Serial Estimation of Survival Prediction Indices Does Not Improve Outcome Prediction in Critically III Dogs with Naturally Occurring Disease
Article first published online: 1 JUL 2007
Journal of Veterinary Emergency and Critical Care
Volume 11, Issue 3, pages 183–189, September 2001
How to Cite
King, L. G., Fordyce, H. and Maislin, G. (2001), Serial Estimation of Survival Prediction Indices Does Not Improve Outcome Prediction in Critically III Dogs with Naturally Occurring Disease. Journal of Veterinary Emergency and Critical Care, 11: 183–189. doi: 10.1111/j.1476-4431.2001.tb00082.x
- Issue published online: 1 JUL 2007
- Article first published online: 1 JUL 2007
- severity of disease scoring;
- risk stratification;
- clinical trials;
- intensive care unit
Objective: The objectives of this study were to test the value of adding serial measurements to the Survival Prediction Index (SPI 2), and to investigate whether time trajectories add predictive information beyond measurements at a single point in time.
Design: Prospective clinical trial.
Setting: Intensive care unit at a Veterinary Teaching Hospital.
Animals: 63 critically ill dogs
Interventions: Physiologic data were collected within 24 hours of admission to the ICU (Day 1), and again on Day 3 of hospitalization.
Measurements: The first analysis applied the SPI 2 equation on Day 1 and again on Day3. Then a prediction model was re-estimated using Day 1 measurements, and the incremental predictive value of adding Day 1 to Day 3 change scores was evaluated. the third analysis tested the incremental predictive value of change scores in models containing only one prognostic variable. The final analysis compared the re-estimated Day 1 model to an analogously re-estimated Day 3 model.
Main Results: Using the SPI 2 equation, the AUC was 7.7% higher using Day 3 measurements than that obtained using Day 1 measurements (P = 0.515). Starting with the re-estimated Day 1 model (AUC = 0.925), forward stepwise addition of the difference score for each variable did not result in an improvement in the AUC. The AUC for the re-estimated Day 1 model was not statistially different from that of the re-estimated model using Day 3 measurements.
Conculusion: This study shows no benefit to repeated calculation of the SPI 2 later in hospitalization.