analysis comes at the end of the process, the team cannot leave the statistician out of protocol development until the last minute. Unfortunately, it is often up to the statistician to communicate this message. 

I am also a big believer in finalizing the statistical analysis plan, known as SAP, as early as possible. I often have a draft available while the database is being produced and strive to get it signed off on as early as possible. Why go to such extraordinary lengths? Because of the following: 

• It gets everyone thinking about the analysis early, especially while the protocol is fresh in their minds. 
• It ensures the data required to perform the final analysis is collected in an appropriate manner in the study database. 
• It makes the rest of the study team less likely to change the analysis since the SAP has already been signed off on. 

Sometimes, individuals dither finalizing the SAP. Many people make the claim that the SAP needs to be signed prior to database lock and will delay in finalizing the document. While this is true, it represents the worst-case scenario. If the methodology, analyses, and assumptions are changing up until a few weeks before database lock, the quality of the analysis suffers, the team producing the analysis suffers, and the timelines often suffer. 

Like the example above in which I can predict with near certainty whether a statistician was involved in writing the protocol, it is just as easy to predict those instances in which the SAP was finalized just prior to database lock. In these instances, finalizing the database will be chaos, producing initial results will be chaos, and the statisticians and statistical programmers will be miserable. 

The take-home message is this: SAPs need to be finalized sooner. If the argument against finalizing a SAP early is that the protocol is constantly changing through amendments, this can be addressed through improving the quality of the original protocol (Hint: Include a statistician from the start.). Otherwise, the statistician needs to adjust the team’s expectations, given the disruption of downstream processes.  

Hopefully these two examples highlight a crucial point: Statisticians need to be leaders because of the following: 

• Nonstatisticians may not recognize the importance of our unique skill set, even when it comes to the seemingly nonstatistical aspects of multidisciplinary work. 
• Nonstatisticians may not recognize the huge implications minor decisions may have on the final analysis. 
• Nonstatisticians may not know what data is needed to conduct an appropriate analysis. 

So again, statisticians need to be leaders. They need to communicate. They need to build trust through developing and nurturing relationships with key members of the team. Then, and only then, will statisticians be able to influence others and the direction of their collaborations. Unfortunately, much of statisticians’ education and training focuses on technical prowess, and less so on leadership, communication, and other interpersonal skills. 

The good news is that anyone at any level can be a leader. It can be as easy as moving from the corner of the room to the conference table and being actively engaged in the discussion. Or by raising a hand when someone asks for a volunteer. Or by looking for opportunities to simply make things better. Consider the following important skills statisticians generally already possess: 

• Organized  
• Methodical 
• Thoughtful 
• Precise 
• Forward-thinking, with plans for contingencies 
• Philomathic 
• Data-driven

These are critical skills for any leader to be effective.  

Leadership begins with small steps. For example, sharing knowledge and expertise through presentations at conferences, workshops, or webinars or drafting scientific articles, book chapters, or software. Consider opportunities to engage with nonstatistical members of the team by helping them understand statistical concepts (using everyday language) or using your unique blend of skills. Developing or chairing sessions, serving as a referee for a scientific journal, or volunteering for American Statistical Association section or chapter activities are excellent ways to build a network and develop an intuition for how to get things done.  

These small opportunities will inevitably lead to larger opportunities such as leading major ASA initiatives; serving as a chapter, section, or board officer; teaching short courses at major conferences; leading scientific working groups; chairing conferences; serving as an editor for a scientific journal; or leading biometrics departments. Statisticians who have no interest in pursuing major leadership activities will benefit from leadership skills in their day-to-day work as part of a multidisciplinary team. Statisticians represent a unique and distinct point of view, so they need to make themselves heard. 

One does not wake up a leader. And attending a single course or reading a single book will not make one a leader, either. Like any other set of skills, leadership takes time and practice to grow and develop. Following are some things to consider:

• Get involved with the ASA to develop leadership skills, broaden your network, and support the statistics discipline. 
• Exhibiting leadership before having the title makes it possible to get the title. And it’s okay to not have interest in the title; leadership skills are useful at any level. 
• Being able to effectively communicate ideas and convince others to consider implementing them makes good use of knowledge and technical expertise. 
• Identify a mentor or peer group to discuss professional challenges and identify potential solutions. 
• Become more comfortable with the uncomfortable by stretching boundaries. Take risks, and do not become stagnant because of the fear of making mistakes.

So, hop to it! It is never too early to think about statistical leadership. 

Editor’s Note: This article originally appeared in the Biopharmaceutical Section’s spring report.