433 - Combined Use of Electrocardiogram-derived Heart Rate and Pulse Rate Signals for Wireless Cardiac Monitoring in the Neonatal Intensive Care Unit: A Feasibility and Accuracy Study

Publication Number: 3418.433

Daniel J. Radeschi, McGill University Faculty of Medicine and Health Sciences, Pointe Claire, PQ, Canada; Eva Senechal, McGill University Faculty of Medicine and Health Sciences, Montreal, PQ, Canada; Emily Jeanne, McGill University, Montreal, PQ, Canada; Ana Gabriela Saavedra Ruiz, RI-MUHC, Laval, PQ, Canada; Wissam Shalish, McGill University Health Center, Montreal, PQ, Canada; Robert Kearney, McGill University Faculty of Medicine and Health Sciences, Montreal, PQ, Canada; Guilherme Santanna, McGill University Health Center, MONTREAL, PQ, Canada

Background: In the neonatal intensive care unit (NICU), heart rate (HR) is continuously monitored as a standard of care, using wired sensors placed on the infant and connected to bedside displays. However, these wires may tangle around the infant and impede care, making wireless technology an attractive alternative. In a prospective study, we noted strong accuracy for wireless electrocardiogram-derived HR (ECG-HR) monitoring, but with episodes of Bluetooth disconnections that resulted in HR data loss. To overcome this limitation, we investigated the use of oximeter-derived pulse rate (PR) to replace missing ECG-HR data and increase HR availability.

Objective: To evaluate the feasibility and accuracy of the combined use of ECG-HR and PR signals for wireless cardiac monitoring in the NICU.

Design/Methods: Patients admitted to the NICU at the Montreal Children's Hospital were monitored for eight hours/day for four consecutive days, using both the wireless Anne One system by Sibel Health and a wired standard (Philips Intellivue MX450). Informed consent was obtained from parents prior to enrolment. The Anne One system comprised a chest sensor for ECG-HR monitoring and limb pulse oximeter for PR monitoring (Figure 1). We generated combined ECG-HR & PR records by filling periods of missing data in wireless ECG-HR signals with corresponding wireless PR data. Feasibility of ECG-HR & PR synthesis was assessed as the percentage of recording time for which signals were available. Accuracy was assessed by the bias and 95%-limits of agreement (LoA) between wired ECG-HR and wireless PR in two cases: (1) when wireless ECG-HR was available, and (2) when it was not. Finally, the clinical implications of errors were assessed using a modified Clarke error grid.

Results: Twenty-five patients were enrolled with a median postmenstrual age of 33.3 weeks [IQR: 31.3-36.1] and corrected weight of 1,450 g [IQR: 1,151-1,930]. Across 760 recording hours, wireless ECG-HR was available for 652 h (86%), compared to 756 h (99.5%) for the wired. There were 61 h (8%) where wireless PR was available but not ECG-HR; thus, the combined coverage of wireless ECG-HR & PR was 713 h (94%). The wireless PR signals had a median bias of -2 bpm and 95%-LoA of [-19.5, 15.5] bpm (Table 1). In Figure 2(A)-(B), 94% of sample pairs were contained in regions A+B.

Conclusion(s): Combined use of wireless ECG-HR & PR signals increased the availability of HR data; this strategy should be considered to reduce the extent of data loss in wireless cardiac monitoring. Ongoing work will examine other strategies to improve the coverage and accuracy of wireless monitoring in the NICU.

Figure 1. Wireless HR monitoring system: (A) the chest sensor measured ECG-HR, and (B) the limb sensor measured PR.


Table 1. Statistical agreement between ECG-HR and PR signals across n = 96 recordings.
Legend: Results are reported as median [25th-75th percentiles].

Figure 2. Modified Clarke error grids for wireless PR vs wired ECG-HR in two cases: (A) when wireless ECG-HR is available (n = 1,715,459 sample pairs), and (B) when wireless ECG-HR is not available (n = 227,408 sample pairs).
Region descriptions: A = wireless PR within ±10% of wired ECG-HR and yielded same clinical interpretation; B = wireless PR greater than ±10% of wired ECG-HR but still yielded same clinical interpretation; C = wireless PR indicated clinical event when patient was normal; D = wireless PR missed clinical event when they occurred, and E = wireless PR confused clinical events when they occurred (i.e., bradycardias and tachycardias).