Predicting risk of mortality in patients undergoing cardiovascular surgery
Author(s)
Advisor
Güvenir, Halil AltayDate
2008Publisher
Bilkent University
Language
English
Type
ThesisItem Usage Stats
164
views
views
53
downloads
downloads
Abstract
It is very important to inform the patients and their relatives about the risk
of mortality before a cardiovascular operation. For this respect, a model called
EuroSCORE (The European System for Cardiac Operative Risk Evaluation) has
been developed by European cardiovascular surgeons. This system gives the risk
of mortality during or 30 days after the operation, based on the values of some
parameters measured before the operation. The model used by EuroSCORE
has been developed by statistical data gathered from large number of operations
performed in Europe.
Even though due to the surgical techniques that have been developed recently
and the risk of mortality has been reduced in a large extent, predicting that
risk as accurately as possible is still primary concern for the patients and their
relatives in cardiovascular operations. The risk of operation also essentially tells
the surgeon how a patient with similar comorbidity would be expected to fare
based on a standard care. The risk of patient is also important for the health
insurance companies, both public or private. In the context of this project, a
model that can be used for mortality is developed.
In this research project, a database system for storing data about cardiovascular
operations performed in Turkish hospitals, a web application for gathering
data, and a machine learning system on this database to learn a risk model,
similar to EuroSCORE, are developed. This thesis proposes a risk estimation
system for predicting the risk of mortality in patients undergoing cardiovascular
operations by maximizing the Area under the Receiver Operating Characteristic
(ROC) Curve (AUC).
When the genetic characteristics and life styles of Turkish patients are taken into consideration, it is highly probable that the mortality risks of Turkish patients
may be different than European patients. This thesis also intends to investigate
this issue.