Session SB2.3

Early Detection of Decompensation Conditions in Heart Failure Patients by Knowledge Discovery: The HEARTFAID Approaches

A Candelieri*, D Conforti, F Perticone, A Sciacqua,
K Kawecka-Jaszcz, K Styczkiewicz

Università di Calabria
Cosenza, Italy

The EU FP6 Project HEARTFAID (www.heartfaid.org) aims to develop an innovative knowledge based platform of services for more effective and efficient clinical management of heart failure within elderly population. Under this respect, in this paper we present a specific Knowledge Discovery task which has been implemented as a service of the HEARTFAID platform. For the clinical management of chronic heart failure (CHF) patient, a crucial mid-long term goal is the early detection of new acute decompensation events. A well tuned and personalized therapy, high quality outcomes and reduction of the health care costs may be achieved if patient decompensation is early identified and appropriately tackled. Within the relevant clinical protocols and guidelines, a general consensus has not been reached about the definition and assessment of criteria on how to predict when a patient will further decompensate, even though many different evidence-based indications are known. Knowledge Discovery approaches may be a practical and effective solution in order to extract new potentially useful models about this clinical problem from repositories of pertinent clinical data. To this end, we collected relevant data from 49 CHF patients, each of them visited by the cardiologists every two weeks, for a total of 333 visits. During each visit, the cardiologist assessed patient health status: a new decompensation event was occurred or no decompensation event was occurred. The intelligent data analysis was based on both classical parameters taken from the clinical guidelines and from the evidence-based clinical knowledge. Monitored parameters during each visit were: Systolic Blood Pressure, Heart Rate, Respiratory Rate, Body Temperature, Body Weight, Body Water. Also other information about each patient was taken into account for the analysis: gender, age, NYHA class, alcohol use and smoking. The entire set of parameters was used as input of different Knowledge Discovery algorithms: Decision Trees, Decision Lists, Support Vector Machines and Radial Basis Function Networks. We obtained different binary classifiers performing the early detection of acute decompensation events, where the results of some of these models are “easy-to-understand” (Decision Trees and Decision Lists) and their consistency was directly evaluated by the cardiologists. Moreover, high percentage of correct classifications (above 87%) was obtained by using suitable validation approaches.

(Abstract Control Number: 313)