1: %\documentclass[preprint,superscriptaddress,showpacs,preprintnumbers,amsmath,amssymb,aps,pre]{revtex4}
2: \documentclass[twocolumn,superscriptaddress,showpacs,preprintnumbers,amsmath,amssymb,aps,pre]{revtex4}
3: \usepackage{graphicx}% Include figure files
4: \usepackage{dcolumn}% Align table columns on decimal point
5: \usepackage{bm}% bold math
6: \begin{document}
7: \title{Natural entropy fluctuations discriminate similar looking electric signals emitted from systems of different dynamics\footnote{Published in Physical Review E {\bf71}, 011110 (2005).}}
8: \author{P. A. Varotsos}
9: \email{pvaro@otenet.gr}
10: \affiliation{Solid State Section, Physics Department, University of Athens, Panepistimiopolis, Zografos 157 84, Athens, Greece}
11: \affiliation{Solid Earth Physics Institute, Physics Department, University of Athens, Panepistimiopolis, Zografos 157 84, Athens, Greece}
12: \author{N. V. Sarlis}
13: \affiliation{Solid State Section, Physics Department, University of Athens, Panepistimiopolis, Zografos 157 84, Athens, Greece}
14: \author{E. S. Skordas}
15: \affiliation{Solid Earth Physics Institute, Physics Department, University of Athens, Panepistimiopolis, Zografos 157 84, Athens, Greece}
16: \author{M. S. Lazaridou}
17: \affiliation{Solid State Section, Physics Department, University of Athens, Panepistimiopolis, Zografos 157 84, Athens, Greece}
18: \begin{abstract}
19: Complexity measures are introduced, that quantify the change of the natural
20: entropy fluctuations at different length scales in time-series emitted from systems operating far from
21: equilibrium. They identify impending sudden cardiac death (SD) by analyzing fifteen minutes
22: electrocardiograms, and comparing to those of truly healthy humans (H). These measures seem to be
23: complementary to the ones suggested recently [Phys. Rev. E {\bf 70}, 011106 (2004)]
24: and altogether enable the classification of individuals into three
25: categories: H, heart disease patients and SD. All the SD individuals,
26: who exhibit critical dynamics, result in a common
27: behavior.
28: \end{abstract}
29: \pacs{05.40.-a, 91.30.Dk, 05.45.Tp, 87.19.Nn}
30: \maketitle
31:
32:
33: \section{Introduction}
34: The problem of distinguishing electric signals which, although look to be similar, they are
35: emitted from systems of different dynamics, still attracts a strong interest.
36: Two characteristic cases of major practical importance are: First, Seismic Electric
37: Signals (SES) activities, which are low frequency ($\leq$1 Hz) signals of dichotomous
38: nature that have been found in Greece\cite{VAR86A,VAR86B,VAR03x} and Japan\cite{UYE02} to precede earthquakes, may look to be similar
39: to ``artificial'' noises (AN), which are electrical disturbances emitted
40: from nearby man-made sources. It has been argued\cite{VAR86A,VAR02,VAR03x} that
41: SES activities are emitted when the stress reaches a {\em critical} value in the EQ focal area.
42: Second, sudden cardiac death (SD), which is the primary cause of mortality in
43: the industrialized world\cite{BRA03}, may occur even if the electrocardiogram
44: (ECG) looks to be similar to that of truly healthy (H) humans. Sudden cardiac arrest may also be considered as a
45: dynamic phase transition (critical phenomenon)\cite{VAR03y,ref8}.
46:
47: Both cases have been treated in Ref.\cite{ref8}, but here we only focus on the second one.
48: The time-series will be analyzed in the natural time-domain.
49: The natural time $\chi$ is introduced\cite{VAR01,VAR02} by ascribing to the $m$-th
50: pulse of an electric signal consisting of $N$ pulses, the value $\chi_m=m/N$
51: and the analysis is made in terms of the couple $(\chi_m, Q_m)$, where $Q_m$
52: denotes the duration of the $m$-th pulse. The entropy $S$ in the natural
53: time-domain\cite{VAR01,VAR03} is defined as
54: $S=\langle \chi \ln \chi \rangle - \langle \chi \rangle \ln \langle \chi \rangle$,
55: where $\langle \chi \ \ln \chi \rangle=\sum_{k=1}^N p_k \chi_k \ln \chi_k$ , $\langle \chi \rangle=\sum_{k=1}^N p_k \chi_k$
56: and $p_k=Q_k/\sum_{n=1}^N Q_n$. It is {\em dynamic} entropy depending on the {\em sequential}
57: order of pulses\cite{ref8}.
58: Here we calculate the value of $S$ for a number of consecutive pulses and study
59: how it varies within the recording (i.e., using a time-window of certain length $N_w$
60: sliding, each time by one pulse, through the whole time-series). Thus, for a window of length $N_w$,
61: when starting from the $m_0$-th pulse, we have
62: $
63: S(m_0,N_w)=\langle \chi \ln \chi \rangle_w - \langle \chi \rangle_w \ln \langle \chi \rangle_w,
64: $
65: where $\langle \chi \ln \chi \rangle_w=\sum_{k=1}^{N_w} p_{k,w} \chi_{k,w} \ln \chi_{k,w}$ ,
66: $\langle \chi \rangle_w=\sum_{k=1}^{N_w} p_{k,w} \chi_{k,w}$
67: with $p_{k,w}=Q_{m_0-1+k}/\sum_{n=1}^{N_w} Q_{m_0-1+n}$ and $\chi_{k,w}=k/N_w$.
68: This variation is quantified by the standard deviation $\delta S$ ($= \delta S_{N_w}$) of
69: $\{ S(m_0,N_w), m_0=1,2, \ldots N-N_w \}$. The value of $\delta S$ may change to
70: a different value $\delta S_{shuf}$ when repeating the same calculation but after {\em shuffling}
71: the $Q_m$ randomly. In Ref.\cite{ref8} we showed that a distinction between
72: SD and H can be achieved when calculating both $\delta S_{shuf}$ and $\delta S$
73: at the {\em same} (time-window) length $N_w$ and then studying their ratio
74: $\delta S_{shuf}/ \delta S$(which is labeled by $\nu$).
75: Here we show that a similar distinction may be alternatively achieved
76: if we introduce {\em appropriate}
77: measures that quantify the $\delta S$-variability upon {\em changing} the time-window length
78: and, interestingly, their
79: values approach the value of the Markovian case in SD, who exhibit critical
80: dynamics. Furthermore, we show that the measures
81: suggested in this paper exhibit a certain type of complementarity when compared to
82: those discussed in \cite{ref8}.
83:
84: \begin{figure}
85: \includegraphics{Figure1}% Here is how to import EPS art
86: \caption{\label{fig1} (Color) (a) Schematic diagram (not in scale)
87: of a three heartbeat excerpt of an ECG in the usual (conventional)
88: time-domain. Only the durations $Q_m$,$Q_{m+1}$,$Q_{m+2}$ of the
89: QT-interval (marked in each single cycle of the ECG corresponding
90: to one heartbeat) are shown. (b) The QT-interval time-series of
91: (a) read in natural time; the vertical bars are {\em equally}
92: spaced, but the length of each bar denotes the duration of the
93: corresponding QT-interval marked in (a).}
94: \end{figure}
95:
96:
97:
98: In ECG, the turning points are traditionally labeled with the letters Q, R, S, T,
99: see Fig.\ref{fig1}(a). (cf. In Fig. \ref{fig1}(b) we show,
100: for example, how the QT interval time-series can be read in natural time).
101: The RR- (beat-to-beat) and QRS-intervals (cf. mainly the RR)
102: can be automatically detected\cite{LAG90,LAG94,LAG97,JAN97} (which was followed here)
103: more easily than the QT-. In spite of this fact, we intentionally study here all these
104: three types of intervals for the following reasons: It has been clinically observed that the
105: QT interval usually exhibits prolonged values before cardiac death (see Ref.\cite{KHA02} and references therein).
106: Interestingly, this clinical observation was found\cite{ref8} to be consistent with the fact that
107: in all SD the $\delta S$- (and $\delta S_{shuf}$-) values themselves of the QT-intervals
108: exceed those of H, see Fig. \ref{fig2} (cf. the latter distinction between SD and H cannot be attributed
109: to the allocation error of the QT interval, see Section VIII of Ref.\cite{epaps}).
110: Since the latter systematic behavior is not found when
111: studying the RR- or the QRS- intervals\cite{ref8}, it is interesting to investigate
112: here whether a systematicity occurs when employing the complexity measures suggested
113: in this paper. Actually, we find that the latter measures seem to enable the
114: distinction between SD and H when using the RR- and QRS- intervals of the original time-series.
115: Furthermore, and most interestingly, we pinpoint that, even when solely using the
116: most easily accessible values of the RR-intervals, such a distinction seems to
117: be possible if we apply these measures to both the original time-series and the
118: one obtained after shuffling the $Q_m$ randomly. We use here the QT Database
119: from physiobank\cite{GOL00}, which includes fifteen minutes recordings of
120: 10 H and 24 SD (as well as recordings from four groups of heart disease patients,
121: see below). Examples of the $\delta S$-values, calculated for the RR-, QRS- and QT-intervals
122: in the range 3 to 100 beats are plotted in Figs. \ref{fig3}(a) and (b) for one H and one SD,
123: respectively. As for the symbols, we use the same convention as in Ref.\cite{ref8}, i.e.,
124: $\delta S$ is used only when the calculation is made by a single time-window (e.g., 5 pulses),
125: while the symbol $\overline{\delta S}$ stands for the average of the $\delta S$-values
126: calculated for a sequence of single windows (e.g., 3, 4 pulses).
127: Finally, $\langle \delta S \rangle$ denotes the $\delta S$-values averaged over a
128: group of individuals, e.g., 10 healthy subjects.
129:
130:
131: Before proceeding, however, it might be useful to
132: recapitulate the main differences of our procedure
133: compared to several other earlier attempts by other groups.
134: The reasons why the concept of entropy should be
135: preferred (compared to other quantities) as
136: discriminating statistics in physiological time-series have been explained in detail in Ref.\cite{ref8}.
137: Furthermore, the advantages of using complexity measures based on
138: {\em dynamic} entropy
139: (and not on {\em static} entropy, e.g., Shannon entropy), as for example the Kolmogorov-Sinai
140: entropy (K-S entropy), have been clarified\cite{ref8}. Earlier attempts in the ECG analysis
141: have actually used measures related to dynamic entropy. For example, the so called
142: approximate entropy (AE)\cite{PIN91} or sample entropy (SE)\cite{RIC00}
143: have been introduced and later used by other authors (e.g., see Ref. \cite{NIK03}
144: where AE is applied beyond other measures; see also Ref.\cite{epaps}). Also,
145: Costa et al.\cite{COS02}
146: introduced the multiscale entropy (MSE) approach,
147: the algorithm of which is based on AE or SE, calculating the entropy
148: at different scales. As for the $S$, which is also a dynamic entropy,
149: as already mentioned, differs essentially from the other ones, because it is defined\cite{VAR01,VAR03}
150: in an entirely different time-domain (see Fig.\ref{fig1}(b)). Moreover, the following
151: has been found: When studying the $S$-values themselves, most SES activities
152: can be clearly distinguished\cite{VAR03} from the majority of AN,
153: because they have S-values smaller and larger, respectively, than the value $S_u=0.0966$
154: of the ``uniform''
155: distribution (as the latter was defined in Refs.\cite{VAR03feb,VAR03});
156: on the other hand, when dealing with ECG they all have $S$-values comparable, more or less,
157: to $S_u$ \cite{ref8}, see also \cite{epaps}, thus not allowing a
158: clear distinction among their principal
159: categories (cf. the entropy values themselves have been used in earlier attempts).
160: This is achieved, however, when we quantify the $S$ {\em fluctuations}\cite{ref8} and use
161: ratios of ``shuffled'' and ``unshuffled'' $S$ fluctuations on fixed time scales\cite{ref8}
162: or ratios on different time scales that will be introduced here in Section II. Thus, in order
163: to discriminate similar looking electric signals emitted from systems of different dynamics,
164: the following seems to hold: signals that have $S$-values more or less comparable to $S_u$
165: (which is the case of all ECG) can be better classified by the
166: complexity measures relevant to the fluctuations $\delta S$ of the
167: entropy; if the $S$-values {\em markedly} differ from $S_u$ (which
168: is usually -but {\em not} always- the case of SES and AN), the
169: classification of these signals should be preferably made by the
170: use of the $S$-values themselves.
171:
172:
173:
174: \section{THE NEW COMPLEXITY MEASURES PROPOSED. THE DISTINCTION BETWEEN SD AND H}
175:
176:
177: In classical Thermodynamics, the systems are studied close to equilibrium
178: and the relevant quantities have a clear physical meaning. In non-equilibrium systems,
179: however, the meaning of entropy and its treatment should be handled with
180: great caution (e.g., \cite{VAR86A}), because there is at present
181: (e.g., see Ref. \cite{BLY04}) no unified statistical mechanical theory underlying
182: these systems. (cf. In transformations between non-equilibrium stationary states,
183: entropy might be a not well defined concept\cite{GAL03}; the connection of the entropy to
184: microscopic dynamics is still a matter of intensive research (e.g.,\cite{TSA03} and references therein)).
185: In complex systems operating far from equilibrium (like the case
186: of heart dynamics\cite{GOL02}), long-range correlations
187: play an important role (cf. such correlations are, of course, of
188: prominent importance in equilibrium systems as well,
189: when approaching a critical point, e.g., the ``critical''
190: temperature $T_c$, i.e., $T \rightarrow T_c$).
191: Thus, in the latter
192: systems {\em both} correlations (i.e., short- and long-range), in general, is advisable to
193: be studied carefully and hence appropriate complexity measures should be envisaged. This
194: is, in simple words, the physics underlying the present paper and stimulated the procedure
195: followed.
196:
197: Along these lines, we introduce the ratios
198: $ \delta S_i(RR)/ \delta S_j(RR)$, $\delta S_i(QRS)/ \delta S_j(QRS)$ and
199: $\delta S_i(QT)/ \delta S_j(QT)$ for the RR-, QRS- and QT-intervals, respectively,
200: where $i,j$ denote the time-window length used in the calculation of $\delta S$. Assuming
201: that $j<i$, these three ratios provide measures of the $\delta S$- variability when a
202: scale $i$ changes to a scale $j$. We select as a common scale (for all RR-, QRS- and QT-)
203: the {\em smallest} $j$-value allowed for the natural time-domain analysis, i.e., $j=3$
204: beats, and for the short range (s) $i=5$, while for the longer (L) $i=60$ beats.
205: Thus, the following ratios are
206: studied: $\lambda_s(\tau)\equiv \delta S_5(\tau )/ \delta S_3 (\tau )$ and $
207: \lambda_L(\tau)\equiv \delta S_{60} (\tau ) / \delta S_3(\tau )$,
208: where $\tau$ denotes the type of interval, i.e., $\tau=$RR, QRS or QT.
209: We also define the ratios $\rho_i(\tau )=\delta S_i(RR)/ \delta S_i (\tau )$,
210: which provide a {\em relative} measure of the $\delta S$-values of the RR-intervals
211: compared to either QRS- or QT- (for the {\em same} number of beats $i$). Here,
212: we will use for the short range $\rho_s(\tau ) \equiv \rho_3(\tau )$ and
213: for the long range $\rho_L(\tau)\equiv \rho_{60}(\tau )$.
214:
215: The calculated values for the complexity measures $\lambda_\kappa, \rho_\kappa$ (where $\kappa$ denotes
216: either the short, $\kappa=s$, or the longer, $\kappa=L$, range) are given, for all H and SD,
217: in Table \ref{tab1}. The minima $\min_{H}[\lambda_\kappa(\tau)]$ and maxima
218: $\max_{H}[\lambda_\kappa(\tau)]$ among the healthy individuals
219: for the RR ($\tau=$RR) and QRS ($\tau=$QRS) intervals
220: are also inserted in this Table.
221: We also include the corresponding minima $\min_{H}[\rho_\kappa(\tau)]$
222: and maxima $\max_{H}[\rho_\kappa(\tau)]$ for (the relative
223: $\delta S$-variability measure) $\rho$. For the sake of simplicity,
224: they are labelled $H_{min}$ and $H_{max}$, respectively
225: (and jointly named {\em H-limits}).
226: The superscripts `a' and `b' show the
227: cases of SD which have smaller and larger values than $H_{min}$ and $H_{max}$,
228: respectively. In two individuals, i.e., sel41 and sel51, it is uncertain whether their
229: measure $\lambda_s(QRS)$ violates the value
230: $H_{min}$=1.15.
231:
232:
233: \begin{figure}
234: \includegraphics{Figure2}% Here is how to import EPS art
235: \caption{\label{fig2} (color) (a) The $\delta S(QT)$-value for
236: each of the 24 SD and 10 H (see Table \ref{tab1}) and (b) the
237: average of the $\delta S$(QT) values -designated by $\langle
238: \delta S (QT) \rangle$- along with their standard error deviation
239: for each of the two groups SD and H versus the time-window
240: length.}
241: \end{figure}
242:
243: \begin{figure}
244: \includegraphics{Figure3}% Here is how to import EPS art
245: \caption{\label{fig3} (Color) The $\delta S$-value versus the time-window length for one H
246: (a) and one SD (b). Intervals: QT (solid red), QRS (broken green), and RR (dotted blue).}
247: \end{figure}
248:
249: \begin{figure}
250: \includegraphics{Figure4}% Here is how to import EPS art
251: \caption{\label{fig4} (Color) The average (denoted by the
252: brackets) values of (a): the
253: $\delta S (RR)$ and (b): $\delta S (RR)/ \delta S_{3} (RR)$ for the
254: SD (solid black) and H (red circles) versus the time-window length;
255: the bars correspond to the standard error of the mean.
256: The results for a Markovian time-series are also plotted (green squares),
257: but the bars here denote the standard deviation. }
258: \end{figure}
259:
260:
261: \begin{figure}
262: \includegraphics{Figure5}% Here is how to import EPS art
263: \caption{\label{fig5} (color) The average of the $\delta S(QT)$
264: values -labelled $\langle \delta S (QT) \rangle$- for each of the
265: six groups labelled H, MIT, MSV, MST, EST and SD (see the text)
266: versus the time-window length. The bars denote the standard error
267: of the mean. (The corresponding standard deviations overlap
268: considerably and hence are not shown for the sake of clarity). The
269: lowermost- and the uppermost- curve correspond to H and SD,
270: respectively and hence coincide with the two curves depicted in
271: Fig. \ref{fig2}(b).}
272: \end{figure}
273:
274:
275: Table \ref{tab1} reveals that
276: {\em all} SD violate one or more $H-$limits of $\lambda_s(RR)$, $\lambda_L(RR)$, $\rho_s(QRS)$ and $\rho_L(QRS)$,
277: and hence can be distinguished from H.
278: In other words, the $\delta S$-variability measures of the RR-intervals, together
279: with their relative ones with respect to the QRS- (i.e., four parameters in total), seem to achieve a
280: distinction between SD and H. Note that $\lambda_\kappa(RR)$ {\em
281: alone } can classify the vast majority of SD. Furthermore,
282: attention is drawn to the point that if we also consider the
283: $\lambda_\kappa (\tau )$ values calculated ( {\em not } in the
284: original, but) in the randomized (``shuffled'') sequence of $Q_m$,
285: %which
286: % are designated in Table \ref{tab1} by the subscript ``$sh$'' in the symbols used above,
287: we find that {\em all} SD violate one or more {\em H-limits} of $\lambda_\kappa ( RR )$ and
288: $\lambda_{\kappa,shuf} ( RR )$ (see Table VII of Ref.\cite{epaps}). This allows (cf. using again four
289: parameters in total) the distinction of SD from H by using the RR-intervals {\em only}.
290:
291: Thus, we found that among the 10 parameters defined in the original time-series extracted
292: from each ECG (or 20 parameters,
293: in total, if we also account for the corresponding parameters defined in the series obtained after
294: shuffling the $Q_m$ randomly), {\em only} four are required for the distinction between SD and H.
295: We clarify that this seems to be extremely difficult to be achieved by chance. In order to visualize it,
296: if we assume (for the sake of convenience only) independent and identically distributed (iid) distributions
297: of the parameters for one subject, we find that the probability that {\em all} 4 parameters are
298: within the bounds (minima and maxima) set by 10 other subjects (i.e., the healthy ones) is
299: $(1-2/11)^4\approx 0.448$.
300: Thus, the probability that all 24 additional subjects are classified as SD by pure chance is
301: $(1-0.448)^{24}\approx 6.4\times 10^{-7}$, i.e., extremely small. Concerning the validity
302: of this statistical argument, we clarify that it
303: does not remain valid if one just picks 4 parameters out of the original 20 ones. Only if one
304: decides which parameters wants to use {\em before} the calculation of the values, the argument is
305: valid(This is the reason why blind evaluation- defining all methods, parameters and criteria studying
306: one set of data, and {\em then} testing the significance using an additonal set of independent
307: data- is considered very important in medical applications and/or publications).
308:
309:
310:
311:
312: We now attempt a physical interpretation of the present results,
313: the main feature of which focuses on the fact that both ratios
314: $\lambda_s(RR)$ and $\lambda_L(RR)$ become
315: smaller, in the vast majority of SD, compared to H. Recall that the $\delta S(RR)$-values
316: themselves cannot
317: distinguish SD from H, see Fig. \ref{fig4}(a), in contrast to the
318: ratios $\delta S_i(RR) / \delta S_3 (RR)$, see Fig. \ref{fig4}(b).
319: Before proceeding, we mention two points: First, for
320: individuals at high risk of sudden death, fractal organization
321: (long range correlations) breaks down (see Refs.\cite{IVA01,GOL02}
322: and references therein). The breakdown of fractal physiologic
323: complexity is often accompanied by emergence of {\em uncorrelated
324: randomness} or {\em excessive} order (e.g., periodic oscillations
325: appear in the heart rate recordings of
326: ``frequency''$\approx$1/min, which are associated with
327: Cheyne-Stokes breathing) \cite{GOL02}. Second, if we calculate\cite{ref8,VAR03} the
328: $\delta S$-values in a (dichotomous) Markovian (${\cal
329: M}$) time-series (exponentially distributed pulses), for a total
330: number of N=$10^3$ pulses (i.e., length comparable to that of the
331: ECG analyzed here), we find that these values: (a) lead to
332: $\lambda_s({\cal M})=1.20\pm0.03$ and (b) differ drastically, see Fig.
333: \ref{fig4}(a), from the $\delta S(RR)$-values themselves of {\em
334: both} SD and H (thus indicating that they exhibit non-Markovian
335: behavior on the whole; this is consistent with the
336: aspects that bodily rhythms, such as the heart beat, show complex
337: dynamics, e.g., \cite{IVA01,GOL02}). The fact that $\lambda_s(RR)$ in SD
338: becomes smaller than in H can now be understood as follows: Since
339: H exhibit a high order of complexity, it is expected that (even)
340: their $H_{min}$-value (=1.43), should markedly exceed
341: $\lambda_s({\cal M})$. On the other hand, in SD this high
342: complexity is lost, and hence their $\lambda_s(RR)$ values
343: naturally approach $\lambda_s({\cal M})$, thus becoming smaller.
344: Interestingly, the SD average value of $\lambda_s(RR)$ in Table
345: \ref{tab1} is 1.19, i.e., coincides with $\lambda_s({\cal M})$.
346: (Such a coincidence also occurs for the QRS-intervals in {\em
347: both} H and SD, which agrees with the observations\cite{KHA02} mentioned above
348: that the prolonged
349: QT-intervals in SD mainly originate from enlarged ST-values,
350: while their QRS- may remain the {\em same}.)
351: We now proceed to the interpretation of our results related to the
352: ratio $\lambda_L(RR)$. In H, it is expected that
353: (in view of the RR long range correlations\cite{GOL02}) the
354: corresponding values must be appreciably larger than
355: $\lambda_L({\cal M})=0.64\pm0.05$, calculated in the Markovian
356: case (Fig. \ref{fig4}(b)). We now examine the SD: If, in SD ``{\em
357: uncorrelated randomness}'' appears, this reflects that their
358: $\lambda_L(RR)$ values naturally approach $\lambda_L({\cal M})$,
359: thus becoming smaller (compared to H); this actually occurs in the
360: vast majority of SD in Table \ref{tab1}. {\em If} in SD the
361: aforementioned periodicities appear, it is naturally
362: expected to find {\em large} (see Ref.\cite{epaps}) $\delta
363: S$-values when a time-window of length around 60 beats, or so
364: (i.e., related to the aforementioned ``frequency''$\approx$1/min)
365: sweeps through the RR time-series, thus resulting in $\delta
366: S$-values even larger than those in H (since in H {\em no} such
367: periodicities appear). The latter actually occurs in the few
368: cases marked with superscript `b' (i.e., those exceeding
369: $H_{max}$) in Table \ref{tab1} (For additional arguments on the
370: interpretation see \cite{epaps}).
371:
372:
373:
374: The fact that the overall behavior of the complexity measures introduced
375: in this paper (i.e., clear distinction of SD from H) is more or less similar to that of the measures
376: discussed in Ref.\cite{ref8}
377: does not mean that the former measures are similar to the latter, because,
378: as we shall explain below they exhibit a certain type of complementarity
379: in the following sense: if in the frame of the one procedure an ambiguity emerges in the distinction between SD and H, the other procedure gives a clear answer.
380: (Recall that, as mentioned in Section I, in Ref.\cite{ref8}
381: we discussed entropy fluctuations -and ratios of ``shuffled'' and ``unshuffled''
382: entropy fluctuations- on fixed time scales, while here we study
383: entropy fluctuations on different time scales.) This is consistent with
384: the findings of Ashkenazy et al.\cite{ASH01} that an approach
385: dealing with ratios on the same time scale and an
386: approach dealing with ratios on different time scales
387: (or corresponding scaling exponents) are somewhat complementary.
388: We now study, as an example, the following two procedures: i.e., the
389: one that uses $\delta S$(QT)\cite{ref8} and the other which combines
390: the measures $\lambda$,$\rho$.
391: The -values of SD and H given in the last column of Table \ref{tab1} are
392: classified into two classes: the larger values correspond to SD,
393: and the lower ones correspond to H (see also Figs. \ref{fig2} and \ref{fig5} ).
394: Let us focus on the two lowermost SD values and the uppermost H value.
395: The former two correspond to sel33 and sel34 $\overline{\delta S}_{3-4}$(QT)=0.00076 and 0.00069, respectively) and the latter one to sel16795 ($\overline{\delta S}_{3-4}$(QT)=0.00056). In view of their
396: $\overline{\delta S}_{3-4}$(QT)-values proximity, one may wonder whether these two SD
397: could be confused with H. This ambiguity can be dissolved in the light of
398: the other procedure (i.e., $\lambda$, $\rho$), as follows: Table I reveals that
399: sel33 markedly violates both the $H_{min}$-limit for $\lambda_s$(QRS) as well as $H_{min}$ for $\lambda_s$(RR) (the latter can be visualized in Fig. \ref{fig6}). As for sel34, the $H_{max}$-limit of $\lambda_L$(QRS) is strongly violated. We now turn to an alternative example, i.e.,
400: sel47, who, by means of the method using the complexity measures $\lambda$, $\rho$ (of the RR- and QRS-intervals), could be confused with H, because a deviation of only around 12\% from the $H_{min}$-limit of $\min_{H} [\rho_s(QRS)]=0.18$ is noticed. This ambiguity can be
401: dissolved by means of the procedure using $\delta S$(QT) as follows: sel47 has $\overline{\delta S}_{3-4}$(QT)=0.0029 , which exceeds significantly, i.e., by a factor 5, the corresponding value
402: of sel16795, who has the largest $\overline{\delta S}_{3-4}$(QT)=0.00056 value among the H.
403:
404: \section{THE PROCEDURE TO DISTINGUISH SD FROM PATIENTS}
405:
406: This Section aims at distinguishing SD from patients, where the latter terminology
407: refers to individuals suffering only from heart diseases. The QT-Database of physiobank
408: we use here, includes the following four groups of patients (a fifth group that consists
409: of 4 individuals only, was disregarded for the reasons discussed in Ref.\cite{ref8}): 15 individuals from MIT-BIH Arrhythmia Database (labeled hereafter MIT), 13 from MIT-BIH Supraventricular
410: Arrhythmia Database (MSV), 33 from the European ST-T Database (EST) and 6 from MIT-BIH
411: ST change Database (MST). The values of $\lambda$, $\rho$, $\nu$,$\overline{\delta S}_{3-4}$(QT), $\lambda_{shuf}$, $\rho_{shuf}$ and $\overline{\delta S}_{3-4,shuf}$(QT) of all these patients are given in Ref.\cite{epaps}.
412:
413:
414: An inspection of the measures $\lambda$, $\rho$, $\nu$ shows three facts: First,
415: all SD and all patients violate one or more $H$-limits. Second, {\em none} of the measures $\lambda$, $\rho$, $\nu$ alone, or a combination of two of them, can effectively differentiate the SD from the patients.
416: Third, if we consider the three measures $\lambda$, $\rho$, $\nu$ (i.e., 16 parameters) altogether,
417: we find that twenty SD out of 24 violate some of the limits of both patients and H, thus allowing in principle a distinction of the vast majority of SD from the other individuals.
418: Thus, in summary, the consideration of the quantities ($\lambda$, $\rho$, $\nu$) only,
419: does not lead to a distinction between {\em all} SD and patients. The same conclusion is drawn if we alternatively consider the quantities ($\lambda$, $\lambda_{shuf}$,$\rho$) only.
420:
421:
422: We now turn to the investigation of the $\delta S$(QT) values. In Fig. \ref{fig5}, the average
423: $\langle \delta S\rangle$(QT) value for each group is plotted versus the time-window length.
424: It is intriguing that the results of the four groups (MIT, MSV, MST, EST) of patients
425: are located between H (the lowermost curve) and SD (the uppermost curve). We emphasize,
426: however, that if we plot the curves for each one of the 101 individuals
427: (in a way similar to that of Fig.\ref{fig2}(a)), we find that there are
428: some patients the results of which overlap with either SD or H. Let us consider
429: only the limiting cases -i.e., the lowermost and the uppermost curve, to be called
430: hereafter $\delta S$(QT)$_{min}$ and $\delta S$(QT)$_{max}$, respectively-
431: obtained in each groups of patients. In order to distinguish SD from patients,
432: we must appropriately discriminate the overlap which refers to those of the patients
433: that lie above the uppermost $\delta S$(QT) curve of H; the latter curve from
434: now on will be called $\delta S$(QT)$_{max,H}$. Thus, the limits
435: of the patients we are currently interested in, do not
436: extend from $\delta S$(QT)$_{min}$ to $\delta S$(QT)$_{max}$, since
437: they must exceed $\delta S$(QT)$_{max,H}$, i.e.,
438: \begin{equation}
439: \delta S (QT) > \delta S(QT)_{max,H}.
440: \label{con1}
441: \end{equation}
442: The curve which corresponds to the one of the patients,
443: that has $\delta S$(QT) lying just above the $\delta S(QT)_{max,H}$ corresponds to a value, which will be labeled hereafter $\delta S$(QT)$_{min'}$. Thus, if we apply the condition
444: \begin{equation}
445: \delta S(QT)_{min'} \leq \delta S (QT) \leq \delta S(QT)_{max}
446: \label{con2}
447: \end{equation}
448: to each group of patients, we are left only with those of the patients that actually overlap with SD.
449:
450: We now recall that, as mentioned above, the measures $\lambda$, $\rho$, $\nu$ altogether,
451: which are in fact ratios of $\delta S$ values, enable the discrimination
452: of the vast majority of SD from all the others (i.e., patients and H),
453: while the $\delta S$(QT) values themselves efficiently distinguish\cite{ref8} all SD from H. This motivates us to investigate whether a proper combination of these
454: two facts can serve our purpose, which refers to the identification
455: of all SD among the other individuals (patients and H). Thus, we now compare
456: the quantities $\lambda$, $\rho$, $\nu$, $\delta S$(QT) altogether,
457: of each SD, to the corresponding parameters of only
458: those among the patients that happen to have $\delta S$(QT) values
459: exceeding the corresponding values of H, i.e., obey the condition (\ref{con1}), or preferably the more accurate condition (\ref{con2}). Such a comparison reveals that some
460: of the 17 parameters of $\lambda$, $\rho$, $\nu$, $\delta S$(QT), in all SD,
461: lie outside the limits of these patients (cf. the same happens, of course,
462: if we compare each SD to the limits of H). These results point to the
463: conclusion that all 24 SD are distinguished from the patients (and H).
464: The same conclusion is drawn if we consider instead, the 17 parameters $\lambda$, $\lambda_{shuf}$, $\rho$, $\delta S$(QT). We emphasize, however, that the study of the
465: estimation errors (see the Appendix) reveals that the confidence level for the distinction of
466: all SD from the patients becomes appreciably larger if we combine all the measures $\lambda$, $\lambda_{shuf}$, $\rho$ $\rho_{shuf}$, $\nu$ (of all intervals) with the condition (\ref{con2}) applied to both $\delta S$(QT) and $\delta S_{shuf}$(QT) (i.e., in reality, we then consider the
467: limits of those patients whom {\em both} $\delta S$(QT)- and $\delta S_{shuf}$(QT)-values
468: are larger than those in H, as shown in Fig. 6 of Ref.\cite{ref8}).
469:
470:
471: We finally comment on three points. First, once the identification of SD has been completed,
472: the distinction between patients and H can be made by identifying as patients the individuals
473: whom one or more of the aforementioned parameters violate the $H$-limits. Second, since it is known
474: that heart rate variability depends strongly on the age, it is highly recommended that when
475: comparing values of the aforementioned complexity measures, the corresponding limits should
476: be taken from subjects (patients, H) of comparable age. Third, we now focus on the importance
477: of the sequential order of $Q_m$ on the aforementioned complexity measures. We prefer to
478: deal with the results related to the RR-intervals since it is known that the healthy heart
479: beats irregularly and that the intervals between beats (i.e., the RR-intervals) fluctuate
480: widely, following complicated patterns\cite{CHI02}. Let us investigate, for example, the possibility
481: of using $\lambda_\kappa$(RR) alone to distinguish the SD as well as the four groups of
482: patients from H, i.e., examine whether the $\lambda_\kappa$(RR)-values of each
483: individual violate one (at least) of the relevant $H$-limits. The
484: results show (see Table \ref{tab2}) that the vast majority of SD and
485: of each group of patients is well distinguished from H
486: by means of $\lambda_\kappa$(RR). The situation drastically changes,
487: however, if we use, instead of $\lambda_\kappa$(RR), the $\lambda_{\kappa ,shuf}$-values
488: (see the Tables V to VII in \cite{epaps}):
489: only the minority of SD and of each group of patients can be differentiated
490: from H. Since the calculation of the $\lambda_\kappa$(RR)-values takes into
491: account the sequential order of $Q_m$, while the $\lambda_{\kappa ,shuf}$(RR)-values
492: do not, this points to the following conclusion: It is the sequential order of beats
493: that inherently contains the primary information which enables the distinction between
494: the SD and patients, on the one hand, and the H, on the other. This might explain why
495: procedures based on the entropy in natural time (which is dynamic entropy, affected by
496: the sequential order\cite{VAR03y,ref8}, see Section I) -and hence they consider the complexity
497: measures mentioned in the preceding Sections- can achieve such a distinction, while a static
498: entropy (e.g., Shannon entropy, see Ref. \cite{ref8}) cannot.
499:
500: \section{Conclusions}
501:
502: First, in SD, the $\delta S$-values depend on the length scale in
503: a way significantly different from that in H. Hence these two groups
504: of humans can be well distinguished. Second, the SD, who exhibit
505: critical dynamics, have $\lambda$-values (being, in fact,
506: ratios of $\delta S$-values, as mentioned above) which approach those of the Markovian case.
507: This should {\em not} be misinterpreted as showing that the corresponding time-series
508: are of Markovian nature, because the $\delta S$-values themselves are
509: approximately one order of magnitude smaller than those of
510: the (dichotomous) Markovian time-series (see Fig. 4(a) and Ref. \cite{ref8}).
511: Third, the quantities $\lambda$, $\lambda_{shuf}$, $\rho$, $\rho_{shuf}$, $\nu$, $\delta S$(QT) and
512: $\delta S_{shuf}$(QT) {\em altogether}, seem to
513: enable the classification of individuals into the three categories: H, patients and SD.
514:
515:
516: \appendix*
517: \section{THE INFLUENCE OF THE Estimation ERRORS ON THE PROCEDURES FOR THE DISTINCTION OF SD}
518: Beyond the error introduced by the use of an automatic threshold
519: detector for the allocation of the corresponding intervals (cf.
520: this is largest for the QT- and smallest for the RR-intervals),
521: the following two sources of errors must be considered\cite{VAR03y,ref8}: First, an
522: estimation error emerges when analyzing -instead of the original
523: time series of length $l \approx 10^3$- smaller lengths $l'$,
524: which, however, still significantly exceed the time-window lengths
525: used, for example $l' \approx 2 \times 10^2$ (the errors
526: associated with the measures in the short range,$s$, are smaller
527: from those corresponding to the longer range, $L$, because for the
528: latter range the $l/l'$ values -due to the restricted length of
529: the records available- are small, thus not allowing more reliable
530: statistics). Second, a source of (statistical) error in the
531: results emerges when considering the ratio(s) $\delta
532: S_{shuf}$/$\delta S$ (i.e., when dealing with $\nu$ and
533: $\lambda_{shuf}$) instead of $\delta S$ itself. While $\delta S$
534: may be considered to have a {\em unique} value for a (given)
535: original $Q_m$ time-series, the value of $\delta S_{shuf}$ depends
536: on the randomly shuffled $Q_m$ series each time selected (cf. such
537: differences are well known\cite{kan02} when dealing with randomized series
538: of {\em finite} length). This is why the $\nu$-values given in
539: Ref.\cite{ref8} for SD and H do not fully coincide with those
540: tabulated in the present paper. To account roughly for the extent of this
541: statistical error, we averaged here the $\delta S_{shuf}$-values
542: calculated over a number (e.g. 20) of randomly shuffled
543: $Q_m$-series generated from the {\em same} original series and
544: then the corresponding standard deviation was estimated.
545:
546: The final results on the above sources, could be summarized as
547: follows: The (percentage)
548: estimation error was found to be around 10\% (cf. this is an {\em average} value)
549: for the complexity measures $\lambda$,$\lambda_{shuf}$, $\rho$, $\rho_{shuf}$, $\nu$
550: associated with the RR- and QRS-intervals.
551: Furthermore, since the error in the $\delta S$(QT) may reach 20\%,
552: the estimation error
553: in those of the complexity measures that involve $\delta S$(QT) may be as high as $\approx$ 30\%.
554: Upon considering such error-levels, hereafter called ``plausible
555: estimation errors'' $\epsilon_p$, a study of each of the methods
556: for the distinction of SD was made. The study was repeated by
557: assuming larger (percentage) estimation errors, hereafter labeled
558: ``modified estimation errors''$\epsilon_m$, calculated from
559: \begin{equation}
560: \epsilon_m=\epsilon_p \left( 1+\frac{H_{max}-H_{min}}{H_{max}+H_{min}} \right)
561: \end{equation}
562: for each parameter (see Table VIII in Ref. \cite{epaps}). Both
563: studies led, more or less, to the same results. The
564: calculation, in each study, was made as follows: Each parameter
565: was assumed to be equal to its value (initially estimated from the
566: original time-series available) multiplied by a number randomly
567: selected in the range $1 \pm \epsilon_p$ or $1 \pm \epsilon_m$,
568: respectively) and then each of the methods for the distinction of
569: SD was applied. This application was repeated, for each method,
570: $10^3$ times via Monte Carlo and relevant conclusions have been
571: drawn for both studies. The extent to which these conclusions
572: hold, was also investigated in the following {\em extreme} case:
573: the limits of the parameters of H (and patients), which are
574: automatically adjusted for each ``random'' selection of the values
575: described above, have been assumed to {\em additionally} relax by
576: (extra) amounts equal to $\epsilon_p$ or $\epsilon_m$. (Such a
577: ``relaxation'' faces the {\em extreme} possibility that the
578: populations of H and patients treated here are not considered
579: large enough to allow a precise determination of their limits, and
580: hence future increased populations' studies could somehow broaden
581: these limits by {\em extra} amounts as large as $\epsilon_p$ or
582: $\epsilon_m$).
583:
584: The following conclusions were finally drawn concerning the
585: distinction between SD and H (see also Table \ref{aca7}): Among the four
586: methods suggested (i.e., two in Ref.\cite{ref8} and two in Section II),
587: the one that uses the
588: measures $\lambda$, $\rho$ (associated, however, with {\em all}
589: three types of intervals, i.e., 10 parameters in total) seems to
590: be robust in the following sense: when assuming the error-levels
591: mentioned above, the use of $\lambda$, $\rho$ still allow with a
592: confidence level above 99\% the distinction of {\em all} SD from
593: H. ( Then a calculation similar to that given in Section II
594: concerning the probability that all 24 subjects are classified,
595: by means of 10 parameters, as SD
596: by {\em pure chance} -based on the limits set by 10 other subjects-
597: results in $[1-(1-2/11)^{10}]^{24}\approx 0.03$, i.e., too small.)
598: The confidence level decreases to 63\%, 49\%, 32\% and 59\% ,
599: respectively, when using four parameters or one parameter only as
600: follows: First: $\lambda_\kappa$(RR) and $\rho_\kappa$(QRS);
601: second: $\lambda_\kappa$(RR) and $\lambda_{\kappa,shuf}$(RR);
602: third: $\nu_\kappa$(RR) and $\nu_\kappa$(QRS); fourth:
603: $\overline{\delta S}_{3-4}$(QT). If we investigate the aforementioned
604: extreme case of the additional ``relaxation'' of the {\em
605: H-limits}, the capability for
606: the distinction of {\em all} SD still remains with the following
607: results: In the case of using $\lambda$, $\rho$ (of all intervals),
608: the confidence level in distinguishing {\em all} SD is 88\%, while it
609: becomes {\em appreciably higher}, i.e., larger than 99\%,
610: if we use the quantities $\lambda$,
611: $\rho$, $\lambda_{shuf}$, $\rho_{shuf}$, $\nu$,
612: $\overline{\delta S}_{3-4}(QT)$, $\overline{\delta S}_{3-4, shuf}(QT)$ {\em altogether}. When
613: using, however, four parameters only in the first three
614: combinations mentioned above, the confidence level decreases to
615: 90\%, 36\% and 8\% , respectively (and to 77\% when using
616: $\overline{\delta S}_{3-4}$(QT)), even when allowing two at the
617: most SD -out of 24- to be misinterpreted as being H.
618: As for the corresponding conclusions related to the distinction of SD from the patients,
619: these can be drawn on the basis of the values given in the lower part of Table \ref{aca7}.
620:
621: In summary, the study of the estimation errors reveals that
622: ({\em if} the limits of the parameters will {\em not} be
623: broadened by future investigations) we can satisfactorily
624: distinguish the {\em totality} of SD from H as well as
625: discriminate the totality of SD from patients,
626: upon employing the quantities $\lambda$,
627: $\lambda_{shuf}$, $\rho$, $\rho_{shuf}$, $\nu$,$\overline{\delta
628: S}_{3-4}$(QT), $\overline{\delta S}_{3-4, shuf}$(QT) {\em
629: altogether}, i.e., the sixth and the last method, respectively in
630: Table \ref{aca7}. These quantities also allow the distinction of the
631: {\em totality} of SD from H (as well as distinguishes the {\em vast majority}
632: of SD from the patients), {\em even if } their limits will be eventually
633: broadened (by $\epsilon_m$).
634:
635:
636: The following remark should be added concerning the number
637: of parameters required to achieve the desired distinction: In
638: reality, only twelve {\em independent} quantities, (i.e., the six:
639: $\delta S_{\kappa}(\tau)$ and the six $\delta
640: S_{\kappa,shuf}(\tau)$, where $\kappa$=s,L and $\tau$=RR,QRS, QT)
641: are extracted from the
642: experimental data. Thus, for example, beyond $\overline{\delta S}_{3-4}(QT)$
643: {\em or} $\overline{\delta S}_{3-4,shuf}(QT)$, eleven
644: additional parameters (out of 26) of the
645: ratios: $\lambda$, $\lambda_{shuf}$, $\rho$, $\rho_{shuf}$, $\nu$,
646: are in principle required to be used for the distinction. These
647: twelve quantities , however, should {\em not} be fortuitously
648: selected, but the following points must be carefully considered:
649: (i) priority should be given to the eight parameters associated
650: with $\lambda$-values and $\lambda_{shuf}$- (or $\nu$-) values of
651: RR and QRS, (ii) using, at least, one $\rho$-parameter (involving
652: $\overline{\delta S}_{3-4}(QT)$ or $\overline{\delta S}_{3-4,shuf}(QT)$), and (iii)
653: examining whether the totality of the parameters used can actually
654: reproduce the aforementioned twelve $\delta S$-values determined directly from
655: the data. However, in order to avoid the
656: difficulty arising from the completeness (or not) of the
657: aforementioned selection, at the present stage (i.e., until an
658: appreciably larger number of H and patients will be analyzed to
659: allow a better precision in the determination of the corresponding
660: limits), the preceding paragraph
661: recommends to use -instead of twelve- {\em all} the
662: 28 parameters associated with the quantities $\lambda$,
663: $\lambda_{shuf}$, $\rho$, $\rho_{shuf}$, $\nu$,
664: $\overline{\delta S}_{3-4}(QT)$ and $\overline{\delta S}_{3-4, shuf}(QT)$.
665:
666: \bibliographystyle{apsrev}
667: \begin{thebibliography}{30}
668: \expandafter\ifx\csname natexlab\endcsname\relax\def\natexlab#1{#1}\fi
669: \expandafter\ifx\csname bibnamefont\endcsname\relax
670: \def\bibnamefont#1{#1}\fi
671: \expandafter\ifx\csname bibfnamefont\endcsname\relax
672: \def\bibfnamefont#1{#1}\fi
673: \expandafter\ifx\csname citenamefont\endcsname\relax
674: \def\citenamefont#1{#1}\fi
675: \expandafter\ifx\csname url\endcsname\relax
676: \def\url#1{\texttt{#1}}\fi
677: \expandafter\ifx\csname urlprefix\endcsname\relax\def\urlprefix{URL }\fi
678: \providecommand{\bibinfo}[2]{#2}
679: \providecommand{\eprint}[2][]{\url{#2}}
680:
681: \bibitem[{\citenamefont{Varotsos and Alexopoulos}(1986)}]{VAR86A}
682: \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{Varotsos}} \bibnamefont{and}
683: \bibinfo{author}{\bibfnamefont{K.}~\bibnamefont{Alexopoulos}},
684: \emph{\bibinfo{title}{Thermodynamics of Point Defects and their relation with
685: the Bulk Properties}} (\bibinfo{publisher}{North-Holland},
686: \bibinfo{address}{Amsterdam}, \bibinfo{year}{1986}).
687:
688: \bibitem[{\citenamefont{Varotsos et~al.}(1986)\citenamefont{Varotsos,
689: Alexopoulos, Nomicos, and Lazaridou}}]{VAR86B}
690: \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{Varotsos}},
691: \bibinfo{author}{\bibfnamefont{K.}~\bibnamefont{Alexopoulos}},
692: \bibinfo{author}{\bibfnamefont{K.}~\bibnamefont{Nomicos}}, \bibnamefont{and}
693: \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Lazaridou}},
694: \bibinfo{journal}{Nature (London)} \textbf{\bibinfo{volume}{322}},
695: \bibinfo{pages}{120} (\bibinfo{year}{1986}).
696:
697: \bibitem[{\citenamefont{Varotsos
698: et~al.}(2003{\natexlab{a}})\citenamefont{Varotsos, Sarlis, and
699: Skordas}}]{VAR03x}
700: \bibinfo{author}{\bibfnamefont{P.~A.} \bibnamefont{Varotsos}},
701: \bibinfo{author}{\bibfnamefont{N.~V.} \bibnamefont{Sarlis}},
702: \bibnamefont{and} \bibinfo{author}{\bibfnamefont{E.~S.}
703: \bibnamefont{Skordas}}, \bibinfo{journal}{Phys. Rev. Lett.}
704: \textbf{\bibinfo{volume}{91}}, \bibinfo{pages}{148501}
705: (\bibinfo{year}{2003}{\natexlab{a}}).
706:
707: \bibitem[{\citenamefont{Uyeda et~al.}(2002)\citenamefont{Uyeda, Hayakawa,
708: Nagao, Molchanov, Hattori, Orihara, Gotoh, Akinaga, and Tanaka}}]{UYE02}
709: \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Uyeda}},
710: \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Hayakawa}},
711: \bibinfo{author}{\bibfnamefont{T.}~\bibnamefont{Nagao}},
712: \bibinfo{author}{\bibfnamefont{O.}~\bibnamefont{Molchanov}},
713: \bibinfo{author}{\bibfnamefont{K.}~\bibnamefont{Hattori}},
714: \bibinfo{author}{\bibfnamefont{Y.}~\bibnamefont{Orihara}},
715: \bibinfo{author}{\bibfnamefont{K.}~\bibnamefont{Gotoh}},
716: \bibinfo{author}{\bibfnamefont{Y.}~\bibnamefont{Akinaga}}, \bibnamefont{and}
717: \bibinfo{author}{\bibfnamefont{H.}~\bibnamefont{Tanaka}},
718: \bibinfo{journal}{Proc. Natl. Acad. Sci. USA} \textbf{\bibinfo{volume}{99}},
719: \bibinfo{pages}{7352} (\bibinfo{year}{2002}).
720:
721: \bibitem[{\citenamefont{Varotsos et~al.}(2002)\citenamefont{Varotsos, Sarlis,
722: and Skordas}}]{VAR02}
723: \bibinfo{author}{\bibfnamefont{P.~A.} \bibnamefont{Varotsos}},
724: \bibinfo{author}{\bibfnamefont{N.~V.} \bibnamefont{Sarlis}},
725: \bibnamefont{and} \bibinfo{author}{\bibfnamefont{E.~S.}
726: \bibnamefont{Skordas}}, \bibinfo{journal}{Phys. Rev. E}
727: \textbf{\bibinfo{volume}{66}}, \bibinfo{pages}{011902}
728: (\bibinfo{year}{2002}).
729:
730: \bibitem[{\citenamefont{Bray}(2003)}]{BRA03}
731: \bibinfo{author}{\bibfnamefont{M.~A.~P.} \bibnamefont{Bray}},
732: \bibinfo{journal}{Med. Phys.} \textbf{\bibinfo{volume}{30}},
733: \bibinfo{pages}{3045} (\bibinfo{year}{2003}).
734:
735: \bibitem[{\citenamefont{Varotsos
736: et~al.}(2003{\natexlab{b}})\citenamefont{Varotsos, Sarlis, Skordas, and
737: Lazaridou}}]{VAR03y}
738: \bibinfo{author}{\bibfnamefont{P.~A.} \bibnamefont{Varotsos}},
739: \bibinfo{author}{\bibfnamefont{N.~V.} \bibnamefont{Sarlis}},
740: \bibinfo{author}{\bibfnamefont{E.~S.} \bibnamefont{Skordas}},
741: \bibnamefont{and} \bibinfo{author}{\bibfnamefont{M.~S.}
742: \bibnamefont{Lazaridou}}, \bibinfo{journal}{Practica of Athens Academy}
743: \textbf{\bibinfo{volume}{78}}, \bibinfo{pages}{281}
744: (\bibinfo{year}{2003}{\natexlab{b}}).
745:
746: \bibitem[{\citenamefont{Varotsos et~al.}(2004)\citenamefont{Varotsos, Sarlis,
747: Skordas, and Lazaridou}}]{ref8}
748: \bibinfo{author}{\bibfnamefont{P.~A.} \bibnamefont{Varotsos}},
749: \bibinfo{author}{\bibfnamefont{N.~V.} \bibnamefont{Sarlis}},
750: \bibinfo{author}{\bibfnamefont{E.~S.} \bibnamefont{Skordas}},
751: \bibnamefont{and} \bibinfo{author}{\bibfnamefont{M.~S.}
752: \bibnamefont{Lazaridou}}, \bibinfo{journal}{Phys. Rev. E}
753: \textbf{\bibinfo{volume}{70}}, \bibinfo{pages}{011106}
754: (\bibinfo{year}{2004}).
755:
756: \bibitem[{\citenamefont{Varotsos et~al.}(2001)\citenamefont{Varotsos, Sarlis,
757: and Skordas}}]{VAR01}
758: \bibinfo{author}{\bibfnamefont{P.~A.} \bibnamefont{Varotsos}},
759: \bibinfo{author}{\bibfnamefont{N.~V.} \bibnamefont{Sarlis}},
760: \bibnamefont{and} \bibinfo{author}{\bibfnamefont{E.~S.}
761: \bibnamefont{Skordas}}, \bibinfo{journal}{Practica of Athens Academy}
762: \textbf{\bibinfo{volume}{76}}, \bibinfo{pages}{294} (\bibinfo{year}{2001}).
763:
764: \bibitem[{\citenamefont{Varotsos
765: et~al.}(2003{\natexlab{c}})\citenamefont{Varotsos, Sarlis, and
766: Skordas}}]{VAR03}
767: \bibinfo{author}{\bibfnamefont{P.~A.} \bibnamefont{Varotsos}},
768: \bibinfo{author}{\bibfnamefont{N.~V.} \bibnamefont{Sarlis}},
769: \bibnamefont{and} \bibinfo{author}{\bibfnamefont{E.~S.}
770: \bibnamefont{Skordas}}, \bibinfo{journal}{Phys. Rev. E}
771: \textbf{\bibinfo{volume}{68}}, \bibinfo{pages}{031106}
772: (\bibinfo{year}{2003}{\natexlab{c}}).
773:
774: \bibitem[{\citenamefont{Laguna et~al.}(1990)\citenamefont{Laguna, Thakor,
775: Caminal, Jan\'{e}, and Yoon}}]{LAG90}
776: \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{Laguna}},
777: \bibinfo{author}{\bibfnamefont{N.~V.} \bibnamefont{Thakor}},
778: \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{Caminal}},
779: \bibinfo{author}{\bibfnamefont{R.}~\bibnamefont{Jan\'{e}}}, \bibnamefont{and}
780: \bibinfo{author}{\bibfnamefont{H.~R.} \bibnamefont{Yoon}},
781: \bibinfo{journal}{Medical \& Biological Engineering \& Computing}
782: \textbf{\bibinfo{volume}{28}}, \bibinfo{pages}{67} (\bibinfo{year}{1990}).
783:
784: \bibitem[{\citenamefont{Laguna et~al.}(1994)\citenamefont{Laguna, Jan\'{e}, and
785: Caminal}}]{LAG94}
786: \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{Laguna}},
787: \bibinfo{author}{\bibfnamefont{R.}~\bibnamefont{Jan\'{e}}}, \bibnamefont{and}
788: \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{Caminal}},
789: \bibinfo{journal}{Computers and Biomedical Research}
790: \textbf{\bibinfo{volume}{27}}, \bibinfo{pages}{45} (\bibinfo{year}{1994}).
791:
792: \bibitem[{\citenamefont{Laguna et~al.}(1997)\citenamefont{Laguna, Mark,
793: Goldberger, and Moody}}]{LAG97}
794: \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{Laguna}},
795: \bibinfo{author}{\bibfnamefont{R.~G.} \bibnamefont{Mark}},
796: \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{Goldberger}},
797: \bibnamefont{and} \bibinfo{author}{\bibfnamefont{G.~B.} \bibnamefont{Moody}},
798: in \emph{\bibinfo{booktitle}{Computers in Cardiology}}
799: (\bibinfo{publisher}{IEEE Computer Society Press},
800: \bibinfo{address}{Piscataway, NJ}, \bibinfo{year}{1997}),
801: vol.~\bibinfo{volume}{24}, p. \bibinfo{pages}{673}.
802:
803: \bibitem[{\citenamefont{Jan\'{e} et~al.}(1997)\citenamefont{Jan\'{e}, Blasi,
804: J.~Garcia, and Laguna}}]{JAN97}
805: \bibinfo{author}{\bibfnamefont{R.}~\bibnamefont{Jan\'{e}}},
806: \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{Blasi}},
807: \bibinfo{author}{\bibfnamefont{J.}~\bibnamefont{J.~Garcia}},
808: \bibnamefont{and} \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{Laguna}},
809: in \emph{\bibinfo{booktitle}{Computers in Cardiology}}
810: (\bibinfo{publisher}{IEEE Computer Society Press},
811: \bibinfo{address}{Piscataway, NJ}, \bibinfo{year}{1997}),
812: vol.~\bibinfo{volume}{24}, p. \bibinfo{pages}{295}.
813:
814: \bibitem[{\citenamefont{Khan}(2002)}]{KHA02}
815: \bibinfo{author}{\bibfnamefont{I.~A.} \bibnamefont{Khan}},
816: \bibinfo{journal}{Am. Heart J.} \textbf{\bibinfo{volume}{143}},
817: \bibinfo{pages}{7} (\bibinfo{year}{2002}).
818:
819: \bibitem[{epa()}]{epaps}
820: \eprint{See EPAPS Document No. E-PLEEE8-71-134501 for additional information. A direct link
821: to this document may be mfound in the online article's HTML reference section. The document
822: may also be readed via via the EPAPS homepage
823: (\url{http://www.aip.org/pubservs/epaps.html}) or from \url{ftp.aip.org} in
824: the directory /epaps/. See the EPAPS homepage for more information.}
825:
826: \bibitem[{\citenamefont{Goldberger et~al.}(2000)\citenamefont{Goldberger,
827: Amaral, Glass, Hausdorff, Ivanov, Mark, Mictus, Moody, Peng, and
828: Stanley}}]{GOL00}
829: \bibinfo{author}{\bibfnamefont{A.~L.} \bibnamefont{Goldberger}},
830: \bibinfo{author}{\bibfnamefont{L.~A.~N.} \bibnamefont{Amaral}},
831: \bibinfo{author}{\bibfnamefont{L.}~\bibnamefont{Glass}},
832: \bibinfo{author}{\bibfnamefont{J.~M.} \bibnamefont{Hausdorff}},
833: \bibinfo{author}{\bibfnamefont{P.~C.} \bibnamefont{Ivanov}},
834: \bibinfo{author}{\bibfnamefont{R.~G.} \bibnamefont{Mark}},
835: \bibinfo{author}{\bibfnamefont{J.~E.} \bibnamefont{Mictus}},
836: \bibinfo{author}{\bibfnamefont{G.~B.} \bibnamefont{Moody}},
837: \bibinfo{author}{\bibfnamefont{C.-K.} \bibnamefont{Peng}}, \bibnamefont{and}
838: \bibinfo{author}{\bibfnamefont{H.~E.} \bibnamefont{Stanley}},
839: \bibinfo{journal}{Circulation} \textbf{\bibinfo{volume}{101}},
840: \bibinfo{pages}{e215 (see also \url{\http://www.physionet.org})}
841: (\bibinfo{year}{2000}).
842:
843: \bibitem[{\citenamefont{Pincus}(1991)}]{PIN91}
844: \bibinfo{author}{\bibfnamefont{S.~M.} \bibnamefont{Pincus}},
845: \bibinfo{journal}{Proc. Natl. Acad. Sci. USA} \textbf{\bibinfo{volume}{88}},
846: \bibinfo{pages}{2297} (\bibinfo{year}{1991}).
847:
848: \bibitem[{\citenamefont{Richman and Moorman}(2000)}]{RIC00}
849: \bibinfo{author}{\bibfnamefont{J.~S.} \bibnamefont{Richman}} \bibnamefont{and}
850: \bibinfo{author}{\bibfnamefont{J.~R.} \bibnamefont{Moorman}},
851: \bibinfo{journal}{Am. J. Physiol. Heart. Circ. Physiol.}
852: \textbf{\bibinfo{volume}{278}}, \bibinfo{pages}{2039} (\bibinfo{year}{2000}).
853:
854: \bibitem[{\citenamefont{Nikolopoulos et~al.}(2003)\citenamefont{Nikolopoulos,
855: Alexandridi, Nikolakeas, and Manis}}]{NIK03}
856: \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Nikolopoulos}},
857: \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{Alexandridi}},
858: \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Nikolakeas}},
859: \bibnamefont{and} \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{Manis}},
860: \bibinfo{journal}{J. Biomed. Informatics} \textbf{\bibinfo{volume}{36}},
861: \bibinfo{pages}{202} (\bibinfo{year}{2003}).
862:
863: \bibitem[{\citenamefont{Costa et~al.}(2002)\citenamefont{Costa, Goldberger, and
864: Peng}}]{COS02}
865: \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Costa}},
866: \bibinfo{author}{\bibfnamefont{A.~L.} \bibnamefont{Goldberger}},
867: \bibnamefont{and} \bibinfo{author}{\bibfnamefont{C.~K.} \bibnamefont{Peng}},
868: \bibinfo{journal}{Phys. Rev. Lett.} \textbf{\bibinfo{volume}{89}},
869: \bibinfo{pages}{068102} (\bibinfo{year}{2002}).
870:
871: \bibitem[{\citenamefont{Varotsos
872: et~al.}(2003{\natexlab{d}})\citenamefont{Varotsos, Sarlis, and
873: Skordas}}]{VAR03feb}
874: \bibinfo{author}{\bibfnamefont{P.~A.} \bibnamefont{Varotsos}},
875: \bibinfo{author}{\bibfnamefont{N.~V.} \bibnamefont{Sarlis}},
876: \bibnamefont{and} \bibinfo{author}{\bibfnamefont{E.~S.}
877: \bibnamefont{Skordas}}, \bibinfo{journal}{Phys. Rev. E}
878: \textbf{\bibinfo{volume}{67}}, \bibinfo{pages}{021109}
879: (\bibinfo{year}{2003}{\natexlab{d}}).
880:
881: \bibitem[{\citenamefont{Blythe et~al.}(2004)\citenamefont{Blythe, Janke,
882: Johnston, and Kenna}}]{BLY04}
883: \bibinfo{author}{\bibfnamefont{R.~A.} \bibnamefont{Blythe}},
884: \bibinfo{author}{\bibfnamefont{W.}~\bibnamefont{Janke}},
885: \bibinfo{author}{\bibfnamefont{D.~A.} \bibnamefont{Johnston}},
886: \bibnamefont{and} \bibinfo{author}{\bibfnamefont{R.}~\bibnamefont{Kenna}}
887: (\bibinfo{year}{2004}), \bibinfo{note}{cond-mat/0401385}.
888:
889: \bibitem[{\citenamefont{Gallavoti and Cohen}(2003)}]{GAL03}
890: \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{Gallavoti}} \bibnamefont{and}
891: \bibinfo{author}{\bibfnamefont{E.~G.~D.} \bibnamefont{Cohen}}
892: (\bibinfo{year}{2003}), \bibinfo{note}{cond-mat/0312306}.
893:
894: \bibitem[{\citenamefont{Tsallis}(2003)}]{TSA03}
895: \bibinfo{author}{\bibfnamefont{C.}~\bibnamefont{Tsallis}}
896: (\bibinfo{year}{2003}), \bibinfo{note}{cond-mat/0312500}.
897:
898: \bibitem[{\citenamefont{Goldberger et~al.}(2002)\citenamefont{Goldberger,
899: Amaral, Hausdorff, Ivanov, Peng, and Stanley}}]{GOL02}
900: \bibinfo{author}{\bibfnamefont{A.~L.} \bibnamefont{Goldberger}},
901: \bibinfo{author}{\bibfnamefont{L.~A.~N.} \bibnamefont{Amaral}},
902: \bibinfo{author}{\bibfnamefont{J.~M.} \bibnamefont{Hausdorff}},
903: \bibinfo{author}{\bibfnamefont{P.~C.} \bibnamefont{Ivanov}},
904: \bibinfo{author}{\bibfnamefont{C.-K.} \bibnamefont{Peng}}, \bibnamefont{and}
905: \bibinfo{author}{\bibfnamefont{H.~E.} \bibnamefont{Stanley}},
906: \bibinfo{journal}{Proc. Natl. Acad. Sci. USA} \textbf{\bibinfo{volume}{99}},
907: \bibinfo{pages}{2467} (\bibinfo{year}{2002}).
908:
909: \bibitem[{\citenamefont{Ivanov et~al.}(2001)\citenamefont{Ivanov, Amaral,
910: Goldberger, Havlin, Rosenblum, Stanley, and Struzik}}]{IVA01}
911: \bibinfo{author}{\bibfnamefont{P.~C.} \bibnamefont{Ivanov}},
912: \bibinfo{author}{\bibfnamefont{L.~A.~N.} \bibnamefont{Amaral}},
913: \bibinfo{author}{\bibfnamefont{A.~L.} \bibnamefont{Goldberger}},
914: \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Havlin}},
915: \bibinfo{author}{\bibfnamefont{M.~G.} \bibnamefont{Rosenblum}},
916: \bibinfo{author}{\bibfnamefont{H.~E.} \bibnamefont{Stanley}},
917: \bibnamefont{and} \bibinfo{author}{\bibfnamefont{Z.~R.}
918: \bibnamefont{Struzik}}, \bibinfo{journal}{Chaos}
919: \textbf{\bibinfo{volume}{11}}, \bibinfo{pages}{641} (\bibinfo{year}{2001}).
920:
921: \bibitem[{\citenamefont{Ashkenazy et~al.}(2001)\citenamefont{Ashkenazy,
922: Lewkowicz, Levitan, Havlin, Saermark, Moelgaard, Bloch~Thomsen, Moller,
923: Hintze, and Huikuri}}]{ASH01}
924: \bibinfo{author}{\bibfnamefont{Y.}~\bibnamefont{Ashkenazy}},
925: \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Lewkowicz}},
926: \bibinfo{author}{\bibfnamefont{J.}~\bibnamefont{Levitan}},
927: \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Havlin}},
928: \bibinfo{author}{\bibfnamefont{K.}~\bibnamefont{Saermark}},
929: \bibinfo{author}{\bibfnamefont{H.}~\bibnamefont{Moelgaard}},
930: \bibinfo{author}{\bibfnamefont{P.~E.} \bibnamefont{Bloch~Thomsen}},
931: \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Moller}},
932: \bibinfo{author}{\bibfnamefont{U.}~\bibnamefont{Hintze}}, \bibnamefont{and}
933: \bibinfo{author}{\bibfnamefont{H.~V.} \bibnamefont{Huikuri}},
934: \bibinfo{journal}{Europhys. Lett.} \textbf{\bibinfo{volume}{53}},
935: \bibinfo{pages}{709} (\bibinfo{year}{2001}).
936:
937: \bibitem[{\citenamefont{Chialvo}(2002)}]{CHI02}
938: \bibinfo{author}{\bibfnamefont{D.~R.} \bibnamefont{Chialvo}},
939: \bibinfo{journal}{Nature} \textbf{\bibinfo{volume}{419}},
940: \bibinfo{pages}{263} (\bibinfo{year}{2002}).
941:
942: \bibitem[{\citenamefont{Kantelhardt et~al.}(2002)\citenamefont{Kantelhardt,
943: Zschiegner, Koscienly-Bunde, Bunde, Havlin, and Stanley}}]{kan02}
944: \bibinfo{author}{\bibfnamefont{J.}~\bibnamefont{Kantelhardt}},
945: \bibinfo{author}{\bibfnamefont{S.~A.} \bibnamefont{Zschiegner}},
946: \bibinfo{author}{\bibfnamefont{E.}~\bibnamefont{Koscienly-Bunde}},
947: \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{Bunde}},
948: \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Havlin}}, \bibnamefont{and}
949: \bibinfo{author}{\bibfnamefont{H.~E.} \bibnamefont{Stanley}},
950: \bibinfo{journal}{Physica A} \textbf{\bibinfo{volume}{316}},
951: \bibinfo{pages}{87} (\bibinfo{year}{2002}).
952:
953: \end{thebibliography}
954:
955: \begin{turnpage}
956: \squeezetable
957: \begin{table*}
958: \caption{The variability measures ($\lambda$), the relative ones
959: ($\rho$), and the ratios $\nu \equiv$ $\overline{\delta
960: S}_{shuf}/ \overline{{\delta S}}$ in the short (s) range and in
961: the longer (L) range in H (sel16265 to sel17453) and SD (sel30 to
962: sel17152) along with their $\overline{\delta S}_{3-4}(QT)$-values.} \label{tab1}
963: \begin{ruledtabular}
964:
965: \begin{tabular}{c cc cc cc cc cc ccc ccc c}
966:
967: individual & RR & & QRS & & QT & & RR over QRS & & RR over QT
968: \vspace{0.05cm} & & & 3-4 beats ($\nu_s$)\footnotemark[3] & & & 50-70 beats
969: ($\nu_L$)\footnotemark[3] &
970: \\
971:
972: \cline{2-3}
973: % \cline{4-5}
974: \cline{6-7}
975: % \cline{8-9}
976: \cline{10-11}
977: %\cline{12-14}
978: \cline{15-17}
979:
980: & \vspace{0.05cm}$\lambda_s(RR)$ &
981: $\lambda_L(RR)$ &
982: $\lambda_s(QRS)$
983: & $\lambda_L(QRS)$ &
984: $\lambda_s(QT)$
985: & $\lambda_L(QT)$ &
986: $\rho_s(QRS)$ &
987: $\rho_L(QRS)$ & $\rho_s(QT)$ & $\rho_L(QT)$ &
988: RR & QRS & QT & RR & QRS & QT & $\overline{\delta S}_{3-4}(QT) \times 10^{3}$\\
989: \hline
990: sel16265 & 1.72 & 2.38 & 1.19 & 0.52 & 1.27 & 0.88 & 0.88 & 4.01 & 2.44 & 6.62 & 1.87 & 0.98 & 1.29 & 0.48 & 1
991: .02 & 0.75 & 0.38\\
992: sel16272 & 1.69 & 1.35 & 1.29 & 0.61 & 1.21 & 0.50 & 0.18 & 0.40 & 0.67 & 1.79 & 1.65 & 0.88 & 0.94 & 0.77 & 1
993: .10 & 1.07 & 0.48\\
994: sel16273 & 1.61 & 2.69 & 1.16 & 0.59 & 1.30 & 1.11 & 1.11 & 5.05 & 3.17 & 7.65 & 2.18 & 0.99 & 1.46 & 0.50 & 0
995: .88 & 0.71 & 0.24\\
996: sel16420 & 1.51 & 1.74 & 1.22 & 0.48 & 1.37 & 0.66 & 0.96 & 3.46 & 1.97 & 5.21 & 1.60 & 0.99 & 1.07 & 0.53 & 1
997: .09 & 0.90 & 0.36\\
998: sel16483 & 1.43 & 2.37 & 1.23 & 0.49 & 1.31 & 0.68 & 0.25 & 1.22 & 0.96 & 3.37 & 2.27 & 0.99 & 1.17 & 0.52 & 1
999: .15 & 0.92 & 0.35\\
1000: sel16539 & 2.00 & 1.94 & 1.26 & 0.50 & 1.41 & 1.08 & 1.85 & 7.10 & 5.57 & 10.04 & 1.43 & 1.07 & 1.27 & 0.50 & 1
1001: .08 & 0.65 & 0.52\\
1002: sel16773 & 1.92 & 2.61 & 1.21 & 0.49 & 1.31 & 0.70 & 0.90 & 4.84 & 1.49 & 5.54 & 1.85 & 1.01 & 0.91 & 0.44 & 1
1003: .05 & 0.97 & 0.55\\
1004: sel16786 & 1.71 & 1.57 & 1.19 & 0.51 & 1.31 & 0.84 & 1.16 & 3.56 & 3.97 & 7.43 & 1.39 & 1.01 & 1.19 & 0.55 & 1
1005: .04 & 0.77 & 0.23\\
1006: sel16795 & 1.77 & 0.99 & 1.24 & 0.55 & 1.16 & 0.56 & 0.77 & 1.37 & 2.87 & 5.08 & 1.10 & 0.98 & 1.05 & 0.74 & 0
1007: .95 & 1.00 & 0.56\\
1008: sel17453 & 1.87 & 1.67 & 1.26 & 0.54 & 1.22 & 0.68 & 1.49 & 4.59 & 2.91 & 7.12 & 1.46 & 1.01 & 1.02 & 0.57 & 0
1009: .98 & 0.81 & 0.34\\
1010: $H_{min}$ & 1.43 & 0.99 & 1.16 & 0.48 & 1.16 & 0.50 & 0.18 & 0.40 & 0.67 & 1.79 & 1.10 & 0.88 & 0.91 & 0.44 & 0.88 & 0.65 & 0.23\\
1011: $H_{max}$ & 2.00 & 2.69 & 1.29 & 0.61 & 1.41 & 1.11 & 1.85 & 7.10 & 5.57 & 10.04 & 2.27 & 1.07 & 1.46 & 0.77 & 1.15 & 1.07 &
1012: 0.56\\
1013: \\
1014: sel30 & 1.11\footnotemark[1] & 0.89\footnotemark[1] & 1.20 & 1.05\footnotemark[2] & 1.28 & 0.56 & 0.51 & 0.43 & 1.73 & 2.73 & 1.15 & 1.08\footnotemark[2] & 1.13 & 0.66 & 0.71\footnotemark[1] & 1.10\footnotemark[2] & 1.04\footnotemark[2]\\
1015: sel31 & 0.96\footnotemark[1] & 0.34\footnotemark[1] & 1.39\footnotemark[2] & 0.89\footnotemark[2] & 1.30 & 0.84 & 1.10 & 0.42 & 0.80 & 0.32\footnotemark[1] & 0.90\footnotemark[1] & 1.06 & 1.15 & 1.23\footnotemark[2] & 0.97 & 0.63\footnotemark[1] & 3.01\footnotemark[2]\\
1016: sel32 & 0.96\footnotemark[1] & 0.67\footnotemark[1] & 1.26 & 0.96\footnotemark[2] & 1.16 & 0.65 & 0.23 & 0.16\footnotemark[1] & 0.63\footnotemark[1] & 0.64\footnotemark[1] & 1.31 & 1.11\footnotemark[2] & 1.13 & 1.02\footnotemark[2] &
1017: 0.69\footnotemark[1] & 0.90 & 1.14\footnotemark[2]\\
1018: sel33 & 1.14\footnotemark[1] & 0.77\footnotemark[1] & 0.96\footnotemark[1] & 0.52 & 1.21 & 0.53 & 0.79 & 1.17 & 2.41 & 3.50 & 1.07\footnotemark[1] & 1.00 & 1.08 & 0.85\footnotemark[2] & 0.83\footnotemark[1] & 1.00 & 0.76\footnotemark[2]\\
1019: sel34 & 1.87 & 3.04\footnotemark[2] & 1.33\footnotemark[2] & 1.22\footnotemark[2] & 1.15\footnotemark[1] & 0.85 & 0
1020: .40 & 1.00 & 1.16 & 4.12 & 2.13 & 1.11\footnotemark[2] & 1.12 & 0.41\footnotemark[1] & 0.77\footnotemark[1] & 0.67 &
1021: 0.69\footnotemark[2]\\
1022: sel35 & 1.12\footnotemark[1] & 0.52\footnotemark[1] & 1.24 & 0.66\footnotemark[2] & 1.12\footnotemark[1] & 0.44\footnotemark[1] & 1.72 & 1.36 & 0.83 & 0.99\footnotemark[1] & 1.02\footnotemark[1] & 0.97 & 0.97 & 1.02\footnotemark[2] &
1023: 1.05 & 1.07 & 6.45\footnotemark[2]\\
1024: sel36 & 1.31\footnotemark[1] & 0.62\footnotemark[1] & 1.12\footnotemark[1] & 0.51 & 1.26 & 0.60 & 2.35\footnotemark[2] & 2.88 & 1.45 & 1.52\footnotemark[1] & 1.03\footnotemark[1] & 1.01 & 1.08 & 0.93\footnotemark[2] & 0.99 & 0.89 &
1025: 2.08\footnotemark[2]\\
1026: sel37 & 0.92\footnotemark[1] & 0.71\footnotemark[1] & 1.26 & 0.87\footnotemark[2] & 1.11\footnotemark[1] & 0.78 & 0.71 & 0.58 & 1.19 & 1.07\footnotemark[1] & 1.11 & 1.17\footnotemark[2] & 1.07 & 0.56 & 0.75\footnotemark[1] & 0.64\footnotemark[1] & 3.30\footnotemark[2]\\
1027: sel38 & 0.91\footnotemark[1] & 0.34\footnotemark[1] & 1.27 & 0.65\footnotemark[2] & 1.03\footnotemark[1] & 0.50 & 0.65 & 0.34\footnotemark[1] & 0.37\footnotemark[1] & 0.25\footnotemark[1] & 1.15 & 1.08 & 1.12 & 1.33\footnotemark[2] &
1028: 0.89 & 1.03 & 2.71\footnotemark[2]\\
1029: sel39 & 0.81\footnotemark[1] & 0.11\footnotemark[1] & 1.23 & 0.72\footnotemark[2] & 1.17 & 0.58 & 0.80 & 0.12\footnotemark[1] & 1.53 & 0.28\footnotemark[1] & 0.97\footnotemark[1] & 0.97 & 0.99 & 2.93\footnotemark[2] & 0.93 & 0.89 &
1030: 2.44\footnotemark[2]\\
1031: sel40 & 1.66 & 0.81\footnotemark[1] & 1.14\footnotemark[1] & 0.55 & 1.19 & 0.43\footnotemark[1] & 0.12\footnotemark[1] & 0.18\footnotemark[1] & 0.20\footnotemark[1] & 0.38\footnotemark[1] & 1.03\footnotemark[1] & 1.01 & 0.93 & 0.79\footnotemark[2] & 0.94 & 1.30\footnotemark[2] & 3.43\footnotemark[2]\\
1032: sel41 & 1.14\footnotemark[1] & 0.48\footnotemark[1] & 1.18 & 0.70\footnotemark[2] & 1.22 & 0.56 & 0.21 & 0.15\footnotemark[1] & 0.80 & 0.68\footnotemark[1] & 0.91\footnotemark[1] & 1.04 & 1.06 & 1.05\footnotemark[2] & 0.84\footnotemark[1] & 0.96 & 1.53\footnotemark[2]\\
1033: sel42 & 1.10\footnotemark[1] & 1.81 & 1.16 & 0.51 & 1.31 & 1.01 & 0.95 & 3.40 & 1.62 & 2.89 & 1.63 & 1.09\footnotemark[2] & 1.26 & 0.43\footnotemark[1] & 1.06 & 0.66 & 0.95\footnotemark[2]\\
1034: sel43 & 1.69 & 3.04\footnotemark[2] & 1.24 & 0.77\footnotemark[2] & 1.26 & 0.68 & 0.06\footnotemark[1] & 0.23\footnotemark[1] & 0.11 & 0.48\footnotemark[1] & 2.79\footnotemark[2] & 1.12\footnotemark[2] & 1.08 & 0.56 & 0.77\footnotemark[1] & 0.89 & 2.23\footnotemark[2]\\
1035: sel44 & 1.18\footnotemark[1] & 0.18\footnotemark[1] & 1.52\footnotemark[2] & 0.43\footnotemark[1] & 1.02\footnotemark[1] & 0.34\footnotemark[1] & 0.59 & 0.25\footnotemark[1] & 1.08 & 0.58\footnotemark[1] & 0.91\footnotemark[1] & 0.92 &
1036: 0.90\footnotemark[1] & 2.25\footnotemark[2] & 1.46\footnotemark[2] & 1.33\footnotemark[2] & 4.12\footnotemark[2]\\
1037: sel45 & 0.92\footnotemark[1] & 0.42\footnotemark[1] & 1.16 & 0.73\footnotemark[2] & 1.37 & 0.68 & 1.46 & 0.85 & 1.14 & 0.71\footnotemark[1] & 0.97\footnotemark[1] & 1.05 & 1.11 & 0.98\footnotemark[2] & 0.88 & 0.79 & 1.71\footnotemark[2]\\
1038: sel46 & 0.94\footnotemark[1] & 0.43\footnotemark[1] & 1.05\footnotemark[1] & 0.71\footnotemark[2] & 1.12\footnotemark[1] & 0.55 & 1.35 & 0.82 & 1.59 & 1.26\footnotemark[1] & 1.01\footnotemark[1] & 0.99 & 1.01 & 0.99\footnotemark[2] & 0.85\footnotemark[1] & 1.01 & 3.44\footnotemark[2]\\
1039: sel47 & 1.54 & 2.07 & 1.19 & 0.54 & 1.36 & 0.57 & 0.16\footnotemark[1] & 0.63 & 0.14\footnotemark[1] & 0.49\footnotemark[1] & 1.60 & 0.97 & 0.97 & 0.45 & 0.96 & 1.02 & 2.85\footnotemark[2]\\
1040: sel48 & 0.84\footnotemark[1] & 0.30\footnotemark[1] & 1.23 & 1.08\footnotemark[2] & 1.14\footnotemark[1] & 1.00 & 0.91 & 0.26\footnotemark[1] & 1.36 & 0.41\footnotemark[1] & 0.84\footnotemark[1] & 1.24\footnotemark[2] & 1.42 & 1.49\footnotemark[2] & 0.68\footnotemark[1] & 0.74 & 1.75\footnotemark[2]\\
1041: sel49 & 0.93\footnotemark[1] & 0.33\footnotemark[1] & 1.17 & 0.83\footnotemark[2] & 1.16 & 0.50 & 1.27 & 0.50 & 1.08 & 0.71\footnotemark[1] & 0.86\footnotemark[1] & 1.15\footnotemark[2] & 0.96 & 1.21\footnotemark[2] & 0.71\footnotemark[1] & 1.11\footnotemark[2] & 3.96\footnotemark[2]\\
1042: sel50 & 1.32\footnotemark[1] & 0.59\footnotemark[1] & 1.28 & 0.46\footnotemark[1] & 1.21 & 0.32\footnotemark[1] & 1.78 & 2.31 & 1.21 & 2.26 & 1.07\footnotemark[1] & 1.00 & 0.91 & 0.93\footnotemark[2] & 1.20\footnotemark[2] & 1.62\footnotemark[2] & 5.21\footnotemark[2]\\
1043: sel51 & 1.83 & 0.72\footnotemark[1] & 1.14\footnotemark[1] & 0.42\footnotemark[1] & 1.24 & 0.66 & 0.16\footnotemark[1] & 0.27\footnotemark[1] & 0.30\footnotemark[1] & 0.33\footnotemark[1] & 1.30 & 1.04 & 1.00 & 1.05\footnotemark[2] &
1044: 1.24\footnotemark[2] & 0.90 & 1.83\footnotemark[2]\\
1045: sel52 & 1.40\footnotemark[1] & 0.73 & 1.32\footnotemark[2] & 1.02\footnotemark[2] & 1.29 & 1.01 & 0.14\footnotemark[1] & 0.10\footnotemark[1] & 0.42\footnotemark[1] & 0.31\footnotemark[1] & 1.51 & 1.13\footnotemark[2] & 1.17 & 1.02\footnotemark[2] & 0.73\footnotemark[1] & 0.67 & 1.66\footnotemark[2]\\
1046: sel17152 & 1.06\footnotemark[1] & 0.93\footnotemark[1] & 1.31\footnotemark[2] & 0.58 & 1.13\footnotemark[1] & 0.54 & 0.06\footnotemark[1] & 0.10\footnotemark[1] & 0.23\footnotemark[1] & 0.40\footnotemark[1] & 1.68 & 1.01 & 1.03 & 0.91\footnotemark[2] & 1.01 & 0.97 & 1.15\footnotemark[2]\\
1047: $min$ & 0.81 & 0.11 & 0.96 & 0.42 & 1.02 & 0.32 & 0.06 & 0.10 & 0.11 & 0.25 & 0.84 & 0.92 & 0.90 & 0.41 & 0.68 & 0.63 & 0.69\\
1048: $max$ & 1.87 & 3.04 & 1.52 & 1.22 & 1.37 & 1.01 & 2.35 & 3.40 & 2.41 & 4.12 & 2.79 & 1.24 & 1.42 & 2.93 & 1.46 & 1.62 & 6.45\
1049: \
1050: %\\
1051: %\hline
1052:
1053: \end{tabular}
1054: \footnotetext[1]{These values are smaller than the $H_{min}$ given
1055: in each column}
1056: \footnotetext[2]{ These values are larger than the $H_{max}$
1057: given in each column}
1058: \footnotetext[3]{These values do not fully coincide with those given
1059: in Ref.\cite{ref8} for the reasons discussed in the Appendix}
1060: \end{ruledtabular}
1061: \end{table*}
1062: \end{turnpage}
1063:
1064:
1065: \squeezetable
1066: \begin{table}
1067: \caption{The number of SD and patients that can be distinguished from H when
1068: using $\lambda_\kappa$(RR) or $\lambda_{\kappa,shuf}$(RR) alone.} \label{tab2}
1069: \begin{ruledtabular}
1070: \begin{tabular}{c c c c c }
1071: % \hline
1072:
1073: % after \\: \hline or \cline{col1-col2} \cline{col3-col4} ...
1074: group & Total number & $\lambda_\kappa$(RR) & $\lambda_{\kappa,shuf}$(RR) & $\lambda_\kappa$(RR) and $\lambda_{\kappa,shuf}$(RR) \\
1075: \hline
1076: SD & 24 & 23 & 10 & 24
1077: \\
1078: MIT & 15 & 14 & 6 & 14 \\
1079: MSV & 13 & 13 & 2 & 13 \\
1080: EST & 33 & 29 & 8 & 29 \\
1081: MST & 6 & 5 & 0 & 5 \\
1082: \end{tabular}
1083: \end{ruledtabular}
1084: \end{table}
1085:
1086:
1087: \begin{turnpage}
1088: \squeezetable
1089: \begin{table*}
1090: \caption{The confidence levels to distinguish SD from either H or
1091: patients when considering the estimation errors $\epsilon_m$ discussed in the Appendix and given
1092: in Table VIII of Ref.\cite{epaps}} \label{aca7}
1093: \begin{ruledtabular}
1094: \begin{tabular}{cccc ccc cccc}
1095: % \hline
1096:
1097: % after \\: \hline or \cline{col1-col2} \cline{col3-col4} ...
1098: Method Employed & & & & Confidence levels to distinguish SD & &
1099: & & & & \\
1100: \cline{1-4}
1101: \cline{5-11}
1102: \\
1103: & & & & Using the limits & & & & Using &
1104: & \\
1105: & & & & from the data analyzed & & & & broader limits\footnotemark[3] &
1106: & \\
1107: \cline{5-7}
1108: \cline{8-11}
1109: \\
1110: Aim & Measures & Type of & No. of & All SD & All but & All but &
1111: All SD & All but & All but & All but
1112: \\
1113: & & intervals & para- & & one SD &
1114: two SD\footnotemark[4] & & one SD & two SD &
1115: five SD\footnotemark[4] \\
1116:
1117: & & & meters & \% & \% &
1118: \% & \% & \% & \% & \%
1119: \\
1120: \hline Distinction & $\lambda$, $\rho$ & RR, QRS, QT & 10 & $>$99
1121: & $>$99 & $>$99 &
1122: 88 & 99 & $>$99 & $>$99 \\
1123: of SD & $\lambda$, $\rho$ & RR, QRS & 4 & 63 & 95 & $>$99 & 8 & 43 & 90 & $>$99 \\
1124: from H & $\lambda$, $\lambda_{shuf}$ & RR & 4 & 49 & 90
1125: & 99 & 1 & 11 & 36 & 97 \\
1126: & $\nu$ & RR,QRS & 4 & 32 & 74 & 96 & $<$0.5 & 1 & 8 & 60 \\
1127: & $\delta S_{3-4}(QT)$ & QT & 1 & 59 & 93 & $>$99 & 11 & 39 & 77 & $>$99 \\
1128: & $\lambda$, $\rho$,$\lambda_{sh}$, $\rho_{sh}$, $\nu$, & RR, QRS, QT & 28 & $>$99
1129: & $>$99 & $>$99 &
1130: $>$99 & $>$ 99 & $>$99 & $>$99 \\
1131: &$\delta S_{3-4}(QT)$,$\delta S_{sh, 3-4}(QT)$ & & & & & & & & & \\
1132: \\
1133: Distinction & $\lambda$, $\rho$, $\nu$, $\delta
1134: S_{3-4}(QT)$\footnotemark[1] & RR, QRS, QT & 17 & 51 &
1135: 83 & 95 & $<$0.1 & $<$0.1 & $<$0.1 & 1 \\
1136: of SD & $\lambda$, $\rho$, $\lambda_{sh}$, $\delta
1137: S_{3-4}(QT)$\footnotemark[1] &
1138: RR, QRS, QT & 17 & 62 & 91 & 98 & $<$0.1& $<$0.1&$<$0.1 & 1 \\
1139: from patients & $\lambda$, $\rho$, $\lambda_{sh}$, $\rho_{sh}$,
1140: $\nu$, & RR, QRS, QT & 28 & 95 & $>$99 &$>$99 & 16 & 41
1141: & 68 & 98 \\
1142: & $\delta S_{3-4}(QT)$, & & & & & & & & & \\
1143: & $\delta S_{sh, 3-4}(QT)$\footnotemark[2] & & & & & & & & & \\
1144:
1145: \end{tabular}
1146: \footnotetext[1]{Considering the limits of those patients that
1147: have $\delta S_{3-4}(QT)$ larger than those in H}
1148: \footnotetext[2]{Considering the limits of those patients that
1149: have {\em both} $\delta S_{3-4}(QT)$ and $\delta S_{sh, 3-4}(QT)$
1150: larger than those in H} \footnotetext[3]{by amounts $\epsilon_m$
1151: given in Table VIII of Ref. \cite{epaps}} \footnotetext[4]{When stating, e.g., ``All but
1152: one'', it means that when allowing {\em at the most}, one SD -out
1153: of 24- to be misinterpreted as being H or patient, respectively}
1154: \end{ruledtabular}
1155: \end{table*}
1156: \end{turnpage}
1157:
1158: \begin{figure}
1159: \includegraphics{Figure6}
1160: \caption{\label{fig6} (color) The $\overline{\delta S}_{3-4}$(QT) values along with those of $\lambda_s$(RR)-
1161: and $\lambda_L$(RR)- for SD (red) and H (black).
1162: The individual sel33, who is discussed as an example in the text, is marked with a green column. }
1163: \end{figure}
1164:
1165:
1166:
1167: \end{document}
1168:
1169: