Automated QT interval measurement in Holter ECGs recorded at 180 and 1000 samples/second

Abstract

To study the effect of sampling rate on automated QT measurements, Holter ECGs were recorded at 180 and 1000 samples/second (s/s) using 2 recorders; 30 ECG snapshots were extracted at varying heart rates from 16 healthy subjects and re-sampled to 180, 500 or 1000s/s using the Antares software. QT interval by CalECG algorithm was longer (5.0±6.3 ms, p<;0.001) in 180s/s ECGs than in 1000s/s ECGs. It decreased to 2.1±5.8ms when 180s/s ECGs were re-sampled to 500s/s, and to 2.6±6.2ms at 1000s/s. It also decreased progressively on resampling both sets of ECGs to 1000s/s (2.6±6.2ms), 500s/s (1.8±5.5ms) and 180s/s (0.4±5.9ms). Differences in QT interval were independent of the QT measurement algorithm used: University of Glasgow (Uni-G) program and CalECG for 500s/s ECGs; Veritas and CalECG for 1000s/s ECGs. Thus, QT interval is longer in ECGs with lower sampling rates; resampling them to a higher resolution partially compensates for this.

Full text

Automated QT Interval Measurement in Holter ECGs Recorded at 180 and 1000
samples/second
GK Panicker
1
, V Salvi
1
, DR Karnad
1
, P Macfarlane
2
, E Clark
2
, A Ramasamy
1
, S Kothari
1
, D Narula
1
1
Quintiles Cardiac Safety Services, Mumbai, India
2
University of Glasgow, Glasgow, UK
Abstract
To study the effect of sampling rate on automated QT
measurements, Holter ECGs were recorded at 180 and
1000 samples/second (s/s) using 2 recorders; 30 ECG
snapshots were extracted at varying heart rates from 16
healthy subjects and re-sampled to 180, 500 or 1000s/s
using the Antares software. QT interval by CalECG
algorithm was longer (5.0±6.3 ms, p<0.001) in 180s/s
ECGs than in 1000s/s ECGs. It decreased to 2.1±5.8ms
when 180s/s ECGs were re-sampled to 500s/s, and to
2.6±6.2ms at 1000s/s. It also decreased progressively on
resampling both sets of ECGs to 1000s/s (2.6±6.2ms),
500s/s (1.8±5.5ms) and 180s/s (0.4±5.9ms). Differences
in QT interval were independent of the QT measurement
algorithm used: University of Glasgow (Uni-G) program
and CalECG for 500s/s ECGs; Veritas and CalECG for
1000s/s ECGs. Thus, QT interval is longer in ECGs with
lower sampling rates; resampling them to a higher
resolution partially compensates for this.
1. Introduction
Digital 12-lead resting ECGs are used in clinical
research or drug trials for studying changes in various
intervals in the ECG. Regulatory guidelines require that
studies designed to detect QTc prolongation by a new
drug are able to detect a mean prolongation of 5
milliseconds (ms).
1
Electrocardiographs with a sampling
rate of 500 or 1000 s/s are used for this purpose.
Improvements in acquisition and storage technology have
permitted recording of longer durations of continuous 12-
lead Holter ECG recordings at sampling rates of up to
1000 samples per second (s/s). However, due to cost
constraints, 12-lead Holters with lower sampling rates are
still used in many studies.
Holter ECGs recorded at 180 s/s have data points that
are 5.6 ms apart. Is the QT interval in these ECGs
comparable with that in Holter ECGs recorded at 1000
s/s where data points are 1 ms apart? This question is
more pertinent when a computer algorithm is used for
measurement of various intervals as automated QT
measurement algorithms place annotations on sample
points and not between them. Moreover, many computer
programs can only analyze ECGs at a specified sampling
rate. Consequently, digital 12-lead Holter recordings
acquired at lower sampling rates are often up-sampled to
a higher sampling rate, before further analysis. This
involves interpolation of data values between actual
samples. How up-sampling affects automated QT
measurement in digital ECGs acquired at a lower
sampling rate is not clear. We, therefore, studied QT
interval measurements in Holter ECGs recorded at 180
and 1000 s/s with and without resampling.
2. Material and methods
Two 12-lead Holter recorders (Model H12+, Mortara
Instrument, Milwaukee, WI) were connected using dual-
snap electrodes and 5 hour recordings acquired
simultaneously from 16 healthy volunteers. One Holter
device recorded the digital ECG signal at a sampling
frequency of 180 s/s and the other at 1000 s/s, with a 16-
bit amplitude resolution (2.5 µV). 10-second ECG
snapshots were extracted from the simultaneous Holter
recordings at 30 identical time-points from each subject.
Snapshots were extracted at heart rates between 50-60
bpm, 61-70 bpm, 71-80 bpm, 81-90 bpm, 91-100 bpm
and ≥101 bpm. Thus, 480 ECGs at a sampling rate of 180
s/s and 480 simultaneous ECGs at a sampling rate of
1000 s/s from 16 subjects were obtained.
ECG resampling
ECGs recorded at 180 s/s were up-sampled to 500 s/s
and 1000 s/s and those recorded at 1000 s/s were down-
sampled to 500 s/s and 180 s/s using commercially
available software (Antares version 2.2.3, AMPS LLC,
New York)
2
and converted to HL7 compliant XML files.
Thus, six sets of ECGs were created – 180 s/s without
resampling; 180 s/s resampled to 500 s/s, 180 s/s
resampled to 1000 s/s and 1000 s/s without resampling,
1000 s/s resampled to 500 s/s and 1000 s/s resampled to
180 s/s. ECGs recorded at 180 s/s were also up-sampled
to 1000 s/s using another software application (H-Scribe,
Version 4.3, Mortara Inc), thereby creating seven sets.
ISSN 0276−6574 761 Computing in Cardiology 2010;37:761−764.

762

Table 1. Comparison of QT intervals measured by the CalECG algorithm in 12-lead digital Holter ECGs recorded at 180
s/s resampled to 500 and 1000 s/s compared to the ‘gold standard’ i.e. ECGs recorded at 1000 s/s

ECG
interval
Comparison between ECG sets
Mean
Standard
Deviation
Min
Diff
Max
Diff
LOA
Range
of LOA
P value
QT
180 s/s vs 1000 s/s
5.0
6.3
-35.2
23.1
-7.6 to 17.7
25.3
<0.0001
180 s/s@500s/s vs 1000s/s
2.1
5.8
-35
25
-9.5 to 13.8
23.3
<0.0001
180 s/s@1000s/s vs 1000s/s
2.6
6.2
-37
38
-9.8 to 15.0
24.8
<0.0001
QRS
180 s/s vs 1000 s/s
4
5.2
-15.7
33.7
-6.4 to 14.3
20.7
<0.0001
180 s/s @ 500 s/s vs 1000 s/s
2.1
4.2
-16
17
-6.3 to 10.5
16.8
<0.0001
180 s/s @ 1000 s/s vs 1000 s/s
2.2
4.6
-16
20
-7 to 11.3
17.3
<0.0001
JT
180 s/s vs 1000 s/s
1.1
5.1
-21.8
17.9
-9.1 to 11.3
20.4
<0.0001
180 s/s @ 500 s/s vs 1000 s/s
0
4.7
-22
30
-9.3 to 9.4
18.7
0.88
180 s/s @ 1000 s/s vs 1000 s/s
0.4
4.5
-21
28
-8.7 to 9.5
18.2
0.04
All values in milliseconds. Min: Minimum, Max: Maximum
Table 2. Comparison of QT intervals measured by the CalECG algorithm in 12-lead digital Holter ECGs recorded at 180
s/s and 1000 s/s and resampled to identical sampling rates of 1000 s/s, 500 s/s and 180 s/s
Comparison between ECG sets
Mean
Standard
Deviation
Min
Max
LOA
Range
of LOA
P value
180 s/s@1000s/s vs 1000s/s
2.6
6.2
-37
38
-9.8 to 15.0
24.8
<0.0001
180 s/s @ 500 s/s vs 1000 s/s@ 500 s/s
1.8
5.5
-16
32
-9.3 to 12.9
22.1
<0.0001
180 s/s vs 1000 s/s @180 s/s
0.4
5.9
-22.2
22.2
-11.3 to 12.2
23.5
0.10
Table 3. Comparison of QT interval measurements re-sampled to the common sampling rate of 500 s/s (by CalECG and
Uni-G) and 1000 s/s (by CalECG and Veritas)
ECG sampling rates
QT measurement
software used
Mean
Standard
Deviation
LOA
Range
of LOA
P value
180 s/s@500s/s vs 1000 s/s @ 500 s/s
CalECG
5.0
6.3
-7.6 to 17.7
25.3
<0.0001
Uni-G
2.1
5.8
-9.5 to 13.8
23.3
<0.0001
180 s/s @ 1000 s/s vs 1000 s/s
CalECG
4
5.2
-6.4 to 14.3
20.7
<0.0001
Veritas
2.1
4.2
-6.3 to 10.5
16.8
<0.0001
4. Discussion
Using the Holter ECGs acquired at 1000 s/s as the
gold standard, we found that the mean automated QT
interval measurement in corresponding ECGs recorded at
180 s/s was greater than the gold standard by 5.0 ms.
This difference decreased to 2.1 ms on up-sampling the
180 s/s ECGs to 500 s/s and to 2.6 ms at 1000 s/s. In
order to identify why QT measurements are longer in
ECGs recorded at 180 s/s than in corresponding ECGs
acquired at 1000 s/s, we compared the QRS duration and
JT interval in the same sets of ECGs. While the JT
intervals were comparable in ECGs recorded at 180 s/s
and 1000 s/s, the QRS duration was greater in the 180 s/s
ECGs by a mean of 4 ms, suggesting that the difference
in the QT intervals was almost entirely accounted for by
the QRS duration and not the JT interval.
Previous studies have shown that sampling rate
significantly influences the amplitude of high-frequency
components of the ECG waveform; the QRS amplitude is
lower in ECGs recorded at lower sampling rates.
6,7
The
present study revealed that a lower sampling rate also
affects the duration of high frequency components of the
ECG waveform like the QRS complex; the QRS duration
in ECGs acquired at 180 s/s was 5 ms longer than that in
ECGs acquired at 1000 s/s. One possible explanation for
this is that automated algorithms can place fiducial points
only on sampling points (Figure 2).
3
Since the QRS onset
is identified as the last data point on the PR interval and
QRS offset as the first data point on the ST segment,
these will be further apart on ECGs recorded at 180 s/s
(Figure 3).
ECGs acquired at different sampling rates may also
have to be re-sampled to a common rate because
automated algorithms are programmed to perform at a
specific sampling rate. The Uni-G algorithm measures
QT interval only at 500 s/s while the Veritas algorithm
measures QT intervals only at 1000 s/s. We found that
the difference between QT intervals in ECGs recorded at
180 s/s and 1000 s/s decreased when both sets were
resampled to the same sampling rate; the difference
decreased progressively from 1000 s/s to 500 s/s to 180
s/s. Again, this is possibly because fiducial points are
placed only on sample points.
3
Therefore, agreement
between the two sets of ECGs is apparently best at 180
s/s where the sample points are 5.6 ms apart rather than
when sample points are 1 or 2 ms apart at 1000s/s or
500s/s respectively. However, it must be remembered
that ECGs at 180 s/s have a longer measured QT interval
than the same ECG recorded at 1000 s/s. Therefore,
763

764