Background and Aims Continuous physiological parameter monitoring is routine in NICU. However, high false alarm rates exist that can lead to inappropriate responses from clinical staff.
At present, parameters are assessed independently to generate alarms. The positive predictive value can be increased in adult patients by combining the physiological parameters using statistical models1.
We have developed a multi-parameter model, designed for sick newborn infants, which produces alarms based on an integrated assessment of patient physiology. We hypothesized that this model would have greater specificity than conventional single channel alerts.
Methods Continuous physiological data (heart rate, respiration rate, oxygen saturation, blood pressure and temperature) were collected from 6 preterm infants, median gestation 26.1 weeks (range 24.3–28.9). The median period of data collection was 16.5 days (range 10.9–23.2). A mathematical model was developed using Matlab, which ‘learnt’ 1 hour of normal data in order to subsequently identify abnormal events in the remaining dataset. Adverse clinical events were identified retrospectively from patient notes.
157 clinical events were obtained for analysis. The proposed model increases specificity of event detection, reduces false positives by a factor of 18 and maintains high sensitivity.
Conclusions This pilot study demonstrates that combining existing physiological data using a multi-parameter model improves the specificity of adverse event detection in the NICU.
1Tarassenko L et al. Br J Anaesth. 2006; 97:64–8.