Signal merging and signal fusion to enhance robustness and reduce false alarms during multichannel patient monitoring

It is recognised that false alarms lead to habituation among clinical staff and to distress for patients. The goal of our work is to reduce the number of false alarms produced by a standard patient monitor and to provide early warning of patient deterioration based on multiple vital signs.

Two approaches have been used. In the first ('signal merging'), multiple signals are merged in order to produce a more robust derived measure. For example, cardiac and respiratory information are extracted from the electrocardiogram and photoplethysmogram, and are combined to produce more robust estimates of heart and respiration rates.

The second ('signal fusion') uses a combination of advanced signal processing methods to 'integrate' multiple vital signs, to develop a global representation of patient status. The hypothesis is that if a monitoring system can identify periods of physiological instability preceding clinically apparent adverse events, then early warning may result in reduced monitoring alarms, prompt treatment and improved outcome.

Over a 3-year period at the John Radcliffe hospital in Oxford, UK, we collected noninvasive physiological data for over 24 hours in 150 'high-risk' patients, including the continuous heart rate, respiratory rate, oxygen saturation, temperature and intermittent noninvasive blood pressure. The system was 'trained' to distinguish periods of physiological stability from instability.

Two evaluation studies have taken place. The first assessed the signal merging technology during simulated patient transport. Results showed that signal merging gives a more precise estimate of respiration rate than a standard transport monitor, suggesting that signal merging may reduce the number of false alarms caused by external movement. The second study assessed the alarms of a standard patient monitor versus our signal fusion technology. Numerous standard monitor alarms were recorded that did not correspond to physiological events, often due to signal artefact or nonclinical events such as patient movement, but typically these did not trigger an alert from our signal fusion technology. Signal fusion identified physiological events that included atrial fibrillation, elevated blood pressure, extreme distress and oxygen desaturation.

The conclusions are that signal merging and fusion techniques may improve robustness, reduce false-positive alarms and potentially provide early warning of adverse physiological events.