Session S61.1

Increasing Patient Safety in Drug Trials with Computer Based ECG Analysis and Classification: A Study with 13,000 Resting ECGs

TK Zywietz*, A Khawaja, R Petrovic, R Fischer, J Reinstaedtler

Institute for Biosignal-Processing
Munich, Germany

Today, the pharma industry relies on ECG recordings (digital Holter or 12 lead resting ECG) to ensure cardiac safety of subjects in drug trials. State of the art is a central ECG analysis in ECG core labs, where the ECGs are analyzed fully manual by specially trained technicians or experienced cardiologists. The ECGs are commonly analyzed within 48 hours after upload to the core lab. In a worst case scenario this means a patient has to wait two days to be withdrawn of a trial if pathologic signs are developed. In order to reduce the response time in case of critical ECGs, we have developed a 12 lead resting ECG analysis algorithm to quantitatively classify the ECGs (according to typical criteria applied by pharma sponsors). As bottom line statement, the ECGs are classified into ”normal” (N), “abnormal clinical non significant” (ACNS), and “abnormal clinical significant” (ACS). The classification is based on the Hannover ECG System HES, where approximately 1200 parameters are calculated per 12 lead resting ECG. To derive the bottom line statement, we apply the measurement results of a representative beat such as P, PQ, QRS, QT and STT, as well as the rhythm analysis and a detailed morphologic QRST evaluation. The algorithm was tested and validated against a data base of ~13.000 ECGs of 700 patients, recorded in a drug trial and analyzed in a professional core lab. The data base contained 296 records classified as ACS, 1897 as ACNS and the rest classified as normal. We were able to detect the ECGs classified as ACS with a sensitivity of 98, 7%. Four ECGs were falsely classified as “normal” – the HES code failed to detect very slight signs of a preexitation syndrome. We also compared the quantitative results of the manual and automatic measurements. The results clearly show that the variability of the measurements is significantly reduced by the HES analysis. For example, the standard deviation of QTc Bazett over the patient population is 20, 8 ms (HES analysis) versus 50, 8 ms (manual measurements). In conclusion, by identifying all ECGs, which might be critical, in real time, patient safety can be increased by reducing the response time in drug trials using computer based analysis.

(Abstract Control Number: 235)