A method is described for the assessment of the complexity of short data sets by nonlinear dynamics. The method was devised for and tested on human heart rate recordings approximately 2000 to 9000 RR intervals long which were extracted from the memory of implantable defibrillator devices (ICD). It is, however, applicable in a more general context. The ICDs are meant to control life-threatening episodes of ventricular tachycardia and/or ventricular fibrillation by applying a electric shock to the heart through intracardiac electrodes. It is well known that conventional ICD algorithms yield approximately 20--30 \% of spurious interventions. The main aim of this work is to look for nonlinear dynamics methods to enhance the appropriateness of the ICD intervention. We first showed that nonlinear dynamics methods first applied to 24-hour heart rate variability analysis were able to detect the need for the ICD intervention. To be applicable to future ICD use, the methods must also be low in computational requirements. Methods to analyse the complexity of the short and non-stationary sets were devised. We calculated the Shannon entropy of symbolic words obtained in a sliding 50 beat window and analysed the dependence of this complexity measure on the time. Precursors were found extending much earlier time than the time the standard ICD algorithms span.

PACS numbers: 05.45.--a, 05.45.Tp, 87.19.Hh, 05.10.--a