Search

Browse

• Communities
• Authors
• Year
• Titles
• Subjects

My Space

• Login

Data mining from medical databases and machine learning based mortality prediction for venous thromboembolism, myocardial infarction and ischemic stroke

Nikolakakis Stylianos

Full record

---

Summary

Intensive care unit (ICU) patients with arterial or venous thrombosis suffer from high mortality rates. Mortality prediction in the ICU has been a major medical challenge for which several scoring systems exist but lack in specificity. This study focuses on three target groups, namely patients with thrombosis, ischemic stroke or myocardial infraction. The main goal is to develop and validate interpretable machine learning (ML) models to predict mortality, while exploiting all available data stored in the medical record. To this end, retrospective data from one freely accessible database, eICU, were used. Well-established ML algorithms were implemented utilizing automated and purposely built ML frameworks for addressing class imbalance. Prediction of early mortality showed excellent performance in all disease categories, in terms of the area under the receiver operating characteristic curve (AUC–ROC ): VTE 0.87, myocardial infraction 0.95, ischemic stroke 0.90. The predictive model of mortality developed from 4,385 VTE patients ended up with a signature of 475 features, 10,543 patients with myocardial infarction using 317 features and 4,326 patients with ischemic testing on 338 features. Our model outperformed traditional scoring systems in predicting mortality.

Available Files

Services