Live Premier League Monitoring Shows Meaningful Separation of Injury Exposure

Out-of-sample monitoring across the 2025/26 Premier League season shows recurring pre-injury escalation and clear separation between risk bands.

Proven: Players in the Critical class recorded 5.6× the observed injury frequency of the league average in this sample.
Proven: Observed injury frequency rises monotonically across the five risk classes, from below the league average at the bottom to 5.6× at the top.
Proven: Player trajectories show risk building ahead of injury events and falling during absence and recovery windows.
Live: The system is useful as an attention-ranking layer, helping staff decide which players merit earlier review.

Be-Healthy.AI ran its early-warning system live across the 2025/26 English Premier League season. The model was trained only on data available up to 30 June 2025, so it ran out-of-sample on a season it had never seen, using publicly available data only: no club systems, no wearables, no integrations.

Two results

Risk classes separate, and they separate in order. Players in the Critical class recorded 5.6× the observed injury frequency of the league average across the sample. The class below it sits at 1.8×. The two middle classes sit at or below the league rate, and the Low class recorded no injuries at all.

Observed Injury Frequency by Risk Class

Premier League 2025/26, monitored live from a 30 June 2025 training cut-off. Observed injury frequency relative to the league average in this sample.

That ordering matters more than the headline figure. A single high number can come from one unlucky group. A gradient that holds across five classes, with the lower ones below the league rate and the concentration at the top, is what a ranking system is supposed to look like when it is working.

Escalation is visible before the event. Selected trajectories show the same operational pattern repeating: risk builds ahead of an injury, falls during the absence, then rebuilds after return. A single spike is noise. A recurring build-up, reset and renewed climb is something staff can act on.

What this does not claim

The study is deliberately narrow. It does not claim deterministic injury prediction for any individual player, it is not a full technical validation pack, and it does not replace medical, performance or coaching judgement.

The Low class recording no injuries is a result in this sample, not a claim that the class carries no risk. The published document reports relative segment performance and selected trajectories; full denominators, labelling rules and statistical detail sit outside its scope.

Its value is practical, and it is about order rather than certainty: helping staff decide where to look first.

Players are anonymised throughout.

These results are from the model as it stood during the 2025/26 monitoring period. The modelling framework has since moved on, and a study of the current models will follow once they have been running for long enough to report on.

Read the full document

The complete study, including the player trajectories behind it.

Download the study

PDF, 7 pages, 811 KB