Symbolic Analysis of 24h Holter Heart Period Variability Series: Comparison between Normal and Heart Failure Patients A Porta 1 , G D'Addio 2 , GD Pinna 3 , R Maestri 3 , T Gnecchi-Ruscone 4 , R Furlan 4 , N Montano 4 , S Guzzetti 4 , A Malliani 4 1 DiSP LITA di Vialba, Universita' di Milano, Milan, Italy 2 S Maugeri Foundation, IRCCS, Rehabilitation Institute of Telese, Italy 3 S Maugeri Foundation, IRCCS, Rehabilitation Institute of Montescano, Italy 4 DiSC “Luigi Sacco”, Universita’ degli Studi di Milano, Milano, Italy Abstract This study proposes an application of symbolic analysis to beat-to-beat heart rate variability data derived from 24h Holter recordings both in healthy and heart failure populations. Heart rate series are transformed in a sequence of six symbols via a uniform quantization procedure. Symbols are grouped in patterns lasting three cardiac cycles and these patterns are grouped in few (four) families without any loss. The rate of occurrence of these families are found useful to distinguish two physiological conditions characterised by a different status of the autonomic nervous system (i.e. day-time and nigh-time) in healthy subjects and to discriminate between healthy and pathological populations. Indexes derived from symbolic analysis deserve to be added to traditional time and frequency domain parameters in standard analysis of heart rate variability obtained from 24h Holter recordings and tested over larger databases. 1. Introduction Symbolic analysis applied to heart rate variability is providing new parameters independent of those derived from time and frequency domains and helpful in interpreting the encoded physiological information [1-4]. Voss et al [1] demonstrated that the addition of indexes originated from symbolic analysis to traditional time and frequency domain indexes improves the accuracy of risk stratification after myocardial infarction. Wessel et al [2] found that parameters from symbolic analysis discriminate chronic heart failure patients with high risk to develop life-threatening arrhythmias from those with marginal risk, while time and frequency domain parameters cannot separate the two populations. Guzzetti et al [3] showed that symbolic analysis in healthy subjects is sensible to the activation of the autonomic nervous system as much as power spectral analysis with the additional possibility to permit the detection of concomitant activation of sympathetic and parasympathetic autonomic nervous systems. The aim of this study is to propose an application of symbolic analysis to heart rate variability series obtained from standard 24h Holter recordings. In this application we have considered healthy subjects and heart failure patients to verify whether symbolic analysis can detect in normal population the well-known differences between day-time and night-time and can discriminate the two populations. The core of the approach lies in the application of the previously published symbolic analysis method [4] searching for the rates of occurrence of specific short patterns (i.e. sequences of symbols lasting 3 cardiac cycles). This approach, designed to reliably work on short beat-to-beat series (about 300 beats), is iteratively applied to 24h Holter heart rate variability recordings. 2. Symbolic analysis Given the series x={x(i), i=1,…,N}, where i is the progressive sample counter and N is the series length, x is transformed in a sequence of symbols using a coarse graining approach based on a uniform quantization procedure [4]. The full range of the series is spread over ξ symbols with a resolution of (x max -x min )/ξ, where x max and x min are the maximum and the minimum of the series. After quantization x becomes a sequence x ξ ={x ξ (i), i=1,…,N} of integer values ranging from 0 to ξ-1. The series x ξ is transformed in a sequence of patterns of L delayed samples, x ξ,L ={x ξ,L (i), i=L,…,N} with x ξ,L (i)=(x ξ (i),x ξ (i-1),…, x ξ (i-L+1)) and the number of possible x ξ,L (i) is ξ L . The rate of occurrence of each x ξ,L (i) can be calculated by dividing the number of times that x ξ,L (i) is found in x ξ by ξ L (i.e. the sample frequency of x ξ,L (i)). As ξ L grows very rapidly with L and ξ, both parameters have to be small, thus limiting the length of 0276-6547/05 $20.00 © 2005 IEEE 575 Computers in Cardiology 2005;32:575-578.