Using Machine Learning to Guide Movement-Related Hypotheses and Remembering Louie Simmons

📝 Weekly paper summary

Characterising initial sprint acceleration strategies using a whole-body kinematics approach (Wild et al., 2022)

Category

Cross-Sectional, Exploratory Analysis

Context

Task, environmental, and personal constraints shape the emergence of movement behaviours in biologically degenerate systems. Capturing the high-level organizational structure of a movement system while sprinting can provide richer information for researchers and practitioners to assess athletes and guide future training. This goal is pending the assumption that specific acceleration profiles tend to exist between performers, and thus individuals may benefit from a shift towards these general patterns of motion. Unsupervised machine learning techniques (such as clustering algorithms) on "full-body" motion data provide a potential exploratory analysis technique to identify clusters to assess their associations with sprinting performance. Therefore, this study's primary purpose was to investigate whether it is possible to classify subgroups of professional rugby players based on full-body kinematic data and determine whether they're associated with sprint performance or joint kinematics. The study's secondary purpose was to examine how "stable" these kinematic strategies were at the intra-individual level.

Correctness

For the strengths:

• I liked that the authors had a subsample return for multiple testing to assess the reliability of their movement data. I was initially concerned with the small number of strides they included in their analysis, so the additional data demonstrating that this movement data was reliable between testing sessions went a long way for me to feel more comfortable about their results.
• It's not common to have datasets with professional athletes, so it's great that they could recruit professional rugby union players for their study.

Now, for the delimitations:

• What the authors referred to as "full-body" kinematics was step length, step rate, contact time, flight time, touchdown distance, toe-off distance, contact length, and flight length (as well as step length/step rate and contact time/flight time ratios). I thought the authors may have clustered individuals based on their angular kinematics but instead clustered individuals based on these discrete measures. While this isn't wrong, I think it makes the next steps of guiding training a bit more challenging when all we have is outcome data.
• I do somewhat worry about the analysis technique used on this dataset. In some research exploring clustering algorithms for market segmentation analyses, the researchers recommended sample sizes of up to 70x the number of input variables (which, for this study, would be about 700 people). In a more recent pre-print aimed explicitly at biomedical research, the researchers argued samples of 20-30 could be sufficient depending on the clustering algorithm used and if the analyst anticipates large subgroup differences. Perhaps this came up in the review process, and the authors didn't have sufficient space in their manuscript to elaborate and still be within the word limit. Still, I am not wholly convinced that the findings from this research are robust beyond the current sample. If the goal was to show proof of principle, that is one thing. However, we need more research to yield knowledge about how people move and how we can change the way we physically prepare people for their sport. Nevertheless, this was still an interesting technique to explore for this work, and I see lots of potential with these approaches in the future.
• Not to be an open-source software zealot, but I would have appreciated it if the authors conducted their statistical analyses in R or Python rather than SPSS. This way, I would know exactly how the researchers handled their data and could have reproduced it if I was so inclined. Given the novelty of their work and the degrees of freedom afforded in these analyses, it would be nice to sink my teeth into their code.

Contributions

• The researchers identified four clusters, but no differences in sprint performance or strength between the clusters (only exception is that clusters "A" and "B" had significantly higher hip extensor torque/contact time ratios relative to clusters "C" and "D").
• Individuals' movement strategies were similar across sprint efforts.
• Individuals adopted different movement strategies but yielded similar movement outcomes, providing further evidence that there is no single "ideal" movement strategy.
• Although movement outcomes were similar between clusters, it is still possible that the design of training programs to elicit further improvements would differ between clusters based on the needs of individuals in that cluster. For example, specific individuals may achieve higher step rates by reducing contact time, flight time, or both. However, future work is necessary to test this method of designing training programs.

🧠 Fun fact of the week

Your ears and nose never stop growing! After puberty, bones stop dividing, but cartilage continues to grow for the rest of our lives.

🎙 Podcast recommendation

Putting this here so I get around to listening to it :)

🗣 Quote of the week

Louie Simmons was one of the prominent people I read and listened to when I started my journey in this field over a decade ago. Never one to shy away from a controversial opinion, Louie often spoke his mind without any semblance of a filter (for better and for worse).

Unfortunately, Louie passed away at the age of 74 on March 24. I never got a chance to head down to Westside Barbell and meet him to talk shop about strength training and powerlifting, so perhaps this is also a lesson to be more proactive in life. However, I want to leave you with some of Louie's quotes that stuck out to me over the years:

"If you run with the lame you will develop a limp."

"Just when your body has all the answers, you have to change the questions."

"Normal people only give you normal results..I don’t need that."

"Don’t have $100 shoes and a 10 cent squat."

"Everything works, but nothing works forever."

"If you turn your car left all the time, some parts will be worn out and some won’t be touched. Your body is the same way."

"A pyramid is only as tall as its base."

Rest in Peace, Louie. Thanks for introducing us to Siff, Zatsiorsky, Verkhoshansky, Bompa, and many others all those years ago. When everyone else was on bodybuilding.com, you got us onto Russian training manuals. Louie was truly a pioneer in this space and will be missed by countless others as well.