Background: Hall et al. used an expert AI system based on 42,631 starts by 15,779 horses including 144 races by 61 horses which suffered either a condylar or proximal sesamoid fracture, to correctly identify >75% of potential fractures. Additional cases of fatal musculoskeletal injuries (FMI) were made available for modelling.
Objectives:
• Can an expert AI system use data from a single run to assess the overall risk for any FMI?
• For an individual horse, can AI estimate the likelihood of FMI by using longitudinal data?
Methods: Data was collected on 54,000 runs including 353 fracture runs of 157 horses with 15 different FMIs.
A proprietary expert AI algorithm (StrideSAFE version 4.5) assigned a Risk Factor for each run on a scale of 1 to 500 (lowest to highest risk). For each horse a Risk Factor Average (RFA) was calculated using all runs over the previous 12-months.
Risk Factor ranges were used to define risk category ratings 1-5 where category 5 were at the highest risk.
Results: Risk score category 1 was associated with 65% of starts. Other proportions were 20% category 2, 10% category 3, 3% category 4, and 2% category 5. RFA and FMI incidences were exponentially correlated (R2 = 0.95 p<0.001) 27.8% of 36 horses with a RFA>125 suffered FMIs compared to 7.7% of 78 horses with RFA 65-125 and <1% of horses with RFA<5.
Conclusion: Only 2% of runs were category 5. To reduce FMI a category 5 run, particularly if involving a horse with a RFA>125, should receive a thorough clinical examination. Longitudinal data and the RFA may provide the most accurate assessment of FMI risk as horses with an RFA 126-500, had a >25% chance of FMI.