Correlates of physical activity in the ECG

Among the candidates for inclusion in an ECG-based activity index are:

• mean heart rate
• total power of the observed heart rate signal
• ``stationarity'' of the heart rate signal
• ectopic beat rate
• ST segment deviation
• QT interval
• noise content of the ECG

Mean heart rate is the most obvious of these, since it responds directly to changes in the level of physical activity. Many other factors influence mean heart rate, however, so that mean heart rate in isolation is inadequate as an index of activity. It should be noted that systematic detector errors can occasionally result in erroneous measurements of mean heart rate; this observation alone suggests the need for additional components in an activity index.

The total power measurement is strongly influenced by QRS detector errors, which affect virtually all HRV measurements. These errors include not only missed or extra detections, but also mismeasured fiducial points. Any of these errors is most likely to occur when the ECG is noise-corrupted, which in turn is most likely to occur when the subject is relatively active. The inclusion of total power in an activity index is thus motivated primarily by the need to account for imperfect signal processing rather than any intrinsic property of the signal source. In this study, total power is defined as

$$P(t) = \frac{1}{2N} \sum_{n=-N}^{N-1} \left( H(t + n \Delta t) - \overline{H}_{N}(t) \right)^{2}$$ (4)

where $N$, $H(t)$, $\Delta t$, and $\overline{H}_{N}(t)$ are defined as above.

``Stationarity'' of the heart rate signal refers to the likelihood that a trend is present during the window of observation; such a trend is likely to be associated with a change in activity level. For this study, a simple index of stationarity is defined as

$$S(t) = \left\vert \overline{H}_{N/2}(t - \frac{N}{2} \Delta t) - \overline{H}_{N/2}(t + \frac{N}{2} \Delta t) \right\vert$$ (5)

where $N$ and $\Delta t$ are defined as above, and $\overline{H}_{N/2}(t)$ is the mean heart rate over a window only half the length of that in equation (3).

Although changes in ectopic beat rate, ST deviation, and QT interval are often associated with activity, no such changes are observable during the normal activities of many subjects, reducing the value of these indices for the purposes of this study. Similarly, although an increase in the noise content of the ECG is strong evidence for an increase in the level of physical activity, the absence of significant noise may simply indicate good practice in electrode application.

More significantly, it is possible to derive mean heart rate, total power, and stationarity from an RR interval sequence without reference to the original ECG. An activity index based on these three measures may be obtained easily from the computer-readable RR interval files produced by many modern ECG analyzers. The other measures are not always available, and when they are provided tend to be less reliable than are RR interval measurements. For these reasons, the present study focussed on an activity index based on mean heart rate, total power, and stationarity only.