Next Article in Journal
The Development of User-Centric Design Guidelines for Web3 Applications: An Empirical Study
Next Article in Special Issue
Enhancing Cryptographic Solutions for Resource-Constrained RFID Assistive Devices: Implementing a Resource-Efficient Field Montgomery Multiplier
Previous Article in Journal
Optimizing Loss Functions for You Only Look Once Models: Improving Object Detection in Agricultural Datasets
Previous Article in Special Issue
Exploring the Design for Wearability of Wearable Devices: A Scoping Review
 
 
Article
Peer-Review Record

Heart Rate Variability-Based Stress Detection and Fall Risk Monitoring During Daily Activities: A Machine Learning Approach

by Ines Belhaj Messaoud 1 and Ornwipa Thamsuwan 2,*
Reviewer 1: Anonymous
Reviewer 2:
Reviewer 4: Anonymous
Submission received: 2 December 2024 / Revised: 17 January 2025 / Accepted: 20 January 2025 / Published: 30 January 2025
(This article belongs to the Special Issue Wearable Computing and Activity Recognition)

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The manuscript is well presented, but has a major problem related to the age of the sample. The authors report that the focus is on older people. Nevertheless, the age of the sample is far below the classical age for older people (i.e. 65 years). This severely compromises the validity of the study results.

Author Response

Comment 1: The manuscript is well presented, but has a major problem related to the age of the sample. The authors report that the focus is on older people. Nevertheless, the age of the sample is far below the classical age for older people (i.e. 65 years). This severely compromises the validity of the study results.

Reply 1: Thank you for your observation regarding the age of our sample and its implications for the study focus. We have reframed the article to cover healthy adults in general, rather than exclusively elderly. Our study recruited participants aged 40+ years, a population that includes both middle-aged adults and older adults (aged 65 and above) since balance and fall risks are less common in younger individuals under 40.

To ensure clarity, we have updated the manuscript to consistently reflect this broader. These changes are now reflected in the following places:

  • Line 19, remove “among the elderly”.
  • Line 28, remove “in the elderly”.
  • Line 140, remove “in the older adults” from the main objective, and the objective is framed to “healthy adults”.
  • Line 162, the experimental activities are those that “healthy persons”
  • Line 250, remove “in the elderly”.
  • Line 278, remove “among elderly people”.
  • Line 582, our dataset “from people over 40 years old”.
  • Line 601, remove “in the elderly”.
  • Line 610, remove “elderly”.
  • Line 617, add “vulnerable people such as” in front of the elderly.
  • Line 622, remove “in the elderly”.

Reviewer 2 Report

Comments and Suggestions for Authors

The paper studies machine learning approaches for fall detection from heart rate variability and studies the feasibility of using machine learning for this problem. Overall I like the idea and the paper is reasonably well written. However, the sample size of 14 is too small to have any significance in terms of the reliability of the results. I would like to see the results for a larger sample size. With the current sample size, I am not convinced about the efficacy given that the reliability if questionable. 

Since it might be a lot of work to collect more data, I would recommend that this be published as some sort of a vision/idea paper with preliminary results, rather than a complete journal paper. It's up to the authors to withdraw and resubmit to submit a major revision with significantly more data points.

Author Response

Comment 1: The paper studies machine learning approaches for fall detection from heart rate variability and studies the feasibility of using machine learning for this problem. Overall I like the idea and the paper is reasonably well written. However, the sample size of 14 is too small to have any significance in terms of the reliability of the results. I would like to see the results for a larger sample size. With the current sample size, I am not convinced about the efficacy given that the reliability if questionable.

Since it might be a lot of work to collect more data, I would recommend that this be published as some sort of a vision/idea paper with preliminary results, rather than a complete journal paper. It's up to the authors to withdraw and resubmit to submit a major revision with significantly more data points.

Reply 1: Thank you for your thoughtful feedback and acknowledgment of the potential contributions to the field. We agree that the small sample size of 14 participants limits the generalizability and reliability of the results. The study was conducted as an initial exploratory effort to investigate the feasibility of using machine learning approaches in the domains of fall detection based on heart rate variability (HRV).

To address your concern, we have clarified in the first paragraph of the limitations, line 590-591, that “Nevertheless, this study provided methods and preliminary results that could serve as a proof of concept to guide future research”.

Reviewer 3 Report

Comments and Suggestions for Authors

The aim of this paper is to establish relationships between patterns or characteristics observed in (1) HRV and HR and (2) loss of balance and risk of falling. This is an intriguing hypothesis that involves a multifaceted approach in terms of the timescales involved. In this regard, the researchers make a fairly complete assessment of the spectrum of time scales, including tools and measures to reveal specific features of the time sequences (i.e. RR sequences), such as the 2-D Poincaré plots. The work seems to me to be a fairly solid and well-documented research, and an organized and well-written manuscript. Therefore, I will be in favor of publishing this work if the authors can address the following observations and comments:

1-      For the lay reader, it would be useful to better describe the objectives and specific basis of the classification schemes described.

2-      In relation to the hypothesis of this work, the four performance indices used (e.g. accuracy, recall, precision and F1 score) on HRV and HR data must at some point be related to falling events or at least to a subjective perception of fall risk or loss of balance.

3-      It would be very useful for future studies if the authors could provide more specific descriptions and associated intensity levels of these subjective perceptions in a table.

4-      These loss of balance and perceptions of fall risk (or actual falls) may indeed be strongly related to impairments in the VOR (vestibulo-ocular reflex). Did the authors try to establish a relationship between these potential impairments and the higher presence of certain HRV or HR characteristics?

5-      In order to assess the value of the central hypothesis, it would be necessary to see the precise relationships found between the risk of falling and HRV and/or HR. In this respect, the conclusions could provide a better understanding of the study if, for example, the results could provide information such as:

a.      What is a better predictor of fall risk: higher or lower HRV or HR?

b.      Do the authors see an optimal predictor based on a non-trivial combination of HR and HRV based on their findings?

6-      In the initial assessment of HRV features, the authors correctly use Poincare plots, which provide at least a three-beats sequential information. However, the authors should be aware of existing generalized graphical Poincare multidimensional methods (e.g., Ganan-Calvo & Fajardo 2016; Ganan-Calvo et al. 2018, Sci. Rep. 8:9897) specifically developed for HRV that can provide much more qualitative and quantitative information. Since the authors did not use these advanced methods, it would be highly recommended that they acknowledge their potential use in their research, along with their assessment of the potential value that the data on more complex sequential HRV features can offer.

Author Response

Comment 1: The aim of this paper is to establish relationships between patterns or characteristics observed in (1) HRV and HR and (2) loss of balance and risk of falling. This is an intriguing hypothesis that involves a multifaceted approach in terms of the timescales involved. In this regard, the researchers make a fairly complete assessment of the spectrum of time scales, including tools and measures to reveal specific features of the time sequences (i.e. RR sequences), such as the 2-D Poincaré plots. The work seems to me to be a fairly solid and well-documented research, and an organized and well-written manuscript. Therefore, I will be in favor of publishing this work if the authors can address the following observations and comments: For the lay reader, it would be useful to better describe the objectives and specific basis of the classification schemes described

Reply 1: Thank you for your positive feedback and for highlighting the importance of clarifying the objectives and basis of the classification schemes for a lay audience. We revised the objective section as shown in lines 139-148 that “this study aimed to explore how HRV can inform our understanding of stress and postural balance in healthy adults. The specific objectives of this study are twofold. First, we identified to identify patterns in HRV that may indicate stress during daily activities, using clustering techniques to uncover natural groupings in the data. Second, through supervised classification, we evaluated whether HRV features can be used to predict fall risk by categorizing balance states into "low risk" (steady balance) and "high risk" (impending loss of balance) through supervised machine learning methods. This work is meant to establish a foundation for developing pragmatic tools, such as wearable devices, to monitor balance continuously and warn individuals of potential falls.”

Comment 2: In relation to the hypothesis of this work, the four performance indices used (e.g. accuracy, recall, precision and F1 score) on HRV and HR data must at some point be related to falling events or at least to a subjective perception of fall risk or loss of balance.

Reply 2: In this study, we employed the Berg Balance Scale (BBS), a widely used tool for assessing functional balance, as a subjective measure of fall risk. BBS scores were categorized into two classes: steady balance ('low risk') and potential balance loss ('high risk').

In line 315-322, we revise clarified that “The performance metrics—accuracy, recall, precision, and F1 score—were used to evaluate the models' ability to classify these binary balance states. Among these, recall was critical to ensure the model captured all high-risk moments, minimizing missed instances of instability in balance. Precision validated the reliability of the model in predicting high-risk moments, ensuring the identified instances were indeed associated with potential balance loss. The F1 score, as the harmonic mean of precision and recall, provided a balanced evaluation of the model’s performance, especially important in the presence of class imbalance.” These metrics are therefore directly linked to the subjective, BBS-based perception of fall risk.

Comment 3: It would be very useful for future studies if the authors could provide more specific descriptions and associated intensity levels of these subjective perceptions in a table.

Reply 3: If we understand you correctly, the subjective perceptions may mean the BBS score. While the BBS itself does not provide subjective intensity levels of fall risk, we acknowledge the benefit of having some associated descriptions. In the line 174-176, the data collection section, we added that “The score of 4 means that the participant exhibits no noticeable imbalance during tasks, whereas the score of 0 means that the participant is unable to maintain balance and needs support.”

Comment 4: These loss of balance and perceptions of fall risk (or actual falls) may indeed be strongly related to impairments in the VOR (vestibulo-ocular reflex). Did the authors try to establish a relationship between these potential impairments and the higher presence of certain HRV or HR characteristics?

Reply 4: Thank you very much for highlighting the potential link between VOR (vestibulo-ocular reflex) impairments and HRV or HR characteristics. In our study, we did not establish a direct relationship between VOR impairments and HRV or HR characteristics. This was primarily due to our final choice of Gradient boosting model, which, unlike logistic regression, do not provide direct measures of the strength of contribution for individual features. Although we initially employed logistic regression to investigate these relationships. the recall and accuracy of the logistic regression model were found to be low, limiting its utility for reliable predictions.

However, we acknowledge the importance of VOR in balance control and its potential impact on HRV patterns, such as decreased parasympathetic regulation or increased sympathetic activity. We include this potential avenue for future work in the lines 633-637 that “Additionally, as loss of balance and perceptions of fall risk may be related to impairments in the vestibulo-ocular reflex, which is crucial for preserving gaze stability during head movements, future research could benefit from integrating vestibular function tests or dynamic visual acuity assessments with HRV and HR analysis.”

Comment 5: In order to assess the value of the central hypothesis, it would be necessary to see the precise relationships found between the risk of falling and HRV and/or HR. In this respect, the conclusions could provide a better understanding of the study if, for example, the results could provide information such as:

  1. What is a better predictor of fall risk: higher or lower HRV or HR?
  2. Do the authors see an optimal predictor based on a non-trivial combination of HR and HRV based on their findings?

Reply 5: Thank you for your insightful suggestions regarding the relationship between fall risk and HRV or HR as predictors. While the main goal of our study was not to pinpoint the HRV or HR parameter that best predicts fall risk, we acknowledge the importance of such analysis in assessing the value of the central hypothesis. In our study, we instead adopted a two-phase strategy. First, we applied a K-means clustering to explore whether participant data exhibited observable HRV patterns, primarily focusing on stress-related variables and their trends in connection to physical activity. Second, we used supervised classification to evaluate the feasibility of predicting fall risk based on these features. To identify which HRV or HR parameters are the most reliable predictors of fall risk, future analyses could incorporate feature importance techniques Thus, we added in the conclusion and future work section lines 621-627 that “However, while this study focused on model performance as a whole, identifying the specific HRV or HR parameters most indicative of fall risk remains an open question. Future analyses could address this by incorporating feature importance techniques that would allow for a more detailed quantification of the contribution of individual HRV or HR parameters to predictive models. Although outside the scope of the current study, such analyses could provide valuable insights into the specific role of each parameter in fall risk assessment.”

Comment 6: In the initial assessment of HRV features, the authors correctly use Poincare plots, which provide at least a three-beats sequential information. However, the authors should be aware of existing generalized graphical Poincare multidimensional methods (e.g., Ganan-Calvo & Fajardo 2016; Ganan-Calvo et al. 2018, Sci. Rep. 8:9897) specifically developed for HRV that can provide much more qualitative and quantitative information. Since the authors did not use these advanced methods, it would be highly recommended that they acknowledge their potential use in their research, along with their assessment of the potential value that the data on more complex sequential HRV features can offer.

Reply 6: Thank you for pointing out advanced multidimensional Poincaré methods for the analysis of HRV. We appreciate the reviewer’s suggestion and have cited the relevant references (Gañán-Calvo & Fajardo, 2016; Gañán-Calvo et al., 2018) to provide a broader context for readers interested in advancing HRV analysis in this domain. In the line 556-564, we added a paragraph about these work that “While our investigation used only traditional forms of Poincaré plots to analyze HRV, we realize there are more advanced methods. Specifically, in two studies presented by Ganan-Calvo and co-authors (cite), generalized graphical Poincaré multidimensional methods presented more informative and complex sequential HRV features with qualitative and quantitative assessments of the HRV data. The scope of the present study focused on established, widely-used HRV analysis methods; therefore, these multidimensional Poincaré methods were not incorporated. However, future research could leverage these methods to enhance the understanding of the intricate dynamics underlying HRV and their relationship to fall risk.”

Reviewer 4 Report

Comments and Suggestions for Authors

Please see the PDF attached. 

Comments for author File: Comments.pdf

Author Response

Comment 1: balance of what

Reply 1: In line 36, we clarified that it is “postural balance”.

Comment 2: it is clear to see a connection between HRV and postural balance as evidence in the following section, therefore it is not suitable to use the word of hypothesis

Reply 2: In line 36, we removed “it is hypothesized that”.

Comment 3: this is confusing. In RRSD, SD is a postfix, while in this case I assume the RMS is a prefix, and SD is not refering standard deviation any more. I prefer to use subscriptions.

Reply 3: In line 63-64, we used the standard abbreviations for the SDRR and RMSSD and mentioned the full names of the terms as “Standard Deviation of RR Intervals” and “Root Mean Square of Successive Differences of the RR Intervals”. We believe this approach maintains alignment with standard practices in HRV analysis.

Comment 4: why? although you have confirmed this finding in this study, this is yet not convincing. maybe try to tune the hyper-parameters. meanwhile this statement wont weaken or enhance your novelty but only leads to doubtful comments.

Reply 4: We agree with you that the performance might have to deal with the hyperparameter tuning. We thus removed the statement “Notwithstanding, when it comes to generalization across various datasets, HRV-based models, such as those employing random forests and gradient boosting, performed better than deep learning models.” This is reflected in the line 112.

Comment 5: definition of the acronym for standard deviation

Reply 5: In line 151, we added “standard deviation” before the abbreviation SD.

Comment 6: it is also very interesting to show if the dataset is balanced. Are those activities equally distributed?

Reply 6: We added in the lines 348-349 that “In our study, activities were not equally represented in the dataset. The counts of each activity type were as follows: MOVEMENT: 98 cases, SIT: 28 times, STAND: 98 times.” We acknowledge that this may lead to imbalance, especially on the lower occurrences of the SIT activity, and may bias our results. Still, our data on STAND and MOVEMENT activities are more of the interest when it comes to fall and balance loss.

Comment 7: worth to be mentioned in the abstract and conclusion

Reply 7: Thank you and we agree with you. In the conclusion, line 618, we add “and the cluster labels”. In the abstract, lines 10-13, we revised the text that “In the second phase, integrating the cluster labels obtained from the first phase together with HRV features into machine learning models for fall risk classification, we found that…”

Comment 8: specifications of the wavelet filter you used? And why the use of a wavelet filter can ensure your signal informative and clear. You have mentioned "an improper wavelet basis might result in signal distortions."

Reply 8: Thank you again for your feedback. In lines 547-551, we added the details of wavelet filter that “The Daubechies 4 (db4) wavelet is successful in biomedical signal processing, we used it in our investigation to filter IBI cardiac data. Multiple decomposition levels provide a detailed depiction of HRV data using the db4 wavelet, which was selected for its harmony between smoothness and sensitivity to abrupt changes in signal properties.”

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The authors have made a significant change modifying the reference population. The manuscript is now well-centred.

Author Response

Comment 1 : The authors have made a significant change modifying the reference population. The manuscript is now well-centred.

Response 1 : Thank you very much for reviewing the article and providing your feedback twice.

Reviewer 3 Report

Comments and Suggestions for Authors

The authors have satisfactorily addressed virtually all of my observations in their new version, and I now consider this interesting manuscript publishable.

Author Response

Comment 1 : The authors have satisfactorily addressed virtually all of my observations in their new version, and I now consider this interesting manuscript publishable.
Response 1 : Thank you very much for your time reviewing our study twice and for considering this work sufficient for publication.

Back to TopTop