Make sure to consider statistical methods during the design phase of trials

Jamie Collins, PhD
Jamie Collins, PhD

Statistical methods are a mystery to many rheumatologists, including many experienced researchers. This year’s Stats Bootcamp series can help dissolve the mystery and bring clarity to statistics, if not comfort and ease.

“As the field changes, it’s important for people to stay on top of the subject. There are always new methodologies coming out,” said Jamie Collins, PhD, Senior Biostatistician at the Orthopaedic and Arthritis Center for Outcomes Research and Assistant Professor of Orthopedic Surgery at Harvard Medical School.

Dr. Collins will co-moderate Stats Bootcamp 2: Statistical Modeling of Categorical Data from 9:00 – 10:00 am Monday in Room B204, Building B of the Georgia World Congress Center. She will share moderating duties with Dana Guglielmo, MPH, Epidemiology Fellow at the Centers for Disease Control and Prevention.

The challenge for statisticians is that too many wait too long to consider their statistical methods. The time to think about statistics is during the study concept and design phase, not after the trial is up and running. And never after the trial has finished.

“The quote from Ronal Fisher, one of the founders of modern statistics, is that ‘going to a statistician after you’ve already completed the study is like asking for a postmortem,’” Dr. Collins said. “At that point, I can tell you what the study died of, but nothing can bring it back to life. You need to understand those clockwork details and incorporate the statistics into the study starting from the concept and design phases if you want to keep the whole mechanism on track.”

Statistics may seem inaccessible or even stodgy from the outside, Dr. Collins said, but statistical methods are critical to interpreting data and communicating results.

In recent years, controversies as basic as the appropriate use of p-values has shaken the worlds of statistics, medical journals, rheumatology research, and rheumatologists searching the literature for an appropriate treatment. Understanding the basic concepts of the field and how these concepts have evolved over time is central to both conducting and evaluating research.

And anyone writing a research grant application who doesn’t understand the basics of statistics is likely to fail.

It is not just statisticians who commonly ponder the fine distinctions between different models. Researchers, journal reviewers, and clinicians should also be paying attention. Bootcamp is less about the grand theories that underlie statistical methods and more about the practical details of which method is more appropriate for which kinds of variables, which kinds of data, which kinds of studies.”

“Categorical data is not Stats 101, more Stats 201,” Dr. Collins said. “You need to be able to identify categorical response variables and understand how they are different from continuous variables.”

The complication is that there is no one-size-fits-all modeling method, she noted. Every method has its own advantages and limitations. One of the keys to a successful study is to select the method that is most appropriate for that specific study and set of data.

“Hands-on, practical examples will help you understand some of the salient features of different models for categorical response variables and some of the differences between models that are commonly used,” Dr. Collins said, “Attendees should leave the session able to read and understand the literature in rheumatology with a more informed eye.”