The reality of sports injury prediction: lots of effort with little reward

Qatar’s successful bid to host the 2022 football World Cup was atypical in that the World Cup usually takes place in June and July. The World Cup typically falls in the gap between the end and the beginning of most northern hemisphere professional football leagues, and players would have a short rest before returning to their clubs for another league season. However, the June-July World Cup window was unfeasible due to the extreme heat of the Qatari summer. So instead, the competition will take place from 20 November to 18 December, in the (relative) cool of the Qatari winter. This is unique because the major European Leagues will take a break throughout the World Cup, but players will return to action for their clubs immediately after Christmas.

This arrangement provides a unique challenge for footballers and their clubs worldwide, with many voicing concern over the capacity of footballers to maintain performance and remain injury free in the face of such a demanding schedule. These concerns are not unfounded. In 2020, a group of Israeli researchers published research quantifying the estimated financial costs of injuries to footballers in the British Premier League competition⁽¹⁾. They estimated that clubs lose, on average, one league position for every 271 days lost to injury. The financial cost of this to clubs in terms of reduced prize money, tv viewership, and access to more lucrative competitions is estimated to be £45 million. No wonder teams are nervous about their players returning from the World Cup tired and injured.

Sports science research has conclusively demonstrated that teams that manage to keep their players’ injury free perform better over the long term⁽²⁾. In an effort to mitigate the negative effects of injuries, many professional sports organizations now employ sports and data scientists to monitor their athletes and guide them on preventing injuries. For example, in March 2021, Nature.com published an article describing how sports scientists working in professional football used artificial intelligence (AI) and advanced algorithms to “predict” sports injuries. This is an appealing concept! In addition, a modern understanding of training science suggests that injuries occur when players exceed a particular training load and become fatigued. Surely then, if we can measure all the loads that players are exposed to, we could “predict” injury risk and manage them away from it. Again, it is an appealing concept, but this assumption has several problems.

Complex nature of sports

The relationship between training load, fatigue, and injury is not linear. There are a lot of variables to consider, and most of these variables affect other variables. For example, let’s say that a team’s sports scientist has determined that when a player exceeds 25 000m of running in a week, they are at increased risk of injury. This seems like a reasonable assertion, but what if a player is sleeping poorly or is dehydrated? Does that number go up or down? What if they stretch regularly, use ice baths, and follow a nutritionist’s directions? Could we then squeeze more out of that player? What if they are older or younger? Or have more or less playing experience? Or its week one of the season or the week leading up to a big final? The interrelationships between these different factors create a complex system⁽³⁾. A system with any hope of predicting sports injuries would need to account for most of the interrelationships in the complex system. Interestingly, this is one of the major appeals of the use of AI applications – they can account for a vast array of different variables. But AI applications depend entirely on the data entered into them, which provides some challenges.

The process of injury prediction

Developing algorithms that might be able to predict injury is a long and complex process. Norwegian sports scientist Roald Bahr has accurately described this process in his 2016 publication “Why screening tests to predict injury do not work—and probably never will…: a critical review” ⁽⁴⁾.

1. Identify risk factors

The first step in predicting injury is identifying potential risk factors in a prospective cohort study. This step is initiated by identifying a group of athletes and then measuring their exposure to training over time. In the context of professional football clubs hoping to reduce injuries in their team, this would likely involve several measures. Clubs will likely engage in preseason screening, where they might record individual characteristics such as age, body mass, injury history, and muscle imbalances. In addition, teams would record daily measures of training outputs such as total distance, high-speed distance, and the number of accelerations and decelerations performed. Teams may also record individual daily measures such as sleep quality and time, mood state, or hydration status (see figure 1).