Review Reports - Association of Air Pollution and Socioeconomic Deprivation with Depression and Pain

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This manuscript presents a robust and timely investigation into the synergistic effects of environmental and social stressors on public health. The study’s primary strength lies in its integrative approach, successfully bridging the gap between geospatial environmental data and individual psychological outcomes. By employing a pre-registered design and a large, diverse sample (), the authors demonstrate a high level of methodological rigor and a commendable commitment to open science and transparency.

The paper’s innovation is most evident in its multi-level analysis of socioeconomic deprivation. The distinction between individual-level (USiDep) and neighborhood-level (ADI) factors, alongside the novel concept of "relative deprivation," provides a sophisticated framework for understanding health disparities. Furthermore, the findings regarding the cumulative impact of PM2.5 on pain and the significant moderating role of age are highly impactful. These results offer crucial evidence for targeted public health interventions, particularly for the aging population living in high-exposure areas.

While the study is fundamentally sound, the following points are intended to further strengthen the manuscript and clarify certain technical aspects for the readership:

- Given the transparent use of ChatGPT-5 for data analysis and visualization coding, could the authors specify the manual verification protocols used to ensure the absolute accuracy of the statistical outputs?

- To what extent were potential mitigating factors, such as proximity to green spaces or access to healthcare facilities, considered as covariates in the relationship between deprivation and health?

- Given that the PM2.5 data spans a decade (2012–2022), it would be beneficial to discuss whether recent fluctuations in air quality might have a more acute influence on current symptomatic reports compared to the long-term average.

- A brief reflection on how "relative deprivation" might inherently bias the self-reporting of pain and depression (PROMIS scales) would add valuable depth to the Discussion section.

Author Response

Comment 1: While the study is fundamentally sound, the following points are intended to further strengthen the manuscript and clarify certain technical aspects for the readership:

Response 1: We agree that this would be beneficial for everyone to know. In our Acknowledgements section (Page 17, lines 503-505), we have added the following clarification:

“Outputs were manually verified for accuracy by C.S., A.D., M.C.K., and D.V.S., who independently verified all statistical outputs, tables, and figures against the raw model results.”

Comment 2: To what extent were potential mitigating factors, such as proximity to green spaces or access to healthcare facilities, considered as covariates in the relationship between deprivation and health?

Response 2: Unfortunately, these analyses were beyond the scope of our pre-registered analyses. However, this is very true that these factors may impact the relationship between deprivation and health. We have added the following sentences to our Discussion section under limitations of this study (Page 16, lines 454-457):

“Participants also have varying access to resources, such as green spaces and healthcare facilities, which may be mitigating factors between deprivation and health (Wolch et al., 2014; Lusk et al., 2023). These factors should be considered in the future to clarify these relationships.”

Comment 3: Given that the PM2.5 data spans a decade (2012–2022), it would be beneficial to discuss whether recent fluctuations in air quality might have a more acute influence on current symptomatic reports compared to the long-term average.

Response 3: Thank you for bringing this up. To clarify this, we have added the following to the Discussion section (page 15, lines 436-439):

“Our use of a long-term exposure average is also beneficial in accounting for seasonal fluctuations and downplaying the impact of outlier years, like the COVID-19 pandemic. It also allowed us to assess cumulative exposure alongside the residential time lengths we had available.”

Comment 4: A brief reflection on how "relative deprivation" might inherently bias the self-reporting of pain and depression (PROMIS scales) would add valuable depth to the Discussion section.

Response 4: Thank you for bringing up this point. We agree that this may be the case for many participants. We have added the following sentences to Page 16, lines 458-462):

“Additionally, we acknowledge that perceiving oneself as more deprived may bias self-reports of depression and pain to be higher. Therefore, future studies should expand the relative deprivation measure to incorporate more objective indicators, such as directly comparing individual income to median income of one’s neighborhood, to combat self-report bias.”

Reviewer 2 Report

Comments and Suggestions for Authors

The topic is timely and relevant, and the dataset is sufficient. The paper is generally well written. However, several issues require attention and revision by the authors.

Issues:

Specification of data used, spatial linkage and averaging window.

The paper mentioned the use of the Atmospheric Composition Analysis Group’s satellite-derived PM2.5 datasets. However, the version, product name or retrieval year was not stated.

The coordinate-to-grid method is not clear.

It is unclear whether the authors used nearest neighbour, bilinear interpolation or area weighted extraction to address points.

Justification on long term averaging (2012 – 2022) and potential exposure misclassification.

The long-term averaging (2012 – 2022) may not reflect participants’ residence-time-weighted exposure, especially without the residential histories. The lockdown due to COVID-19 may also affect the results. The manuscript acknowledges this, but should quantify sensitivity.

Spatial resolution in metric units.

0.01^o x 0.01^o should be expressed in km for readability. Can add approximate km equivalents and clarify that the grid cell area varies with latitude.

Figure 1: confidentiality radius

The caption mentioned in Figure 1 says “20km radius around each participant address’. This requires explanation (random jitter, buffers, density mapping, etc.) or else it may imply that exposure was averaged within 20km, which contradicts the point-based extraction described.

Controlling depression when predicting pain (and vice versa)

This can constitute over-adjustment if both are on a causal pathway or share unmeasured causes; it can also induce collider bias. Maybe can present primary models without mutual adjustment and move mutually adjusted models to supplementary sensitivity.

Inconsistent spacing in PM2.5 and PM 2.5 throughout the manuscript. Please check.

Author Response

Comments 1: The topic is timely and relevant, and the dataset is sufficient. The paper is generally well written. However, several issues require attention and revision by the authors.

Issues:

Specification of data used, spatial linkage and averaging window.

The paper mentioned the use of the Atmospheric Composition Analysis Group’s satellite-derived PM2.5 datasets. However, the version, product name or retrieval year was not stated.

Response 1: Thank you for pointing this out. We have made the following change to reflect the version name (Page 4, lines 130-134):

“Participants’ geocoded addresses were used to retrieve their mean PM_2.5 concentrations via the Atmospheric Composition Analysis Group’s North American Regional Estimates with Composition (V5.NA.05 / V5.NA.05.02) (available online: https://sites.wustl.edu/acag/surface-pm2-5/#V5.NA.04.02, accessed 04-03-2026).”

Comments 2: The coordinate-to-grid method is not clear.

It is unclear whether the authors used nearest neighbour, bilinear interpolation or area weighted extraction to address points.

Response 2: Thank you for pointing this out. We have edited the following sentences to clarify this in the following areas:

“Time-averaged estimations of PM_2.5 concentrations at participant address coordinate point locations were calculated using ArcGIS Pro (version 3.1.0).” (Page 5, lines 186-187).

“Due to seasonal changes in pollutant concentrations [32,33] and unusual reductions in PM_2.5 emissions during the COVID-19 pandemic [34,35], we averaged together the annual mean PM_2.5 concentrations from 2012-2022 to more accurately estimate pollutant trends. We then used point-based extraction (nearest neighbor) to provide an estimate for long-term pollutant exposure at each participant’s address coordinates.” (Page 4, lines 137-142)

Comments 3: Justification on long term averaging (2012 – 2022) and potential exposure misclassification.

Response 3: Thank you for this comment. To clarify our rationale for long-term averaging, we have made the following edit on page 6, lines 236-239:

“We selected 2012-2022 as our timeframe to focus on chronic exposure and reduce sensitivity to short-term fluctuations and anomalous trends (e.g. the COVID-19 pandemic, major wildfires). As a sensitivity analysis, we tested alternative averaging windows (see Supplement Tables S1-S3) to show that results were materially unchanged.”

Our sensitivity analysis in the Supplement averages 2012-2014, as well as 2020-2022, for three-year average sensitivity checks. We reran our primary models with these 3-year averages, and the results remain consistent with our long-term 11-year average.

Comments 4: Spatial resolution in metric units.

0.01^o x 0.01^o should be expressed in km for readability. Can add approximate km equivalents and clarify that the grid cell area varies with latitude.

Response 4: We agree this is necessary for clarification and readability. We have made the following edit on Page 4, lines 128-130:

“Air pollution was measured in terms of mean PM_2.5 concentrations (µg/m³) present annually in participants’ neighborhoods, using a spatial resolution of 0.01° × 0.01° (approximately 1.11 km x 1.11 km), with grid cells varying slightly by latitude [31].”

Comments 5: Figure 1: confidentiality radius

Response 5: Thank you for pointing this out. We have made the following edit in the caption of Figure 1 for clarity (page 7, lines 241-244):

“Figure 1. Map of the time-averaged annual PM_2.5 concentrations from 2012-2022 for the study region covering the sample residential locations. Our map displays high spatial variability in PM_2.5 concentration throughout the study region, ensuring representation of both high and low concentrations within our sample.”

Comments 6: Controlling depression when predicting pain (and vice versa)

Response 6: This is a valid concern. We have added the following sentences to our methods section to account for this and have included the results from unadjusted models in our Supplement (page 6, lines 213-218):

“Additionally, we planned to adjust for one when investigating the other to isolate outcome-specific variance. This mutual adjustment aligns with our pre-registration, but we recognize that overadjustment is a valid concern. Therefore, results from unadjusted models can be found in the Supplement (Table S5), confirming that our conclusions are unchanged and ruling out collider-induced distortion.”

Comments 7: Inconsistent spacing in PM2.5 and PM 2.5 throughout the manuscript. Please check.

Response 7: Thank you for pointing this out. We have fixed the spacing issues throughout.

Reviewer 3 Report

Comments and Suggestions for Authors

Check the attachment

Comments for author File: Comments.pdf

Author Response

Comments 1: This section is carefully handled. The abstract reflects all the aspects of the investigation. The introduction (background), aim and objectives, methodology, results, and conclusion are embedded in the abstract. However, the only suggestion I wish to make is in:

Page 1 Line 29: 1113 adults (start this in a sentence)

Response 1: We have made the following edit to the abstract to reflect this (Page 1, lines 29-31): “A total of 1113 adults (aged 18-88 years) completed a series of surveys about their mental and physical health and living environment.”

Comments 2: Page 4, lines 130-134: the Atmospheric Composition Analysis Group’s satellite derived PM2.5 datasets – Add the link or website to this. This will assist readers.

If this is available, I suggest you provide it in the article.

Response 2: This is a good point. We have added a citation in the following sentence: “Participants’ geocoded addresses were used to retrieve their mean total PM2.5 concentrations via the Atmospheric Composition Analysis Group’s North American Regional Estimates with Composition (V5.NA.05 / V5.NA.05.02) (available online: https://sites.wustl.edu/acag/surface-pm2-5/#V5.NA.04.02, accessed 04-03-2026).”

Reviewer 4 Report

Comments and Suggestions for Authors

Although the manuscript acknowledges its correlational design, some wording (e.g., “effects of PM2.5,” “consequences of deprivation”) may imply causality.
- Replace “effects” with “associations”
- Emphasize cross-sectional nature
- Clarify that reverse causation cannot be ruled out (e.g., depressed individuals may move into certain neighbourhoods)
The use of averaged PM2.5 (2012–2022) based on current address is reasonable but introduces potential exposure misclassification.
- Quantify how many participants lived at address >5 years
- Clarify rationale for 2012–2022 averaging window
- Discuss potential bias direction (non-differential misclassification?)
The objective relative deprivation variable (zUSiDep ‚zADI) is novel and interesting. However, since it is derived directly from two predictors already included in models, multicollinearity risk should be discussed. Clarify whether USiDep and ADI were included simultaneously when relative deprivation was tested.
Cohen's values are reported, which is commendable. However, interpretation of magnitude is minimal. Some effects labeled "strong" may be small-to-moderate.

Include brief interpretation (small/medium/large benchmarks)
Emphasize practical vs. statistical significance

Clarify whether PROMIS raw scores were converted to T-scores (if not, justify).
The figure captions (Figures 2 & 3) could more clearly specify covariates included.

Author Response

Comments 1: Although the manuscript acknowledges its correlational design, some wording (e.g., “effects of PM2.5,” “consequences of deprivation”) may imply causality.

Replace “effects” with “associations”
Emphasize cross-sectional nature
Clarify that reverse causation cannot be ruled out (e.g., depressed individuals may move into certain neighbourhoods)

Response 1: Thank you for pointing this out, and we apologize for the oversight. We have changed causal wording throughout the manuscript to emphasize correlation instead. In the Discussion section, we have made the following revisions to discuss cross-sectionality and reverse causation (Page 16, lines 463-466):

“Lastly, we emphasize that this is a cross-sectional study, so our results may reflect reverse causation (e.g. individuals with depression may differentially move into more deprived neighborhoods). Future studies should examine our results longitudinally to further investigate age effects and develop experimental conditions to determine causality.”

Comments 2: The use of averaged PM2.5 (2012–2022) based on current address is reasonable but introduces potential exposure misclassification.

Quantify how many participants lived at address >5 years
Clarify rationale for 2012–2022 averaging window
Discuss potential bias direction (non-differential misclassification?)

Response 2: Thank you for mentioning this. We agree that additional discussion on these points would be beneficial.

On Page 5, lines 202-203, we added the following edit to quantify the participants: “The remaining participants reported current residential times from 5 to 62 years (n = 534, Mtime = 16.49).”

To clarify our rationale for the 2012-2022 average, we have made the following change on page 6, lines 236-239:

As for misclassification, we have added the following to our limitations section on page 16, lines 444-449:

“We did not have access to many of the participants’ full residential histories, only their current addresses, which would have provided a more accurate summary of their long-term PM2.5 exposure. Having detailed residential histories would reduce the risk of exposure misclassification, although any exposure misclassification that occurred in this sample likely just attenuated our associations.”

Comments 3: The objective relative deprivation variable (zUSiDep ‚zADI) is novel and interesting. However, since it is derived directly from two predictors already included in models, multicollinearity risk should be discussed. Clarify whether USiDep and ADI were included simultaneously when relative deprivation was tested.

Response 3: Thank you for this comment. We have clarified this on page 13, lines 337-339:

“For these models, ADI and USiDep were not included as covariates to avoid risk of multicollinearity.”

Comments 4: Cohen's values are reported, which is commendable. However, interpretation of magnitude is minimal. Some effects labeled "strong" may be small-to-moderate.

Include brief interpretation (small/medium/large benchmarks)
Emphasize practical vs. statistical significance

Response 4: This is a good point. We have adjusted the following sentence to include the benchmarks for the Cohen’s values (Page 5, lines 195-196):

“Effect size was evaluated using Cohen’s f²to determine if effects were small (f² = .02), medium (f² = .15), or large (f² = .35) [42].”

Comments 5: Clarify whether PROMIS raw scores were converted to T-scores (if not, justify).

Response 5: We have included the following edit to our Methods section (Page 6, lines 211-213):

“For both PROMIS measures, our analyses used raw summed scores over T-scores for transparency and simplicity. We have included our main models using the T-scores in the Supplement to show the minimal impact this had on our results (Table S4).”

Comments 6: The figure captions (Figures 2 & 3) could more clearly specify covariates included.

Response 6: Thank you for the feedback. To Figure 2, we have adjusted the caption to read:

“Figure 2. Bar graph with error bars depicting mean PROMIS depression scores for each USiDep score and an asterisk indicating significance. Age, gender, and race/ethnicity were used as covariates. USiDep scores at the highest level (5) were significant predictors of PROMIS depression scores, p < .001, indicating that experiencing higher levels of individual deprivation are associated with higher levels of depressive symptoms in everyday life.”

To Figure 3, we have adjusted the caption to read:

“Multiple regression models for (a) PROMIS depression on relative deprivation and (b) PROMIS pain on relative deprivation in our main sample (N = 1113). A relative deprivation value less than 0 indicates being less deprived than the neighborhood, and a value more than 0 indicates being more deprived. The shaded areas display the 95% confidence interval. After corrections and controlling for covariates (age, gender, and race/ethnicity), both results were significant, p < .001, indicating that being more deprived than one’s neighborhood is associated with increased depression and pain symptoms.”