Predicting Serious Outcomes in Syncope Patients Using Data Mining Techniques

Abstract

Syncope or fainting refers to a temporary loss of consciousness usually related to insufficient blood flow to the brain and can be due to several causes, which are either simple or serious conditions. Syncope can be caused by life-threatening conditions not evident in the first evaluations, which can lead to serious outcomes, including death, after discharge from the hospital. We have developed a decision tool to identify syncope patients with 18 years of age or higher who are at risk of a serious event within 30 days after discharge from the hospital.We used the data provided by the Tehran Heart Clinic. In this dataset adults with 18 years old or above with syncope signs are enrolled. The patients presented themselves within 24 hours after the event to the THC. Standardized variables from clinical evaluation and investigations have been collected. Serious adverse events included death, Intracerebral hemorrhage (ICH) or Subarachnoid hemorrhage (SAH), Cerebrovascular accident (CVA), Device Implantation, myocardial infarction, arrhythmia, traumatic syncope or cardiac surgery within 30 days. 356 patients were enrolled with syncope; the mean age was 44.5 years and 53.6% were women. Serious events occurred among 26 (7.3%) of the patients within 30 days of discharge from the hospital.Different machine learning algorithms such as Decision Tree, SMO, Neural Networks, Naïve Bayes and Random Forest have been used on the dataset to predict patients with serious adverse outcomes and the WEKA program has been used to validate the results.Results show that when using Random Forrest Algorithm, the accuracy rate and ROC Area reached 91.09% and 0.90. However, previous statistical risk scores such as the San Francisco Score resulted in lower ROC Area readings.