There’s a chance many physicians, and many others, dozed off during Statistics 101 or missed the course entirely. They are at disadvantage when it comes to reading research, conducting research, and writing research grant applications.
“Statistics is the fundamental language of science across disciplines,” said Todd A. Schwartz, DrPH. “Statistics permeate the way researchers talk about their studies and what they found. Statistics are how you quantify the evidence and explain what the data say. Statistics are the common language we all use to communicate with other researchers and with clinicians.”
Dr. Schwartz will help clarify and demystify two of the most basic concepts in Stats Bootcamp 1: Sample Size & Power Calculations, on Sunday from 11:00 am – 12:00 pm in Room B403, Builiding B of the Georgia World Congress Center. He is Associate Professor of Biostatistics at the University of North Carolina Gillings School of Global Public Health and works closely with rheumatology researchers in the UNC Thurston Arthritis Research Center.
The Sunday session is the first of three Stats Bootcamps. The other two sessions explore statistical modeling of categorical data (Monday, see session preview below) and the current uses of multi-state models for time-to-event data analysis (Tuesday, see Monday’s ACR Daily News).
“These are all stand-alone sessions and do not build off of each other,” said moderator Rebecca Cleveland, PhD, Assistant Professor of Rheumatology, Allergy and Immunology at the UNC School of Medicine. “We do three sessions knowing that attendees have different levels and statistics training. We want to have sessions that are useful for a wide range of members attending the Annual Meeting.”
Researchers and clinicians come from a wide variety of backgrounds when it comes to the whys and hows of statistical methods.
Take sample size and power calculation, which form the foundation that supports every successful study.
“There are certain aspects that must be considered before beginning your study, and even before designing your study,” said Alyssa B. Dufour, PhD, Instructor of Medicine at Harvard Medical School and Assistant Director of the Biostatistics and Data Sciences Core at the Marcus Institute for Aging Research. “Obtaining meaningful results at the end of your study strongly relies on having thoughtfully established your sample size by doing a power calculation long before the study begins.”
Sample size is simply the number of patients or other participants in your study. The greater the sample size, the more likely the study will be able to uncover meaningful associations.
Statistical power is the probability of detecting an effect of a given size in your study population, assuming that effect exists. Power calculations depend on the type of outcome being analyzed, its variability, the targeted effect size, the level of significance you want to achieve, and sample size, along with study design, statistical hypotheses, and other relevant factors.
Journals don’t often reject studies simply because they are underpowered, Dr. Dufour said, but the lower the power, the less authoritative the findings. If there truly is a statistically significant association in your study, you may not be able to detect it if you do not have sufficient power.
It’s also possible to overpower a study. Larger sample sizes and higher levels of power can be beneficial from a statistical perspective, but this can also lead to achieving statistical significance for meaninglessly small effects. Properly powering a study allows for a balance between statistical significance and clinical significance. And setting enrollment targets too high can slow progress, even stall a trial.
“If you overpower the study, recruit too many people or conduct too many assessments, it could cost your extra time and money,” Dr. Dufour said. “Which is why we always calculate ahead of time the ideal number of participants to maximize your chance of finding significant results — if there truly is a significant association to be found.
“Additionally, if you are conducting a high-risk study of a new drug or other potentially harmful exposure, minimizing the number of people exposed to that potentially negative factor is critical.”
Sample size and power calculations don’t have to be a mystery. There are a number of online calculators that researchers or study readers can use on their own. And armed with a good idea of the concepts behind sample size and power calculations, researchers can work with biostatisticians more effectively to fine-tune those initial calculations.
“We want to demystify what goes into these kinds of calculations and how the different pieces fit together,” Dr. Schwartz said. “This is a crucial design issue in every research study. If you have conducted your study and your findings were not as strong and as meaningful as you had hoped, that’s not the time to be wondering about power calculations. We are here to help researchers think about those critical issues when they are putting a trial together and can make adjustments that will help strengthen their eventual findings.”