Presentation
Self-Modulation of Heart Rate Variability with Haptic Biofeedback
SessionPoster Session 2
DescriptionBackground Information
Autonomic Nervous System (ANS), as the part of the nervous system that controls automatic and involuntary activities such as breathing and cardiovascular activity, plays a vital role in maintaining the body’s internal stability and responsiveness to external stressors (McCorry 2007). Dysregulation of ANS can disrupt the balance which imposes a risk to the regular performance of vital organs such as heart and lungs, while also creating mental risks such as heightened anxiety, lower stress resilience and cognitive dysfunction (Grippo 2017; McCorry 2007).
Biofeedback training has been shown to restore ANS balance by regulating the activity of the sympathetic and parasympathetic branches through simple and repetitive tasks such as paced breathing (Perry and Hambrick 2008; de Souto Barbosa et al. 2023). The aim of our study is to evaluate the effectiveness of haptic biofeedback, involving guided breathing, in improving ANS regulation. For this purpose, the trends related to heart rate variability (HRV), one of the key biomarkers of ANS function, are analyzed.
Methodology
Device and Intervention: The training is designed based on breathing patterns that are delivered, controlled and adjusted through a commercial wearable device, Lief Patch, synced with a smartphone application. Heart metrics are collected continuously via the patch and analyzed in real-time. When the user’s HRV drops below a threshold (indicating stress), the user is prompted via the patch to practice a “dose” of guided breathing for 180 seconds or until their HRV improves above the threshold. In addition, challenges participants encountered with adhering to the program were collected through a survey and telehealth usability questionnaire. Semi-structured interviews were conducted at the end of the program to gain more detail and deeper insights into participant’s experience.
Participants and Data: In total, 13 participants were recruited through [recruitment method] at a large public hospital in the southern United States. Each participant enrolled in a 12-week program, in which they were asked to wear the patch for as long as possible each day, and respond to the “doses” of biofeedback delivered through the patch. Of these, five participants did not complete the program. Datasets from the remaining eight participants were included in the analysis.
The dataset from each participant consisted of heart metrics collected continuously by the patch included:
● Instantaneous Heart Rate (IHR) (beats per minute),
● RR interval (time between two consecutive R-ways on ECG/EKG, in milliseconds),
● HRV,
● and “Zone” – an integer ranging from 1 to 5 that represents status of the heart function, with 1 demonstrating high risk to 5 for low risk or safe state
● Timestamp for each recorded datapoint (recording frequency was every half second on average)
Data features associated with each breathing intervention or “dose”, including:
● Pre-dose and Post-dose IHR
● Pre-dose and Post-dose HRV,
● Pre-dose and Post-dose Zone,
● Duration of the intervention (ranging from 0 to 180 seconds),
● Dose type (that is related to the type of stimulation or sensory cue that the user received as a label. The user could choose between four different types offered)
Survey and telehealth usability questionnaire were designed to extract information on the user’s experience with the patch, including topics such as consistency with the program, perceived effectiveness of the program or improved access to healthcare services. The participants were asked to fill out the combined questionnaires three times during the program. The questions in the semi-structured interview protocol were mainly targeting the barriers and the adaptations users employed to overcome those barriers. Total of 8 responses to the questionnaire from five participants were collected and two participants participated in the interview.
Data Processing and Analysis: The rows with noisy data were removed. For the analysis of the continuous data, Heart Rate Variability (HRV) was calculated manually from the RR intervals using the RMSSD (Root Mean Square of Successive Differences) method. RMSSD is the square root of the mean of squared successive differences between adjacent RR intervals (DeGiorgio et al., 2010). For analysis of long-term trends, the following variables were defined and computed:
● Base HRV: HRV over the first five minutes of wearing the patch each day.
● End HRV: HRV over the last five minutes of wearing the patch each day.
● Delta HRV: the difference between Base HRV and End HRV
● Average HRV: the daily average HRV based on data collected by the patch.
These variables were calculated for each day. To study the short-term effect of biofeedback training, pre- and post-dose data were analyzed, and new variables were defined as follows:
● Average pre-dose HRV,
● Average post-dose HRV,
● DeltaHRV (as the difference between the average pre and post-dose HRV),
● Average pre-dose IHR,
● Average post-dose IHR,
● DeltaIHR (as the difference between the average pre and post-dose IHR),
● Average pre-dose zone (as numeric average of pre-dose zone),
● Average post-dose zone (as numeric average of post-dose zone),
● Dose frequency (as the number of interventions per day)
● Average duration of all recorded interventions per day
A final dataset comprising the above variables was developed for all participants. Rows with missing data were omitted, and each feature was then tested for normality (using Shapiro-Wilk test), autocorrelation (using Durbin-Watson test) and stationarity (using Dickey-Fuller unit root test). The results indicated that data did not follow the normal distribution, and no evidence of autocorrelation was observed. The Dickey–Fuller test results showed mixed outcomes, with some data features demonstrating stationarity across participants, while others remained non-stationary.
Linear Mixed-Effects Models (LMM) were used to analyze the long-term effect of the program, specifically, the effects of the duration of dose (categorized into six 30-second intervals within the 0–180s span) on HRV; and of dose type in relation to changes in HRV. For analyzing the short-term (immediate) effects of duration and dose type, Wilcoxon signed-rank test was conducted. Finally, survey data were analyzed descriptively and interview transcriptions were analyzed for prominent themes.
Findings and Outcomes
The LMM results highlighted a significant decrease in dose frequency (coef=-0.083 , p-value=0.026), increased average pre-dose zone (coef=0.014 , p-value=0.000) and increased average post-dose zone (coef=0.011 , p-value=0.009) over time (days). The positive trend in pre- and post-dose zone alongside the deceasing trend in the number of doses over time suggests improvement in ANS function, requiring fewer dose-interventions to maintain the stable state. The results of the Wilcoxon signed-rank test confirmed an immediate positive change in HRV after intervention (V = 245,647, p < 0.001). With respect to the intervention characteristics, the LMM found a significant increases in HRV with each consecutive 30 sec window. Results also demonstrated a significant effect of the type of the dose over the change in HRV with type ‘Practice’ leading to higher post HRV.
While a systematic thematic analysis of participant responses to the survey was not possible due to the small volume of data, the responses indicated that most participants used the patch regularly and reported some benefits in stress awareness and control. In contrast, telehealth usability feedback was more mixed and polarized. The qualitative analysis of the interviews demonstrated some of the challenges participants faced such as technical challenges with the patch/application, difficulty in doing the doses due to external/environmental factors, and discomfort with wearing the patch.
Autonomic Nervous System (ANS), as the part of the nervous system that controls automatic and involuntary activities such as breathing and cardiovascular activity, plays a vital role in maintaining the body’s internal stability and responsiveness to external stressors (McCorry 2007). Dysregulation of ANS can disrupt the balance which imposes a risk to the regular performance of vital organs such as heart and lungs, while also creating mental risks such as heightened anxiety, lower stress resilience and cognitive dysfunction (Grippo 2017; McCorry 2007).
Biofeedback training has been shown to restore ANS balance by regulating the activity of the sympathetic and parasympathetic branches through simple and repetitive tasks such as paced breathing (Perry and Hambrick 2008; de Souto Barbosa et al. 2023). The aim of our study is to evaluate the effectiveness of haptic biofeedback, involving guided breathing, in improving ANS regulation. For this purpose, the trends related to heart rate variability (HRV), one of the key biomarkers of ANS function, are analyzed.
Methodology
Device and Intervention: The training is designed based on breathing patterns that are delivered, controlled and adjusted through a commercial wearable device, Lief Patch, synced with a smartphone application. Heart metrics are collected continuously via the patch and analyzed in real-time. When the user’s HRV drops below a threshold (indicating stress), the user is prompted via the patch to practice a “dose” of guided breathing for 180 seconds or until their HRV improves above the threshold. In addition, challenges participants encountered with adhering to the program were collected through a survey and telehealth usability questionnaire. Semi-structured interviews were conducted at the end of the program to gain more detail and deeper insights into participant’s experience.
Participants and Data: In total, 13 participants were recruited through [recruitment method] at a large public hospital in the southern United States. Each participant enrolled in a 12-week program, in which they were asked to wear the patch for as long as possible each day, and respond to the “doses” of biofeedback delivered through the patch. Of these, five participants did not complete the program. Datasets from the remaining eight participants were included in the analysis.
The dataset from each participant consisted of heart metrics collected continuously by the patch included:
● Instantaneous Heart Rate (IHR) (beats per minute),
● RR interval (time between two consecutive R-ways on ECG/EKG, in milliseconds),
● HRV,
● and “Zone” – an integer ranging from 1 to 5 that represents status of the heart function, with 1 demonstrating high risk to 5 for low risk or safe state
● Timestamp for each recorded datapoint (recording frequency was every half second on average)
Data features associated with each breathing intervention or “dose”, including:
● Pre-dose and Post-dose IHR
● Pre-dose and Post-dose HRV,
● Pre-dose and Post-dose Zone,
● Duration of the intervention (ranging from 0 to 180 seconds),
● Dose type (that is related to the type of stimulation or sensory cue that the user received as a label. The user could choose between four different types offered)
Survey and telehealth usability questionnaire were designed to extract information on the user’s experience with the patch, including topics such as consistency with the program, perceived effectiveness of the program or improved access to healthcare services. The participants were asked to fill out the combined questionnaires three times during the program. The questions in the semi-structured interview protocol were mainly targeting the barriers and the adaptations users employed to overcome those barriers. Total of 8 responses to the questionnaire from five participants were collected and two participants participated in the interview.
Data Processing and Analysis: The rows with noisy data were removed. For the analysis of the continuous data, Heart Rate Variability (HRV) was calculated manually from the RR intervals using the RMSSD (Root Mean Square of Successive Differences) method. RMSSD is the square root of the mean of squared successive differences between adjacent RR intervals (DeGiorgio et al., 2010). For analysis of long-term trends, the following variables were defined and computed:
● Base HRV: HRV over the first five minutes of wearing the patch each day.
● End HRV: HRV over the last five minutes of wearing the patch each day.
● Delta HRV: the difference between Base HRV and End HRV
● Average HRV: the daily average HRV based on data collected by the patch.
These variables were calculated for each day. To study the short-term effect of biofeedback training, pre- and post-dose data were analyzed, and new variables were defined as follows:
● Average pre-dose HRV,
● Average post-dose HRV,
● DeltaHRV (as the difference between the average pre and post-dose HRV),
● Average pre-dose IHR,
● Average post-dose IHR,
● DeltaIHR (as the difference between the average pre and post-dose IHR),
● Average pre-dose zone (as numeric average of pre-dose zone),
● Average post-dose zone (as numeric average of post-dose zone),
● Dose frequency (as the number of interventions per day)
● Average duration of all recorded interventions per day
A final dataset comprising the above variables was developed for all participants. Rows with missing data were omitted, and each feature was then tested for normality (using Shapiro-Wilk test), autocorrelation (using Durbin-Watson test) and stationarity (using Dickey-Fuller unit root test). The results indicated that data did not follow the normal distribution, and no evidence of autocorrelation was observed. The Dickey–Fuller test results showed mixed outcomes, with some data features demonstrating stationarity across participants, while others remained non-stationary.
Linear Mixed-Effects Models (LMM) were used to analyze the long-term effect of the program, specifically, the effects of the duration of dose (categorized into six 30-second intervals within the 0–180s span) on HRV; and of dose type in relation to changes in HRV. For analyzing the short-term (immediate) effects of duration and dose type, Wilcoxon signed-rank test was conducted. Finally, survey data were analyzed descriptively and interview transcriptions were analyzed for prominent themes.
Findings and Outcomes
The LMM results highlighted a significant decrease in dose frequency (coef=-0.083 , p-value=0.026), increased average pre-dose zone (coef=0.014 , p-value=0.000) and increased average post-dose zone (coef=0.011 , p-value=0.009) over time (days). The positive trend in pre- and post-dose zone alongside the deceasing trend in the number of doses over time suggests improvement in ANS function, requiring fewer dose-interventions to maintain the stable state. The results of the Wilcoxon signed-rank test confirmed an immediate positive change in HRV after intervention (V = 245,647, p < 0.001). With respect to the intervention characteristics, the LMM found a significant increases in HRV with each consecutive 30 sec window. Results also demonstrated a significant effect of the type of the dose over the change in HRV with type ‘Practice’ leading to higher post HRV.
While a systematic thematic analysis of participant responses to the survey was not possible due to the small volume of data, the responses indicated that most participants used the patch regularly and reported some benefits in stress awareness and control. In contrast, telehealth usability feedback was more mixed and polarized. The qualitative analysis of the interviews demonstrated some of the challenges participants faced such as technical challenges with the patch/application, difficulty in doing the doses due to external/environmental factors, and discomfort with wearing the patch.
Event Type
Poster Presentation
TimeTuesday, March 244:45pm - 6:15pm EDT
LocationRhinelander Gallery
Digital Health