Data-driven safety limits for assessing perianesthetic mortality risk in dogs and cats undergoing elective procedures

OBJECTIVE: To create data-driven clinical pathology thresholds for routine blood tests to assess perianesthetic mortality risk in dogs and cats undergoing elective procedures. STUDY DESIGN: Retrospective analysis. ANIMALS: Data from dogs and cats visiting Banfield hospitals between January 2019 and March 2023. METHODS: Using routinely collected veterinary data from dogs and cats undergoing elective procedures at primary care veterinary practices, data-driven clinical pathology thresholds were calculated to identify perianesthetic mortality risk. This process involved simulation from a generalized additive model that estimated perianesthetic mortality risk, based on signalment data, procedure type, physical status classification and blood analyte values, with data collected between January 2019 and March 2023. The simulated data was used to test 400 potential thresholds. Optimal thresholds were chosen to maximize sensitivity (proportion of deaths correctly predicted) subject to limits on false positive rate (number of non-deaths incorrectly identified). These optimal limits were evaluated on observed data. RESULTS: Two levels of data-driven clinical pathology thresholds were determined, 'stop' and 'critical stop', with the latter more stringent. For the 'stop' limits, the overall median sensitivity on observed data was 7.16% (interquartile range 4.29%-9.88%) and 11.80% (11.00%-14.60%) for dogs and cats, respectively, and the overall median false positive rate was 2.65% (2.40%-3.31%) and 8.25% (6.86%-9.66%) for dogs and cats, respectively. Albumin, hematocrit, neutrophils and white blood cells were identified as having strong evidence for their utility in creating thresholds for dogs. Neutrophils showed strong evidence for thresholds for cats. These new data-driven thresholds offer more consistent and quantifiable false positive rates and increased sensitivities compared with existing limits. However, weaker evidence resulted in fewer useful thresholds for cats owing to lower data availability. CONCLUSIONS AND CLINICAL RELEVANCE: It is possible to use routinely collected data to create data-driven clinical pathology thresholds for blood tests to identify animals at increased risk of perianesthetic mortality.