This was a single-center, observational, exploratory study. We enrolled hospitalized patients aged ≥18 years with a confirmed diagnosis of HF in the Department of Cardiology, West China Hospital, Sichuan University. Patients were excluded if they met any of the following criteria: (1) severe lung diseases, such as pulmonary fibrosis, emphysema, or severe chronic obstructive pulmonary disease (COPD), as these conditions are associated with abnormal breathing patterns and could confound PC assessment via lung ultrasound (LUS); (2) clinician-determined ineligibility, including active malignancies or pregnancy, due to potential risks or study interference; (3) acute pulmonary edema or severe dyspnea requiring intensive care or mechanical ventilation, as these conditions necessitate urgent intervention and may distort respiratory pattern analysis; and (4) patient refusal to participate.

All the enrolled patients were required to wear the wearable device for at least 24 hours since the day of admission. Within the first 24 hours after admission, a comprehensive evaluation including vital signs, laboratory tests, echocardiography, and PC assessment by LUS was performed.

The study was approved by the Biomedical Research Ethics Committee of West China Hospital, Sichuan University (20211045). The study qualified as minimal-risk observational research involving routine clinical care augmentation. All participants provided written informed consent before enrollment. All collected data were deidentified using the following measures: clinical data were stored with pseudonymized codes, and the master linkage file is maintained separately under password protection. Participants received nonmonetary compensation as a full waiver of outpatient follow-up consultation fees (270 CNY, approximately US $43 at the exchange rate during the study period) and complimentary device-derived health reports. No direct financial payments were made to avoid undue inducement. The fee waiver mechanism was preapproved by the ethics committee as equitable compensation.

Within 24 hours postadmission, the following clinical data were recorded: (1) demographic, comorbidities (diabetes, hypertension, hyperlipidemia, chronic kidney disease, and sleep disorders), and medications (Angiotensin-Converting Enzyme Inhibitors, Angiotensin II Receptor Blockers, Angiotensin Receptor-Neprilysin Inhibitors, beta-blocker, sodium-glucose cotransporter-2 inhibitors, Mineralocorticoid Receptor Antagonists, and diuretics) by history taking and verified through medical record review; (2) blood pressure, weight, height, and BMI measured by the nurse; (3) the echocardiographic data of left ventricular ejection fraction (EF); (4) New York Heart Association (NYHA) functional class, Borg score, EuroQol Five Dimensions Questionnaire assessed by researchers.

LUS was performed by an experienced investigator, who was blinded to the results of clinical congestion assessment and other clinical data. Patients were asked to lie in a recumbent position during the LUS examination performed using a pocket ultrasound device (Lumify, Phillips Healthcare) with a phased array transducer in sagittal orientation at an imaging depth of 18 cm. A 28-zone protocol was used for image acquisition as previously recommended [19]. The highest number of B-lines visualized during the entire clip was quantified for each zone, and the sum of the 28 zones was used for analysis; PC(+) was defined as the presence of >5 comets [20].

The wearable device used is the SensEcho physiological monitoring system, as illustrated in Figure 1A, which could continuously collect physiological data [3]. The device uses Respiratory Inductive Plethysmography technology to capture chest and abdominal respiratory movement signals at a sampling frequency of 25 Hz, as well as electrocardiogram signals at 200 Hz, and 3-axis accelerometer signals at 25 Hz. Chest and abdominal movement signals were collected using wearable devices for 24 hours on the day of admission. However, only signals collected at nighttime were used for analysis. Because nighttime monitoring is more comparable than daytime monitoring, as the nighttime environment generally remains consistent, involving fewer external interference factors, such as visits from relatives, medical activities, or variations in external environments [21].

Only the signals during the nighttime period (11 PM to 5 AM) were used for analysis. During nighttime, patients frequently engaged in movements such as sitting up, turning over, or adjusting clothing. To address the resulting artifacts caused by significant amplitude changes during these posture shifts, we identified movement periods using the 3-axis accelerometer signal and excluded the corresponding respiratory data. The chest respiratory signals and abdominal respiratory signals were aligned by timestamp and added together to generate the total respiratory signal. Subsequently, a linear detrending process was used to remove slow baseline drift or fluctuations and to center the total respiratory signals around zero. A fifth-order, 2 Hz low-pass infinite impulse response Butterworth filter was then used to eliminate high-frequency noise and obtain clean respiratory signals. In total, 527.19 hours of respiratory signals were included, accounting for 91.51% of the total dataset. The extremum-point-search algorithm was used to detect and identify the peaks and valleys of the respiratory wave [22], thus obtaining the breath-to-breath intervals (BBI) sequence, as well as the respiratory amplitude (RA) sequence through the calculation of distances between the peaks and valleys of each breath.

Based on the BBI and RA, several parameters could be derived as listed in the following. All these parameters have been validated and could discriminate patients with HF from healthy participants in our previous study [23].

Continuous variables are presented as mean (SD) or median (IQR), while categorical variables are presented as n (%). Independent sample t tests are used for the comparison of continuous variables with normal variance, while the Mann-Whitney U test is used for skewed continuous variables, and the chi-square test or Fisher exact test is used for categorical variables. Logistic regression analysis is used to identify risk factors associated with PC. Considering that age and gender may have an impact on breathing patterns [5], age, gender, and factors identified as significant in univariate regression analysis are then sequentially included in the model for adjustment to determine the corrected risk factors for PC. The receiver operating characteristic (ROC) curve is used to evaluate the ability of respiratory pattern parameters to identify HF combined with PC. A significance level of P<.05 was considered statistically significant. Data processing and statistical analysis were performed using Python 3.7 and IBM SPSS 25.

A total of 62 patients with HF (mean age 60.53, SD 14.92 y, 38/62, 61% male) were enrolled between May 2021 and November 2022, including 19 (31%) patients with reduced EF and 43 (69%) with preserved EF. The comorbidities included 30 (48%) patients with hypertension, 17 (27%) with diabetes. By LUS examination, there were 44 (80%) patients with PC (PC group) and 18 (20%) patients without PC (non-PC group). By comparison, the PC group had lower diastolic blood pressure (DBP) and higher N-terminal pro-B-type natriuretic peptide than the non-PC group, while the rest of the parameters listed in Table 1 were comparable.

Table 2 shows the comparison of respiratory parameters between patients with HF with PC and non-PC. With regard to the respiratory rate and time, the PC group had longer mean TE but a higher mean TE_ratio when compared with the non-PC group, as illustrated in Figure 2. No difference was found in the parameters of RA. MSE analysis indicates similar SampEn values of BBI on various time scale factors among the 2 patient groups with HF (Figure 3A). However, the SampEn values of RA across the scale factors are significantly higher in the PC group compared with those without PC, with higher RA area_1_5 and RA area_6_20 (Figure 3B).

By univariate logistic regression analysis, only respiratory parameters of mean TE_ratio, RA area_1_5, and RA area_6_20, and clinical parameters of DBP, NYHA IV, and logNT-proBNP (logarithmically transformed N-terminal pro-B-type natriuretic peptide) were associated with PC (P<.05) (Table 3). By multivariate logistic regression, the RA area_1_5 and RA area_6_20 were independently related to PC after adjusting for age, sex, DBP, NYHA IV, and logNT-proBNP (Table 4).

Through ROC curve analysis, the ability of all relevant indicators to distinguish between participants with and without PC was compared. The AUC for RA area_1_5 was the largest at 0.75 (95% CI 0.63-0.88, P=.002, sensitivity of 65.9%, and specificity of 83.3%), as presented in Table 5. Figure 4 displays the ROC curve of a multivariate logistic regression model constructed using combined parameters of DBP, logNT-proBNP, NYHA IV, mean TE_ratio, RA area_1_5, and RA area_6_20, with an AUC of 0.91 (95% CI 0.84-0.98, P<.001), sensitivity of 77.3%, and specificity of 88.9%.

In the current observational exploratory study, we demonstrated the feasibility and capability of a wearable device (“SensEcho” by Beijing SensEcho Science & Technology Co) to differentiate PC severity by detecting nocturnal respiratory signals (cycle and amplitude) in hospitalized patients with HF. First, we found the hospitalized patients with HF with PC had a longer mean TE and subsequently an increased TE_ratio than those without PC. Second, by using the MSE algorithm to assess respiratory complexity, the increased RA complexity could independently indicate a PC condition.

It was well-known that patients with HF showed increased rest respiratory rate and decreased amplitude than healthy controls, characterized as a “rapid and shallow” pattern. However, our study found that these parameters about respiratory rate and amplitude recorded at nighttime failed to differentiate patients with HF with and without PC. Unexpectedly, patients with PC were found to show a longer expiratory time and a ratio, which is commonly seen in patients with asthma or chronic pulmonary disease who had increased airway resistance but seldom reported for patients with HF. Theoretically, a prolonged expiratory time is reasonable and seen in patients with HF and PC. At first, the PC-induced airway compression or bronchial congestion–induced mucosal edema could result in bronchial flow limitation, increased ventilation-perfusion mismatch, and increased work of breathing [26]. In addition, experimental studies have shown fluid overload leads to a decrease of the diameter of both small and large airways [27]. All these factors could contribute to the prolonged expiratory time as seen in patients with obstructive lung disease [28]. Second, ventilation at rest requires only the inspiratory muscles, but the expiratory is a passive process in normal conditions. When with disease or increased ventilatory demands, the expiratory muscles begin to play a role. For heart failure, Zoccal and Machado [29] found that volume overload in rats with HF displayed active expiration and the recruitment of abdominal muscles. Furthermore, enhanced central chemoreflex gain was observed in the volume overload HF model and played a role in the development of respiratory dysfunction. After establishing a volume-overload HF model in rats via an arteriovenous fistula, episodic hypercapnic stimulation could further increase active expiration [30]. Taken together, prolonged TE or an elevated expiratory ratio may serve as promising indicators of HF-induced PC. However, these findings should be interpreted with caution in patients with HF with coexisting obstructive lung disease (eg, COPD or asthma)—a population excluded from this study. Such comorbidities share overlapping pathophysiological mechanisms that alter breathing patterns, including bronchial flow limitation, increased ventilation-perfusion mismatch, and elevated work of breathing. Thus, the specificity of these respiratory metrics for PC detection could be compromised in patients with concurrent airflow obstruction.

Periodic breath is a consequence of respiratory control system instability and is well-recognized in patients with HF due to prolonged circulatory time, increased carbon dioxide chemosensitivity, lower carbon dioxide setpoint, activation of pulmonary C-fibers due to congestion, increased sympathetic nerve activity, or modulation of the ergoreflex, etc [26]. It is characterized by waxing and waning of tidal volume with or without interposed apnea, which contributes to the breath pattern (time, rate, and amplitude) complexity. Recognizing the complexity of periodic breathing could serve as a prognosticator. For instance, Asanoi et al [31] calculated the respiratory instability based on the frequency distribution of the respiratory spectral components and the ultra-low frequency component and found that it has prognostic significance independent of sympathetic activation. Furthermore, respiratory complexity is also closely related to disease severity, a valuable guide in the management and treatment of patients with HF. Similarly, Corrà et al [32] assessed cyclic fluctuations in minute ventilation and found that patients with higher variability in respiratory rate had worse clinical status and exercise capacity.

For the first time, this study demonstrates the feasibility of using wearable devices to continuously monitor respiratory signals in patients with HF, coupled with MSE analysis—a robust method for quantifying physiologic time-series complexity [33-35]. Our findings reveal that elevated MSE amplitude, reflecting increased respiratory complexity, is strongly associated with PC. Given that worsening PC drives HF exacerbations and hospitalizations, this home-based MSE amplitude monitoring approach may offer a novel, noninvasive tool for early detection of worsening PC, enabling timely interventions to reduce rehospitalization rates [36]. However, several critical steps are needed to translate these findings into clinical practice. First, prospective validation in larger, diverse HF cohorts (eg, across NYHA classes, etiologies, and comorbidities) is essential to confirm generalizability. Notably, future studies should include patients with concurrent conditions like COPD or obesity, which may confound respiratory signal interpretation. Second, longitudinal studies should assess whether day-to-day MSE amplitude fluctuations precede symptomatic HF deterioration and set a rule to alert, potentially serving as a predictive biomarker. Third, seamless integration of wearable-derived MSE data into electronic health records and telehealth platforms, leveraging machine learning models to analyze MSE patterns alongside multimodal data (eg, daily vital signs and weight trends) for enhanced exacerbation prediction and risk stratification.

This study has several limitations that warrant consideration. First, as an exploratory investigation, our findings are derived from a small, single-center cohort, which inherently limits the generalizability of the results. External validation in larger, multicenter populations—particularly those encompassing broader demographic and clinical spectra—is necessary to confirm the robustness of a respiratory complexity biomarker in HF. Second, we excluded patients with concurrent pulmonary diseases to isolate HF-specific respiratory patterns. Consequently, the applicability of wearable device–based respiratory pattern analysis in patients with HF with comorbid pulmonary conditions remains uncertain. Third, to mitigate potential confounders from daytime physical activity, our analysis focused exclusively on nocturnal respiratory signals. While this approach enhances signal purity, it precludes assessment of whether daytime respiratory patterns—which may capture dynamic responses to exertion or postural changes—hold superior predictive value for HF exacerbations. Future studies should evaluate 24-hour respiratory monitoring to determine the optimal sampling window for clinical prediction.

In this small-scale exploratory study, wearable-based MSE analysis successfully distinguished between hospitalized patients with HF with and without PC, as evidenced by prolonged expiratory phases and increased RA complexity in the PC group.