A point-based laminitis risk scoring system and machine learning framework for prediction of laminitis in horses.

Laminitis is a painful and potentially life-threatening inflammatory condition of the equine hoof, and its detection remains a major challenge in veterinary practice. Although several risk factors have been reported, the prediction of the onset of laminitis in a subclinical stage using routine clinical data is still limited. This study aims to address this gap by developing an interpretable machine learning (ML) framework for the prediction of the risk of laminitis in horses. A point-based risk scoring system was first constructed using multivariate logistic regression to quantify the contribution of clinical, physiological, and conformational variables. Subsequently, this statistical model was combined with multiple ML classifiers to improve predictive performance. Explainability techniques, including ELI5 and feature importance analysis, were applied to each model to improve transparency and support the clinical interpretation of their predictions. Among the nine classifiers evaluated, SVM achieved the highest F1-score of 0.858 ± 0.152 and MCC of 0.811 ± 0.211, with TabPFN recording the highest AUC of 0.932 ± 0.094 and Random Forest achieving the highest precision of 0.917 ± 0.180. The risk scores reached maximum values of up to 0.97, with the lameness examination right fore (LERF), hoof testers right hind (HTRH), digital pulses, rectal temperature, and age identified as the most influential predictors across all evaluated classifiers. In general, the results demonstrate that the integration of statistical risk modeling with interpretable ML enables an accurate and clinically meaningful assessment of laminitis risk. The proposed approach provides a practical decision-support tool that may help veterinarians in preventive management and improve the results of equine welfare. All the source code is available at github.