Integration of machine learning and viscoelastic testing to improve survival prediction in horses experiencing acute abdominal pain at a veterinary teaching hospital.

This research investigates how the combination of machine learning models and viscoelastic coagulation testing can improve survival predictions in horses suffering from acute abdominal pain.

Research Objective

• The purpose of this research was to determine the usefulness of machine learning models that incorporate viscoelastic coagulation testing and clinical data to predict the survival outcomes in horses with acute abdominal pain.

Methodology

• The study used a retrospective observational cohort study methodology.
• Viscoelastic coagulation testing was performed on 57 horses with acute abdominal pain upon admission, and clinical data was collected retrospectively.
• A diagnosis of coagulopathy was defined when two or more abnormal VCT parameters were present.
• The researchers developed two types of models: Generalised Linear Models (GLM) and random forest (RF) classification models to predict the horses’ short-term survival.
• They used a training cohort of 40 horses for model development, and validated the performance of these models with the remaining 17 horses.
• The RF models were deployed on a web-based application to showcase how they could be used clinically.

Results

• Out of the 57 horses, 31 survived and 26 did not. The most common case among the horses was colitis (47.7%), with fewer instances of impactions, strangulating obstructions, and other causes of colic.
• Using coagulopathy diagnosis alone was not efficient for survival prediction (sensitivity 81%, specificity 31%, AUC=0.515).
• In the final GLM model, Systemic Inflammatory Response Syndrome (SIRS) score, L-lactate, clot time (CT), and clot amplitude at 10 min (A10) were the variables used.
• The final RF model incorporated parameters such as heart rate, Pack Cell Volume (PCV), L-lactate, white blood cell count, neutrophil count, clot amplitude at 20 min (A20), and CT.
• The RF models displayed better sensitivity (83-91%) and specificity (both 83%) compared to the GLM model (sensitivity 65% and specificity 42%).

Limitations

• The main limitation of the study was a small sample size and the use of convenience sampling.
• Model validation using an independent population is required to further support the use of this model in a clinical setting.

Conclusions

• The level of L-lactate remains a crucial predictor of survival in horses suffering from colic.
• The integration of Viscoelastic Coagulation Testing data with clinical data in machine learning models potentially enhances the quality of prognostication.