At-home validation of a non-contact, radar-based breathing monitor for long-term care of patients with respiratory diseases: A proof-of-concept study

Abstract

Background

Long-term monitoring of respiratory rate (RR) is an important component in the management of chronic respiratory diseases (CRDs). Specifically, predicting acute exacerbations of chronic obstructive pulmonary disease (AECOPD) is of significant scientific and clinical interest. This study aimed to evaluate the long-term validity of a novel contactless sleep monitor (CSM) in the home environment of CRD patients receiving ventilatory support. Additionally, we assessed patient acceptance, device usability, and RR fluctuations associated with AECOPD to establish a robust foundation for future research.

Patients and Methods

In this prospective proof-of-concept study, nineteen patients requiring non-invasive ventilation (NIV) were provided with the CSM in their home environment for six months and seven patients requiring invasive mechanical ventilation (IMV) for one month. The primary indication for NIV therapy was chronic obstructive pulmonary disease (COPD).

The CSM was validated under real-life conditions by comparing its nocturnal RR values with software data from both types of ventilators. Acceptability and usability of the sensor were assessed using a questionnaire. Additionally, COPD exacerbations occurring during the study period were analyzed for potential RR fluctuations preceding these events.

Results

Mean absolute error (MAE) of median RR between the NIV device and the CSM, based on 2326 nights, was 0.78 (SD: 1.96) breaths per minute (brpm). MAE between the IMV device and the CSM was 0.12 brpm (SD: 0.52) for 215 nights. The non-contact device was accepted by the patients and proved to be easy in use. In some of the overall only 13 cases of AECOPD, RR time courses showed variations of increased nocturnal respiratory activity a few days before the occurrence of such events.

Conclusion

The present CSM is suitable for valid long-term monitoring of nocturnal RR in patients’ home environment and is well accepted by the patients. The exploratory findings related to AECOPD events may serve as a starting point for larger studies aimed at developing robust prediction rules.

Zusammenfassung

Hintergrund

Die langfristige Überwachung der Atemfrequenz (RR) ist ein wichtiger Baustein für das Management chronischer Atemwegserkrankungen (CRD). Dabei ist insbesondere die Vorhersage akuter Exazerbationen der chronisch obstruktiven Lungenerkrankung (AECOPD) von großem wissenschaftlichem und klinischem Interesse. Ziel der vorliegenden Studie war es, die Langzeitvalidität eines neuen kontaktlosen Schlafmonitors (CSM) in der häuslichen Umgebung von CRD-Patienten mit Beatmungsunterstützung zu evaluieren. Zudem wurden die Patientenakzeptanz und die Benutzerfreundlichkeit des Gerätes erfasst sowie RR-Schwankungen im Zusammenhang mit AECOPD untersucht, um eine robuste Grundlage für zukünftige Forschung zu schaffen.

Patienten und Methoden

In dieser prospektiven Proof-of-Concept-Studie wurden 19 Patienten, die eine nicht-invasive Beatmung (NIV) benötigten, und 7 Patienten, die eine invasive mechanische Beatmungstherapie (IMV) benötigten, jeweils 6 bzw. 1 Monat lang mit dem CSM in ihrer häuslichen Umgebung versorgt. Die Hauptindikation für die NIV-Therapie war die Diagnose einer chronisch obstruktiven Lungenerkrankung (COPD).

Der CSM wurde unter realen Bedingungen validiert, indem seine nächtlichen RR-Werte mit den Softwaredaten der Beatmungsgeräte verglichen wurden. Die Akzeptanz und Benutzerfreundlichkeit des Sensors wurden mittels eines Fragebogens bewertet. COPD-Exazerbationen, die während des Studienzeitraums auftraten, wurden auf mögliche RR-Schwankungen untersucht, die diesen Ereignissen vorausgingen.

Ergebnisse

Der mittlere absolute Fehler (MAE) der medianen Atemfrequenz zwischen den NIV-Geräten und dem CSM betrug 0,78 (SD: 1,96) Atemzüge pro Minute (brpm), basierend auf 2326 Nächten. Der MAE zwischen den IMV-Geräten und dem CSM betrug für 215 Nächte 0,12 brpm (SD: 0,52). Der CSM wurde von den Patienten akzeptiert und erwies sich als leicht bedienbar. In einigen der nur insgesamt 13 Fälle von AECOPD wiesen die Atemfrequenzverläufe in den Tagen vor dem Auftreten der Ereignisse Schwankungen mit erhöhter nächtlicher Atemaktivität auf.

Schlussfolgerung

Der CSM eignet sich für die valide Langzeitüberwachung der nächtlichen Atemfrequenz im häuslichen Umfeld und wird von den Patienten gut akzeptiert. Die explorativen Ergebnisse im Zusammenhang mit AECOPD-Ereignissen können als Grundlage für größere Studien dienen, die auf die Entwicklung robuster Vorhersagemodelle abzielen.

Keywords

telehealth - respiratory rate - AECOPD - exacerbation - non-invasive ventilation - PREM

Schlüsselwörter

Telemedizin - Atemfrequenz - AECOPD - Exazerbation - nicht-invasive Beatmung - PREM

Introduction

Technological innovations enable continuous monitoring and digital transmission of vital signs and disease-specific parameters to healthcare providers. The increasing acceptance of such telemedical services, accelerated by the COVID-19 pandemic, creates promising opportunities for personalized smart healthcare in chronic diseases [1]. In addition to the adequate validation of new technologies, active involvement from both patients and practitioners is crucial to gather practical insights, thereby achieving effective and well-accepted integration into the healthcare system. While respiratory rate (RR) is an integral part of clinical care and assessment, e.g. for pneumonia or sepsis [2] [3], ongoing research explores its utility beyond hospital settings for chronic conditions. For example, investigations include RRʼs predictive role in chronic heart failure [4], cystic fibrosis [5], and chronic obstructive pulmonary disease (COPD) [6] [7] [8] [9].

COPD is the most prevalent among chronic respiratory diseases (CRD), which are acknowledged as one of the top leading causes of death worldwide [10] [11]. Over the past decade, considerable effort has been made in COPD research to improve disease management, with the focus on early detection and prevention of acute exacerbations of COPD (AECOPD), as they are known to have a negative impact on disease progression and the patient’s health-related quality of life [12] [13]. Various studies suggest that there are fluctuations in nocturnal RR [6] [7] [8] [9] occurring during a brief prodrome, which is known to precede COPD exacerbations [14] [15]. Thus, early recognition of significant RR changes would allow for timely intervention, potentially reducing the need for hospitalization or leading to faster recovery from a worsening of symptoms [16]. This could be, for example, accomplished by the implementation of continuous RR monitoring in the home environment [6] [7] [8] [9].

Up to now, various attempts have been made to monitor RR at home, e.g. via undergarment waistband-adhered physiologic monitors [17], via monitors embedded in the domiciliary oxygen supply system [9], by analyzing RR estimates from non-invasive ventilator software [7] [8], by minimal-contact sensors placed under the mattress [18], or by using a non-contact biomotion sensor to measure thoracic excursions [6] [19]. Home-based sleep sensors offer distinct advantages. Confounding factors like mental strain, heat, cold, or physical effort [20], are expected to be minimized. Additionally, contactless approaches with non-impairing designs, coupled with automated data collection and transmission, may enhance compliance and enable extended data acquisition in familiar environments, facilitating trend analyses.

Novel devices designed to monitor vital parameters typically undergo initial validation in controlled laboratory settings for a limited usage time. Despite the intrinsic strength of these devices for extended monitoring in the home environment, re-evaluation over prolonged periods is rarely carried out. The main reason may be that the established reference methods for measuring RR, such as manually counting breaths or using a thoracic respiratory effort belt (TREB), are not well suited for long-term home monitoring. Along with the frequent lack of analysis of patient compliance, these factors may contribute to the fact that none of such devices have yet become widely accepted.

The present proof-of-concept evaluates the Sleepiz One+ non-contact biomotion sensor (Sleepiz AG, Zurich, Switzerland; hereafter referred to as contactless sleep monitor or CSM) as an outpatient approach to monitor nocturnal RR. Its RR estimation has recently been validated in a separate in-lab study against the TREB of a polysomnography (PSG) setup, with an accuracy of 99.5% (+/– 3 breaths per minute; brpm) and a mean absolute error (MAE) of 0.48 brpm for median RR of 139 whole night recordings [21]. The main objective of the present investigation was to assess the long-term validity of these findings within a home setting by comparing RR estimates with built-in software data in a cohort of patients with domiciliary ventilator support. Within a subset of patients undergoing non-invasive ventilation (NIV) due to COPD, an exploratory analysis was conducted to determine the potential of the CSM in detecting variations in RR preceding impending exacerbations. The identification of such variations could hold implications for clinical applications. Including a second group of patients treated with invasive mechanical ventilation (IMV) aimed to mitigate the negative impact of facemask leakage on the reliability of NIV data [22]. Considering both patient experience and operability measures, the present study is the first to demonstrate the feasibility of this CSM for valid long-term home RR monitoring.

Methods

Patients

Between September 2020 and May 2022, a total of 19 patients receiving NIV therapy (“NIV patients”) were included at the study hospital (University Medicine Essen, Ruhrlandklinik, Germany). Additionally, seven patients treated with IMV were recruited from a collaborating outpatient critical care facility (“IMV patients”). All patients or their legal representatives were fully informed about the study and provided written informed consent upon being given adequate time to consider participation. The study protocol was approved by the ethics committee of the University of Duisburg-Essen (19-8961-BO) and was performed in accordance with the Declaration of Helsinki.

Patients had to be ≥18 years of age and showed the willingness and the ability to comply with the study protocol. Owing to the exploratory nature of the present study, patient selection based on the availability of eligible long-term NIV users at the study hospital with focus on COPD patients. Gravidity or presence of a pacemaker were considered as exclusion criteria. Detailed characteristics of enrolled patients are shown in [Table 1].

While all 19 patients received NIV therapy at study enrollment, three patients had their therapy modified over time. In two patients, therapy was changed from NIV to long-term oxygen therapy (LTOT) and one patient switched from NIV to continuous positive airway pressure (CPAP) therapy. Conditions for NIV treatment were COPD (n=15), diaphragmatic paresis (n=3), and severe obstructive sleep apnea (OSA; n=1). Main condition for IMV treatment was respiratory insufficiency due to hypoxic or hemorrhagic brain damage (n=7). The smaller size of the IMV cohort was due to the significant logistical and administrative effort required, as well as the limited availability of suitable patients within a reasonable distance from the study hospital.

Table 1 Patient characteristics (data reported as mean ± SD).

Parameter	NIV patients (n=19)	IMV patients (n=7)
Notes: a12 (75.0%) after treatment adjustment of three patients during study monitoring. bat study enrolment. Abbreviations: SD, standard deviation; n, number; NIV, non-invasive ventilation; IMV, invasive mechanical ventilation; BMI, body mass index; LTOT, long-term oxygen therapy; COPD, chronic obstructive pulmonary disease; GOLD, global initiative for chronic obstructive lung disease; FEV1, forced expiratory volume in one second; FVC, forced vital capacity; CAT, COPD assessment test.
Age (years)	67.3 (±7.2)	60.0 (±16.2)
Sex
Male, n (%)	8 (42.1%)	3 (42.9%)
Female, n (%)	11 (57.9%)	4 (57.1%)
BMI (kg/m2)	28.4 (±7.6)	24,9 (±3.9)
LTOT, n (%)	11 (57.9%)	–
Reason for NIV therapy
COPD	15 (78,9%)a	–
Smoking (pack-years)	42,2 (±30.0)	–
GOLD B, n (%)	3 (20.0%)	–
GOLD E, n (%)	12 (80.0%)	–
FEV1 (l)	0.92 (±0.6)	–
FEV1 (% predicted)	33.3 (±19.6)	–
FVC (l)	1.90 (±0.7)	–
FEV1/FVC	0.49 (±0.19)	–
CAT–Score	22.3 (±8.1)b	–
Diaphragmatic paresis, n (%)	3 (15.8%)	–
Obstructive sleep apnea, n (%)	1 (5.3%)	–
Comorbidities
Chronic heart failure, n (%)	3 (15.8%)	–
Hypertension, n (%)	14 (73.7%)	–
Cardiac arrhythmia, n (%)	4 (21.1%)	–
Type II diabetes, n (%)	4 (21.1%)	–
Asthma, n (%)	3 (15.8%)	–
Obstructive sleep apnea, n (%)	5 (26.3%)	–

Study design and experimental procedures

According to the study protocol of this prospective monocentric proof-of-concept trial, nocturnal home RR measurements were expected to be collected for at least six months for NIV patients, and at least one month for IMV patients. The shorter IMV monitoring period was based on the assumption of more usable 24/7 data from these ventilators. Additionally, the IMV cohort was used solely for validation and feasibility assessment, which reduced the need for an extended monitoring period.

The NIV patients’ hospitalizations, during which they were informed about the experiment, were either scheduled for stationary routine control of NIV-settings or following AECOPD treatment. Upon the patients’ agreement, a PSG was additionally recorded during these nights. The PSG-derived TREB signal was used to assess agreement between built-in software RR estimates of the NIV devices and traditional respiratory effort measures in this patient cohort. Good agreement between these two methods would indirectly support the validity of the CSM RR estimates when compared to NIV data in the home environment.

NIV patients received the CSM after hospital discharge. They were instructed to place it at body height, 40–50 cm from the body next to the bed, directed towards the lower thoracic and upper abdominal region. The transmission area needed to be clear of interfering objects, except for clothing or bedspreads. Patients were advised to turn on the device and associated hotspot only once, as the CSM was programmed for automatic recording between 9 pm and 9 am.

During monthly follow-up calls, COPD patients reported details on doctor visits and episodes of subjective health deterioration for the retrospective identification of potential exacerbation events.

AECOPD were defined as periods of significantly higher symptom load that required the use of prescribed rescue medication (oral steroids and/or antibiotics) or led to hospitalization.

Finally, to gather information about the patients’ experiences with the device, a patient-reported experience measurement (PREM) questionnaire was handed out to all 19 NIV patients at the end of the study. The questionnaire comprised 38 items covering different experience categories and has been developed for this device in German language. Categories included handling, acceptance, privacy concerns, benefits and actions they expect from such telehealth programs, as well as overall satisfaction. Patients also had the opportunity to express problems and suggest improvements through free-text responses.

IMV patients were given the contactless device at the collaborating outpatient critical care facility. As these patients were unable to complete the PREM questionnaire, the caregivers were asked about their experience with the CSM in their daily work routine instead. A flowchart illustrating the study procedure can be found in the online supplemental Fig. S1.

Device specifications

The CSM is able to provide RR estimates in a completely contactless manner. Based on doppler radar technology, it measures thoracic movements with sub-millimeter resolution and uses these measurements to estimate RR via signal processing and machine learning algorithms. Measured data is transmitted daily via hotspot to a cloud server where RR statistics can be monitored by authorized persons, such as the attending physicians or the patients themselves. A more detailed description of technical specifications and algorithmic RR computation can be found in the clinical validation report of this device [21]. To minimize missing data, patients were offered assistance in rebooting the device or the hotspot, as well as optimizing the device positioning as necessary. To be considered for analysis, recordings required a minimum duration of 120 minutes where RR estimates where available. The CSM does not generate RR output when device positioning is poor or when the monitored person is tossing and turning.

Ventilator built-in software data were either collected via a cloud-based telehealth platform or by manual read out using the manufacturers’ software. Ventilator type specifications and quantity are summarized in the online supplemental Table S1.

For PSG data collection, the Nox A1 recording system (Nox Medical, Reykjavík, Iceland) was used.

Data preprocessing and analysis

In the NIV patient group, the CSM recorded an average of 203 nights per patient with 5.7 hours of suitable contactless data per night, which remained constant throughout the study. NIV data were not available for three patients, resulting in 2326 overlapping nights from 16 patients. Comparison between NIV and TREB data was possible for 12 nights from 10 patients who underwent PSG measurement during the study period, with a maximum of two consecutive nights per patient included.

In the IMV patient group, a total of 334 valid CSM recordings (on average 47 nights per patient) and 505 ventilator datasets were evaluated, resulting in 215 overlapping nights from 4 patients, as three IMV-devices failed to store RR data for export. The online supplemental Fig. S2 illustrates both the quantity of collected data and the reasons for missing recordings (e.g., suboptimal positioning of the CSM, or patients deliberately not using the CSM or the NIV device).

Outcome comparison was performed twice, first with time-synchronized data using NIV recordings as reference, and once without time-synchronization. Synchronization of the time-series data limited the RR analysis of the TREB and the CSM data to the two longest periods of continuous ventilator use overnight. Despite this, the ventilator data represents nightly averages, as detailed assessment for individual time periods was not possible. A minimum valid overlap duration of 60 minutes was set as a requirement for analysis.

Without time-synchronization, RR estimates were analyzed based on device output, irrespective of total recording time or mandatory overlap, reflecting uncontrolled conditions.

RR estimates of interest comprised the 5^th, 50^{t h} (median), and 95^th percentile. The comparison of the different measurement techniques based on the mean absolute error (MAE) of these RR statistics. Bland-Altman- and scatterplots were used to illustrate their relationship.

Descriptive statistics (frequency, median, mean ± standard deviation) were used for the analysis of the PREM questionnaire. All statistics were performed with IBM SPSS Statistics (Version 28).

Results

Respiratory rate statistics

Outcome comparisons between TREB and NIV-software data yielded a MAE of 0.92±1.10 brpm for the 5^th percentile of the RR and 0.33±0.47 brpm for the median RR, respectively. However, the analysis revealed rather poor agreement between NIV and TREB for the 95^th percentile of the RR with a MAE of 4.25±4.51 brpm ([Table 2]). The analyses with the time synchronized data showed similar results for the 5^th percentile and for the median, but slightly better agreement for the 95^th percentile of the RR with a MAE of 3.57±4.78 brpm ([Table 2]).

Table 2 Respiratory rate statistics comparison.

	NIV vs. TREB			CSM vs. IMV			CSM vs. NIV
athree NIV devices did not provide 5th percentile values, MAE for 50th percentile values in bold numbers. Abbreviations: NIV, non-invasive ventilation; vs., versus; TREB, thoracic respiratory effort belt; CSM, contactless sleep monitor; IMV, invasive mechanical ventilation; RR, respiratory rate; 5%, 5th percentile; 50%, 50th percentile (=median); 95%, 95th percentile; n, number of recordings included; MAE, mean absolute error (breaths per minute); SD, standard deviation.
RR	5%	50%	95%	5%	50%	95%	5%	50%	95%
Without time synchronization
n	10a	12	12	215	215	215	2116a	2326	2326
MAE ±SD	0.92 ±1.10	0.33 ±0.47	4.25 ±4.51	0.44 ±0.63	0.12 ±0.52	1.90 ±2.63	0.43 ±0.84	0.78 ±1.96	2.85 ±3.12
With time synchronization
n	10a	12	12	215	215	215	2008a	2200	2200
MAE ±SD	0.84 ±1.17	0.38 ±0.54	3.57 ±4.78	0.44 ±0.63	0.12 ±0.52	1.90 ±2.63	0.36 ±0.65	0.65 ±1.90	2.81 ±3.10

Results from the IMV group showed excellent agreement for median RR between CSM and IMV data with a MAE of 0.12±0.52 brpm. MAE was 0.44±0.63 brpm and 1.90±2.63 brpm for the 5^th percentile and for the 95^th percentile, respectively ([Table 2]). The results did not change with time series synchronization. Bland-Altman analysis and the corresponding scatterplot for IMV and CSM data without time synchronization are displayed in [Fig. 1] a.

Analysis of the home night results derived from the NIV group yielded a MAE for the median RR of 0.78±1.96 brpm. MAE was 0.43±0.84 brpm and 2.85±3.12 brpm for the 5^th and for the 95^th percentile, respectively ([Table 2]). When the time-synchronized data based on the NIV usage time were used, the RR statistics comparison showed a marginally higher level of agreement. Bland-Altman analysis and the corresponding scatterplot for data without time synchronization are displayed in [Fig. 1] b. This figure further highlights that the agreement between the CSM and NIV declines with increasing RR, with higher values detected by the CSM compared to the NIV devices. A closer look on the outliers shows that this is predominantly the case for two patients (colored data points in [Fig. 1] b).

Fig. 1 RR comparison between (a) CSM and IMV and between (b) CSM and NIV [brpm]. Notes: The main body of the panels (a, b) illustrate the Bland-Altman-Analysis. The solid lines indicate the bias (MAE), dashed lines indicate the limits of agreement (mean ± SD*1.96). Additionally, the respective scatter plots are displayed. Jitter has been added in both charts for illustration purposes. The data point clustering in (a) is explained by the respiratory backup-rates of the IMV devices. The diamond and dot-shaped data points in (b) highlight that most outliers originate from two patients. Abbreviations: RR, respiratory rate; brpm, breaths per minute; IMV, invasive mechanical ventilator; NIV, non-invasive ventilator; CSM, contactless sleep monitor; MAE, mean absolute error; SD, standard deviation.

Patient-reported experience measurement (PREM)

The evaluation of the PREM showed positive results regarding the general usage of the CSM. The overall satisfaction score was 7.9 (±2.2) out of 10. Mean values for the patient responses within the categories of experience, expectations, concerns, and handling are illustrated in [Fig. 2]. Especially the handling of the device was rated positively, although patients reported that the device sometimes tipped over due to the light construction of the stand base. The potential concerns with the device were strongly dismissed by the patients. IMV patient caregivers indicated that the device was easy to use, although it had to be moved occasionally during nursing procedures. Results for all items are provided in the online supplemental Table S2.

Fig. 2 Patient reported experience and operability measures. Notes: Each point in the polar plot presents the mean PREM response-value for the corresponding category in the NIV group (n=19). In contrast to the other categories, the scale of the category concerns comprised only four points. Abbreviations: PREM, patient-reported experience measurement; NIV, non-invasive ventilation.

Exploring RR in relation to AECOPD

Overall, ten outpatient and three inpatient AECOPD events were observed in a total of six patients. CSM data were unavailable for one AECOPD event, while NIV data were missing for seven events. Two re-exacerbation events were excluded from analysis as they fell within the guard band of the previous exacerbation. [Fig. 3] illustrates the median RR measured by the CSM for ten AECOPD events, divided in three time periods: (i) three days before exacerbation onset (prodromal), (ii) three days after exacerbation onset (exacerbation), and (iii) a stable period defined as the nearest consecutive three-day period to the event, which is guarded by a band of seven stable days to another event (baseline). A modest increase in the median RR of 0.75 brpm can be detected during the prodromal phase. As the observed RR variabilities differ notably in terms of their amplitude, onset and duration between patients, three individual examples of RR time courses related to AECOPD events are given in the online supplemental Fig. S3.

Fig. 3 RR assessment in the context of AECOPD events. Notes: Each violin plot illustrates the distribution of RR across three-day time periods: when symptoms were stable (Baseline); three days preceding the event (Prodromal); three days starting from the onset day of the exacerbation (Exacerbation). Two re-exacerbation events were disregarded as they occurred within the guard band of the previous exacerbation. Abbreviations: RR, respiratory rate; brpm, breaths per minute; AECOPD, acute exacerbation of chronic obstructive pulmonary disease.

Discussion

The present study was the first to measure contactless recorded RR values across multiple nights in the home environment of patients with CRDs requiring additional ventilator support. The at-home performance metrics of the CSM reported here are consistent with those observed in its clinical validation trial [21], demonstrating accurate respiratory rate monitoring and favorable patient experiences in the home environment.

Within the discussion of this recent clinical validation report, the RR estimate outcomes of the CSM were compared with those of four other contactless respiration monitors reported in the literature [21]. Although the performance of all five sensors was broadly similar, the key distinction of the CSM used in this study lies in its status as a certified medical device, with its validity now successfully confirmed in non-clinical settings.

To the best of our knowledge, only one other attempt has been made so far to validate a novel RR monitor in the home environment. Do et al. used the TREB signal of a polygraph system as reference to validate a contactless RR monitor at home [23]. Mean accuracy was 99.0% (SD: 1.9) and MAE was 0.6 brpm (SD: 0.1) for median RR, which is not much different to the validation results presented here. However, the analysis time was restricted to 15 or 120 minutes per patient due to the manual scoring of the TREB data. In addition, the study by Do and colleagues covered only one night of monitoring with both measuring methods, rather than multiple nights as in the present investigation. Rubio et al. compared five minimal-contact home monitoring devices with a vest-mounted metabolic sensor system [24]. The acceptability and performance of the two best-performing monitors (a chest-band and an accelerometer) were assessed in 23 stable COPD patients at home, but without any further referencing, unlike in the present study.

A common problem with validation studies in the home environment is the selection of an acceptable reference standard that can be used comfortably at home. For the present study, it was decided to use NIV instead of TREB as a reference to compare long-term CSM RR statistics with an established technology. The rationale for this decision was that (i) integrated NIV software solutions have demonstrated adequate reliability for most ventilation parameters [22] [25], (ii) no other interfering sensors are required, and (iii) this method is particularly suited for recording several consecutive nights, especially taking intraindividual variation into account. However, in addition to these advantages, there are also some issues to consider when applying this approach. Differences in measurements may result from different operation times, since the CSM measures RR continuously regardless of NIV usage. Especially time periods directly before and/or after NIV use may be accompanied by abnormal RR. Therefore, an additional analysis using time synchronization was performed on the data, which was supposed to put this problem into perspective. However, the fact that only average values for NIV and IMV are available for synchronized CSM data poses as a limiting factor. Furthermore, data obtained from ventilators can vary across manufacturers and especially the RR and the percentage of cycles triggered by the patient lack validation through bench test studies [22] [25]. We addressed this issue by adding a within-subject comparison between NIV and TREB data to the study protocol, offering a specific assessment of the individual reference accuracy for this study collective (please note that this additional in-lab measurement was not possible for all study participants due to personal reasons). The analysis showed sufficient agreement between NIV and TREB for median RR, but revealed poor agreement for the 95^th percentile RR, possibly because NIV RR estimates can be compromised by facemask leakage or obstructive events [22]. This possible explanation is firstly supported by the IMV group results, where the 95^th percentile agreement of IMV and CSM was notably higher ([Table 2]). Secondly, the MAE from the previously reported in-lab validation study was also higher, at 0.67 brpm between CSM and TREB for the 90^th percentile [21], indicating that the lower agreement between CSM and NIV in the home environment reported here may partly depend on NIV inaccuracy.

Apart from that, non-contact Doppler radar sensors are inherently susceptible to motion artifacts, which can degrade data quality due to factors such as unpredictable body movements or changes in sleep position [26]. The performance of the CSM could therefore be somewhat biased, as patients with ventilator support may move less during the night and sometimes have a greater tidal volume compared to those without support. However, a subgroup analysis within the referenced in-lab study of this device yielded a MAE of 0.29±0.28 brpm for the median RR in 12 COPD patients without ventilator support, and 0.39±0.30 brpm across all 139 participants with various conditions, but also without ventilator support [21], which seems to qualify this possible explanation. Therefore, the lower reliability at higher RR needs to be addressed in more detail in future studies, by considering all possible sources of noise that might challenge the signal processing of radar biomotion-sensors.

With respect to the acceptability and operability of the CSM, PREM results show that the device can be easily used by a representative group of patients and medical staff. Patients neither felt restricted in their privacy nor bothered by the device. They would appreciate independent access to their data and personal contact to healthcare providers when necessary. Regarding the latter, the mobile hotspot used in this study has recently been replaced by a tablet, which now provides a communication interface for the patient, enhancing status assessment and intervention opportunities. Results for general satisfaction are encouraging, especially when considering the lack of any direct patient advantages in this study. Rubio et al. reported a few difficulties that patients had with the wearable devices in their study, such as correct positioning on the body or skin problems caused by the adhesive patches [24]. Naturally, these issues were not a problem in our study due to the non-intrusive, contactless design.

As an exploratory part of our analysis, we assessed RR fluctuations in relation to AECOPD to confirm the feasibility of detecting these fluctuations with the evaluated CSM. The averaged RR values show a discrete increase of 0.75 brpm in the median RR prior to an exacerbation. For comparative purposes, the analysis followed Hawthorne et al., who observed a 2.0 brpm increase in the median RR during the prodromal phase [27]. In our analysis, however, we chose to flank the baseline period by a seven-day stable phase similar to the proposal by Shah et al. [15]. In summary, individual time courses either revealed variations in elevated RR, which partly align with already proposed AECOPD prediction rules, or showed no elevations at all. The examples of three patients, shown in the online supplemental Fig. S3, illustrate this range of outcome variation. The second patient (Fig. S3B), for example, experienced an elevated RR three days prior to an AECOPD, as well as a higher RR variability in general, which Blouet et al. considered to be predictive for AECOPD [7]. Interestingly, this severely affected patient accounts for most of the outliers observed in the Bland-Altman analyses ([Fig. 1] b, red diamond-shaped data points), for which it remains speculative which measurement method is more accurate, as the time synchronized analysis showed no differences for these nights. The third patient (Fig. S3C) had an increase of 3.0 brpm two days before hospitalization, which is comparable to the findings of Yañez et al [9]. Although similarities to previous studies can be identified, it is important to emphasize that these potential parallels should be treated with caution. The number of AECOPD events in this proof-of-concept was too small to support generalization or the development of a robust prediction rule. Further evaluation through additional studies is needed.

Besides RR, oxygen saturation [28], heart rate [27], spirometry [29], and periodic assessment of symptom load are of high interest for the development of AECOPD prediction models [30]. Notably, the CSM evaluated in this study is also able to measure heart rate without contact. Furthermore, oxygen saturation can now be registered by adding a wireless wearable to preserve the nonintrusive design of this system as best as possible. Symptom questionnaires and patient education programs can be conducted via the tablet. These circumstances provide a unique opportunity to evaluate these indicators in a larger study using a standardized study protocol.

At first glance, using an additional breathing monitor for ventilator-dependent patients, whose devices often provide automatic RR estimation, may seem redundant. However, the fact that five ventilators did not provide any data (Fig. S2) and that one AECOPD event was potentially recorded solely by the CSM (Fig. S3) suggests that additional monitoring of specific patient groups may be warranted. Nevertheless, the patient selection based on the validation approach in our study overlooks the potential of the CSM for other, less severely affected patient groups, such as non-ventilated COPD individuals. Furthermore, in the context of a hospital-at-home concept, options such as additional daytime monitoring, as well as assessments of physical activity, body temperature and lung function could be considered. A multiparameter approach, expected to enhance robustness significantly, will then require defining the value and limitations of each individual component within the model. This could facilitate the development of robust parameters and patient-individual prediction rules for AECOPD, including for patients in earlier disease stages without ventilator support, as remote vital sign monitoring holds the potential to prevent even an initial exacerbation, a paramount goal in COPD management [31]. A natural progression of this approach would be the evaluation of similar deterioration prediction models in other diseases such as asthma or cystic fibrosis.

Conclusion

The present study has been the first attempt to examine the long-term validity of contactless recorded RR estimates in the patients’ home environment. The findings are in line with the clinical validation of this device, indicating accurate respiratory rate monitoring accompanied by high patients’ acceptance. Therefore, this study suggests that the device is suitable for future clinical research, focusing on multiparameter approaches and subgroup analyses, as telehealth remains promising in the management of CRD in general and COPD in particular.

Abbreviations

AECOPD: acute exacerbation of COPD

ASB: assisted spontaneous breathing

BMI: body mass index

brpm: breaths per minute

CAT: COPD assessment test

COPD: chronic obstructive pulmonary disease

CPAP: continuous positive airway pressure

CRD: chronic respiratory disease

CSM: contactless sleep monitor

FEV₁ : forced expiratory volume in one second

FVC: forced vital capacity

GOLD: global initiative for COPD

IMV: invasive mechanical ventilation

LTOT: long-term oxygen therapy

MAE: mean absolute error

NIV: non-invasive ventilation

OSA: obstructive sleep apnea

PREM: patient-reported experience measurement

PSG: polysomnography

RR: respiratory rate

SD: standard deviation

TREB: thoracic respiratory effort belt

vs.: versus

Funding

TF received a general project funding granted by the German Association of Sleep Medicine (DGSM), which did not influence the study design, analysis, interpretation or the writing of the manuscript.

Conflict of Interest

C.S. declares no personal conflicts of interest, but institutional grants for scientific projects from Bayer, Nox Medical, Onera, ResMed and Sleepiz. T.F., T.E., A.W., S.D.T. and R.V. state no conflicts of interest.

Acknowledgement

The authors would like to thank all study participants as well as the staff at the sleep laboratory at the study hospital (Ruhrlandklinik, Essen, Germany) and the collaborating outpatient critical care facility (amicu – Außerklinische Intensivpflege, Mühlheim, Germany) for their support in conducting the study. In addition, we appreciate the assistance of Sleepiz AG in the time synchronization and export of RR estimates from the CSM.

• References

• 1 Lee SM, Lee DH. Opportunities and challenges for contactless healthcare services in the post-COVID-19 Era. Technol Forecast Soc Change 2021; 167: 120712
• 2 van Dam PMEL, Lievens S, Zelis N. et al. Head-to-head comparison of 19 prediction models for short-term outcome in medical patients in the emergency department: a retrospective study. Ann Med 2023; 55: 2290211
• 3 McNally M, Curtain J, O’Brien KK. et al. Validity of British Thoracic Society guidance (the CRB-65 rule) for predicting the severity of pneumonia in general practice: Systematic review and meta-analysis. Br J Gen Pract 2010; 60: e423-e433
• 4 Forleo GB, Santini L, Campoli M. et al. Long-term monitoring of respiratory rate in patients with heart failure: the Multiparametric Heart Failure Evaluation in Implantable Cardioverter-Defibrillator Patients (MULTITUDE-HF) study. J Interv Card Electrophysiol 2015; 43: 135-144
• 5 Straßburg S, Linker CM, Brato S. et al. Investigation of respiratory rate in patients with cystic fibrosis using a minimal-impact biomotion system. BMC Pulm Med 2022; 22: 59
• 6 Ballal T, Heneghan C, Zaffaroni A. et al. A pilot study of the nocturnal respiration rates in COPD patients in the home environment using a non-contact biomotion sensor. Physiol Meas 2014; 35: 2513-2527
• 7 Blouet S, Sutter J, Fresnel E. et al. Prediction of severe acute exacerbation using changes in breathing pattern of COPD patients on home noninvasive ventilation. Int J Chron Obstruct Pulmon Dis 2018; 13: 2577-2586
• 8 Borel JC, Pelletier J, Taleux N. et al. Parameters recorded by software of non-invasive ventilators predict copd exacerbation: A proof-of-concept study. Thorax 2015; 70: 284-285
• 9 Yañez AM, Guerrero D, Pérez De Alejo R. et al. Monitoring breathing rate at home allows early identification of COPD exacerbations. Chest 2012; 142: 1524-1529
• 10 Adeloye D, Song P, Zhu Y. et al. Global, regional, and national prevalence of, and risk factors for, chronic obstructive pulmonary disease (COPD) in 2019: a systematic review and modelling analysis. Lancet Respir Med 2022; 10: 447-458
• 11 GBD Chronic Respiratory Disease Collaborators . Prevalence and attributable health burden of chronic respiratory diseases, 1990–2017 – a systematic analysis for the Global Burden of Disease Study 2017. Lancet Respir Med 2020; 8: 585-596
• 12 Wedzicha JA, Seemungal TAR. COPD exacerbations: defining their cause and prevention. Lancet 2007; 370: 786-796
• 13 Hurst JR, Siddiqui MK, Singh B. et al. A systematic literature review of the humanistic burden of copd. Int J Chron Obstruct Pulmon Dis 2021; 16: 1303-1314
• 14 Seemungal TAR, Donaldson GC, Bhowmik A. et al. Time Course and Recovery of Exacerbations in Patients with Chronic Obstructive Pulmonary Disease. Am J Respir Crit Care Med 2000; 161: 1608-1613
• 15 Shah SA, Velardo C, Farmer A. et al. Exacerbations in chronic obstructive pulmonary disease: Identification and prediction using a digital health system. J Med Internet Res 2017; 19: e69
• 16 Wilkinson TMA, Donaldson GC, Hurst JR. et al. Early therapy improves outcomes of exacerbations of chronic obstructive pulmonary disease. Am J Respir Crit Care Med 2004; 169: 1298-1303
• 17 Polsky MB, Moraveji N. Early identification and treatment of COPD exacerbation using remote respiratory monitoring. Respir Med Case Rep 2021; 34: 101475
• 18 Dietz-Terjung S, Geldmacher J, Brato S. et al. A novel minimal-contact biomotion method for long-term respiratory rate monitoring. Sleep Breath 2021; 25: 145-149
• 19 Lauteslager T, Maslik M, Siddiqui F. et al. Validation of a new contactless and continuous respiratory rate monitoring device based on ultra-wideband radar technology. Sensors 2021; 21: 4027
• 20 Nicolò A, Massaroni C, Schena E. et al. The importance of respiratory rate monitoring: From healthcare to sport and exercise. Sensors (Switzerland) 2020; 20: 1-45
• 21 Bujan B, Fischer T, Dietz-Terjung S. et al. Clinical validation of a contactless respiration rate monitor. Sci Rep 2023; 13: 3480
• 22 Borel JC, Palot A, Patout M. Technological advances in home non-invasive ventilation monitoring: Reliability of data and effect on patient outcomes. Respirology 2019; 24: 1143-1151
• 23 Do W, Russell R, Wheeler C. et al. Performance of Contactless Respiratory Rate Monitoring by Albus HomeTM, an Automated System for Nocturnal Monitoring at Home: A Validation Study. Sensors 2022; 22: 7142
• 24 Rubio N, Parker RA, Drost EM. et al. Home monitoring of breathing rate in people with chronic obstructive pulmonary disease: Observational study of feasibility, acceptability, and change after exacerbation. Int J Chron Obstruct Pulmon Dis 2017; 12: 1221-1231
• 25 Janssens JP, Borel JC, Pépin JL. Nocturnal monitoring of home non-invasive ventilation: The contribution of simple tools such as pulse oximetry, capnography, built-in ventilator software and autonomic markers of sleep fragmentation. Thorax 2011; 66: 438-445
• 26 Tran VP, Al-Jumaily AA, Islam SMS. Doppler radar-based non-contact health monitoring for obstructive sleep apnea diagnosis: A comprehensive review. Big Data and Cognitive Computing 2019; 3: 1-21
• 27 Hawthorne G, Richardson M, Greening NJ. et al. A proof of concept for continuous, non-invasive, free-living vital signs monitoring to predict readmission following an acute exacerbation of COPD: a prospective cohort study. Respir Res 2022; 23: 102
• 28 Mehdipour A, Wiley E, Richardson J. et al. The Performance of Digital Monitoring Devices for Oxygen Saturation and Respiratory Rate in COPD: A Systematic Review. COPD 2021; 18: 469-475
• 29 Miłkowska-Dymanowska J, Białas AJ, Obrębski W. et al. A pilot study of daily telemonitoring to predict acute exacerbation in chronic obstructive pulmonary disease. Int J Med Inform 2018; 116: 46-51
• 30 Sanchez-Morillo D, Fernandez-Granero MA, Jiménez AL. Detecting COPD exacerbations early using daily telemonitoring of symptoms and k-means clustering: a pilot study. Med Biol Eng Comput 2015; 53: 441-451
• 31 Viniol C, Vogelmeier CF. Exacerbations of COPD. Eur Respir Rev 2018; 27: 170103

Publication History

Received: 03 September 2024

Accepted after revision: 31 January 2025

Article published online:
09 May 2025