In a bid to improve this situation, a team based at various institutes in Virginia, US, has published details of an algorithm that significantly improves the detection of central apnea. The algorithm has now analysed about 100 baby-years of data and can improve the central apnea detection rate, as well as markedly reduce the number of false positives compared with current methods (Physiol. Meas. 33 1).

"In infants, apneas lasting more than 30 seconds, accompanied by loss of oxygen in the blood and slowing of the heart are called 'extreme events'," John Delos, from the College of William and Mary (Williamsburg, VA), told medicalphysicsweb. "We found that about 14% of these extreme events are entirely missed by today's monitors. Our algorithm misses about 5% of apneas. Our false alarm rate is half what the present monitors give and we hope to further reduce that."

Confusing the detector

Respiration is currently measured by monitoring the chest impedance (CI). Using the same leads that monitor the electrocardiogram, a small high-frequency voltage is applied, and the resulting high-frequency current is measured. The CI is the applied voltage divided by the observed current. As the infant inhales, its chest fills with air, which has a low electrical conductivity. In turn, this increases the CI reading and provides a way in which to monitor respiration cycles.

The problem is that the beating heart also causes rhythmic changes in CI, which creates a cardiac artefact large enough to confuse the monitor's apnea alarm. Current monitors employ a breath-detection algorithm that marks the time of each inhale and exhale, and then computes the respiration rate.

"When an infant stops breathing, the time since the last breath steadily increases and should set off an alarm," explained Delos. "In practice, the monitor sometimes confuses the heartbeat with respiration. During apnea, the heart slows down, and the rate might come down close to the normal respiration rate. Our method is better at filtering out this problematic cardiac artefact."

Tackling the cardiac artefact

The key to more effectively filtering out the cardiac artefact in the CI signal is changing the unit of time to the interval between heartbeats. "Digitally we make a new timescale for the CI," commented Delos. "One 'new-time' unit is always one heartbeat. As the heart rhythm changes, our clock also changes, always keeping exact time with the heartbeat. Therefore in our 'new-time', the heart has a perfectly regular rhythm that the computer can recognise and filter out, leaving just the breathing signal."

The team initially studied data from five very low birth-weight babies in order to optimize the way in which the cardiac artefact was filtered out of the CI signal. The algorithm was meticulously validated by comparing its output with the verdict of three expert clinicians. The researchers randomly selected 237 episodes of coexisting bradycardia and oxygen desaturation, and calculated the probability of apnea from analysis of the filtered CI signal. The clinicians and the algorithm agreed in 212 cases, giving an accuracy of 91%.

"We have subsequently analysed about 100 baby-years of data, which equates to about 40 babies over a two and a half year period," said Delos. "Of that, we have picked out around 1000 apnea events for validation and further study."

Although the effectiveness of the algorithm is not in doubt, the main limiting factor is that it can only be applied retrospectively. Work is now ongoing to optimize real-time operation. "The hardest part will be getting digital data immediately," said Delos.