PRISM: Pre-Injury Sprint Mechanics

PRISM (Pre-Injury Sprint Mechanics) is FitWise's research study of hamstring injuries in Premier League football. It asks whether the way a player sprints changes in the matches before a hamstring injury, using nothing but broadcast footage.

Primary

To find out whether measurable biomechanical changes appear in a player's sprint mechanics before a hamstring injury. If they do, the study will show what they look like and how early they appear.

There are three possible outcomes.
→ Nothing detectable precedes the injury.
→ One common pattern precedes every injury.
→ There are several distinct patterns, each shared by a group of players.

Secondary

To find out whether players who return to play after a hamstring injury get back to their own pre-injury baseline, or settle at a new one. Players who are injured again are analysed as a separate subgroup.

PRISM is a retrospective study. Every player is measured in 3D from existing broadcast footage, with no wearables, markers or extra cameras.

Cohort

• Injured group: 127 Premier League players with 166 hamstring injury episodes across five seasons, 2021/22 to 2025/26. Some players were injured more than once.
• For each injury: the four matches before it and the return-to-play match after it.
• Comparison group: players with no injury, measured over matched periods.
• In total, around 800 broadcast matches.

What we measure

We measure every high-speed run and sprint a player makes in each match: around 45,000 efforts in total. Runs are efforts above 4.5 m/s, and sprints are efforts above 6 m/s. Each effort is cut into stride cycles and each cycle into five phases: touchdown, mid-stance, toe-off, maximal vertical projection and late swing. At every phase we record joint angles and differences between the left and right sides.

Each player is compared against his own healthy baseline, not a population average.

Analysis principles

• The analysis starts from a small set of measures chosen from the published literature and widens from there.
• All analysis runs on one fixed, versioned dataset, so every result can be reproduced and compared with the others.
• We look for trends first and test the link to injury second.

1. Dataset · complete

We identified the injured players and the injury dates. For each injury we selected the four matches before it and the return-to-play match after it, together with matched periods for the comparison group: around 800 broadcast matches in total.

2. Processing · in progress

Every match is processed from broadcast footage into 3D movement data, and each player's high-speed runs and sprints are extracted. Each run is checked for whether the camera view supports reliable measurement; runs that fail are flagged rather than deleted. The result is frozen into a versioned dataset, with a held-out portion kept back for testing.

3. Data analysis

We analyse the pre-injury question first and then return to play, with players injured again as a separate subgroup. Trends come first; the link to injury is tested second.

4. Publication

We publish the results in a peer-reviewed journal, whether they are positive or negative; submission is planned for Q4 2026. A separate paper describing how the dataset was built may come first, so that later studies can build on it.

• Return to play: whether injured players get back to their own baseline. This may be reported as a separate study.
• Calf injuries, using the same design.
• ARC (ACL Risk in Competition): a study in women's football focused on ACL injuries.