← All health signals
Trust signal

The Predicted Injury

Machine-learning models that forecast footballer injuries now face their first evidence audit: a 2026 review rates 17 of 22 studies at high risk of bias.

What is it?

Injury models give each footballer a weekly risk score from training load, strength tests and injury history. A 2026 systematic review finds most rest on weak method and few are checked outside their own data. The reported accuracy and the verified accuracy are drifting apart.

Why now?

In July 2026 a review of 22 studies rated 17 at high risk of bias, mostly in how models were analysed. Earlier that year, one prospective model was tested on a later season at the same six clubs. The field now has both an audit and a template.

Who’s building it?

Club medical and performance staff, sports-science researchers and injury-analytics vendors, with journal reviewers setting the bar.

Club medical teamsSports scientistsAnalytics vendorsJournal reviewersPerformance staff
The full read
Why now, who is building it, where the market goes, and what stays unsettled.
Get weekly →
01 · The market
Who’s in it
Builders — sports scientists and vendors modelling player load data.
Buyers — club medical and performance staff who decide who trains and who rests.
Gatekeepers — journal reviewers and the PROBAST standard for prediction-model bias.
What’s at stake
A risk score that reads as precise can steer who trains and who rests before anyone has checked it.
Players’ trust in a forecast about their own bodies.
Budget spent on models no second club has tested.
What’s unsettled
Whether temporal validation at the same clubs carries to new ones.
Whether the observational 73% reduction survives a randomised trial.
Which metric clubs will trust when injuries are rare and accuracy flatters every model.
02 · How the demand matureda progression, not a timeline
Novelty
Staff try risk scores that promise to name next week’s injury.
Retention
Weekly scores become routine, fed by load and strength testing.
Willingness to pay
Clubs weigh a preliminary €215,800 season saving against models few have validated.
Category
Validated injury forecasting becomes a checked layer in sports medicine, not a claim.
03 · Where the market goesthe forks, not a prediction
→Checked or claimed: clubs ask for calibration and a test at a new club, or keep buying on reported accuracy alone.
→A randomised trial confirms the intervention effect, and risk scores move into routine care.
→The effect fades under trial conditions, and models settle as one input among many for staff.
→Reporting standards from clinical prediction research become the entry ticket for sports models.
Everything turns on whether a model that works at six clubs keeps working at a seventh, with players and injuries it has never seen.
04 · The Checked Body deskadjacent market, moving now
The Body Tells First
Wristbands read stress from several signals.
One wristband tracks heart rate, skin conductance and sweat cortisol.
Spectrum News
The Monitored Scalp
Hair-counting AI gets validated against clinician-checked manual counts.
Shed-hair counts land within 5.4 hairs of manual counts.
Clinical, Cosmetic and Investigational Dermatology (Dove Medical Press)
The Bias Audit
Algorithm audit mandates meet political pushback in 2026.
Colorado repealed its bias-audit law in May 2026.
Colorado General Assembly
05 · The Wire
01
Most reviewed injury-prediction studies rate at high risk of bias.
Journal of International Medical Research ↗
02
The model was tested on a later season of data.
Scientific Reports ↗
03
A 2025 random-forest model reported an AUC of 90.5% for 300 players using cross-validation only.
BMC Medical Informatics and Decision Making ↗
The panel on this signal
They agreed on
–Accuracy is the wrong headline number for rare injuries, and several published models still lead with it.
–The 2026 validation is temporal, within the same clubs.
–No public record shows a model score deciding selection or a contract yet.
The sports-injury epidemiologist disagreed on
–Clubs define an injury differently, so a model validated on one club’s records predicts a different outcome at the next.
Know more about the method →
✦How this signal was built
01 · Panel
A Delphi panel of 4 AI analysts, each reasoning independently.
→
02 · Debate
They challenge and revise each other’s reads over two rounds.
→
03 · Evidence
Every claim is tied to a cited, linked source on The Wire.
→
04 · Synthesis
A moderator writes it into one read, disagreement and all.