Comparing Prediction Models for Hospital Mortality in Critically Ill Cancer Patients
Author Information
Author(s): Verplancke T, Van Looy S, Benoit D, Vansteelandt S, Depuydt P, De Turck F, Decruyenaere J
Primary Institution: Ghent University Hospital
Hypothesis
This study aims to compare the accuracy of predicting hospital mortality in patients with haematological malignancies using multiple logistic regression and support vector machine models.
Conclusion
Both the MLR and SVM models showed good discriminative power, with no significant differences in their ability to predict hospital mortality.
Supporting Evidence
- The study included 352 patients with haematological malignancies.
- The MLR model used 7 variables while the SVM model used only 4 variables for predictions.
- No significant differences were found in the discriminative power of the two models.
Takeaway
The study looked at two different ways to predict if very sick cancer patients would survive their hospital stay, and found that both methods worked well.
Methodology
The study included 352 patients and compared two models: one using multiple logistic regression and the other using support vector machines, evaluating their performance with ROC curves.
Limitations
The study was conducted in a single ICU, which may limit the generalizability of the findings.
Participant Demographics
Patients with haematological malignancies admitted to the ICU for life-threatening complications.
Statistical Information
P-Value
p = 0.19 for model 1 and p = 0.44 for model 2
Confidence Interval
0.768 (± 0.04) for MLR and 0.802 (± 0.04) for SVM in model 1; 0.781 (± 0.05) for MLR and 0.808 (± 0.04) for SVM in model 2
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website