Breaking Down Zone7, Part IV: How to validate injury risk detection

...

Zone7 uses sophisticated technology – Artificial Intelligence and Machine Learning – to help ensures athletes can train towards peak performance with minimal risk of injury. Now, there has been enormous hype about AI  —  but in an age when everything is supposed to be evidence-based, frankly, there hasn’t been a whole lot of it proving the benefits of AI.

To address the natural need for evidence by many of the industry practitioners, I’m sharing a list of top-10 questions about our validation methods and thinking. These questions will detail a recent case study produced for a client in the UK’s EFL Championship. We will be releasing the full case study on our website shortly.

1. “Predicting Injuries” – is that even possible?

Zone7 does not ‘predict injuries’. We apply AI to understand how workload, biomechanics and recovery affect ongoing risk and in certain cases suggest derisking strategies manifested in optimal workload bands, not rest. One can’t predict injuries incidents, just like one cannot predict the exact moment it will rain in a specific street corner. But the right technology and scientific framework can provide a reliable and actionable forecast – quantifying the risk of incident for a well defined time window.

2. How do you KNOW it works? For real?

On top of standard mathematical methods of validation when developing algorithms, we use two ways of testing to validate our algorithms: ‘retrospective and ‘prospective.

  • Retrospective analysis is looking back at data, applying the algorithm to the data, and seeing which incidents would have been identified in time by the algorithm. In our case, we always do an ‘out of sample test’ which is to say the algorithms were NOT allowed to peek at the data before the test.
  • Forecasting is looking into the future —  analyzing athletes’ data in real-time and flashing alerts and concrete recommendations about impending injury. – this does guarantee prevention but is definitely an opportunity to intervene. We’ve tested the algorithm over many man-years of live, meticulously managed forecasting tests. As well as validating that our intervention is effective and leads to reduced incident rates.

Zone7 has both types of evidence – plenty of it. Here’s a quick image indicating our retrospective results across a dozen soccer environments:

A summary of 12 case studies performed in 2020

Today, I will talk about retrospective analysis, for a team with which we worked with, in the EFL Championship, England’s second division of professional soccer, that is considered as one of the most physically demanding leagues in the sport.

3. Set the stage for us, Tal. What did the Zone7 platform do?

We did a retrospective analysis of injuries for this team’s players, during the 2019/20 season. Zone7 analyzed the risk assessment that would have been provided the night before the injury occurred if Zone7’s algorithms were used at the time. The risk factors and injury types that were successfully identified were compiled and documented. They included all types of injuries, based on data supplied by the club and sustained by players in the study.

In this specific case, there were 29 players included, part of the first-team squad at some point during the season.

I want to stress that Zone7 algorithms saw these data for the first time and WERE NOT TRAINED on them.

4. Tell us which data you used and how you collected it.

So here, it gets rather technical. In simple terms, we used data on past injuries, heart rate data, team schedule data, distance covered at various speeds, accelerations and decelerations, and more. In short, a mountain of data, too big to analyze with a spreadsheet.

5. And what were the results?

During the 19/20 season, players suffered over 30 injuries (we are excluding exact numbers to maintain anonymity). Zone7 retrospective analysis shows our platform could have detected 72% of all injuries up to a week in advance. Of the detected injuries 76% would have gotten a “High Risk” tag – meaning, there is a very high probability this injury will indeed occur.

6. How serious were those injuries Zone7 detected?

They accounted for over 600 lost-man-days – days the players did not play or practice —  or almost two-thirds of total absences.

7. What types of injuries did the Zone7 algorithm correctly identify?

The platform performed well in detecting risk for lower extremity injuries – ankle and hamstring [tendons and muscles on the back of the thigh] – highly common in football/soccer. 100% of these injuries were detected in time. For knee and lower leg injuries, 50 % of the injuries were detected – but for those that were, 100% were correctly tagged as risk of a knee or lower leg injury.

8. So far we covered individual injuries. But what about the team?  What did the retrospective analysis show for the team as a whole?

Zone7’s platform generated a comprehensive risk profile for the team as a whole.  Some things we found are highly intuitive – e.g. forceful acceleration or deceleration have a high impact on injury.

9. Zone7 provides teams with intervention plans, but teams differ widely in their training philosophies.  Does Zone7 impose its own strategy on coaches and trainers?

Not in the least.  We offer, for example, customization —  teams can decide on risk management policies that configure Zone7 in the best possible way for their environment. The team has the power to find its own sweet spot in the tradeoff between detecting more injuries, with a higher volume of alerts and highly precise alerts with a lower, yet meaningful portion of injuries detected.

10. Final Comments?

So, just to clarify.  Zone7’s platform tackled this club’s data, having never seen it before. It detected 71% of impending injuries, up to a 7-days in advance. This translates to keeping more of a team’s players fit and on the field in key games, had the algorithms been used live.

And a really last comment – this EFL case study is one of about a dozen that we have done for clients in several top Serie A, MLS, Bundesliga, EPL, La Liga and. Results are consistent and we are sharing more examples on our website so you’re more than welcome to take a look


Injuries are costly to the team, and often disastrous to players. English Premier League players experienced 804 injuries in the 2018/19 season, costing 18,230 days on the sidelines. We think Zone7’s platform could have correctly identified risk for, creating prevention opportunities for 70% of those.

In the next blog post (and probably the last in this series!), we are going to review some of the ‘Prospective’ case studies, where Zone7 was used LIVE and created two important tools for our clients: (A) the opportunity to intervene by identifying risk in time and (B) easy-to-read intervention plans that affected a POSITIVE change – less injuries and more winning.