A computer-derived protocol to aid in selecting medical versus surgical treatment of horses with abdominal pain.

This study presents a new decision protocol, derived by computer analysis, to support clinicians when deciding between surgical or medical treatment for horses with abdominal pain. The research demonstrates that this tool could optimize decision making, yielding few false negatives, thereby reducing unnecessary surgeries.

Study Design and Participants

• The researchers analyzed pre-existing data from 219 horses that presented with abdominal pain at a teaching veterinary hospital.
• This sample size was rigorous enough to generate significant and reliable conclusions.
• The participating horses were already receiving treatment for abdominal pain, which formed the basis for the study.

Methodology

• The researchers employed a multilayered statistical approach: multiple-step discriminant analysis coupled with a recursive partitioning algorithm.
• These methods were used to identify condition variables that would predict whether a horse required surgical intervention.
• The result of this analysis was a decision tree model.

Performance Metrics

• In the examined population, surgical cases represented 79 per cent.
• The utility of the decision tree was evaluated on key performance metrics: sensitivity, specificity, and predictive values.
• Impressively high scores were shown for sensitivity (99 per cent) and negative predictive value (99 per cent), indicating the model had excellent predictive accuracy and an exceptional ability to correctly identify non-surgical cases.
• However, the specificity and positive predictive value were lower at 55 per cent and 90 per cent respectively, showing some limitations in precisely distinguishing surgical cases.
• Compared to clinical decisions made without the model, the decision tree yielded more false positives (an 11 per cent rate) but nearly eliminated false negatives (down to a 1 per cent rate).

Validation and Comparison

• The findings were validated using the jack-knife method which is a statistical technique used in machine learning to estimate the performance of predictive models.
• Furthermore, the decision tree was tested with a new sample at a different veterinary teaching hospital where the prevalence of surgical cases was 55 per cent.
• The model retained impressive sensitivity, specificity and predictive values (93 per cent, 73 per cent, 81 per cent, and 89 per cent respectively), confirming the validity of the model in different settings.

Implications

• Overall, this research offers a potentially beneficial decision-making tool for clinicians treating abdominal pain in horses.
• The decision tree model can help in generating more accurate decisions while reducing the number of unnecessary surgeries, offering a significant potential to save cost and resources.
• The research also opens avenues for further refining and testing of computer-derived protocols in medical decision making for veterinary practices.