by Joshua Kriger, MS ’14, in the Biostatistics Theory and Methods Program
The days are over when a statistician sits in a back room tabulating data. The statistician needs to be a questioner with a passion for the truth and an expert communicator dedicated to efficiency and clarity.
During my time at Columbia, with the support of my colleagues and mentors, I was fortunate to participate in three separate studies in three different roles. I have been the chief statistician on a phase 2 dose finding trial, an associate statistician for the Statistical Analysis Center for Clinical Trials (SAC), and as a project manager on a service contract for a major pharmaceutical company. I loved the experience working in these clinical trials, and I want to share some of my insights from the experience.
1. Keep it simple.
Design should be as simple as possible to answer clinical questions. Beyond this, articulate assumptions for the trial and what will happen if those assumptions are not met.
2. Take your time when it is not intuitive.
Type 1 and type 2 errors are often misunderstood by people who do not have formal training. Take time to go over the basics. It all starts with the fundamentals of blocking and tackling.
3. Agree to disagree.
People often agree with each other not realizing they are talking about two different things. Just because a meeting was full of agreement does not necessarily mean the people involved were agreeing on the same thing. Don’t be afraid to disagree.
4. Speak plainly.
Biostatisticians, and statisticians in general, must be able to put things in plain language, not just rattle off obscure acronyms.
5. Ask questions.
I love asking this last question of the day: “What have we decided and discussed today?” The investigator or clinician working with you should be confident in why the design or analysis choices are made while also being confident in articulating them to others.
6. Sensitivity analyses are key.
What happens if an assumption or target is not met? Example: What will happen if our projected recruitment rate is lower than expected? What if the effect size is not what we previously saw in research? It is important to see what can break and the implications for the research.
7. Get to the heart of the question.
A Biostatistician should be able to articulate, identify and get to the heart of the clinical question. No amount of data (no matter how “Big”) should be substituted for taking the time to ask the right question.
8. Summarize key information.
How was recruitment? How long did it take to report events outside the norm? For each group involved with research there is a different set of priorities. Filter out information not needed to account for patient safety or to answer the desired research question.
9. Respect data management.
Have a quality data manager. Data management is often the step in clinical research that limits the outcomes. Data management is the foundation of well-done studies. Some up front design work will save you multiples on the time invested when you are ready for project analysis.
10. Bring the statistician in during your early planning phase.
As soon as you start to conceptualize the question you want to investigate, call a statistician. The last thing anyone wants (funders, patients etc.) is to find out mid-way through, or worse, at the end of a project, that the data collected or design does not answer the desired central question.
Statisticians today have an ever-growing network of colleagues to learn from. These include other statisticians, clinicians (both those who understand statistics and those who don’t), budget officers, review boards, company representatives, patient representative organizations, and dozens of others. We need statisticians to not just be brilliant theoreticians, but also masterful in communicators. Statisticians who incorporate and understand these two concepts save you precious time, money, and countless headaches.
Special thanks to Dr. Seamus Thompson, Dr. Bruce Levin, Columbia University Mailman School of Public Health, and the Department of Biostatistics.