Early Detection of Worsening Heart Failure in Patients at Home Using a New Telemonitoring System of Respiratory Stability

Abstract

Background: Early detection of worsening heart failure (HF) with a telemonitoring system crucially depends on monitoring parameters. The present study aimed to examine whether a serial follow up of all-night respiratory stability time (RST) built into a telemonitoring system could faithfully reflect ongoing deterioration in HF patients at home and detect early signs of worsening HF in a multicenter, prospective study.

Methods and Results: Seventeen subjects with New York Heart Association class II or III were followed up for a mean of 9 months using a newly developed telemonitoring system equipped with non-attached sensor technologies and automatic RST analysis. Signals from the home sensor were transferred to a cloud server, where all-night RSTs were calculated every morning and traced by the monitoring center. During the follow up, 9 episodes of admission due to worsening HF and 1 episode of sudden death were preceded by progressive declines of RST. The receiver operating characteristic curve demonstrated that the progressive or sustained reduction of RST below 20 s during 28 days before hospital admission achieved the highest sensitivity of 90.0% and specificity of 81.7% to subsequent hospitalization, with an area under the curve of 0.85.

Conclusions: RST could serve as a sensitive and specific indicator of worsening HF and allow the detection of an early sign of clinical deterioration in the telemedical management of HF.

Despite therapeutic advances in heart failure (HF) management, HF patients are still at high risk of rehospitalization and mortality, resulting in a major burden on patients and a strong impact on the medical economy.^1,2 Recent advances in telecommunication technology have created new opportunities to provide a telemonitoring care system as an adjunct to medical management of HF patients. Approaches to telemedical management rely on the concept that regular observation of carefully selected physiological parameters will enable early detection of clinical deterioration and timely intervention to prevent poor outcomes.^3–5 The home telemonitoring system has 2 major clinical benefits: facilitation of patient self-care⁴ and early detection of HF decompensation;⁶ the latter benefit crucially depends on what is best to monitor under the conditions where direct patient contact is not utilized. Although well-known as symptoms and signs of worsening HF, progressive dyspnea, edema, and body weight gain leading to hospitalization usually occur late in the course of HF decompensation and thus failed to detect early deterioration in previous studies.^7–10 In contrast, the CardioMEMS Heart Sensor Allows Monitoring of Pressure to Improve Outcomes in NYHA Class III Heart Failure Patients trial (CHAMPION) reported that wireless implantable monitoring of pulmonary artery pressures significantly reduced the risk of hospital admission by 39%.¹¹ Despite accumulating benefits of implantable devices for the monitoring of pulmonary artery pressures or intrathoracic impedance to evaluate central fluid balance,¹² these device implantations are invasive and costly, and therefore applicable to a limited number of patients with congestive HF. Recently, we have developed respiratory stability time (RST),¹³ a new quantitative and non-invasive measure of respiratory instability because respiratory instability such as Cheyne-Stokes respiration and irregular rapid and shallow respiration without periodicity are related to neurohumoral, hemodynamic, and respiratory derangement, which is associated with worsening HF.¹³

In previous studies, we have reported that augmented respiratory instability, as reflected by a fall in RST, is a strong and independent predictor of poor prognosis in patients with chronic HF and thus serves as an ominous sign of worsening HF.¹³ In a recent multicenter prospective study, we examined changes in respiratory stability during the recovery process of HF and found that the depressed all-night RST in the acute phase of HF decompensation was progressively restored as HF improved,¹⁴ whereas, the RST remained unchanged in patients with no improvement of HF. A multivariate analysis demonstrated that the major determinant for all-night RST was systemic and pulmonary congestion.¹⁴ Given that RST, obtained easily and non-invasively from respiratory signals, reflects the congestive state of HF, this parameter provides potential benefits for the early detection of worsening HF in a telemonitoring environment equipped with non-attached sensor technologies.

The aim of the present study was therefore to examine whether serial follow up of all-night RST built into a telemonitoring system could faithfully reflect ongoing deterioration in HF patients at home and detect early signs of worsening HF in patients from a multicenter, prospective study.

Methods

Subjects

We recruited patients with New York Heart Association (NYHA) functional class ≥II who had a history of admission or therapeutic reinforcement for worsening HF at least once within the latest year. All patients aged ≥20 years had not been started on any respiratory equipment such as positive pressure ventilation or oxygen inhalation within a month. Key exclusion criteria included the following: chronic obstructive pulmonary disease; pneumonia or other infection; sequela of a stroke; dementia; or symptomatic malignancy with limited quality of life. This study was undertaken at 6 centers. Enrollment began in May 2017, and the follow up was completed in March 2019.

Study Design

The study was performed in a blind fashion to both doctors and patients by covering the RST trend throughout the follow-up periods. Outpatient follow-up visits were scheduled every 4 weeks until 48 weeks. Investigators examined the subjects’ HF conditions following Guidelines for Diagnosis and Treatment of Acute and Chronic Heart Failure (JCS 2017/JHFS 2017)¹⁵ and collected physical findings, laboratory data, and information about adverse events. Only 1 delegated individual could browse the computer viewer and confirm whether RSTs were calculated every morning while keeping all information confidential. In this study, HF deterioration was defined as cardiac death or hospitalization due to worsening HF. Based on the investigators’ reports including signs, symptoms, and laboratory data, the HF evaluation committee independently examined HF worsening events and finalized HF conditions in a blinded manner.

Measurements

Our telemonitoring system consisted of a piezo-electric, non-attached sensor installed under a bed sheet and a microcomputer connected via the Internet as a gateway at the patient’s home. The telemonitoring system was capable of collecting respiratory and heartbeat signals and the patient’s lying periods on the bed. During the study period, all-night respiratory signals were continuously obtained at the sampling frequency of 400 Hz every night, transferred to a cloud server for calculation of all-night RST, and stored there. The following background information was collected: gender, age, body weight, underlying heart disease, HF treatment, functional capacity evaluated by NYHA functional class and a specific activity scale,¹⁶ laboratory data including B-type natriuretic peptide (BNP), chest-ray, electrocardiogram, and the baseline cardiac function on echocardiography. Symptoms and signs were followed up every 4 weeks, and laboratory and chest-X ray examinations were followed up every 8 weeks in outpatient clinics.

All data were accumulated in the Academic Research Organization (ARO) of Osaka University Hospital, together with other clinical information, for analysis by statisticians.

Ethical Approval

The study was conducted according to the Declaration of Helsinki (revised in October 2013) and the “Ethical Guidelines for Medical and Health Research Involving Human Subjects” amended on 28 February 2017. Written informed consent to participate in this study was provided by all patients. The study was approved by the Ethical Review Board of Osaka University Hospital and the Ethics Committee at each center.

Analysis of All-Night RST

The data to be included in the analysis were restricted to those for the fixed hours from 23:00 to 05:00 o’clock. All respiratory signals transferred to a cloud server were automatically analyzed to extract RST, respiratory rate, heart rate, and grade of periodic breathing (Figure 1). The method of RST measurements has been reported in previous studies.^13,14 As shown in the pre-processing step of RST calculation (Figure 2), we removed the direct current component of the signals to avoid the zero frequency impulse, applied a zero-phase digital filter to retrieve respiratory signals by excluding high frequency components >0.5 Hz, and finally resampled the all-night signals at 4 Hz. We focused on 2 frequency ranges to estimate RST, as previously described.¹³ One range consisted of the respiratory frequency components retrieved from the instantaneous respiratory signals by removing high- or low-frequency noise through a 5^th order band-pass Butterworth filter with cut-off frequencies of 0.11 and 0.5 Hz. The other components comprised the very low-frequency range of respiration that corresponded to periodic breathing. This periodic breathing was obtained by tracing peaks of the instantaneous respiratory signals with the baseline adjusted to zero, and then applying the band-pass filter with cut-off frequencies of 0.008 and 0.04 Hz. To serially calculate all-night RST, the respiratory signals were divided into serial 5-min segments every 50 s. More than 420 segments of 5-min data (≥350 min) were analyzed. For each epoch, the maximum entropy method (MEM; Figure 2) was applied to these respiratory and periodic breathing curves to extract the specific components of respiratory variations from each wave. All spectral power was normalized using the ratio (%) of the maximum power of the respiratory components. In the evaluation of respiratory instability, we equally adopted all respiratory frequency points that had spectral power >10% of the maximum respiratory power.¹⁴ The very low-frequency points of the periodic breathing curve were also taken into account in evaluating RST, only when the maximum power of the very low-frequency components was >50% of the maximum power of the respiratory components. For each epoch, the distribution of these respiratory frequency points was evaluated using standard deviation, and RST was defined as the reciprocal of the standard deviation. Serial changes in all-night RST were averaged to serve as a representative of all-night respiratory instability. The RST was updated every morning and a 5-day moving average of every day RST was traced on the viewer of the monitoring center (Figure 1). The program to calculate RST, developed using Matlab (The Mathworks, Inc., Natick, MA, USA), underwent validation by the ARO of Kyushu University Hospital.¹⁴ Analysis of the source code and movement inspection using raw data were performed for the RST calculation system, including the RST calculation program, along with that of the peripheral equipment, by the ARO, who confirmed that they worked according to the required specifications.¹⁴

Figure 1.

Respiratory stability time (RST) telemonitoring system. All respiratory signals transferred to a cloud server were automatically analyzed to extract the RST, respiratory rate, heart rate, and grade of periodic breathing. The RST was updated every morning and traced on the viewer of the monitoring center.

Figure 2.

Calculation of all-night respiratory stability time (RST) in a cloud server. Signals to calculate all-night RST were collected from 23:00 to 5:00 o’clock of the next morning every day. In the pre-processing steps for RST measurement, we focused on 2 frequency ranges, which were high-frequency respiratory components (0.11–0.5 Hz) and very low-frequency components corresponding to the periodic breathing (0.008–0.04 Hz). These 2 signals were divided into serial 5-min segments every 50 s. For each epoch of the 5-min high- and very low-frequency data, power spectra were serially calculated with the maximum entropy method and normalized as a percentage of the maximum power of the respiratory components. The high-frequency distribution for spectral power >10% of the maximum and the very low-frequency distribution for spectral power >50% of the maximum were adopted to calculate a standard deviation of respiratory frequency variations. RST was defined as a reciprocal of the standard deviation. For each epoch, RST was serially calculated and averaged to represent all-night respiratory instability.

Study Endpoints

The endpoint of this study was to determine the optimal cut-off level of RST that could achieve the highest sensitivity and specificity for early detection of cardiac death or hospitalization due to worsening HF. In the present study, therefore, we examined RSTs up to 28 days prior to hospitalization due to worsening HF and RSTs prior to stable follow-up visits to the hospital. When patients were rehospitalized due to worsening HF within 28 days after discharge, the preceding RSTs before the second admission were excluded from the analysis of sensitivity and specificity to predict hospitalization.

Statistical Analysis

Continuous variables are presented as median and interquartile ranges (IQR) and categorical variables as the proportion (%). We used the Wilcoxon signed-rank sum test for continuous variables and the 2-tailed Fisher’s exact test for categorical data to evaluate differences in baseline clinical characteristics, laboratory findings, and RSTs between patients with stable conditions during the follow up and those with HF deterioration. In patients of HF deterioration (hospitalization or sudden death), the levels of RST and BNP were compared between at the baseline and at the hospitalization using the Wilcoxon signed-rank test. In these comparisons, BNP during HF deterioration was not measured in 1 patient because he died suddenly, and in another patient, RST could not be measured when HF worsened because she was already hospitalized due to ovarian cancer. We used the latest HF deterioration admissions for statistics in patients who experienced repeated hospitalization for HF. The performance of the RST threshold for the prediction of worsening HF was evaluated by receiver-operating characteristics (ROC) analysis. A 5-day moving average of RST was adopted in this study. Two-sided P<0.05 was considered significant. Statistical analysis was performed using the statistical analysis system (SAS) software version 9.4 (SAS Institute Inc., Cary, NC, USA).

Results

Patient Characteristics

A total of 17 patients were enrolled in the present study. Their age ranged from 45 to 90 years with a median of 67.0 (IQR: 60.5–80.5) years. Twelve patients were male, and the follow-up period was 299 (IQR: 252–356) days. Baseline patient characteristics are shown in Table 1. Seven patients were in NYHA functional class II and the remaining 10 in class III, and the specific activity scale of all patients was <6.5 metabolic equivalents (METs), with a median of 3.5 (IQR: 2.5–5.1) METs. The majority (76.5%) of patients had ischemic heart disease or dilated cardiomyopathy; therefore, the left ventricle was enlarged with a cardiothoracic ratio of 57.0%, end-diastolic dimension of 69.8 mm, left ventricular ejection fraction of 31.0%, and BNP of 285.6 pg/mL. There was no relationship between RST and body mass index. Ten patients had chronic atrial fibrillation, and 6 patients received an implantable cardiac defibrillator or cardiac resynchronized therapy. Diuretics were used for all patients, β-blockers in 15 patients, angiotensin-converting enzyme inhibitors or angiotensin II receptor blockers in 13 patients, and oral inotropic agents in 7 patients.

Table 1. Baseline Patient Characteristics

Variable	Total	No admission	Admission	P value*
n	17	8	9
Age, years	67.0 [60.5–80.5]	65.5 [52.5–82.5]	67.0 [64.8–76.0]	0.61
Males, n (%)	12 (70.6)	5 (62.5)	7 (77.8)	0.62
Body weight, kg	57.6 [49.6–61.9]	54.3 [49.1–60.6]	59.9 [51.7–65.4]	0.48
Body mass index, kg/m2	21.0 [20.0–22.9]	21.6 [20.6–22.5]	20.0 [19.0–23.8]	0.33
Duration of HF, years	6.0 [2.0–10.0]	4.0 [0.5–8.5]	6.0 [4.0–10.0]	0.33
NYHA function class II/III, n (%)	7 (41.2)/10 (58.8)	4 (50.0)/4 (50.0)	3 (33.3)/6 (66.7)	0.64
Specific Activity Scale, Mets	3.5 [2.5–5.1]	3.5 [2.9–5.5]	3.5 [2.4–4.0]	0.65
No. of HF readmissions 1/2/3, n (%)**	10 (58.8)/4 (23.5)/3 (17.7)	5 (62.5)/3 (37.5)/0 (0)	5 (55.6)/1 (11.1)/3 (33.3)	0.22
Cardiothoracic ratio, %	57.0 [55.0–67.5]	57.7 [54.5–65.1]	56.2 [55.0–67.5]	0.76
Echocardiography
LVDd, mm	69.8 [50.0–76.0]	59.3 [50.0–70.8]	74.0 [58.9–77.5]	0.29
LVDs, mm	63.5 [36.2–69.0]	50.8 [36.2–64.5]	67.5 [48.9–70.5]	0.37
LVEF, %	31.0 [21.3–45.0]	36.5 [20.5–50.0]	28.0 [23.3–39.8]	0.74
DcT, ms	170.0 [125.8–221.5]	180.5 [145.0–274.0]	170.0 [123.3–212.8]	0.73
Laboratory values
Serum sodium, mEq/L	139.0 [135.8–141.3]	138.5 [134.5–141.5]	139.0 [136.5–140.5]	0.94
Serum creatinine, mg/dL	1.46 [1.01–1.73]	1.04 [0.98–1.56]	1.56 [1.27–2.04]	0.08
B-type natriuretic peptide, pg/mL	285.6 [169.0–565.0]	178.7 [110.7–306.2]	564.0 [264.5–612.9]	0.02
Respiratory stability time, s	30.6 [19.4–42.1]	34.6 [30.5–44.9]	20.3 [15.0–36.8]	0.07
Etiologies, n (%)
Ischemic heart disease	7 (41.2)	2 (25.0)	5 (55.6)	0.43
Dilated cardiomyopathy	6 (35.3)	4 (50.0)	2 (22.2)
Valvular heart disease	3 (17.7)	1 (12.5)	2 (22.2)
Miscellaneous	1 (5.9)	1 (12.5)	0 (0)
Comorbidities, n (%)
Atrial fibrillation	10 (58.8)	3 (37.5)	7 (77.8)	0.15
Dyslipidemia	13 (76.5)	7 (87.5)	6 (66.7)	0.58
Sleep apnea syndrome	3 (17.7)	2 (25.0)	1 (11.1)	0.58
Implanted device, n (%)
ICD	3 (17.7)	2 (25.0)	1 (11.1)	0.58
CRT-D	3 (17.7)	0 (0)	3 (33.3)	0.21
Pacemaker	2 (11.8)	0 (0)	2 (22.2)	0.47
Medications, n (%)
Diuretics	17 (100)	8 (100)	9 (100)	1.00
ACE inhibitors/ARB	13 (76.5)	6 (75.0)	7 (77.8)	1.00
β-blockers	15 (88.2)	7 (87.5)	8 (88.9)	1.00
Inotropes	7 (41.2)	2 (25.0)	5 (55.6)	0.33

Data are presented as median [25 percentile–75 percentile]. Admission includes 8 patients with hospitalization and 1 patient with sudden death. *For differences between no admission and deterioration. **The number of admissions or therapeutic reinforcements due to worsening HF during 1 year preceding the enrolment. ACE, angiotensin-converting enzyme; ARB, angiotensin II receptor blockers; CRT-D, cardiac resynchronization therapy and defibrillator; DcT, deceleration time; HF, heart failure; ICD, implantable cardioverter defibrillator; LVDd, left ventricular end-diastolic dimension; LVDs, left ventricular end-systolic dimension; LVEF, left ventricular ejection fraction; n, number of subjects; NYHA, New York Heart Association.

Prevalence of Worsening HF

There were 12 episodes of admission due to worsening HF in 8 patients and 1 sudden death in 1 patient. One patient had 3 episodes of HF deterioration; 2 episodes for this patient were excluded from the analysis because she had already been admitted to the hospital first due to pneumonia and second due to ovarian cancer and, therefore, RSTs preceding worsening HF were not measured in hospital. Another episode for the same patient was readmission due to worsening HF within 5 days after discharge; this was not adopted for ROC analysis which requires at least 28 day-RSTs preceding hospitalization. Therefore, the remaining 9 episodes of admission and 1 episode of sudden death that occurred in 9 patients during the follow-up periods were available for ROC analysis to detect the RST threshold of HF deterioration. Table 1 compared baseline characteristics between 8 patients who were stable throughout the follow-up periods and 9 patients with worsening HF who required hospitalization or suffered sudden death. The baseline characteristics were similar between the 2 groups, except that the level of BNP was significantly higher in the admission group than in the no admission group.

Serial Changes in All-Night RST During Worsening HF

In 9 of 10 episodes of admission or sudden death, the preceding decline of RST below 20 s was observed from 7 to 28 days or more prior to the events. The remaining 1 episode was caused by hospitalization due to acute exacerbation of HF, where a 5-day average of RST masked a rapid drop of RST preceding hospitalization. Figure 3 shows the RST trends of 3 representative patients. In the Case 1 patient who was clinically stable (NYHA II) with no worsening episodes throughout follow-up periods, RST fluctuated >20 s. This stable pattern of RST trend was also found in 8 patients who did not have worsening episodes (Table 2). The Case 2 patient was clinically unstable (NYHA III). The RST of this patient showed fluctuating declines that started >30 days prior to admission and eventually fell <20 s with subsequent hospitalization due to worsening HF. In contrast to the RST trend, most of the worsening signs and symptoms became apparent within 1 week prior to admission. This declining pattern of RST toward <20 s preceded hospitalization in 6 patients. The Case 3 patient had been severely ill (NYHA III–IV), hospitalized 3 times a year, and finally died during this hospitalization. His RSTs remained <20 s, even after discharge from the previous hospitalization. This pattern of sustained low RST levels was found in 2 patients who had had a history of frequent hospitalization. Table 2 summarized the baseline characteristics of patients with each pattern of RST trend; type A: no admission with stable high RSTs, type BC: admission with declining RSTs towards <20 s, and type D: admission with RSTs sustained <20 s. Table 2 did not include 1 patient with acute exacerbation because a 5-day-averaged RST of this patient showed a delayed drop and did not correspond to any type of RST trends identified. From type A, to type BC, and to type D, there was a tendency to have a decline in the specific activity scale, left ventricular ejection fraction, deceleration time of early mitral flow velocity and RST, and a tendency to have an increase in the duration of HF, left ventricular end-systolic and end-diastolic dimensions, and BNP levels.

Figure 3.

Respiratory stability time (RST) trends prior to admission. Case 1 patient (NYHA II) was clinically stable with no worsening episode throughout follow-up periods, and the RST trend remained >20 s. Case 2 patient (NYHA III) showed gradual worsening of heart failure (HF) to hospitalization with fluctuating declines towards <20 s. Case 3 (NYHA III–IV) patient was severely ill with frequent readmission, and the RST remained <20 s. NYHA, New York Heart Association.

Table 2. Baseline Patients Characteristics by Patterns, as Shown in Figure 3

Variable	Type A	Type BC	Type D
n	8	6	2
Age, years	65.5 [52.5–82.5]	66.0 [64.0–74.0]	69.5 [66.0–73.0]
Males, n (%)	5 (62.5)	5 (83.3)	2 (100)
Body weight, kg	54.3 [49.1–60.6]	58.8 [55.9–73.0]	62.3 [61.6–62.9]
Body mass index, kg/m2	21.6 [20.6–22.5]	21.7 [19.8–25.4]	21.0 [19.9–22.1]
Duration of HF, years	4.0 [0.5–8.5]	5.0 [2.0–7.0]	14.5 [13.0–16.0]
NYHA function class II/III, n (%)	4 (50.0)/4 (50.0)	1 (16.7)/5 (83.3)	1 (50.0)/1 (50.0)
Specific Activity Scale, Mets	3.5 [2.9–5.5]	3.5 [2.5–5.5]	2.8 [2.0–3.5]
No. of HF readmissions 1/2/3, n (%)*	5 (62.5)/3 (37.5)/0 (0)	4 (66.7)/1 (16.7)/1(16.7)	0 (0)/0 (0)/2(100)
Cardiothoracic ratio, %	57.7 [54.5–65.1]	55.9 [53.4–67.3]	56.3 [55.5–57.0]
Echocardiography
LVDd, mm	59.3 [50.0–70.8]	72.0 [63.3–79.5]	76.0 [76.0–76.0]
LVDs, mm	50.8 [36.2–64.5]	66.0 [54.5–70.3]	70.5 [69.0–72.0]
LVEF, %	36.5 [20.5–50.0]	30.0 [25.0–36.0]	19.0 [10.0–28.0]
DcT, ms	180.5 [145.0–274.0]	170.0 [124.5–230.0]	154.0 [96.0–212.0]
Laboratory values
Serum sodium, mEq/L	138.5 [134.5–141.5]	139.5 [137.0–140.0]	136.5 [135.0–138.0]
Serum creatinine, mg/dL	1.04 [0.98–1.56]	1.61 [1.13–2.40]	1.62 [1.31–1.92]
Brain natriuretic peptide, pg/mL	178.7 [110.7–306.2]	424.8 [201.1–577.5]	534.5 [349.7–719.2]
Respiratory stability time, s	34.6 [30.5–44.9]	21.9 [16.8–33.6]	14.2 [13.3–15.1]
Etiologies, n (%)
Ischemic heart disease	2 (25.0)	4 (66.7)	1 (50.0)
Dilated cardiomyopathy	4 (50.0)	1 (16.7)	1 (50.0)
Valvular heart disease	1 (12.5)	1 (16.7)	0 (0)
Miscellaneous	1 (12.5)	0 (0)	0 (0)
Comorbidities, n (%)
Atrial fibrillation	3 (37.5)	4 (66.7)	2 (100)
Dyslipidemia	7 (87.5)	4 (66.7)	1 (50.0)
Sleep apnea syndrome	2 (25.0)	1 (16,7)	0 (0)
Implanted device, n (%)
ICD	2 (25.0)	0	1 (50.0)
CRT-D	0 (0)	2 (33.3)	1 (50.0)
Pacemaker	0 (0)	1 (16,7)	0 (0)

Data are presented as median [25 percentile–75 percentile]. Type A, no admission with stable high RSTs; Type BC, admission with declining RSTs towards <20 s; Type D, admission with RSTs sustained <20 s. The case of acute exacerbation is not included in this table because it is not characterized by these types of RST. *The number of admissions or therapeutic reinforcements due to worsening HF during 1 year preceding the enrolment. Abbreviations as in Table 1.

An all-night RST indicated by arrow A (Case 1) in Figure 3 is shown in Figure 4A, representing a relatively short sleeping time with a high amplitude of oscillation and an average RST of 43 s. Original respiratory signals that corresponded to the highest RST and that to a low RST demonstrated that RST faithfully reflected the respiratory instability. All-night RST before worsening HF, indicated by arrow B (Case 2) in Figure 3, has an average RST of 37 s (Figure 4B), whereas the all-night RST of day C just before admission fell <20 s with an average of 11 s (Figure 4C). Similarly, all-night RST indicated by arrow D (Case 3) in Figure 3 showed sustained low levels <20 s, with an average of 13 s (Figure 4D). Thus, differences in respiratory stability of respiratory signals were clearly discriminated from the magnitude of RST. Figure 5 shows comparisons of RST and BNP between the baseline and at the worsening events in admission group in Table 1. RST significantly fell towards <20 s at admission or sudden death with small individual differences. In contrast, BNP at the time of hospitalization showed a tendency to increase with a wider range of individual variations, as compared to the baseline BNP. The outlier defined by a red cross in Figure 5 is the case of acute exacerbation of HF. This case was hospitalized before reaching 20 s of RST because a 5-day average of RST delayed a rapid drop of day-by-day RST. Changes in body weight measured every 4 weeks were compared to the decline in RST before admission. During 14–28 days prior to admission, a body weight gain of ≥2 kg was observed in 1 of 10 events, whereas an RST <20 s was found in 5 events at that time. At the time of admission, a body weight gain was found in 4 events and an RST <20 s in 9 events; these had already been observed during ≥7 days prior to admission.

Figure 4.

All-night respiratory stability time (RST) of the 3 cases in Figure 3. (A) All-night RST of the day indicated by the arrow A in Case 1 (Figure 3) has relatively high RST components with an average of 43 s. Respiratory signals were quite different between those with high RST and those with low RST. (B) All-night RST of the day indicated by the arrow B in Case 2 (Figure 3) was obtained under stable conditions with relatively high components of RST. (C) All-night RST of the day indicated by the arrow C in Case 2 (Figure 3) was <20 s, with an average of 11 s just prior to admission. The respiratory pattern remained highly unstable at any time. (D) All-night RST of the day indicated by the arrow D in Case 3 (Figure 3) was <20 s, with an average of 13 s, where a stable respiration was abolished, similar to that seen in (C).

Figure 5.

Changes in all-night respiratory stability time (RST) and B-type natriuretic peptide (BNP) during deterioration of heart failure (HF). RST significantly fell towards <20 s at admission or sudden death with small individual differences. In contrast, BNP at the time of hospitalization showed a tendency of increase with a wider range of individual variations, as compared to baseline BNP. The red line of each boxplot indicates the median on each box, and the bottom and top edges of the box indicate the 25th and 75th percentiles, respectively. The outlier, the case of acute exacerbation of HF, is defined by a red cross 1.5 times the interquartile ranges above the 3rd quartile. In this case, a 5-day average of RST delayed a rapid drop of RST preceding hospitalization.

RST Threshold for Hospitalization

Ten RST trends up to 28 days prior to admission due to worsening HF or sudden death and 104 RST trends up to 28 days prior to follow-up visits with no worsening signs were used for ROC analysis. We examined the optimal RST cut-off level to detect hospitalization using various cut-off levels and found that consecutive decline in RST <20 s during 28 days prior to admission achieved the highest sensitivity of 90.0% and specificity of 81.7%, with an AUC of 0.85 (Figure 6).

Figure 6.

Receiver operating characteristic (ROC) curve for the optimal respiratory stability time (RST). The optimal RST cut-off level to detect hospitalization during 28 days prior to admission was 20 s, which achieved the highest sensitivity of 90.0% and specificity of 81.7% with an area under the curve (AUC) of 0.85. The red circle corresponds to the optimal RST point.

Discussion

The daily home telemonitoring environment we have developed in the present study enabled the measurement of all-night physiological variables in a fully-automated manner without the need to attach any biological sensors to the subject’s body. Although the patient was lying in bed at night, a sheet-type sensor continuously recorded signals of respiration and cardiac beats. The obtained information was fully and automatically processed and transferred to a cloud server via the Internet, where an all-night RST was calculated every morning and transferred to the monitoring center every day. The main finding in the present study was that an RST of 20 s was the optimal threshold to predict subsequent hospitalization with the highest sensitivity and specificity and the maximum AUC. Early detection of worsening HF requires the finding of subclinical signs before the progression of dyspnea, edema, or weight gain, which usually occurs within 1 week before hospitalization. In the present study, declines of RST <20 s were observed more frequently and earlier before admission than the weight gain was. Therefore, we examined RSTs up to 28 days prior to hospitalization due to worsening HF and found that the progressive decline of RST <20 s preceded hospitalization by 7–28 days in most patients. Thus, our innovative telemonitoring environment equipped with daily RST measurements and non-attached sensor technologies potentially facilitate the early detection of worsening HF before clinical signs and symptoms worsen in patients at home.

Early detection of HF decompensation in outpatient settings is critical in preventing development of subclinical HF into overt congestive HF and requires serial monitoring of HF conditions. Therefore, it is crucial to identify which variables are the best to monitor. Although home monitoring of daily body weights remains an important component in the management of patients with HF, telemonitoring of weights and vital signs as an adjunct to routine HF care had no significant impact on early detection of HF decompensation and readmission rate,^8–10 as shown in this study. This is because these parameters are likely to be insensitive to intracardiac filling pressures and late markers of incipient HF decompensation.^7,17 As increases in cardiac filling pressures are often apparent several weeks before symptoms worsen, direct daily measurements of pulmonary artery pressures or right ventricular pressure using wireless implantable monitoring devices have been applied to enable remote monitoring of HF successfully.^13,14 In a substudy of the Chronicle Offers Management to Patients with Advanced Signs and Symptoms of Heart Failure (COMPASS-HF) trial, serial measurements of right ventricular pressure and an estimated pulmonary diastolic pressure using implantable hemodynamic monitoring demonstrated that acute HF decompensation of both diastolic and systolic HF was preceded by a gradual and significant increase in diastolic pulmonary pressure. However, no statistically significant changes in average body weight were detected during this examination period.¹⁸ In the CHAMPION trial, management of HF by use of wireless pulmonary artery hemodynamic monitoring significantly reduced the rate of HF-related hospitalization.¹¹ BNP levels that reflect intra-cardiac filling and pulmonary capillary pressures are quite useful for detecting acutely decompensated HF. In the Heart Failure Assessment With BNP in the Home (HABIT) trial, Maisel et al continuously measured daily BNP levels in HF patients for 60 days using finger-stick blood sampling technology and demonstrated that the slope calculated by ordinary linear regression of lnBNP vs. time represents the risk change for acute HF decompensation; the upward trend increased the risk by 59.8%, whereas the downward trend decreased the risk by 39.0%.¹⁹

These novel approaches indicate the importance and considerable benefits of monitoring cardiac filling pressure for the early detection of worsening HF. In human lungs, increased pulmonary artery pressure and development of congestive pulmonary edema can be intrinsically detected by 4 built-in sensor systems, which are vagal nerve collagen sensors of the lungs.²⁰ The irritant vagal afferents through the lung stretch reflex are activated in different ways by exerting counteracting influences on breathing and could therefore cause respiratory instability such as irregularly rapid and shallow breathing, temporal respiratory inhibition with non-periodic apnea, or sighs. Additionally, an increased central blood volume restricts lung inflation mechanically, leading to a rapid and shallow respiration.²¹ We have developed all-night RST as a quantitative measure of respiratory instability, which could faithfully reflect congestive signs and clinical status during the recovery process from acute decompensation of chronic HF.^13,14 All-night RST measurements have an advantage in that they provide sensitive detection of increases in filling pressure or central blood volume in worsening HF, because overnight rostral fluid displacement from the legs during sleep likely predisposes patients to lung congestion,²² resulting in stretch receptor stimulation and eventually respiratory instability.

Clinical Implications

The present study findings and results of our previous studies suggest that our RST telemonitoring system could provide a new type of medical management of HF patients at home. RST-guided telemedicine allows the detection of early signs of worsening HF, giving advice about an early hospital visit, and evaluating the effects of therapeutic reinforcement by the recovery of RST.¹⁴ Long-term hemodynamic monitoring with implantable devices could have limited applicability due to it being invasive and a costly burden on HF patients. The finger-stick technology for daily BNP monitoring could impose a physical burden on HF patients for a long-term follow-up and is therefore applicable to limited periods. Although BNP has been an established indicator for worsening HF, inter- and intra-individual variability of BNP possibly compromises its utility in identifying early clinical deterioration of HF. The present RST monitoring system seems to have 2 advantages in the early detection of worsening HF. Firstly, this system enabled an everyday RST trend to be provided non-invasively and fully automatically without any manipulation and burden and is, therefore, widely applicable to HF patients for a long-term follow up. Secondly, RST fell <20 s within a small individual difference when HF was worsening to cause hospitalization. Consequently, nurses and co-workers involved in telemedicine could readily recognize the worsening or recovery process of HF and contact the medical doctor at the right time. When downloaded onto a smartphone, the RST algorithm would potentially be applied to monitoring other disorders associated with respiratory instability and to personal health records for better self-management.

Study Limitations

Although RST enabled to monitor the processes of HF worsening, some limitations of the present study should be addressed. First, this study consisted of a low number of patients and was conducted in a multicenter, prospective, observational study manner; therefore, it remains unclear whether RST is applicable to all types of HF. To confirm the clinical usefulness of RST for worsening HF, we need to undertake a prospective, randomized, open, blinded-endpoint or single-blind study that enrolls a large number of HF patients. Second, although difficult to perform and therefore not done in the present study, simultaneous monitoring of daily BNP and RST in the same patient could clearly elucidate their relative advantage in the early detection of worsening HF. Finally, there have been no clinical studies that directly compared the relationship between changes in RST and pulmonary capillary wedge pressure, whereas their relationship was observed indirectly in our previous study.^13,14

Conclusions

Respiratory stability time, a sensitive and specific indicator of worsening HF, could detect early signs of clinical deterioration, potentially enable timely intervention, and finally reduce rehospitalization. Thus, our home telemonitoring system equipped with non-attached sensor technologies and all-night RST evaluation could be a new generation of telemedicine for the home monitoring of patients with HF.

Acknowledgments

We are indebted to Tomoko Mihara, Clinical Trial Data Manager, who assisted with conducting this trial. We gratefully appreciate the contribution of Kenichi Yamakoshi (NPO Research Institute of Life Benefit), Motoi Kosuke (Shizuoka Institute of Science and Technology), and Naoto Tanaka (NPO Research Institute of Life Benefit) who contributed towards the sheet sensor development.

Sources of Funding

This study was supported by a grant from the Japan Agency for Medical Research and Development (18lk1010005 h to S.M.) from September 2016 to March 2019.

Disclosures

H.A. reports patents of the RST algorithm. Y. Sakata is a member of the Associate Editors of Circulation Journal and Y. Sawa is a member of the Editorial Board of Circulation Journal. All other authors have no conflicts of interest to disclose.

IRB Information

This study was approved by the Ethical Review Board of Osaka University Hospital (Approval No. 16141(T3)).

Data Availability

The deidentified participant data analyzed will be shared upon a reasonable request basis after permission from the local ethics committee. Please directly contact the corresponding author to request data sharing or detailed information. Data will be provided under the methods and for periods approved by the ethics committee.

References

• 1.   Roger VL. Epidemiology of heart failure. Circ Res 2013; 113: 646–659.
• 2.   Ambrosy AP, Fonarow GC, Butler J, Chioncel O, Greene SJ, Vaduganathan M, et al. The global health and economic burden of hospitalizations for heart failure: Lessons learned from hospitalized heart failure registries. J Am Coll Cardiol 2014; 63: 1123–1133.
• 3.   Brahmbhatt DH, Cowie MR. Remote management of heart failure: An overview of telemonitoring technologies. Card Fail Rev 2019; 5: 86–92.
• 4.   Van Spall HGC, Rahman T, Mytton O, Ramasundarahettige C, Ibrahim Q, Kabali C, et al. Comparative effectiveness of transitional care services in patients discharged from the hospital with heart failure: A systematic review and network meta-analysis. Eur J Heart Fail 2017; 19: 1427–1443.
• 5.   Anker SD, Koehler F, Abraham WT. Telemedicine and remote management of heart failure. Lancet 2011; 378: 731–739.
• 6.   Louis AA, Turner T, Gretton M, Baksh A, Cleland JG. A systematic review of telemonitoring for the management of heart failure. Eur J Heart Fail 2003; 5: 583–590.
• 7.   Friedman MM. Older adults’ symptoms and their duration before hospitalization for heart failure. Heart Lung 1997; 26: 169–176.
• 8.   Chaudhry SI, Mattera JA, Curtis JP, Spertus JA, Herrin J, Lin ZQ, et al. Telemonitoring in patients with heart failure. N Engl J Med 2010; 363: 2301–2309.
• 9.   Koehler F, Winkler S, Schieber M, Sechtem U, Stangl K, Böhm M, et al. Impact of remote telemedical management on mortality and hospitalizations in ambulatory patients with chronic heart failure: The telemedical interventional monitoring in heart failure study. Circulation 2011; 123: 1873–1880.
• 10.   Galinier M, Roubille F, Berdague P, Brierre G, Cantie P, Dary P, et al. Telemonitoring versus standard care in heart failure: A randomised multicentre trial. Eur J Heart Fail 2020; 22: 985–994.
• 11.   Abraham WT, Adamson PB, Bourge RC, Aaron MF, Costanzo MR, Stevenson LW, et al. Wireless pulmonary artery haemodynamic monitoring in chronic heart failure: A randomised controlled trial. Lancet 2011; 377: 658–666.
• 12.   Abraham WT, Stevenson LW, Bourge RC, Lindenfeld JA, Bauman JG, Adamson PB, et al. Sustained efficacy of pulmonary artery pressure to guide adjustment of chronic heart failure therapy: Complete follow-up results from the CHAMPION randomised trial. Lancet 2016; 387: 453–461.
• 13.   Asanoi H, Harada D, Oda Y, Ueno H, Takagawa J, Ishise H, et al. Independent prognostic importance of respiratory instability and sympathetic nerve activity in patients with chronic heart failure. J Cardiol 2017; 70: 476–483.
• 14.   Takagawa J, Asanoi H, Tobushi T, Kumagai N, Kadokami T, Dohi K, et al. Multicenter, prospective study on respiratory stability during recovery from deterioration of chronic heart failure. Circ J 2018; 83: 164–173.
• 15.   Tsutsui H, Isobe M, Ito H, Ito H, Okumura K, Ono M, et al. JCS 2017/JHFS 2017 guideline on diagnosis and treatment of acute and chronic heart failure: Digest version. Circ J 2019; 83: 2084–2184.
• 16.   Sasayama S, Asanoi H, Ishizaka S, Miyagi K. Evaluation of functional capacity of patients with congestive heart failure.In: Yasuda H, Kawaguchi H, editors. New aspects in the treatment of failing heart syndrome. Tokyo: Springer-Verlag, 1992; 113–117.
• 17.   Bui AL, Fonarow GC. Home monitoring for heart failure management. J Am Coll Cardiol 2012; 59: 97–104.
• 18.   Zile MR, Bennett TD, St John Sutton M, Cho YK, Adamson PB, Aaron MF, et al. Transition from chronic compensated to acute decompensated heart failure: Pathophysiological insights obtained from continuous monitoring of intracardiac pressures. Circulation 2008; 118: 1433–1441.
• 19.   Maisel A, Barnard D, Jaski B, Frivold G, Marais J, Azer M, et al. Primary results of the HABIT Trial (heart failure assessment with BNP in the home). J Am Coll Cardiol 2013; 61: 1726–1735.
• 20.   Roberts AM, Bhattacharya J, Schultz HD, Coleridge HM, Coleridge JC. Stimulation of pulmonary vagal afferent C-fibers by lung edema in dogs. Circ Res 1986; 58: 512–522.
• 21.   Cundrle I, Olson LJ, Johnson BD. Pulmonary limitations in heart failure. Clin Chest Med 2019; 40: 439–448.
• 22.   Yumino D, Redolfi S, Ruttanaumpawan P, Su MC, Smith S, Newton GE, et al. Nocturnal rostral fluid shift: A unifying concept for the pathogenesis of obstructive and central sleep apnea in men with heart failure. Circulation 2010; 121: 1598–1605.