Nonlinear analysis of the heartbeats in public patient ECGs using an automated PD2i algorithm for risk stratification of arrhythmic death

Figures & data

Figure 1 Relative risk (RR)-intervals and associated PD2i’s determined from the electrocardiogram (ECG) of a normal healthy person and two types of PhysioBank patients. A. Shows RR data and corresponding PD2i results from a normal healthy subject (15 min ECG). B. Shows results from an arrhythmia control patient (30 min ECG, PhysioBank). C. Shows results from a VF/VT patient (30 min ECG, PhysioBank). PD2i values are expressed in degrees of freedom (dimensions) and RR values in msec; VF is included in the analysis for illustrative purposes only. Parts A-C each show 4 individual plots for comparing the data and the analytic results: upper left the RR-intervals (RRi), lower left the corresponding PD2i, upper right the joint plot of RRi and PD2i, lower right the histogram of the accepted PD2i values with its associated statistics, including %N. The ECG for the RRi shown in Part C. manifests VF at 17 minutes from the start and the PD2i before VF shows repeated low-dimensional excursions fall below 1.4 (line). The ECG shown in Part B. did not manifest any PD2i below 1.4 throughout the 30-min period. The test result indicated in the upper right plot (underlined) was completely automated by the Vicor 2.0 software once the input file was entered.

Figure 2 PD2i traces from arrhythmia controls and malignant arrhythmia subjects for the first half of the data set. PD2i values are expressed in degrees of freedom (dimensions) and the heartbeats are those obtained in a 15- to 30-min ECG up to the end or point of VF/VT onset. Note the low-dimensional excursions of the PD2i in the VF/VT subjects. The horizontal a priori criterion lines are set at 1.4 degrees of freedom and completely separate the two groups of patients. The rejected PD2i points are due to noise produced by arrhythmias and movement artifacts, but %N remains high enough in these patients for valid analysis. Surrogate- and %N-rejected files (n = 11) are not shown as they were a priori rejected (5 VF/VT, 6 controls). These rejected files contained all subjects with atrial fibrillation and the highest arrhythmia rates (ie, greater than 10% arrhythmias in all beats).

Table 1 PD2i of heartbeats predicts VF/VT in PhysioBank ECGs^a

Coded file	MIT-BIH	PD2i test	VF/VT outcome*	Surrogate test	%N test
1	100	NEG	TN	p < 0.01	86
2	422	POS	TP	p < 0.01	50
3	419	POS	TP	p < 0.01	32
4	107	NEG	TN	p < 0.01	92
5	200	Rej	Exc	ns	17
6	424	Rej	Exc	ns	22
7	611	POS	TP	p < 0.01	78
8	114	NEG	TN	p <0.01	67
9	430	POS	TP	p < 0.01	31
10	426	POS	TP	p < 0.01	41
11	119	Rej	Exc	ns	23
12	602	POS	TP	p < 0.01	33
13	607	POS	TP	p < 0.01	60
14	228	NEG	TN	p < 0.01	37
15	425	Rej	Exc	ns	23
16	104	NEG	TN	p < 0.01	68
17	418	POS	TP	p < 0.01	36
18	428	POS	TP	p < 0.01	87
19	101	NEG	TN	p < 0.01	64
20	421	POS	TP	p < 0.01	67
21	420	POS	TP	p < 0.01	76
22	106	NEG	TN	p < 0.01	33
23	121	NEG	TN	p < 0.01	81
24	610	POS	TP	p < 0.01	33
25	605	Rej	Exc	ns	27
26	231	POS	FP	p < 0.01	54
27	217	NEG	TN	p < 0.01	61
28	201	Rej	Exc	ns	13
29	423	POS	TP	p < 0.01	33
30	429	POS	TP	p < 0.01	38
31	113	NEG	TN	p < 0.01	39
32	116	POS	FP	p < 0.01	92
33	203	Rej	Exc	Ns	22
34	124	NEG	TN	p < 0.01	52
35	427	POS	TP	p < 0.01	68
36	612	POS	TP	p < 0.01	46
37	614	Rej	Esc	ns	23
38	221	Rej	Exc	ns	11
39	213	Rej	Exc	ns	3
40	615	Rej	Exc	ns	17
41	223	NEG	TN	p < 0.01	70
42	233	POS	FP	p < 0.01	33
43	609	POS	TP	p < 0.01	45
44	103	NEG	TN	p < 0.01	82

Notes: Sensitivity = 100%; specificity = 85%; p < 0.01 (Fisher Exact); Coded File: coded file names, as shown in Figure 2, to cross reference with PhysioBank file names; MIT-BIH, PhysioBank file names for the ECG data (ie, from MIT and Beth Israel Hospital archives); PD2i Test, analytic result, where NEG (negative) indicates PD2i > 1.4, POS (positive) indicates PD2i ≤ 1.4, and Rej (rejection) indicates data rejected by the %N criterion; all analytic results were automated by the Vicor 2.0 software; VF/VT Outcome, given the VF/VT or no VF/VT outcome by PhysioBank, the PD2i predicted a True Negative (TN), a True Positive (TP), a False Negative (FN; none indicated), or a False Positive (FP); Exc indicates the prediction could not be made because the Surrogate and %N Tests together could not affirm that the data were noise-free; Surrogate Test, probability that the mean PD2i of the data and the mean PD2i of the randomized-phase surrogate were the same, as tested by a 2-tailed t-test; ns indicates they were not statistically significantly different, and therefore the data could not be presumed to be noise-free; %N Test, % of accepted PD2i values, where values less than 30 lead to rejection of the data by the Vicor 2.0 software for a valid PD2i analysis.