Effect of ECG quality measures on piecewise-linear trend detection for telehealth decision support systems

Abstract

Fledgling clinical decision support systems (DSSs) are being designed on the false assumption that consistent, good-quality signals are created in the unsupervised telehealth environment, but it has in fact been shown that signal quality is often very poor. Hence, it is important to investigate the detrimental impact of failing to recognize erroneous clinical parameter values. This study combines previous work in this area, related to artifact detection in electrocardiogram (ECG) signals, and piecewise-linear trend detection in longitudinal heart rate parameter records, to investigate the impact of choosing to ignore ECG signal quality prior to trend detection in the heart rate (HR) records. Using an artifact detection algorithm to improve the HR estimates from the ECG signals, when compared to reference HR values derived from human annotated 2453 ECGs from nine patients, resulted in a decrease in the estimation bias from 2.54 BPM (beat per minute) to 0.70 BPM and a decrease in the standard error from 0.47 BPM to 0.17 BPM. The application of the same artifact detection also results in a significant improvement in trend fitting, when compared to a fitting of the reference HR values, by reducing the mean RMSE value of the error in the trend fit from 2.14 BPM to 0.78 BPM and standard error from 0.49 BPM to 0.10 BPM. As trend detection will be a component of future telehealth decision support systems, signal quality measures for unsupervised measurements are of paramount importance.