Machine learning-based prediction and quantification of OCD surgery and pedigree effects on racehorse performance.
Authors: An S J, Sohn Y, Forbes E, Ryu S-H
Journal: Veterinary journal (London, England : 1997)
Summary
# Editorial Summary Osteochondritis dissecans (OCD) represents a significant developmental challenge in racing Thoroughbreds, yet uncertainty persists regarding whether arthroscopic surgery genuinely compromises future race performance—a question this retrospective study sought to clarify using machine learning methods on 75 surgically treated horses and 257 unaffected maternal siblings. The researchers built gradient-boosting models (XGBoost and CatBoost) incorporating biometric, pedigree and surgical variables, plus derived features such as earnings per start and sale-price ratios, with performance quantified using the field-adjusted metric 'race_pts_avg'; the optimal CatBoost model achieved an R² of 0.7983, indicating robust predictive accuracy. Surprisingly, surgical history variables including age at operation showed minimal predictive importance, and lesion severity did not emerge as a dominant performance factor—instead, pedigree (particularly family grouping) proved the strongest predictor of long-term racing outcomes. These findings suggest that OCD surgery itself does not substantially impair subsequent race performance, positioning pedigree as the dominant influence on career trajectory in this cohort. For practitioners, these results provide objective evidence supporting the pursuit of surgical correction in young horses with OCD, though regional study limitations and unmeasured confounding factors warrant cautious interpretation when applying findings to individual cases or different populations.
Read the full abstract on PubMed
Practical Takeaways
- •OCD surgery in Thoroughbreds does not appear to compromise future racing ability, supporting treatment as a reasonable option for breeding and racing stock
- •Genetic pedigree factors are more important than surgical details when predicting race performance outcomes; breeding decisions should weigh pedigree heavily
- •Treatment and sales decisions for horses with OCD history can be guided by the evidence that surgery does not significantly impair long-term performance potential
Key Findings
- •OCD surgery does not significantly impair long-term race performance in Thoroughbreds
- •Pedigree (family identifier) was the strongest predictor of race performance (R² = 0.7983 with CatBoost model)
- •Surgical history variables including age at surgery and lesion severity showed limited predictive value for post-operative performance
- •Derived features such as earnings per start and sale-price ratios improved model predictive performance compared to raw variables alone