Many biomechanics studies involve experiments on human subjects. Prior to beginning the study, a critical question arises. How many human participants should be recruited? This is also an important part of convincing the NIH to fund research proposals, such as for an F31 NRSA award. The number of human participants that will be recruited for the study needs to be determined and justified.
Humans carry variability in every biomechanical metric. But what does this mean for our science? It’s no secret that science in general has a reproducibility problem. For a running biomechanics study, several parameters can significantly vary across data collection sessions, suggesting low reproducibility (Larivière et al., 2019). A typical method of combating this is by recruiting more subjects into these kinds of studies. In the age of COVID-19, this has become a challenge. The prospect of leaving the safety of home to participate in a research study, often in a hospital, has become a lot less enticing. For these reasons, it is even more critical in today’s times for researchers to know the minimum number of subjects required to rigidly test the hypothesis.
So how can we ensure that our studies are powered enough to reliably test our hypotheses for our biomechanical studies? For running studies, researchers recently performed a statistical power analysis (Oliveira and Pirscoveanu, 2021) thereby setting a consensus in determining how many subjects are needed to power a biomechanical study. They specifically set out to answer the following questions:
For a study involving multiple biomechanical metrics, how many runners do you need and how many steps should each take?
How does changing the number of runners and the number of steps alter data variability and statistical power?
For their experiment, they recruited a large cohort of runners (N = 44) and had them run at a self-selected speed in 3 sets of 3-minute trials. To introduce a factor that would create expected changes in biomechanical metrics, they asked them to repeat this while trying to run silently.
For data stability, they used a Sequential Estimating Technique. This takes a moving point mean increasing from 1 through the number of trials to determine the number of trials needed for the moving point mean to stay in the bandwidth defining stability for the remaining number of trials. They used a data permutation method to simulate 100,000 runners performing 5 steps each. They then performed statistical power analysis to assess effect sizes and coefficients of variation.
Much of the article features colormaps representing the number of runners and the number of steps for coefficients of variation, interquartile range, and effect sizes for the different variables. They are particularly helpful if you’re interested in specific variables. Figure 12 (included below) provides a nice summary of the results. Most of the running variables required less than 30 steps to reach stability.
Interestingly, the article only included 8 females out of a total of 44 participants (less than 20%). Regarding this, Andersen, corresponding author, said "The environment from which we selected the study sample is dominated by male runners. We selected college students for our study sample, and the majority of students that reported being engaged in running training were male runners… There were other male runners involved in the recruitment for a larger study we did (data not included in this manuscript). So, male runners with male friends were more easily found than females."
Recruitment methods should be carefully considered to obtain a male/female balance. Andersen adds, "It would have been ideal to get the nearly same amount of males/females and I have been striving to reach this balance in future research.”
Achieving a balance is important for generalizing and drawing conclusions from these metrics. Females, for example, have different motor control strategies that can alter knee motion patterns during running (Malinzak et al., 2001). A female/male balance should be the goal to capture this intersex variability.
The authors concluded that, when assessing traditional running biomechanical variables (foot contact time, peak vertical force, foot contact angle, etc), at least 25 subjects to perform 25 steps should be recruited to be able to see a large effect size when comparing running in different conditions. This study could particularly be useful in optimizing subject recruitment into research studies, particularly in the time of COVID-19.
About the author:
Zoe Villamar is a PhD candidate in Biomedical Engineering at Northwestern University in Chicago, IL. Her research interests include joint mechanics and lower extremity musculoskeletal injuries. Her current work focuses on understanding how different biomechanical factors, such as loading and muscle activation, affect ankle joint mechanics particularly as it relates to individuals who frequently experience ankle injuries. Outside of the lab, she enjoys running, exploring Chicago, and hanging out with her foster and adopted pets.
References:
Larivière, O., Provot, T., Valdes-Tamayo, L., Bourgain, M., Chadefaux, D., 2019. Repeatability and reproducibility of stance phase during running. Computer Methods in Biomechanics and Biomedical Engineering 22, S359-S361.
Malinzak, R.A., Colby, S.M., Kirkendall, D.T., Yu, B., Garrett, W.E., 2001. A comparison of knee joint motion patterns between men and women in selected athletic tasks. Clin Biomech (Bristol, Avon) 16, 438-445.
Oliveira, A.S., Pirscoveanu, C.I., 2021. Implications of sample size and acquired number of steps to investigate running biomechanics. Scientific Reports 11, 3083.