Int J Biomed Comput, 21 (1991) 175-192 Elsevier Scientific Publishers Ireland Ltd. 175 zyxwvutsrqp EVALUATION OF RECEIVER OPERATOR CHARACTERISTICS - OPTIMUM TIME OF DAY FOR THE ASSESSMENT OF HEART RATE VARIABILITY AFTER ACUTE MYOCARDIAL INFARCTION MAREK MALIK, THOMAS FARRELL, JAN POLONIECKI and A. JOHN CAMM Department of Cardiological Sciences, St. George’s Hospital Medical School, London (U.K.) (Received June 8th, 1990) (Revision received October 15th, 1990) (Accepted October 31st, 1990) The study reports a mathematical method for deciding which clinical data are of importance for treat- ment decisions in a given clinical setting. The method comprises the following steps: (A) the receiver operator characteristic (ROC) functions of the compared sets of data are computed; (B) the design and aim of the clinical study is expressed as an integral measure on the space of sensitivity values (this reflects the preference of low or high sensitivity values dependent on the clinical targets); (C) the sets of data being compared are characterised by the non-linear integrals of their ROC functions. The approach has been used to compare mean heart rate (HR) and heart rate variability (HRV) data calculated in 5113 different portions of 24-h ECG recordings and assessed in 365 patients surviving acute myocardial infarction, in order to evaluate the utility of Holter recording of varying lengths and starting times for the prediction of sudden cardiac death and/or serious arrhythmic events. The results of the study show that this ap- proach is capable of evaluating and comparing the sets of medical data used for identification of patients who are at increased risk. The experimental part of the study showed that the optimum recording interval for the assessment of HR and HRV data in patients who survived acute myocardial infarction depends on the aim of the identification of increased risk patients. The optimum interval of recording is different for an identification which requires a low number of false negative cases and permits a higher number of false positive cases, than for the situation where a low number of false positive cases are required and a higher number of false negative cases are permissible. Keywords: Risk stratification; Long-term electrocardiograms 1. Introduction In many medical disciplines, the risk of patients deteriorating or developing com- plications can be stratified according to the results of different investigations. It is important to evaluate which investigations are optimum for such a stratification and which values of results of the investigations should be used to identify the patients at higher risk. In practical terms, the power and usefulness of stratification methods are usually expressed in terms of sensitivity, specificity, positive and negative predic- tive accuracy, etc. However, these characteristics depend both on the ap- propriateness of the investigations employed, and on the critical values of the results used to define the physiological and pathological conditions. Correspondence IO: Prof. Marek Malik, Department of Cardiological Sciences, St. George’s Hospital Medical School, Cranmer Terrace, London SW17 ORE, U.K. 0020-7101/91/%03.50 0 1991 Elsevier Scientific Publishers Ireland Ltd. Published and Printed in Ireland