Who re-sprained a chronically unstable ankle over two years

A prospective cohort study published in Sports Medicine – Open on 7 November 2026, by researchers from Brigham Young University (USA), Hong Kong Polytechnic University and Korean universities. Participants were recruited and tested at Brigham Young University between February 2017 and February 2018, then followed for two years.

One hundred physically active university-aged adults with chronic ankle instability (50 men, 50 women, average age 22) were tested once and then followed for two years. Ninety-six answered at least one follow-up and were analysed. Chronic ankle instability was defined using International Ankle Consortium criteria: a significant sprain more than 12 months earlier, at least two giving-way episodes in the past six months, a positive Ankle Instability Index, and reduced FAAM scores.

A re-sprain was a self-reported inversion or supination injury to the same ankle, collected by online survey at 6, 12 and 24 months and by phone for non-responders. No clinician confirmed the injuries, and time off sport wasn't required.

Fourteen possible predictors were recorded at the start, on the more unstable ankle. History and questionnaires were self-reported, and the physical tests were done by one assessor:

Forty-one of 96 people (43%) re-sprained within two years.

Men re-sprained more often than women. Twenty-eight of the 41 re-sprains were in men (odds ratio 3.44, 95% CI 1.37 to 8.99). Sex was the strongest single predictor in the chosen model.

People who re-sprained had more previous sprains (4.4 against 2.9 on average) and more recent giving-way episodes (3.4 against 2.1).

People who re-sprained had less dorsiflexion on the lunge test, 42.6° against 46.5° on average.

People who re-sprained had slightly higher arches (arch height index 0.37 against 0.36) and worse scores on the Biodex stability test and Star Excursion Balance Test.

FAAM scores, age, BMI, activity hours and triple-hop distance added little or nothing.

The chosen model, a penalised regression called Elastic Net, reached an AUC of 0.807 (95% CI 0.714 to 0.893), with a calibration slope of 1.02. It flagged 83% of people who went on to re-sprain and correctly cleared 62% of those who didn't.

Ordinary logistic regression did as well on AUC, at 0.812. Its predicted risks were less well calibrated (slope 0.48), and so were those of most of the other machine learning models. Elastic Net is itself a form of regression. When predictors were filtered before modelling, AUCs fell to between 0.72 and 0.74.

The authors also report exploratory cut-offs: five or more previous sprains, lunge dorsiflexion of 37° or less, arch height index of 0.37 or more, and a Biodex stability score of 1.0 or more. They state these aren't validated.

With one risk threshold for everyone, the model flagged 93% of men who re-sprained and 62% of women who re-sprained. With separate thresholds for men and women, it flagged 71% of men and 77% of women who re-sprained. On the authors' main fairness measure, women were still flagged as high risk less often than men (a ratio of 0.68, below their 0.80 benchmark). The estimates for women rest on only 13 re-sprains, and a model without sex as a predictor wasn't tested.

Funded by a Brigham Young University graduate research grant and a Hong Kong Polytechnic University research centre. The authors declared no conflicts of interest, and no industry funding was reported. No study registration was reported; the paper follows the TRIPOD+AI reporting checklist.

The predictors the model selected most often were sex, arch height index, previous sprains, lunge dorsiflexion and the Biodex stability score. Two of those come straight out of a standard ankle assessment. Sprain count is one question, and the authors' exploratory cut-off was five or more. The lunge test was measured in degrees with a digital inclinometer, and a free clinometer app on a phone held against the shin does the same job. For knee-to-wall distance, a commonly used conversion is about 3.6° per centimetre, which puts the 37° cut-off at roughly 10 cm and the 4° average gap between the groups at about 1 cm. The Star Excursion Balance Test, also worse in people who re-sprained, needs only tape on the floor. Arch height index needs a measuring device or a consistent method, and the Biodex needs the machine. The authors say none of the cut-offs are validated.

The model hasn't been tested on anyone outside this group. It was checked only within the same 96 people, with about three re-sprains per predictor, and is likely to perform worse in a new group.

The sample was young, active university-aged adults in one US city. Older patients, elite athletes and people in other sports may behave differently.

Whether improving dorsiflexion or balance lowers re-sprain risk is untested. That would take an intervention trial.

Re-sprains were self-reported without any severity threshold, so minor rolls and significant injuries counted the same.

No score or calculator was published, so the model can't be applied in clinic as it stands.

In 96 young, active adults with chronic ankle instability, 43% re-sprained within two years. Men, people with more previous sprains, less lunge dorsiflexion, higher arches and worse balance scores were more likely to re-sprain, and ordinary logistic regression separated the groups as well as machine learning. It seems like simple measures such as sprain history and the lunge test carry useful signal, though the model needs testing in a new group before it means much for an individual patient.