The majority of proximal sesamoid and condylar fractures occurred in horses identified as high risk by inertial measurement unit sensors.

Overview

• This study developed and tested algorithms using data from accelerometer-based inertial measurement units (IMUs) worn by racing Thoroughbreds to identify horses at high risk of forelimb fractures.
• The algorithms were able to retrospectively predict which horses would incur proximal sesamoid or condylar fractures with high accuracy, especially when multiple race starts were considered.

Background and Purpose

• Forelimb condylar and proximal sesamoid fractures are common and serious injuries in Thoroughbred racehorses.
• These fractures may be preceded by detectable gait or limb asymmetries during racing that can be captured by IMU sensors.
• The study aimed to develop lesion-specific predictive algorithms using accelerometer data from IMUs worn by horses during races to determine if future fracture risk could be identified.

Data Collection and Algorithm Development

• Data was collected from July 2021 to December 2024, covering 42,623 races involving 15,755 horses.
• Among these, 23 horses sustained condylar fractures after 54 races and 31 horses sustained proximal sesamoid fractures after 90 races.
• Algorithms were created to analyze the accelerometer data to detect patterns of asymmetry or abnormalities linked retrospectively to fractures.

Risk Scoring and Grouping

• Horses were assigned risk scores based on IMU data patterns: green (score = 1, low risk), amber (score = 2, presence of asymmetry), and red (scores 3 to 5, high risk—with 5 being highest risk).
• The scoring system allowed classification of horses’ relative risk of suffering forelimb fractures before the actual injury events occurred.

Predictive Performance and Statistics

• The positive predictive values (PPVs) for the highest risk group were 0.09 for condylar fractures and 0.08 for sesamoid fractures overall, indicating some false positives on single observations.
• When horses had at least 3 race starts with mean risk scores greater than 3 across 3 consecutive starts, the PPV increased dramatically to 0.74.
• For these horses, the odds ratio (OR) of future fracture was 133 (95% confidence interval: 23.5 to 753), showing a very strong association between risk scores and actual fracture occurrence.
• Overall, 89% of the 53 fractures occurred in the red risk group identified by the algorithms.
• More than 98.8% of horses with risk scores less than 5 did not suffer fractures, demonstrating high specificity at this cutoff.
• Conversely, only 1.2% of horses without fractures had a maximum risk score of 5, indicating low false positive rates at the highest risk category.

Conclusions and Implications

• The study demonstrated that accelerometer-based IMU data combined with tailored algorithms can effectively identify racehorses at high risk for serious forelimb fractures well before the injury occurs.
• Repeated measurements (≥3 risk scores per horse) greatly enhanced the precision and predictive power of the models.
• This technology could provide a valuable tool for veterinarians and trainers to monitor horse limb health, implement preventive interventions, or make informed decisions about racing continuation.
• The ability to detect high-risk horses has potential to improve horse welfare and reduce catastrophic injuries by enabling earlier clinical evaluation and management.