A Data Mining Approach for Predicting Delirium After Cardiac Surgery
Date
2014-12-19
Authors
Mufti, Hani N.
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Of particular concern to patients is the effect of surgery upon brain functions following a surgical intervention. Indeed, post-operative neurocognitive complications occur in up to 60% of patients. Delirium is a temporary disturbance of consciousness, attention, cognition, and/or perception, which occurs frequently among hospitalized patients. It develops over a short period and tends to fluctuate. Several risk factors predispose patients to post-operative delirium. Patients who experience delirium after cardiac surgery are at higher risk of multiple adverse outcomes (e.g.: infections). Identification of patients at risk will allow targeted personalized preventive strategies. This thesis demonstrates the development of several predictive models, using a data mining approach, to predict the development of delirium in patients undergoing cardiac surgery. The findings of this research suggest that, applying machine learning and data mining techniques on complex and imbalanced medical data is capable of achieving superior results in comparison to standard statistical approaches.
Description
Of particular concern to patients is the effect of surgery upon brain functions following a surgical intervention. Indeed, post-operative neurocognitive complications occur in up to 60% of patients. These include: stroke, seizures, and delirium. Delirium is a temporary disturbance of consciousness, attention, cognition, and/or perception, which occurs frequently among hospitalized patients. It develops over a short period and tends to fluctuate. Several risk factors predispose patients to post-operative delirium, including: medications, age, male gender, major surgery (cardiac and orthopedic), and others. Delirium occurs relatively frequently (10% to 15%) among patients who undergo cardiac surgery. Patients who experience delirium after cardiac surgeries are at higher risk of multiple adverse outcomes (e.g.: infections, prolonged hospitalization, and death). Identification of patients at risk will allow targeted personalized preventive strategies that might improve the patient transition through the process of care. This thesis demonstrates the development of several predictive models, using a data mining approach, to predict the development of delirium in patients undergoing cardiac surgery. The developed models were derived from a large contemporary registry, and their performance was evaluated on an independent dataset. This work also addresses the issue of class imbalance and its effect on model performance. The findings of this research suggest that, applying machine learning and data mining techniques on complex medical data is capable of achieving superior results in comparison to standard statistical approaches. With increased adaptation of electronic health records, data mining techniques offer novel approaches to aid in the prediction of complex relationships, a typical property of adverse medical events. These models will aid the recovery of high-risk patients by enabling a more proactive approach, initiating preventive measures in a timely fashion.
Keywords
Data mining, Predictive Analytics, Delirium, Cardiac Surgery, Postoperative Complications, Class Imbalance