Case study
The Living Athlete Digital Twin: A Real-World Validation in Professional Football
Real-world validation of PrediTwins across two professional football clubs. PrediTwins detected 31 of 34 recorded muscle injuries, with an average 5.2-day early warning and 83% specificity.
91%
Muscle injuries detected
5.2 days
Average early warning
83%
Specificity
Muscle-specific
Risk identification
The missing layer
The missing layer in athlete monitoring
Professional football already measures almost everything about the athlete. GPS tracks external load. Force platforms measure performance. Wellness monitoring captures fatigue and recovery. Medical systems record clinical information.
But one critical layer remains difficult to observe: how is each athlete’s body responding internally to those demands? Two players can complete the same training session and accumulate similar external loads while experiencing very different responses at muscle and tissue level.
PrediTwins was built to reveal this missing layer. Using Living Athlete Digital Twins, the platform combines athlete data with biomechanically derived, muscle-specific indicators to model how each athlete’s internal state evolves under training and competition. The objective is not simply to flag a player as being at risk. It is to identify where risk is emerging, and early enough for the staff to act.
External Data
GPS · Wellness · Performance
Living Athlete Digital Twin
Individualized biomechanical model
Internal Response
Muscle & tissue state
Muscle-Specific Risk
Where · When
Earlier Decisions
Time to act
The results
31 of 34 injuries detected
During the 2025/26 season, PrediTwins was evaluated using approximately ten months of real-world data from Red Star FC and Valenciennes FC. The analysis included 34 recorded muscle injuries involving 24 injured players.
At the selected six-day prediction horizon, 31 of 34 muscle injuries were preceded by an alert identifying the relevant muscle, a 91% detection rate.
91%
Injuries detected (31 / 34)
5.2 days
Average early warning
Among correctly detected injuries, the first warning appeared an average of 5.2 days before the recorded injury, an actionable window in which performance and medical teams can investigate the athlete’s condition, contextualize the signal and decide whether an intervention is appropriate. PrediTwins does not make that decision for them. It gives them time to make it.
Not just who. Where.
Most injury-risk approaches ultimately produce information at player level: is this athlete at increased risk? PrediTwins goes one level deeper: rather than generating only a global player risk score, it identifies where risk is emerging and how it evolves over time.
Player-specificMuscle-specificDynamic
External data tell us what the athlete did. The Digital Twin helps reveal how the athlete responded.
Alert quality
High detection, without alert overload
High injury detection can be misleading if it comes from continuously classifying athletes as being at risk. That is why specificity matters. At the selected operating point, PrediTwins combined 91% injury detection and 83% specificity with only 0.18 alerts per player per week, approximately 4.5 alerts across an entire 25-player squad, per week.
An alert is only valuable if the staff can trust it.
Does biomechanical information add value?
PrediTwins was compared with an ACWR-based workload-monitoring benchmark using the same GPS information, at the six-day horizon:
| Approach | Injury detection | Specificity |
|---|---|---|
| External workload benchmark (ACWR) | 94% | 44% |
| PrediTwins | 91% | 83% |
Comparable detection, far greater specificity, translating into 3.3× fewer false-alert days than the workload-based benchmark. Modeling internal biomechanical response adds information that external workload alone does not capture.
Validation rigor
Tested on players the model had never seen
Predictive performance can be overestimated when a model is evaluated on athletes whose data were also used during training. Validation was therefore grouped by player: a player included in the test data was never simultaneously included in the training data. Decision thresholds were set on training data and applied to held-out players without retrospective adjustment.
2
Professional football clubs
~10 months
Real-world monitoring
34
Recorded muscle injuries
24
Injured players
3–7 days
Prediction horizons evaluated
Grouped
Cross-validation by player
From measuring to understanding
See risk before it becomes visible.
Sports technology has become exceptionally good at measuring athletes. The next frontier is understanding what those measurements mean inside the individual athlete: the role of the Living Athlete Digital Twin.
91%
Muscle injuries detected
5.2 days
Average early warning
83%
Specificity
3.3×
Fewer false alerts vs. ACWR
Want the full methodology?
See how the validation was run, and what a Living Athlete Digital Twin could surface for your squad.
Contact: contact@preditwins.com