Skip to Main content Skip to Navigation
Master Thesis

Prédiction de la mortalité après chirurgie cardiaque programmée : nouvelles approches par Machine Learning et Decision Curve Analysis

Abstract : Introduction. In cardiac surgery, morbi-mortality is important and the decision to operate is complex. Study purpose was to compare EuroSCORE II and Machine Learning to predict mortality after elective cardiac surgery via a Decision Curve Analysis (DCA). Methods. We conducted a retrospective monocentric study from December 2015 to December 2016, using a prospective data base, from the cardiac surgery unit of a University Hospital in Paris. Non elective cardiac surgery patients were excluded. The different models of prediction of in hospital mortality, including EuroSCORE II, Logistic Regression and Machine Learning, were compared by Receiver Operating Characteristic (ROC) and a Decision Curve Analysis (DCA). Results. The study was carried among 6520 patients. Mortality rate was 6.3%. Average age was 63.4 years old and the average EuroSCORE II was 3.7%. Area under the ROC curve (IC 95%) for the Machine Learning 0.795 (0.755-0.834) model was significantly higher than the one of the EuroSCORE II and the Logistic Regression models (respectively 0.737 (0.691-0.783) and 0.742 (0.698-0.785, p<0.0001). The DCA proved that the Machine Learning model had a greater clinical benefit. Conclusion. According to ROC curve and DCA, Machine Learning model was more efficient than EuroSCORE II to predict in hospital mortality after elective cardiac surgery. This result confirms the growing interest of Machine Learning in establishing predictive models in medicine.
Complete list of metadatas

Cited literature [102 references]  Display  Hide  Download

https://dumas.ccsd.cnrs.fr/dumas-02437224
Contributor : Médecine Ub <>
Submitted on : Monday, January 13, 2020 - 4:12:37 PM
Last modification on : Saturday, January 18, 2020 - 1:30:22 AM
Long-term archiving on: : Tuesday, April 14, 2020 - 6:31:14 PM

File

Med_spe_2017_Belghiti_Alaoui.p...
Files produced by the author(s)

Identifiers

  • HAL Id : dumas-02437224, version 1

Collections

Citation

Myriem Belghiti Alaoui. Prédiction de la mortalité après chirurgie cardiaque programmée : nouvelles approches par Machine Learning et Decision Curve Analysis. Sciences du Vivant [q-bio]. 2017. ⟨dumas-02437224⟩

Share

Metrics

Record views

38

Files downloads

98