This study is derived from a data warehousing and data mining project, including all consecutive patients admitted to our ICU. Our local ethics committee (Comité d’Ethique du CHRU de Brest) approved the data collection study and waived written informed consent according to French legislation. Information was provided to all patients and relatives at ICU admission (ReaSTOC study, ClinicalTrials.gov identifier: NCT02893462).
Patients and measurement set-up
Retrospective evaluation phase
During this evaluation phase of the algorithm, PPG-RR calculations were retrospectively performed on PPG waveforms extracted from the ReaSTOC physiological signals’ data warehouse and compared with RR reference values extracted from metadata files. All recordings from the ReaSTOC database were prospectively performed by the same research assistant, using a standardized protocol.
Prospective evaluation phase
Considering experience from the retrospective evaluation phase, the algorithm was refined and implemented in a biomedical device CE-marked (FreeO2, Oxynov, Québec, QC, Canada) that enables continuous parameters recordings. PPG-RR estimation with the FreeO2 system was performed using the standard SpO2 sensor of the device. For the sake of standardization, we only included spontaneously breathing patients and excluded those with unavailable PPG curves (e.g. patients with severe with hemodynamic instability and vasoconstriction).
Measurements
General data recording
During all phases, measurements were taken with the patients lying in the supine position, with a 30° angulation of the bed head.
Clinical metadata (physiological parameters, clinical status, underlying pathology, medications, etc.) were recorded by the research assistants using our standard monitoring devices and the computerized medical file (ICCA® Philips Healthcare, Amsterdam, the Netherlands).
In all phases, reference RR was measured continuously using both electrical thoracic impedance (ECG-RR) and chronometric evaluation (Manual-RR) over a 30-s period, under spontaneous ventilation or ventilatory rate (Ventilator-RR) for patients under ventilatory assistance. We considered as a reference the mean RR value over three consecutive measurements, taken each 15 min, during a mean 60-min duration, by a dedicated research assistant.
Features extraction
Retrospective evaluation phase
During the retrospective phase, we used a dedicated network that enable direct recording of raw physiological waveforms (ECG at 500 Hz, PPG at 125 Hz and pressure at 62.5 Hz) and a specific data extraction software (SYNAPSE system version 1, LTSI—INSERM U1099, Rennes, France).
After a resting period of 5–10 min, measurements were taken each 15 min during a mean 60-min recording period, for each patient newly admitted to our ICU. Participants were breathing either spontaneously or under mechanical ventilation.
At least three 2-min PPG periods, free from movement-induced disturbances, were selected from each measurement.
The PPG signal includes respiratory synchronous components, seen as frequency modulation of the heart rate (respiratory sinus arrhythmia), amplitude modulation of the cardiac pulse and respiration-induced intensity variation in the PPG baseline. PPG-RR value is extracted via the analysis of respiration-induced frequency components embedded in standard PPG waveform. Advanced statistical signal processing techniques were used to enhance these components and thus reduce the effect of unavoidable disturbance such as artefacts or transient and spontaneous phenomena. A mean PPG-RR value was provided and compared to the mean reference value over the 60-min period.
Prospective evaluation phase
During the prospective evaluation phase, PPG-RR calculations were performed using PPG signals extracted from the Nonin PureSAT® internal OEM III (Plymouth, MN, USA) of the FreeO2 system (Oxynov Inc., Québec, QC, Canada), using a standard reusable Nonin PureLight® finger clip sensor, positioned on the index, middle or ring finger. The algorithm that was used was derived from the previous analysis phase.
PPG-RR calculations were performed at the patients’ bedside during a 45-min period of PPG signal stability, using the dedicated device. Continuous calculations were performed each respiratory cycle, within a 10-min moving average window. PPG-RR was considered as the mean value during the overall 60-min period. Manual-RR count was performed concomitantly to the PPG signals recording.
Statistical analysis
Chronometric measurements of the respiratory rate (Manual-RR), electrical thoracic impedance (ECG-RR) and ventilatory respiratory rate values (Ventilator-RR) were regarded as the reference methods for respiration rate assessment and plethysmographic calculations (PPG-RR) as the method of comparison. Linear correlation between the different methods was performed using Pearson’s correlation coefficient analysis; Intraclass correlation coefficient (ICC) for average measurements is presented as a single-number summary that was used to assess reliability of different measurements averaged together [26].
The agreement between quantitative measure of the reference and the estimated RR values (PPG-RR) was then graphically appreciated according to Bland and Altman [27].
Statistical significance for a bilateral test was set a priori at p < 0.01 for all relevant analyses. Analyses were performed using MedCalc Statistical Software version 18.6 (MedCalc Software bvba, Ostend, Belgium).