Performance of predictive models of survival in horses undergoing emergency exploratory laparotomy for colic.
Authors: Bishop Rebecca C, Gutierrez-Nibeyro Santiago D, Stewart Matthew C, McCoy Annette M
Journal: Veterinary surgery : VS
Summary
# Editorial Summary Predictive models for colic surgery survival have become established tools in equine practice, yet their reliability across different populations remains unclear. Bishop and colleagues retrospectively evaluated how well five previously published survival prediction models performed when applied to 260 horses undergoing emergency laparotomy at a single Midwestern referral centre, calculating sensitivity, specificity, and predictive values for each model. The models showed moderate performance overall—averaging 79% sensitivity but only 48% specificity—with considerable variation between individual models (sensitivity range 44–94%, specificity range 22–83%), and notably, these results fell short of the predictive accuracy reported in the original studies from which the models were derived. Blood lactate concentration ≤6 mmol/L and colic severity score emerged as highly sensitive indicators of survival, though both lacked specificity, meaning whilst they reliably identified horses likely to survive, they generated false positives that could mislead clinical decision-making. These findings highlight that survival prediction models are population-dependent and may not translate reliably between surgical centres, suggesting that practitioners using established models should interpret results cautiously and that development of geographically or institutionally specific models—or collaborative refinement using multi-centre data—would strengthen their clinical utility for prognostic counselling and treatment planning in colic cases.
Read the full abstract on PubMed
Practical Takeaways
- •Published survival prediction models should be used cautiously when applied to your own patient population, as they may overestimate sensitivity; consider developing or validating local prognostic tools specific to your hospital's case mix and outcomes
- •High blood lactate and severe colic scores are useful for identifying horses unlikely to survive, but normal values or low severity scores should not be relied upon alone to predict good outcomes in your population
- •Collaborate with other surgical centers to develop more robust, multi-center validated prediction models that account for regional differences in case selection, surgical capability, and post-operative management
Key Findings
- •Previously published survival prediction models showed mean 79% sensitivity but only 48% specificity when applied to an independent surgical population
- •Blood lactate ≤6 mmol/L and colic severity score were highly sensitive predictors but had poor specificity for survival
- •Single-variable and multivariable models performed similarly (79% sensitivity, 64% accuracy) but substantially worse than in their original validation cohorts
- •Population-specific factors significantly influence the reliability of survival predictions across different surgical centers