Integration of machine learning and viscoelastic testing to improve survival prediction in horses experiencing acute abdominal pain at a veterinary teaching hospital.
Authors: Macleod, Wilkins, McCoy, Bishop
Journal: Equine veterinary journal
Summary
# Editorial Summary Accurate prognostication in horses presenting with acute abdominal pain remains clinically challenging, particularly in cases complicated by systemic inflammation or shock; this retrospective study investigated whether machine learning algorithms incorporating viscoelastic coagulation testing (VCT) could improve survival prediction beyond conventional linear approaches. Researchers performed VCT on 57 colicky horses at admission (predominantly colitis cases) alongside standard clinical data collection, developing predictive models using a training cohort of 40 horses and validating performance on the remaining 17. Whilst coagulopathy diagnosis alone proved a poor prognostic indicator (AUC 0.515), random forest machine learning models substantially outperformed traditional logistic regression, achieving 83–91% sensitivity and 83% specificity by integrating heart rate, packed cell volume, lactate, white blood cell and neutrophil counts alongside VCT parameters (clot amplitude at 20 minutes and clot time). Lactate concentration emerged as the most consistently important predictor of survival across both model types, confirming its established clinical utility in equine colic cases. Whilst validation in an independent population is required before clinical implementation, these findings suggest that integrating viscoelastic coagulation data into machine learning-assisted decision support systems could meaningfully enhance clinicians' ability to stratify risk and guide treatment decisions in acute equine abdominal cases.
Read the full abstract on PubMed
Practical Takeaways
- •Machine learning-based prognostic tools using readily available clinical data plus viscoelastic testing may help identify horses with colic at highest mortality risk earlier in treatment
- •L-lactate concentration should remain a critical component of prognostic assessment in acute colic cases, as it provides strong predictive value across multiple model types
- •Viscoelastic coagulation testing adds predictive value when combined with clinical variables in ML models, but coagulopathy diagnosis alone is insufficient for survival prediction and should not be used as a standalone prognostic indicator
Key Findings
- •Random forest machine learning models achieved 91% sensitivity and 83% specificity for survival prediction compared to GLM at 65% sensitivity and 42% specificity
- •L-lactate remains a key independent predictor of survival in horses with acute abdominal pain
- •Coagulopathy diagnosis alone performed poorly for survival prediction (AUC = 0.515) with 81% sensitivity but only 31% specificity
- •Integration of viscoelastic coagulation testing parameters (CT, A20) with clinical variables (heart rate, PCV, lactate, WBC, neutrophil count) improved prognostication accuracy