The quest for prediction in coaching

Within sport, everyone is now looking at prediction. Coaches, athletes, and support staff are all searching for methods to predict various outcomes, such as injury, talent, performance, or training adaptation. The ability to successfully predict within these areas would obviously be hugely advantageous. Injury prediction could allow you to make interventions to stop that from happening. Talent prediction can allow teams to better focus resources. Predicting adaptations would allow coaches to design better training blocks or alter them based on the predicted response. In other words, prediction is the holy grail of sports science.

An association is not a prediction

As a result of this increased emphasis on prediction, there has been an upturn in research in this field, with many factors associated with an outcome – be that injury, performance, talent, etc. – assumed to be predictors of that same outcome. But are they? This is what a recent editorial in the International Journal of Sports Physiology and Performance explored. In this article, the authors explore the differences between associations – which are often what is found in “prediction” papers – and true prediction, making it a timely read. In short, association helps us explain what has happened previously, whilst prediction allows us to determine a future outcome with a good level of accuracy. Inherent within the predictive powers of a test are the sensitivity and specificity of that test, which allows us to determine how well a test mitigates false positives and false negatives.

This can be quite a difficult concept to understand, so let’s look at an example; that of the gene ACTN3, which I’ve written about before. There are three different versions of this gene a person can have; RR, RX, or XX. Research tells us that the vast majority of elite sprint athletes will have the RR or RX genotype, such that if I were to test all the sprinters who have ever competed in a World Championship 100m final, I’d be almost certain they would all have the RR or RX genotype. So, could we use our knowledge of this gene to test for those people who are likely to become an elite sprinter? On the surface, it seems logical – if essentially all elite sprinters have a certain version of this gene, then testing for it should leave us better informed. However, roughly 5 billion people alive right now also have the RR or RX version of this gene, meaning that the vastly overwhelming majority of people with the RR or RX version of this gene are not elite sprinters, such that testing for it is not going to enable us to predict who will or won’t become an elite sprint athlete with any real sense of accuracy; there are just going to be too many false positives.

Looking backward vs. looking forward

Since associations explain what has happened previously, they’re a good place to start when it comes to trying to prediction and outcome – but they’re not everything. When looking at hamstring injuries, for example, we first need to understand the various factors that were present in a group of athletes that became injured, relative to those that didn’t. Here, we’re explaining what happened previously. But, in order to say we can predict injury, we need to use it to accurately determine future outcomes. So, in the case of hamstring injuries, we would gather the associated risk factors that have explained previous injuries, select those who we think would get injured in future, and see if we’re right.

This was an approach taken by a paper I’m a huge fan of published earlier this year in Medicine and Science in Sports and Exercise. In this study, the authors attempted to find genetic variants that allowed them to predict hamstring injury; such a finding would obviously be hugely important. For five seasons, they collected hamstring injury data on a group of professional soccer players, creating a genetic model that was associated with hamstring injuries. This model included five genetic variants as well as age. Remember, though, that at this point the researchers were explaining previous injuries – the next step was to try and predict them in the future. When they applied the same model to predict which players would and wouldn’t get injured in the future, and then tested this, the model was only as accurate as chance. Which is to say that, although the model explained what had happened in the past, it couldn’t predict what would happen in the future.

From prediction to risk management

In summary, just because something explains what has happened previously doesn’t mean that it will predict what will happen in the future with any certainty. You likely intuitively experience this in your coaching. Perhaps you have some risk factors you associate with injury in your group of athletes; just for the sake of an example, let’s say that you believe not being able to hold a side plank for two minutes is a predictor of back and hamstring injury; do all the athletes you coach who cannot achieve this bench mark get injured? Similarly, do none of the athletes who can hold the side plank for longer than two minutes get injured? The answer, of course, is likely no – indicating the limited predictive power of such tests.

This doesn’t mean that we should throw the baby out with the bath water, however. Being able to explain previous injuries can allow us to identify risk factors associated with that injury type. We can say that a certain test or performance score can increase the chances of being injured – and we should certainly take steps to reduce that risk factor – but that’s not the same as saying the test predicts injury with any certainty, as it likely won’t have the required sensitivity and specificity. Such a distinction is important to keep in mind when it comes to interpreting research that claims to offer predictors of various outcomes.

If you are interested in learning more about this topic, one additional resource I can recommend is “Why screening tests to predict injury do not work – and probably never will . . . a critical review”, which provides a good overview of the use of screening tests in “predicting” future injury.