Heart Rate Variability to Automatically Identify Hyperbaric States Considering Respiratory Component

Round 1

Reviewer 1 Report

The Authors introduced an interesting approach in estimating pressure during a subject’s immersion by measuring HRV parameters, suggesting that a misidentification of the actual pressure can be a warning or risk factor for diving accidents.

The work is very interesting and well written. However, my main concern is about the principal hypothesis of this study: how can we be sure that misclassification is related to the risk of diving accidents? The Authors correctly processed signals in order to filter out ectopic beats, but they say nothing about exclusion criteria of enrolled subjects, like arrhythmias, other autonomous dysfunctions, etc… that may affect estimation of HRV parameters. In other words, misclassification can be due to many other causes, not only to non-normal response to diving, and such causes must be excluded prior to coming to any conclusion. Models may fail also because of their intrinsic limitations in terms of accuracy, due to the quantity and quality of the data they have been trained on. 

Therefore, 

- a better characterization of enrolled people is required

- a larger sample is required, including people that have experienced any significant clinical outcome during diving experiences.

The Authors state, in the abstract, that “The study’s primary objective is to identify individuals whose physiological responses deviate from the rest of the study population by automatically monitoring atmospheric pressure levels to which they are exposed, using parameters derived from their heart rate variability (HRV)”. However, in the last part of the Introduction, they state: “The primary objective of this study is to develop an automated system capable of accurately identifying the atmospheric pressure experienced by subjects.” In the next sentences of this section, they also clarify that the indirect identification/misidentification of pressure can be a marker of the subject’s normal/abnormal reaction to immersion. The information is there, but for the sake of clarity, the two sections need to be congruent.

A few minors:

- line 139: low-pass or, better, high-pass? Moreover, 0.03 seems to be a too low cut-off, as baseline wander can occur at up to 0.7Hz

- line 140: there is a repetition

- line 240: there is a repetition

- names of variables in Figure 1 are not consistent with actual variable names. Please correct, or state that Pres = P┴, Presp = PR, and so on.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

This article about automatic identification of hyperbaric states is useful for safety protocols of divers. I find it appropriate to publish in the way it is presented.
Suggestions to the authors:

1.- Express something more about your future work, what additional measures should be considered or what would you change to make the findings more accurate.

2.-It would be desirable to find the main inquiry addressed by your research work from the beginning in the abstract and in the Introduction sections.

3.- A reference is needed for the Matlab Classification Learner App, maybe a short connection to other research that has used this tool, in case the audience wants to verify how it works for training models to classify data.

4.- About the recording device, can you also provide the accuracy of the pressure measurement and the ECG amplitude resolution? Was there a specific purpose to sample the ECG at 2000 Hz?

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

This study aims to show that the pressure experienced by a diver can be indirectly measured based on respiratory rate and 8 heart rate variability features (time and frequency domain). The experimental design and study population make this study a unique source of information worthy of publication. Additional comments and questions are intended to improve understanding of the text and results.

Revision comments:  

1. The reference list is quite short and essentially old – the main reference studies are around and before 2000. Given that heart rate variability is one of the main areas of research interest in many developments, the reference list should be synchronized with current knowledge, practices and methods.
2. Ln 82-84: Specific for methods. Not applicable for Introduction.
3. Ln 86-87: The aims of the study are not comprehensive. It is clear that the atmospheric pressure can be measured by an automatic system, but the question is what information this system can input and work with.
4. Ln 100: Give information about the number/date of the ethics committee approval, as well as the period of the study (data collection).
5. Ln. 110: “This device records three-lead ECG signals” -> Provide information on the placement of ECG electrodes. A "second frontal bipolar lead" is later mentioned (Ln. 141), but cannot be interpreted without defining the measurement setup.
6. Section “2.2. Respiratory information from the ECG signal” -> Graphical example of recorded data and computed signals/features could greatly help understanding the information (for example, illustrate the signals related to: “These signals are unevenly sampled, so it is necessary to resample them at 4 Hz for standardization”). It is also important to clarify: (1) How are the 3 ECG leads used to improve respiratory rate estimation compared to the case with only one ECG lead? (2) Is it essential to measure ECG at 2 kHz, considering that the diagnostic ECG bandwidth is up to 150 Hz according to standards?
7. Ln 128-130: “Finally an outlier rejection rule based on the median absolute deviation and the application of a band-pass filter (0.075-1 Hz).” -> Clarify the statement. Give an example.
8. Ln 155 (and others): “measure units: s” -> “units of time: s”
9. Ln 173: “variations linearly associated with respiration” -> Provide more information on how linear relationship is determined: what kind of variations, respiration signal and linear regression function.
10. Ln 225-230: Determine whether the trained classifiers have a binary output (e.g. yes/no for the presence of some condition against a 1D reference) or whether a classifier with a single (continuous) output is trained giving the pressure value in function of 9 input features (e.g. like a regression curve). The classification task (Ln 245-248) must be better explained. Make sure that there is not a conflict with formula (1) and general aim to use state 1D as a reference.
11. Ln 225-230: Define the sample size used for the training, validation and test of each classifier. If there are multiple sequential measurements input to the classifier per hyperbaric state, then this must be said. Ensure that training, validation and test are done with independent subjects, so that patient data used in the training are not used in the test if from the same patient.
12. Figure 1: The visualized parameter "ratio" is not defined in the Methods section, therefore it cannot be reported in the Results. If the y-axis parameter plotted is that in equation (1), then its name is "relative change" but not "ratio". Use consistent terms in the text.
13. Figure 1: A legend is required to explain different boxes, whiskers, markers, etc. Otherwise, not interpretable.
14. All tables (Table 2) and figures (Figure 2, Figure 3) must be placed just after their first mention in the text (not several pages/paragraphs later).
15.  Keep sequential number of figures in the text. Currently, Figure 3 is referenced before Figure 2.
16. Figure 2: On the x-axis, instead of the number of the feature (unintelligible – features are not defined by numbers), give the name of the feature.  
17. Figure 2: The y-axis metric “Accuracy” is not defined in Methods, therefore, cannot be interpreted in Results. Make sure that the reported results are on independent TEST set. Readers are not interested in learning results and should not be misled.
18. Figure 3: It is strange that the y-axis are not comparable in subplots (a) and (b). In fact, the caption is not comprehensively written – unclear English phrases. Furthermore, where do 66 cases, 70 cases, 8 cases and 6 cases come from?. Important: make sure that the reported performance metric “number of times misclassified” is defined in Methods, otherwise not interpretable.
19. Figure 3: What is the meaning of the yellow horizontal line? The definition is not interpretable:” horizontal line in the color associated to each set of classes”.  
20. Ln 300-309: Incomprehensive information. Such case-by-case details cannot be followed and is of minor importance. The global statistical analysis is the one which is important and plays a role in the interpretation.
21. The Results section does not correctly report the statistical analyzes defined in the Methods section (Ln 207-218). It is confusing to read about some statistical method that is never applied in analyzing the results, and for drawing some important conclusions.
22. Check the term “misclassified” -> it means “wrongly classified” but the authors have used this term as “correctly classified” in many places in figures/captions/tables/text. Correct all wrong uses of the term.
23. Ln 330: “As a secondary objective…” -> The objectives (primary and secondary) must be clearly defined in the last paragraph of Introduction. Synchronize the information in Introduction and Results/Discussion.
24. Results: Does the accuracy for pressure measurement depend on the pressure value and/or temperature? Can you depict a regression curve which shows the relation of the predicted vs. real pressure value?
25. Compare the results with other studies in the field.
26. Specify the limitations of the study.

I have noted in the review a points of misunderstanding. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

All the flaws have been fixed. However, it is advisable to check a few minors:

- Please use "measurement unit" or "unit of measurement" instead of "measure unit" (e.g.: line 156 and following)

- Please, use the orthogonal symbol (⊥) as subscript or superscript (e.g. line 186) and ensure it is consistent with other occurrences.

English is fine, however a final check is advisable, in order to fix some typos.

Author Response

Thank you very much for your review. All typos indicated by the reviewer have been corrected.

Reviewer 3 Report

The authors have adequately addressed all my revision remarks. The article can be published in the current form. 

Check the content of [?] in: Ln 170 "to be found [? ]", Ln. 174 "based on [?]", Figure 1 (caption) "extracted from [? ].", 

Author Response

Thank you very much for your review.

A thorough revision of the text and corrected all the typos we have found. Errors in the bibliography have also been corrected.