Evaluation of the survival prediction index as a model of risk stratification for clinical research in dogs admitted to intensive care units at four locations

OBJECTIVE: To prospectively evaluate a survival prediction index (SPI) in dogs admitted to intensive care units (ICU) and to generate and test an improved SPI (ie, SPI2). SAMPLE POPULATION: Medical records of 624 critically ill dogs admitted to an ICU. PROCEDURE: Data were collected from dogs within 24 hours after admission to an ICU. Variables recorded reflected function of vital organ systems, severity of underlying physiologic derangement, and extent of physiologic reserve; outcome was defined as dogs that survived or did not survive until 30 days after admission to the ICU. Probabilities of survival were calculated, using an established model (SPI). We then performed another logistic regression analysis, thereby reestimating the variables to create the new SPI2. Cross-validation of the models obtained was performed by randomly assigning the total sample of 624 dogs into an estimation group of 499 dogs and validation group of 125 dogs. RESULTS: Testing of SPI resulted in an area under the curve (AUC) of 0.723. Testing of SPI2 revealed an AUC of 0.773. A backwards-elimination procedure was used to create a model containing fewer variables, and variables were sequentially eliminated. The AUC for the reduced model of SPI2 was 0.76, indicating little loss in predictive accuracy. CONCLUSIONS AND CLINICAL RELEVANCE: The new SPI2 objectively stratified clinical patients into groups according to severity of disease. This index could provide an important tool for clinical research.