Using computer models to predict injury in football
Football and modelling, it’s not what you think!
When the the words “model” and “football” appear in the same sentence, more often than not, it conjures up images of a sultry footballing Adonis rather than a whirring computer that’s busy trawling through bits of data.
So why do we need computer models in football?
It’s no secret that footballers get injured. When they do, this can mean big money for football clubs in the Premier and UEFA leagues. The clubs still have to pay the wages of injured players (and medical bills) and may even need to buy other players to replace them in the season. In some of our previous research we investigated if some of the tests used by sports medicine teams were able to identify footballers at risk of injury. Our research showed that the tests weren’t very good and so we decided to see if we could develop a computer model to try and help sports medicine teams identify footballers at risk of injury.
What actually is a computer model?
In most health care related research, the word model is usually used to describe either a bio mechanical or statistical model. Both involve lots of maths, but in brief, a bio-mechanical model is usually a mathematical representation of the muscles and bones of a person, where as a statistical model is a mathematical equation which uses information about something or someone to either diagnose (diagnostic) or predict (prognostic) something.
These models can be very useful. However, in order for them to work well, you have to make sure that you have the right information . As the saying goes,“rubbish in = rubbish out!”
What did you do in your study?
In our new study, we developed a statistical model that tries to predict who will get injured (prognostic). Predicting injuries in footballers can be difficult as 1) luckily injuries are pretty rare relative to how often they train and play, and 2) footballers don’t always tell you when they are injured as they may not think it is serious enough. Some new (and exciting if you’re a stats nerds like us!) statistical methods, which could potentially help address some of these challenges, have been developed but not been used in football and so we wanted to give them a go!
For our model we used easily collectible information such as how long and on what surfaces footballers trained. We also used information about the footballers themselves, such as their height, weight, body fat and any injuries they had, both before and during the season.
In the first stage of the study, we had to find out which of these bits of information were the most important when trying to predict injuries that made footballers miss days of training or matches. For this part of the study, we compared the new methods against the traditional methods.It turns out the new methods were better. Our model didn’t need as many measurements to make their predictions and all the bit of information it selected could be reasonably linked to injury. Additionally, the traditional methods got a bit muddled up, for example, they said some things both increased and decreased the chance of injury at the same time. Go figure? Fewer things associated with injury is good news, as it means sports and exercise medicine practitioners need to collect fewer bits of information when trying to predict injury.
Whilst we knew which of bits of information were important and which were not, we didn’t know how important each bit of information was. In order to find this out, in the second stage of the study, we had to develop the model using the information identified in the previous step. The model told us that if footballers had more previous injuries, increased their weight, number of training sessions, amount of time training on grass, amount of time playing a match on artificial turf, they were more at risk of getting an injury that would make them miss training or matches. Any activities on an artificial surface of 3G increased the risk of injury and fitter players were more at risk of getting an injury. (We think this is because they can get around the field more which increases their risk of getting an injury.) Unsurprisingly, playing matches and getting injuries during the season were also linked to more severe injuries.
So what do footballers need to do to stop themselves picking up injuries?
It’s going to help if they’ve been injury free previously but there’s not much they can do about that now. This may however have implications for younger footballers as they start progressing through their football careers. Sports performance teams have already started at looking at ways to optimise performance and minimise injuries in youth football and our research shows whey this is important in the longer term. Footballers can also help themselves by doing more conditioning and technical sessions aswell as building up their training and match time gradually. Having a bit more body fat also seems to help but that’s no excuse for a pie on the subs bench!
So what’s next?
Now that we have developed the model and identified the most important bits of information needed for predicting injuries, we hope to test it in other football clubs. We developed the model so that it could be used in a real world setting to try and identify footballers who are at risk of getting injured, before they go out to train or play in a match. We know that some football clubs may collect additional information, but the methods in our study can still be used to help choose the most important information. This reduces the number of things people need to collect, saving time, energy and possibly money. Ultimately we hope this will result in fewer football injuries and make life easier for medical teams by helping to guide decision making. That’s our prediction anyway!
