Review Reports - Objective Assessment of Procedural Pain and Recovery in Preterm Infants Using Time–Frequency Analysis of Electroencephalography

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The introduction does not sufficiently distinguish this study from the authors' own prior work (e.g., the 2025 study mentioned), leaving the specific incremental contribution unclear.

In the last paragraph of the introduction, explicitly state how the current analysis extends the previous one (e.g., "While our prior study characterized the immediate EEG response, this work extends the temporal window to 30 minutes post-procedure to map the complete recovery trajectory and calculate time-integral metrics of pain burden.").

Key parameters for the Continuous Wavelet Transform (CWT) and artifact rejection are omitted, hindering reproducibility.

Add a brief sentence specifying the parameters used (e.g., "The CWT was performed using Morlet wavelets with a central frequency (ω₀) of 6 radians and 6 cycles. Artifact rejection was performed visually by an expert and automatically by rejecting epochs where the signal amplitude exceeded ±200 µV.").

The discussion attributes prolonged power suppression in central channels (C3, C4) primarily to pain-induced neuronal changes, while the potential confounding effect of evolving sleep states over the 30-minute recording is not adequately considered.

Acknowledge this alternative explanation more prominently. Revise the discussion to state: "The sustained suppression of alpha/beta power in central regions may reflect pain-induced neuronal modulation. However, as the baseline and recovery periods likely encompass different vigilance states (e.g., active vs. quiet sleep), further studies controlling for sleep stage are needed to confirm this interpretation."

The conclusion mentions developing a "clinically useful biosensor" but the discussion lacks a concrete proposal on how the derived metric (e.g., time-integral of power) would be operationalized in a real-time monitoring context.

Add a short paragraph in the discussion proposing a potential clinical output. For instance: "To translate these findings into a bedside tool, future devices could calculate a moving-window integral of beta or delta power post-procedure. A return of this integral value to a patient-specific baseline range could provide an objective signal for the cessation of pain-related brain activity, informing caregiver intervention."

Author Response

Reviewer 1
Comments and Suggestions for Authors

The introduction does not sufficiently distinguish this study from the authors' own prior work (e.g., the 2025 study mentioned), leaving the specific incremental contribution unclear.

Response: We thank the reviewer for highlighting the need to clarify the novelty of the present study relative to our prior work. We have revised the final paragraph of the Introduction to highlight that, while our previous study focused on the immediate EEG response to procedural pain, the current work extended the temporal scope to 30 minutes post-procedure to understand the entire impact of procedural pain, which might be more closely related with the long-term outcome of vulnerable infants.

Key parameters for the Continuous Wavelet Transform (CWT) and artifact rejection are omitted, hindering reproducibility.

Response: We appreciate the reviewer’s emphasis on methodological transparency and reproducibility. In response, we have added a concise description of the CWT parameters and artifact rejection procedures in the Methods section.

The discussion attributes prolonged power suppression in central channels (C3, C4) primarily to pain-induced neuronal changes, while the potential confounding effect of evolving sleep states over the 30-minute recording is not adequately considered.

Response: We have revised the Discussion to more explicitly acknowledge this alternative explanation. The updated text now states that sustained suppression of alpha and beta power in central regions may reflect pain-induced neuronal modulation; however, differences in sleep state between baseline and recovery periods (e.g., active versus quiet sleep) may also contribute to these observations.

The conclusion mentions developing a "clinically useful biosensor" but the discussion lacks a concrete proposal on how the derived metric (e.g., time-integral of power) would be operationalized in a real-time monitoring context.

Response: We thank the reviewer for this valuable suggestion. To strengthen the translational relevance of our findings, we have added a new paragraph to the Discussion outlining a potential pathway toward clinical implementation.

Reviewer 2 Report

Comments and Suggestions for Authors

Major Revision Summary

This manuscript addresses an important and sensitive clinical problem by using TF EEG analysis to objectively quantify procedural pain and recovery in preterm infants. The longitudinal design and extended post-procedure window are strengths. However, major concerns remain regarding participant characterization, handling of repeated measures, EEG preprocessing rigor, statistical modeling, interpretation of suppression effects, and overgeneralization of clinical implications. Several methodological details are unclear or insufficiently justified, and key conclusions are stronger than the data allow. Major revisions are required before publication.

Study Design and Participant Characterization
The sample includes repeated recordings from some infants, but this is not clearly integrated into the design description. The rationale for including multiple recordings per infant and how this affects independence assumptions must be clarified.
Inclusion and exclusion criteria are broad, but important clinical variables (e.g., analgesic use, illness severity, sleep state) are not systematically reported.
Table 1 lacks important descriptors such as clinical status, ventilation history, and medication exposure.
The imbalance between male and female infants should be discussed as a potential source of bias.
EEG Recording and Preprocessing Need Stronger Justification
Referencing all channels to Cz and then excluding Cz from analysis is unusual and should be justified.
Artifact handling relies heavily on wavelet denoising and threshold replacement, which may distort physiological signals; validation of this approach is needed.
Replacing large-amplitude segments by linear interpolation may bias power estimates and should be justified or avoided.
The number of rejected channels and segments per dataset should be reported clearly.
Time Segmentation and Experimental Protocol
Time windows differ in length (1 min vs 2 min), which affects power comparability and must be statistically justified.
The exclusion of the 10-minute window due to saliva sampling introduces a gap that may bias recovery trajectory interpretation.
Infant behavioral or sleep state during each segment is not reported, although it strongly affects EEG.
The assumption that baseline represents a true resting state is not supported.
Statistical Modeling and Repeated Measures
Linear mixed-effects models are appropriate, but the model structure is not fully specified (random slopes, covariance structure).
It is unclear whether multiple recordings from the same infant were nested correctly in the model.
Only time is modeled as a fixed effect; important covariates (age, sex, sleep state, medication) are missing.
Effect sizes are not reported, limiting interpretation of clinical relevance.
Power Ratio Metric and Interpretation
Power ratio depends strongly on baseline quality, which is not validated as stable.
Ratios may inflate small baseline values and should be interpreted cautiously.
Both absolute power and ratio are reported, but their relative advantages are not clearly discussed.
The claim that power ratio reflects “dominant” frequencies is descriptive rather than statistically tested.
Results Presentation
The Results section is overly long and difficult to follow, with excessive repetition of significance statements.
Figures are dense and lack clear guidance for readers on key patterns.
Channel-by-channel reporting obscures global trends and should be summarized more clearly.
Tables summarizing main effects across regions and bands would improve clarity.
Interpretation of Recovery and Suppression
The conclusion that recovery occurs by four minutes is oversimplified given later suppression effects.
Post-procedure power suppression is interpreted speculatively without direct evidence.
Sleep state is a likely explanation for suppression but is not measured.
Causal language about pain-induced neural suppression should be softened.
Clinical Claims and Biosensor Implications
Claims about clinical utility as a “biosensor” are premature without validation against behavioral pain scales.
No comparison is made with existing neonatal pain measures.
The link between EEG dynamics and long-term outcomes is speculative and unsupported by data.
The discussion should clearly separate current findings from future possibilities.

Overall Recommendation

Major revision.

Comments for author File: Comments.pdf

Author Response

Study Design and Participant Characterization
The sample includes repeated recordings from some infants, but this is not clearly integrated into the design description. The rationale for including multiple recordings per infant and how this affects independence assumptions must be clarified.

Response: To investigate the age-related change in the pain response, we repeated the observation with an interval of at least two weeks to enable temporal comparisons. Repeated measurements within the same individuals were accounted for so that the findings are not obsqured by some specific cases. This clarification “Repeated measurements were performed in the same individual to examine changes over time in the sub-analysis” has been added to the EEG recording and study protocol.

Inclusion and exclusion criteria are broad, but important clinical variables (e.g., analgesic use, illness severity, sleep state) are not systematically reported.

Table 1 lacks important descriptors such as clinical status, ventilation history, and medication exposure.

Response: We thank the reviewer for the comments. We have newly provided comprehensive clinical information in the table. The definition of entry criteria has also been revised highlighting that infants were recruited only after they had weaned from intensive care interventions, such as venous drip infusion and invasive/non-invasive positive pressure ventilation, and medication excluding supplemental vitamin, iron, sodium and phosphate.

The imbalance between male and female infants should be discussed as a potential source of bias.

Response: This imbalance occurred by chance, however, we accept that this may be a potential source of bias. We have revised the Limitation section to highlight this.

EEG Recording and Preprocessing Need Stronger Justification
Referencing all channels to Cz and then excluding Cz from analysis is unusual and should be justified.

Response: We appreciate this important observation. Cz was used as the reference electrode following prior neonatal EEG studies, including Lavanga et al. (2020), who employed Cz-referenced monopolar recordings and subsequently excluded Cz from analysis. This approach aligns with our study that uses Cz as the reference electrode for analyzing EEG responses to procedural pain in preterm infants.

Artifact handling relies heavily on wavelet denoising and threshold replacement, which may distort physiological signals; validation of this approach is needed.

Response: Wavelet-based denoising was selected due to its suitability for non-stationary neonatal EEG signals. To minimize potential signal distortion, the denoising threshold was chosen based on a previously published neonatal EEG study by Lavanga et al. (2020), which investigated the effects of procedural pain on EEG activity in preterm infants. In line with their methodology, conservative threshold parameters were applied to reduce artifacts while preserving physiological signal characteristics. We have added the relevant reference in the revised manuscript.

Replacing large-amplitude segments by linear interpolation may bias power estimates and should be justified or avoided.

Response: In the present study, artefact handling followed a conservative preprocessing strategy consistent with established neonatal EEG methodologies, including those described by Toole and Boylan (2017), which emphasize removal of major artefacts while preserving temporal continuity of the signal. Specifically, large-amplitude artefacts were identified using objective threshold-based criteria (absolute amplitude, inter-sample difference, and signal variability), and only brief artefact-contaminated segments were replaced using linear interpolation. Channels exhibiting excessive artefacts across segments were excluded entirely from further analysis. We have added the relevant reference in the revised manuscript.

The number of rejected channels and segments per dataset should be reported clearly.

Response: Thank you for this comment. Information on rejected channels is now reported in the Results section. Based on the predefined artifact detection criteria, some channels were excluded from specific datasets. We report the number of recordings retained for each EEG channel (Fp1, Fp2, C3, C4, O1, O2, T3, and T4) across the total 57 datasets, thereby providing transparency regarding channel rejection. Channels that were detected with high levels of artifacts after filtering in at least one of the specified segments were removed entirely from the analysis.

Time Segmentation and Experimental Protocol
Time windows differ in length (1 min vs 2 min), which affects power comparability and must be statistically justified.

Response: The procedural pain segment was shorter due to the brief and clinically constrained nature of the heel lance and skin squeeze, whereas longer pre- and post- procedure segments were selected to improve signal stability during relatively quiescent periods. EEG power was estimated using fixed-length (30-second) windows and averaged within each segment, which mitigates the impact of unequal segment duration. We had discussed with our bio-statistician and concluded that, while the different windows may alter the susceptibility to type-1/2 errors, it is still superior to follow the temporal changes in pain-response throughout the observational period. We hope the reviewer accepts our approach as one of feasible decisions.

The exclusion of the 10-minute window due to saliva sampling introduces a gap that may bias recovery trajectory interpretation.

Response: The 10-minute post-procedure period was not included in the analysis because saliva sampling was conducted at this time point, which was likely to introduce movement-related and non-physiological artifacts into the EEG recordings. Although this results in a discontinuity in the temporal sequence, inclusion of data affected by sampling-related noise could compromise the reliability of the analysis. We hope the reviewer agree to our decision.

Infant behavioral or sleep state during each segment is not reported, although it strongly affects EEG.

Response: We thank the reviewer for this important comment. Infant behavioral/sleep state has now been explicitly included in the manuscript. Behavioral state was assessed from video recordings using an adaptation of Prechtl’s behavioral state categories, as described by Grunau and Craig. This information has been added to both the Methods and Results sections to clarify how infants’ activity altered during the observational period.

The assumption that baseline represents a true resting state is not supported.

Response: In this study, procedures were performed shortly after the infants had been calm or asleep for at least 5 min, and the baseline was defined as the 2-minute period preceding the procedure. Although it is virtually impossible to control the EEG phase in this kind of clinical studies, we consider that this “baseline” period likely corresponded to low-voltage irregular EEG pattern. We hope the reviewer understands the difficulty in collecting comprehensive physiological data from a number of newborn infants and fully control for baseline and background variables. To minimize the bias, unlike previous studies recruiting up to 20-30 participants, we challenged in recruiting a relatively large number of infants to reduce statistical errors.

Statistical Modeling and Repeated Measures
Linear mixed-effects models are appropriate, but the model structure is not fully specified (random slopes, covariance structure).

Response: The statistical analysis section now clarifies that linear mixed-effects models included time as a fixed effect and random intercepts for infant ID to account for repeated recordings. Models were estimated using REML with Dunnett’s correction for comparisons to baseline. For the current analysis exploring for the temporal changes in pain-response, findings were not adjusted for clinical backgrounds, such as sex and treatments during the acute period to highlight the crude tempral trend in the response using a simple model.

It is unclear whether multiple recordings from the same infant were nested correctly in the model.

Response: Repeated recordings in the same individual are explicitly accounted for using linear mixed-effects models with infant ID specified as a random effect.

Only time is modeled as a fixed effect; important covariates (age, sex, sleep state, medication) are missing.

Response: As described in the response to Question 16, we did not adjust our findings for clinical backgrounds because the aim of the current study was to explore for the crude profile of temporal change in the pain response. However, as a clinician myself, I totally agree that future analysis needs to incorporate key clinical variables, such as sex, gestational age, therapies provided during the acute period, and even EEG patterns.

Effect sizes are not reported, limiting interpretation of clinical relevance.

Response: The authors are sure that the reviewer understands the difficulty in precisely estimating the “effect size” of temporal changes in EEG power following heel lancing, because few studies assessed the EEG pain response of newborn infants for up to 30 minutes. Clinical implication of our current study is, however, limited, not because of its insufficient estimation of the effect size, but because of its exploratory nature to elucidate the crude temporal profile of EEG response following procedural pain. We hope the reviewer accepts the difficulty in collecting comprehensive physiological data from newborn infants, unlike in the preclinical experimental settings.

Power Ratio Metric and Interpretation
Power ratio depends strongly on baseline quality, which is not validated as stable.

Response: We acknowledge that power ratio estimates depend on baseline quality. In this study, the baseline was defined as the 2-minute period immediately before the procedure. Based on video assessment, the majority of infants were in active sleep during this baseline period, providing a relatively consistent physiological state across the datasets.

Ratios may inflate small baseline values and should be interpreted cautiously.

Response: I understand that the small denominator has a large impact on the ratio. However, since the ratio was only used for visualization as a topography and not for statistical analysis, I believe this issue does not affect the results. We have revised the manuscript to soften the interpretation of the power ratio findings and to clearly state that these observations are descriptive and exploratory in nature.

Both absolute power and ratio are reported, but their relative advantages are not clearly discussed.

Response: Thank you for this valuable feedback. We have added a clarification in the manuscript discussing the complementary roles of absolute EEG power and the power ratio. Specifically, we note that absolute power captures the magnitude of brain activity, while the power ratio highlights proportional changes relative to baseline. This dual approach allows for a more comprehensive understanding of EEG dynamics during procedural pain. Additionally, we emphasize the potential clinical utility of integrating power measures normalized to individualized baselines for real-time pain monitoring and recovery assessment.

The claim that power ratio reflects “dominant” frequencies is descriptive rather than statistically tested.

Response: Thank you for this insightful comment. We acknowledge that our description of “dominant” frequency bands across different brain regions is based on observed relative changes in EEG power ratio rather than formal statistical testing to confirm dominance. We have revised the manuscript to clarify that these observations are descriptive.

Results Presentation
The Results section is overly long and difficult to follow, with excessive repetition of significance statements.

Response: Thank you for this valuable feedback. We have carefully revised the Results section to improve clarity by reducing repetitive statements about statistical significance.

Figures are dense and lack clear guidance for readers on key patterns.

Response: We thank the reviewer for this constructive comment. To improve figure clarity and readability, we have revised the figures by adding clear titles for each frequency band and expanding the figure captions.

Channel-by-channel reporting obscures global trends and should be summarized more clearly.

Response: While all channels exhibited a similar overall temporal trend in response to procedural pain, the timing and statistical significance varied across channels and frequency bands. We have clarified this point in the text and revised the presentation to emphasize the global patterns.

Tables summarizing main effects across regions and bands would improve clarity.

Response: Thank you for your insightful suggestion. We have added a new summary table presenting the main effects of time on EEG power across all channels and frequency bands. The table reports F-statistics with numerator and denominator degrees of freedom derived from the linear mixed-effects models.

Interpretation of Recovery and Suppression
The conclusion that recovery occurs by four minutes is oversimplified given later suppression effects.

Response: We agree that recovery is a complex process. Although EEG power in several regions returned to baseline by four minutes post-procedure, sustained suppression in theta, alpha, and beta bands at central channels persisted beyond this time. We have updated the manuscript to emphasize this more nuanced recovery pattern rather than suggesting a fixed recovery point.

Post-procedure power suppression is interpreted speculatively without direct evidence.

Response: We agree that our original interpretation of post-procedure power suppression could be perceived as speculative. To address this concern, we have revised the relevant text to adopt more cautious and descriptive language.

Sleep state is a likely explanation for suppression but is not measured.

Response: We agree that sleep state may influence EEG suppression. Sleep/behavioral state was assessed from video recordings using Prechtl’s behavioral state categories, and this information has now been added to the Methods and Results sections of the revised manuscript.

Causal language about pain-induced neural suppression should be softened.

Response: Thank you for this important point. We have revised the manuscript to use more cautious language regarding causal interpretations, emphasizing associations rather than definitive causation.

Clinical Claims and Biosensor Implications
Claims about clinical utility as a “biosensor” are premature without validation against

behavioral pain scales.

Response: We agree that validation against behavioral pain scales is required before EEG measures can be considered a clinical biosensor. The present study was not intended to establish clinical utility, but to characterize the temporal EEG response to procedural pain and recovery. These results may inform the observed patterns of activation and recovery could guide the selection of candidate EEG metrics for subsequent validation against established behavioral pain measures before clinical application.

No comparison is made with existing neonatal pain measures.

Response: We agree that comparison with established neonatal pain measures would strengthen the clinical interpretation of our findings. However, the primary objective of this study was to characterize the temporal dynamics of cortical EEG responses following a noxious procedure, rather than to validate EEG markers against existing behavioral or physiological pain scales. We have acknowledged this limitation in the manuscript and note that future studies incorporating multimodal pain assessments will be necessary to evaluate concordance with established neonatal pain measures and support clinical translation.

The link between EEG dynamics and long-term outcomes is speculative and unsupported by data.

Response: This study represents a starting point. We anticipate that the data obtained here will be valuable for future development. Naturally, the next steps will require further validation and continued investigation.

The discussion should clearly separate current findings from future possibilities.

Response: We thank the reviewer for this insightful suggestion. In the revised manuscript, we have carefully revised the Discussion section to clearly distinguish between our current empirical findings and potential future applications or research directions.

Reviewer 3 Report

Comments and Suggestions for Authors

This paper examines an interesting and rather borderline problem: the search for neurophysiological correlates using EEG data from heel-prick blood draws in premature infants. This topic is somewhat new and borderline. The methodological aspects of the study raise no questions. The EEG was recorded and analyzed using a traditional method using an electroencephalograph from a well-known manufacturer. Below, I will provide a series of recommendations that will improve the understanding of the study.

When conducting such studies on animals, it is customary to talk about nociceptive reactions. When studying adults, we talk about pain. That is, the criterion for pain is consciousness and the verbal component. Therefore, the question of studying pain in newborns is borderline and requires additional justification. Perhaps it would be more appropriate to talk about reactions to a painful stimulus, etc.
The EEG is a very nonspecific indicator of pain reactions. Therefore, when discussing the brain's pain reactions and to justify the choice of EEG in such studies, I recommend considering established neurophysiological concepts and theories about pain mechanisms.
The study results are presented well. However, in addition to the topographic maps, I recommend showing the original dynamic EEGs. Few readers have seen EEGs of newborns. Therefore, such figures will enhance the work.
In the Discussion, I also recommend connecting your data to established neurophysiological theories of pain.

Author Response

When conducting such studies on animals, it is customary to talk about nociceptive reactions. When studying adults, we talk about pain. That is, the criterion for pain is consciousness and the verbal component. Therefore, the question of studying pain in newborns is borderline and requires additional justification. Perhaps it would be more appropriate to talk about reactions to a painful stimulus, etc.

Response: We thank the reviewer for this important conceptual clarification. We fully agree that, in speechless newborn infants, subjective pain experience cannot be directly verified due to the absence of verbal report. In response to this comment, we have revised the terminology throughout the manuscript to avoid implying subjective pain perception. Specifically, we now use the terms “responses to procedural pain,” “responses to painful stimuli,” or “nociceptive-related neural responses” instead of “pain”.

The EEG is a very nonspecific indicator of pain reactions. Therefore, when discussing the brain's pain reactions and to justify the choice of EEG in such studies, I recommend considering established neurophysiological concepts and theories about pain mechanisms.

Response: We appreciate this important recommendation. In response, we have substantially expanded the Introduction to better ground our use of EEG within established neurophysiological theories of pain.

The study results are presented well. However, in addition to the topographic maps, I recommend showing the original dynamic EEGs. Few readers have seen EEGs of newborns. Therefore, such figures will enhance the work.

Response: We agree to the point that most readers are unfamiliar to neonatal EEG and therefore it would be more informative when representative EEG is presented. We now included EEG records from a representative infant at the baseline, during the procedure, immediate, 4 min, 6 min, 8 min, 20 min, and 28 min post-procedure in Figure 2.

In the Discussion, I also recommend connecting your data to established neurophysiological theories of pain.

Response: We fully agree with this recommendation. As in the Introduction section, the Discussion section has also been revised to more explicitly link our findings to established neurophysiological theories of pain processing.

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

The authors addressed all the comments. I have no further obligations or comments. Good work to the authors.

Reviewer 3 Report

Comments and Suggestions for Authors

Accept in present form.