Real world research worth exploring: Machine learning algorithms in injury prevention

Today, I would like to comment on a recent paper reporting the use of machine learning algorithms to estimate injury risk in team sports athletes (López-Valenciano et al., 2018). Machine learning (ML) is a relatively new approach in sports medicine and science that applies certain algorithms, mostly without pre-defined assumptions, to solve complex problems like the sports injury prediction. As the name indicates, machine learning attempts to make computers "learn" and produce more and more accurate algorithms. As a discipline it integrates statistics with computer science.

In the study of López-Valenciano, a total of 132 male professional soccer and handball players underwent pre-season screening evaluation which included personal, psychological and neuromuscular measures. In addition, injury surveillance was employed to all musculoskeletal injuries during the season. The authors employed different learning techniques to check their accuracy in injury prediction.

Their results showed that the machine learning algorithms presented moderate accuracy for identifying players at risk of injury. From other studies, we know that ML accuracy can be improved with more data entered in the analysis. Nevertheless, the novelty of the study of López-Valenciano and colleagues is they showed that machine learning can assist in solving problems like the identification of players at risk. However, one should bear in mind that ML algorithms work well for the population they were created and we cannot predict what will happen with another set of data.

Real world research worth exploring: Accelerometer-based prediction of sports injury

analog.com

The use of accelerometers in studying the non-contact injury risk is a hot topic both in team and individual sports. Recently a research team from the University of California tested the hypothesis that the running-related injuries were the result of a combination of high load magnitude and strides number that result in accumulated microtrauma (Kiernan et al., 2018). During the studying period of 60 days, elite runners wore a hip-mounted activity monitor to record accelerations while training. From these accelerations the researchers estimated the vertical ground reaction forces (vGRFs). Their results showed that  the injured athletes had significantly greater peak vGRFs and weighted cumulative loading per run. 

The beauty of this study is the use of a common accelerometer to derive data associated with injury risk. Of course, these findings should be verified in bigger samples but the main message of this study is that this type of microtechnology, much cheaper that the GPS-embedded accelerometry, may assist in injury risk management for athletes/teams with limited resources.

Player performance metrics: time to reconsider our approach!

There are different performance metrics but most of them are looking at football performance in a fragmented way. For instance, match running distance at different speeds is considered as an important index, sometimes without taking into consideration the match context (player's position, opposition performance etc). 

Also, it seems that some actions, very decisive for the team performance, are not properly evaluated. As an example, in one moment in last night's match Bayern Munich striker defended very effectively against Ronaldo (at 6th minute of the match, match highlights here ). 

How would you rate an attacker's performance doing fantastic work while team is defending? Is it time to reconsider our approach and integrate physical performance  with technical and tactical data? Your thoughts?