The news desk
Study: Running Data Still Hard-Pressed to Reliably Predict Injury
1 sources · 1 reports
On September 30, 2026, Canadian Running magazine reported on a study published in PM&R: researchers tracked 643 runners preparing for the 2022 New York Marathon over 16 weeks, using their Strava training logs and weekly questionnaire data to build a machine learning model in an attempt to predict whether a runner would be injured the following week. The results showed that when using only data from weeks in which runners had not adjusted training due to injury, model performance dropped sharply, indicating that training data plus questionnaires are still insufficient to reliably predict whether an otherwise healthy runner will be injured the next week. The study also noted that injuries were self-reported by runners rather than clinically diagnosed, that the relationship between training patterns and injury risk is complex, and that there is no simple mileage threshold that applies to everyone.
Story reports
A study published in PM&R followed 643 runners training for the 2022 New York City Marathon over 16 weeks, using their Strava training logs and weekly surveys to build machine-learning models predicting injury the following week. When limited to weeks when runners weren't already modifying training due to injury, model performance dropped dramatically, suggesting training data plus surveys aren't yet enough to reliably predict injury in otherwise healthy runners. Injuries were self-reported rather than clinically diagnosed, and the relationship between training patterns and injury risk was complex, with no simple mileage threshold that worked for everyone.
Canadian Running Magazine