Predicting ICU Survival with Data Mining Models
Author Information
Author(s): Gortzis Lefteris, Sakellaropoulos Filippos, Ilias Ioannis, Stamoulis Konstantinos, Dimopoulou Ioanna
Primary Institution: University of Patras
Hypothesis
Can a meta-level predictive approach improve the prediction of ICU survival compared to traditional scoring systems?
Conclusion
Using classic composite assessment indicators as variables can enhance the prediction of ICU survival.
Supporting Evidence
- The DTM achieved an Az score of 0.8773, indicating strong predictive capability.
- The NNM and LRM also showed good predictive performance but were less effective than the DTM.
- The study found significant differences in APACHE II and SOFA scores between survivors and non-survivors.
Takeaway
This study looks at how to better predict if patients in the ICU will survive by using different scoring systems and computer models.
Methodology
The study used retrospective data from 204 ICU patients and applied decision tree, neural network, and logistic regression models to predict survival.
Potential Biases
Potential biases due to the retrospective nature of the study and reliance on existing scoring systems.
Limitations
The study had a small sample size and was conducted at a single center, limiting generalizability.
Participant Demographics
158 men and 46 women, mean age 49.7 years, with various medical conditions.
Statistical Information
P-Value
<0.001
Confidence Interval
(0.7878, 0.9363)
Statistical Significance
p<0.001
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website