In the last blog piece, I wrote about the challenges of introducing and sustaining innovation within a sports performance environment for long term success. In this follow-up I’d like to shift the focus to one such innovation, artificial intelligence (AI). In doing so, I hope to go some way towards and demystifying how AI is rising to the challenge of mitigating the risk of injury in professional sport.
Injuries to professional athletes are first and foremost a major human setback. They can hamper team or individual performance, shorten careers, and often present complex rehabilitation challenges. It is for such reasons that sports medical and performance professionals strive for more effective ways to mitigate its risk. Over the last decade, there has been a vast increase in available performance and medical data, but the industry has been struggling to leverage these datasets effectively to minimise injury risk.
Many opinion pieces and peer reviewed research articles question the current capability of utilising AI to find meaningful insights from the available datasets but do acknowledge further research is required. In parallel, AI systems (such as Zone7) are retrospectively and prospectively demonstrating high levels of capability when it comes to accurately forecasting injury risk. Using a probabilistic forecasting approach, such systems will undoubtedly make significant contributions to the future evidence base.
What does probabilistic forecasting mean?
Few things are 100% certain in this life – this certainly applies to an athlete’s performance and risk of injury. AI systems work by analysing available datasets, searching for data patterns, and making probabilistic forecasts, based on A, B, and C factors. This approach is utilised in many other areas like the weather forecasting and is now effectively being applied in the context of injury risk forecasting.
It is important that practitioners understand that injury risk forecasts from AI systems like Zone7 are probabilistic, meaning in essence it is an estimation of injury risk. AI can only make data derived forecasts based on the data presented for analysis, and as injuries are multi-factorial in nature, injury risk forecasting – whilst currently proving to be highly accurate, cannot factor in all causal elements of injury risk and therefore cannot be binary with 100% accuracy.
What are the concerns?
There exists several legitimate concerns around AI derived injury risk forecasts. Examples include areas such as the size of the dataset in which forecasts are referenced against, the need for data science transparency, necessity for broad validation processes and false positive injury risk forecast rates. Zone7’s recently released validation study addresses many of these concerns and I intend to cover a number of these concerns in future blog posts.
In this post though, I do want to offer a perspective on false positive injury risk forecasts as I believe this is something that exists regardless of AI derived insights to injury risk. It is often overlooked that false positives already exist in the sporting environment, an example being athletes reporting issues that need to be considered and managed but do not lead to injury. So, the real question to be addressed regarding false positives is, “Do AI systems increase the volume of false positives disruptively for practitioners or are the volume of injury risk alerts manageable in normal daily workflows?”
Javier Vidal, Former Head Performance Coach, Getafe CF (now at Valencia CF)
Impact and practical application
AI tools like Zone7 exist to empower practitioners with data derived insights that have previously been invisible to the human eye. With low friction, practitioners can now cross reference deep data insights with their own clinically reasoned professional opinions to simulate thinking and conversation. Ultimately, providing higher levels of psychological security around whether to implement any injury minimisation actions.
Zone7’s solution has been tested in many elite sporting environments and refined over several years through analysing data both retrospectively and prospectively. When used prospectively, injury risk alerts are provided in manageable numbers to allow practitioners to mitigate injury risk by intervening proactively and appropriately (Zone7’s publicly available case studies certainly back this statement up).
I personally feel assisting practitioners in reducing injury incidence is something worth innovating for. As more datasets are fed into AI systems, improved capability and accuracy will accelerate. This perpetual capacity for AI injury risk forecasting systems to improve makes the positive impact AI systems like Zone7 have on mitigating injury truly fascinating.