Pith. sign in

REVIEW 4 major objections 5 minor 46 references

How Your Location Relates to Health: Variable Importance and Interpretable Machine Learning for Environmental and Sociodemographic Data

T0 review · 4 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read NO2 emerges as global predictor for asthma, hypertension, and anxiety

desk verdict A reproducible importance-ranking pipeline with a real NO2 signal, but the ULEZ sign-change story is post-hoc and should be labeled a hypothesis. read the letter →

arxiv 2501.02111 v1 pith:TO2AOLCP submitted 2025-01-03 cs.LG

Ishaan Maitra , Raymond Lin , Eric Chen , Jon Donnelly , Sanja Šćepanović , Cynthia Rudin This is my paper · ORCID
classification cs.LG
keywords MEDSATvariableimportanceinterpretablemachinelearningNO2airpollutiongeneralizedadditivemodelsmultiscalegeographicallyweightedregressionspatialhealthdisparitiesULEZ
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper claims that nitrogen dioxide (NO2) is the single most consistent predictor of prescription-based asthma, hypertension, and anxiety across England, after filtering 154 candidate variables through knockoffs and four variable-importance measures. It builds an interpretable spatial pipeline using a global generalized additive model with a spatial tensor, multiscale geographically weighted regression, and local GAMs for each Local Authority District. The central payoff is a resolution of a known paradox: the negative NO2-asthma association around London is confined almost entirely to districts inside the Ultra Low Emission Zone, where traffic restrictions cut roadside NO2 by more than a third, and global spatial smoothing spreads that local signal into neighboring areas. If correct, the result implies that local traffic policy can change the sign of the pollution-health association and that global models alone can mislocate the effect.

What carries the argument

The argument is carried by a five-stage pipeline: model-X knockoffs as a false-discovery filter, an ensemble of XGBoost regressors sampled to span a Rashomon set (a family of similarly accurate models), four variable-importance metrics averaged into a mean rank, and two spatial models—a GAM with a tensor-product spatial smooth for global effects and MGWR (a geographically weighted regression that lets each predictor have its own spatial bandwidth) for local coefficients. The decisive mechanism for the headline finding is the local GAM variant with spatial smoothing removed, treating each London district independently. It isolates the negative NO2-asthma association inside the Ultra Low Emission Zone, showing that spatially smoothed global models had spread that localized signal into neighboring districts.

What would settle it

Use a non-prescription outcome source for the same Local Authority Districts, such as GP-registered diagnoses or hospital admissions for asthma, hypertension, and anxiety, and rerun the full pipeline; if NO2 does not remain a top-ranked predictor, or the negative NO2-asthma association does not localize to ULEZ districts, the paper's central claims are artifacts of the prescription proxy or of spatial smoothing.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that NO2 is a global predictor of prescription-based asthma, hypertension, and anxiety across England, with the highest mean rank across permutation importance, SHAP, LOCO, and CMR in both pre- and post-COVID years. The paper explains the paradoxical negative NO2-asthma correlation in London as a localized policy effect: the few districts with strong negative associations lie inside the 2019 Ultra Low Emission Zone, where traffic restrictions cut roadside NO2 by about 37%, and spatial smoothing in MGWR and the global GAM then smears that local signal into adjacent areas. The paper also identifies outcome-specific global predictors, including skilled-trades occupation for diabetes, long-term residency for depression, vegetation water content for hypertension, and marital status for depression and anxiety, and reports that PM2.5 associations shifted regionally during COVID. The overall claim is that combining global and local interpretable models can locate where environmental health effects actually operate and can prevent a local policy effect from being misread as a population-wide relationship.

Load-bearing premise

Every health outcome in the paper is measured by the total quantity of prescriptions related to that condition, not by diagnosed disease, so if prescribing behavior or access to care differs across districts, the reported predictors could describe the healthcare system rather than the environment or population.

Editorial extensions

If this is right

  • NO2 should be included in any prescription-based model of asthma, hypertension, or anxiety in England, since it ranks first or near-first on all four importance measures in both periods.
  • The ULEZ result gives a concrete mechanism for why local pollution controls can change measured health associations, implying that expanding such zones could be evaluated by watching NO2-associated prescription changes in newly covered districts.
  • Global spatial models alone are insufficient for identifying where an effect operates, because smoothing can artificially expand a localized signal; local, smoothing-free checks are needed when a coefficient's sign is surprising.
  • Outcome-specific predictors such as skilled-trades occupation, marital status, and vegetation water content identify distinct target populations for diabetes, depression, and hypertension interventions.
  • COVID-19 changed the spatial pattern of PM2.5's association with diabetes and hypertension in England, so pooling pre- and post-COVID data would obscure a real temporal shift.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Beyond the paper: because the outcome is prescription volume rather than diagnosed prevalence, the relative ranks could partly reflect differences in healthcare access or prescribing habits; rerunning the pipeline on GP-diagnosis or hospital-admission data for the same districts would test this directly.
  • Beyond the paper: the ULEZ finding is naturally testable as a quasi-experiment by comparing prescription trends inside and outside the zone across its expansion phases; the paper's explanation predicts that the negative NO2-asthma association will appear in newly covered districts and weaken where restrictions are removed.
  • Beyond the paper: the environmental-over-sociodemographic ranking may be amplified by spatial confounding, because deprived areas tend to have both worse air and different prescribing patterns; adding explicit deprivation controls or using double-robust estimators would show whether NO2's top rank is direct or a proxy for socioeconomic exposure.
  • Beyond the paper: the smoothing-artifact mechanism should be checked in other ecological analyses that report surprising coefficient signs, since any method that borrows strength across spatial boundaries can turn a local policy effect into an apparent regional relationship.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper presents a variable-importance and interpretable-machine-learning pipeline applied to the MEDSAT dataset of English health, environmental, and sociodemographic data. It first filters the 154 candidate variables with model-X knockoffs, then ranks survivors with four importance measures (permutation importance, SHAP, LOCO, and conditional model reliance) across an ensemble of XGBoost models, and retains the top 10 per outcome. Downstream analyses use global Generalized Additive Models with a spatial tensor, Multiscale Geographically Weighted Regression, and local GAMs fit per Local Authority District to describe associations with prescription-based proxies for asthma, hypertension, anxiety, diabetes, and depression. The central claim is that NO2 is a robust global predictor for asthma, hypertension, and anxiety; the paper further offers a post-hoc explanation that the negative NO2-asthma association in London is attributable to the Ultra Low Emission Zone, and reports regional and COVID-period shifts for PM2.5 and solar radiation.

Significance. If the central NO2 claim holds, the paper provides a useful descriptive benchmark on a relatively new, fine-grained public-health dataset, and its methodological combination of multiple importance metrics, Rashomon-set sampling, and global-plus-local spatial models is a reasonable blueprint. The release of code and the use of a public dataset are clear strengths that aid reproducibility. However, several load-bearing claims go beyond what the current evidence supports: the ULEZ sign-change explanation is a post-hoc cross-sectional comparison, the MGWR-based regional claims lack uncertainty quantification, and the COVID-period comparisons are not formally tested. The paper would be a solid descriptive analysis if these claims were either statistically supported or explicitly downgraded to exploratory hypotheses.

major comments (4)
  1. [Related Work] The paper states in Related Work: 'We refer to the health outcome variable ... as the total quantity of prescriptions related to that outcome.' Every outcome model in the paper uses this prescription proxy, but the proxy is never validated against diagnostic prevalence, survey data, or healthcare-access measures. If prescribing intensity, private healthcare use, or registration patterns differ across Local Authority Districts, the reported 'global predictors' could reflect health-system behavior rather than disease burden. Because this assumption is load-bearing for the abstract's health claims, the authors should either validate the proxy or systematically soften all disease-prevalence language to 'prescription volume.'
  2. [Results §2, Figure 3] The claim that the negative NO2-asthma association is 'explainable by' the 2019 Ultra Low Emission Zone is not supported by the evidence presented. The analysis compares London LADs inside versus outside the ULEZ in a single cross-section, with no pre-2019 data, no control group, and no adjustment for income, healthcare access, age structure, or asthma-management practices. The cited 37% roadside NO2 reduction is an external statistic that is not linked to asthma prescriptions in the same data. Moreover, LADs are large administrative units relative to the ULEZ boundary, so a binary inside/outside classification is ecologically imprecise. This section should be reframed as hypothesis-generating, or supported with a difference-in-differences or pre/post design, before any policy conclusion such as 'expanding ULEZ-like zones could be an effective strategy' is drawn.
  3. [Methodology Step 5 and Results §4, Figures 4-7] The local MGWR coefficient maps are presented without confidence intervals or significance tests, so statements such as 'distinctly positive correlations' or 'coefficients shift from ≤ −0.5 to ≥ 0.5' are not statistically supported. This is especially problematic in the COVID comparison (Figures 5-6), where the 2019 versus 2020 differences are assessed visually, and the text itself acknowledges that the local GAMs for two of the three highlighted LADs (Worcester and Wychavon) are 'noisy and likely impacted by spatial smoothing.' The authors should report standard errors, confidence bands, or permutation-based significance for the local coefficients, or explicitly present these maps as exploratory visualizations.
  4. [Methodology Step 4] The aggregation step takes unweighted means of predictors and outcomes within each Local Authority District before fitting the global GAM. Because LADs vary widely in population, an unweighted mean gives equal influence to sparsely and densely populated areas, which can alter both the strength and the sign of estimated associations. The authors should weight by LAD population, use population-weighted centroids, or otherwise justify why unweighted aggregation is appropriate for the research question.
minor comments (5)
  1. [Tables 1-2] The caption of Table 2 says 'Anxiety variable importance and ranks for the top 10 variables after reordering by mean rank,' but the surrounding text describes reinserting demographic variables into the anxiety model; please clarify whether Table 2 reports the reinserted model or a reordering of the original rankings.
  2. [Figure 3] The figure caption should clearly indicate which LADs are inside and outside the ULEZ boundary; the current text names Westminster, Lambeth, Tower Hamlets, and Camden, but the reader cannot verify the inside/outside classification from the figure alone.
  3. [Methodology Step 2] The 'top 10' cutoff is described as a balance between simplicity and performance, but no sensitivity analysis or plot of predictive performance versus the number of retained variables is provided; a brief sensitivity check would strengthen the robustness of the filtering step.
  4. [Methodology Step 1] The knockoffs section does not report the actual FDR thresholds q or the number of variables retained per outcome and time period; these details are needed to assess how much dimensionality reduction occurred.
  5. [Results §4] There is a typo in the sentence 'despite a ≈15% decrease in actual PM2.5 levels in the high-correlation LADsThis counter-intuitive result...' where 'LADsThis' should be 'LADs. This'; also, 'Hamlets' should be 'Tower Hamlets' in the Figure 3 discussion.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the reported importance rankings and spatial associations are descriptive in-sample fits with external corroboration, not predictions derived from the paper's own fitted constants.

full rationale

The derivation chain is descriptive rather than circular. Variable selection (knockoffs, then SHAP/LOCO/permutation/CMR mean ranks) chooses candidate predictors; the global GAMs and MGWR then estimate conditional association shapes and spatially varying coefficients on the same LAD-aggregated data. These are effect estimates, not out-of-sample forecasts, and the paper does not present a fitted parameter as if it were an independent prediction. The NO2-asthma ULEZ discussion is a post-hoc cross-sectional explanation that leans on an external Greater London Authority report of a 37% roadside NO2 reduction; it is not derived from a coefficient fitted in this paper, so it is causally underidentified but not circular. The MEDSAT citation is to a published dataset that includes co-author Sanja Šćepanović, and the remark that the negative association 'is also found in the MEDSAT paper' is corroborative rather than load-bearing, since the present analysis independently re-derives the sign from the same public data. The passage cautioning that individual sociodemographic variables such as ethnicity may be low-ranked due to intercorrelation is a validity concern, not a circular step. No equation or fitted quantity in the paper is reused as its own prediction, and no load-bearing claim reduces to a self-citation chain. A non-circularity score of 0 is therefore appropriate.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The central claims rest on several domain assumptions about the outcome proxy, spatial aggregation, and model validity, plus hand-chosen analytical thresholds (top-10 cutoff, correlation threshold, FDR level). The paper introduces no new physical or conceptual entities.

free parameters (6)
  • Top-10 variable cutoff = 10
    The paper selects the 10 highest-ranked variables per outcome based on mean importance rank; this hand-chosen cutoff determines which variables enter the global GAM and MGWR models and thus shapes all reported results.
  • FDR threshold q for knockoffs
    The knockoff filter threshold is chosen by tuning via cross-validation; the specific value is not reported, so the variable subset depends on an unreported parameter.
  • Correlation threshold for variable removal = 0.8
    Variables correlated above 0.8 are removed, with the less important variable dropped; this hand-chosen threshold affects the composition of the final predictor set.
  • GAM smoothing parameters (lambda, n splines)
    Grid-searched to maximize GCV for each model; these fitted hyperparameters determine the shape functions and reported confidence intervals.
  • MGWR bandwidths b_k
    Estimated per predictor via cross-validation; the bandwidths control the spatial scale of local coefficients and are central to the local variation findings, but are not reported.
  • XGBoost hyperparameters
    Number of estimators, max depth, and learning rate are tuned via randomized grid search; these affect the variable importance rankings that determine the top-10 predictors.
assumptions (5)
  • domain assumption Prescription counts are a valid proxy for health outcome prevalence.
    The outcome variable is defined as the total quantity of outcome-related prescriptions (Related Work). If prescriptions reflect access or prescribing practices, all associations are biased.
  • domain assumption Unweighted LAD-level means preserve the population-level relationships of interest.
    Step 4 aggregates LSOA data to LAD centroids by taking the mean of predictors and outcomes, ignoring population sizes and within-LAD distributions, which can induce ecological bias.
  • domain assumption Knockoffs control FDR and produce a valid variable subset under the treatment of all 154 predictors.
    The knockoffs filter is applied without verifying the model-X conditions or reporting the exact FDR level tuned; the resulting subset is used as the basis for all further analysis.
  • domain assumption GAM smooth terms and MGWR kernel weighting adequately capture non-linearity and spatial heterogeneity without confounding from correlated predictors.
    The paper notes MGWR struggles with correlated predictors and yet includes correlated environmental variables; no multicollinearity diagnostics are reported for the MGWR fits.
  • domain assumption LAD centroids are appropriate coordinates for spatial smoothing.
    Centroids are used as aggregated coordinates in both GAM tensor terms and MGWR distance weighting, implicitly assuming that within-LAD spatial variation is negligible.

how reviews work

0 comments
Cite this review

Pith. "Pith review of How Your Location Relates to Health: Variable Importance and Interpretable Machine Learning for Environmental and Sociodemographic Data." pith.science (2026). https://pith.science/paper/TO2AOLCP

@misc{pith2026250102111,
  author       = {Pith},
  title        = {Pith review of: How Your Location Relates to Health: Variable Importance and Interpretable Machine Learning for Environmental and Sociodemographic Data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TO2AOLCP}},
  note         = {Machine review of arXiv:2501.02111}
}
read the original abstract

Health outcomes depend on complex environmental and sociodemographic factors whose effects change over location and time. Only recently has fine-grained spatial and temporal data become available to study these effects, namely the MEDSAT dataset of English health, environmental, and sociodemographic information. Leveraging this new resource, we use a variety of variable importance techniques to robustly identify the most informative predictors across multiple health outcomes. We then develop an interpretable machine learning framework based on Generalized Additive Models (GAMs) and Multiscale Geographically Weighted Regression (MGWR) to analyze both local and global spatial dependencies of each variable on various health outcomes. Our findings identify NO2 as a global predictor for asthma, hypertension, and anxiety, alongside other outcome-specific predictors related to occupation, marriage, and vegetation. Regional analyses reveal local variations with air pollution and solar radiation, with notable shifts during COVID. This comprehensive approach provides actionable insights for addressing health disparities, and advocates for the integration of interpretable machine learning in public health.

Figures

Figures reproduced from arXiv: 2501.02111 by the authors.

Figure 1
Figure 1. Overview of the full methodology pipeline. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 4
Figure 4. MGWR Coefficients for NO2 with Asthma (left), [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 3
Figure 3. Shape Functions with 95% CIs for NO2 with [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (2 more)
Figure 5
Figure 5. Figure 5: MGWR Coefficients for PM2.5 with Diabetes in [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: MGWR Coefficients for PM2.5 with Hypertension [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

46 extracted references · 42 canonical work pages

  1. [1]

    Altmann, A.; Toloşi, L.; Sander, O.; and Lengauer, T. 2010. Permutation importance: A corrected feature importance measure. Bioinformatics, 26

    work page 2010

  2. [2]

    N.; Biagioni, B.; D'Amato, G.; and Cecchi, L

    Annesi-Maesano, I.; Maesano, C. N.; Biagioni, B.; D'Amato, G.; and Cecchi, L. 2021. Call to action: Air pollution, asthma, and allergy in the exposome era

    work page 2021

  3. [3]

    F.; and Candès, E

    Barber, R. F.; and Candès, E. J. 2015. Controlling the false discovery rate via knockoffs. The Annals of Statistics, 43(5)

    work page 2015

  4. [4]

    S.; and Charlton, M

    Brunsdon, C.; Fotheringham, A. S.; and Charlton, M. 1996. Geographically weighted regression: A method for exploring spatial nonstationarity. Geographical Analysis, 28(4): 281--298

    work page 1996

  5. [5]

    Byeon, H. 2021. Exploring factors for predicting anxiety disorders of the elderly living alone in south korea using interpretable machine learning: A population-based study. International Journal of Environmental Research and Public Health, 18

    work page 2021

  6. [6]

    Carlsson, S.; Andersson, T.; Talbäck, M.; and Feychting, M. 2020. Incidence and prevalence of type 2 diabetes by occupation: results from all Swedish employees. Diabetologia, 63

    work page 2020

  7. [7]

    K.; Azahadi, M.; Phang, S

    Cheah, Y. K.; Azahadi, M.; Phang, S. N.; and Manaf, N. H. A. 2020. Sociodemographic, Lifestyle, and Health Factors Associated With Depression and Generalized Anxiety Disorder Among Malaysian Adults. Journal of Primary Care and Community Health, 11

    work page 2020

  8. [8]

    Chen, T.; and Guestrin, C. 2016. XGBoost: A Scalable Tree Boosting System. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 785--794. ACM

    work page 2016

Show all 46 references
  1. [9]

    C.; and Lin, C

    Chien, I. C.; and Lin, C. H. 2016. Increased risk of diabetes in patients with anxiety disorders: A population-based study. Journal of Psychosomatic Research, 86

    2016

  2. [10]

    N.; Joppa, L.; Vasiliou, V.; and Kleinstreuer, N

    Comess, S.; Akbay, A.; Vasiliou, M.; Hines, R. N.; Joppa, L.; Vasiliou, V.; and Kleinstreuer, N. 2020. Bringing Big Data to Bear in Environmental Public Health: Challenges and Recommendations. Frontiers in Artificial Intelligence, 3

    2020

  3. [11]

    Datadance. 2023. Understanding Generalized Additive Models (GAMs): A Comprehensive Guide. Datadance

    2023

  4. [12]

    ECMWF . 2023. ERA5-Land: data documentation . https://confluence.ecmwf.int/display/CKB/ERA5-Land

    2023

  5. [13]

    H.; and Sakr, S

    Elshawi, R.; Al-Mallah, M. H.; and Sakr, S. 2019. On the interpretability of machine learning-based model for predicting hypertension. BMC Medical Informatics and Decision Making, 19

    2019

  6. [14]

    Fisher, A.; Rudin, C.; and Dominici, F. 2019. All Models are Wrong, but Many are Useful: Learning a Variable's Importance by Studying an Entire Class of Prediction Models Simultaneously. arXiv:1801.01489

    2019 arXiv

  7. [15]

    Gergov, V.; Prevendar, T.; Vousoura, E.; Ulberg, R.; Dahl, H. S. J.; Feller, C.; Jacobsen, C. F.; Karain, A.; Milic, B.; Poznyak, E.; Sacco, R.; Tulbure, B. T.; Camilleri, N.; Liakea, I.; Podina, I.; Saliba, A.; Torres, S.; and Poulsen, S. 2023. Sociodemographic Predictors and...

    2023

  8. [16]

    Greater London Authority . 2020. Central London ULEZ - Six Month Report. Technical report, Greater London Authority, London, UK

    2020

  9. [17]

    J.; and Tibshirani, R

    Hastie, T. J.; and Tibshirani, R. J. 1990. Generalized Additive Models. Chapman and Hall/CRC

    1990

  10. [18]

    Hooker, G.; Mentch, L.; and Zhou, S. 2021. Unrestricted permutation forces extrapolation: variable importance requires at least one more model, or there is no free variable importance. Statistics and Computing, 31: 1--16

    2021

  11. [19]

    Hu, N.; Zhang, Z.; Duffield, N.; Li, X.; Dadashova, B.; Wu, D.; Yu, S.; Ye, X.; Han, D.; and Zhang, Z. 2024. Geographical and temporal weighted regression: examining spatial variations of COVID-19 mortality pattern using mobility and multi-source data. Computational Urban Science, 4

    2024

  12. [20]

    Jia, X. 2022. US Trends in Diabetes and Hypertension: New Year Resolutions for CVD Prevention Improvement. American College of Cardiology

    2022

  13. [21]

    J.; Kendon, E

    Keat, W. J.; Kendon, E. J.; and Bohnenstengel, S. I. 2021. Climate change over UK cities: the urban influence on extreme temperatures in the UK climate projections. Climate Dynamics, 57

    2021

  14. [22]

    J.; Mudway, I

    Kelly, F. J.; Mudway, I. S.; and Fussell, J. C. 2021. Air Pollution and Asthma: Critical Targets for Effective Action

    2021

  15. [23]

    B.; Kjær, J

    Kristiansen, C. B.; Kjær, J. N.; Hjorth, P.; Andersen, K.; and Prina, A. M. 2019. Prevalence of common mental disorders in widowhood: A systematic review and meta-analysis. Journal of Affective Disorders, 245

    2019

  16. [24]

    J.; and Wasserman, L

    Lei, J.; G'Sell, M.; Rinaldo, A.; Tibshirani, R. J.; and Wasserman, L. 2018. Distribution-Free Predictive Inference For Regression. Journal of the American Statistical Association

    2018

  17. [25]

    Li, Z. 2022. Extracting spatial effects from machine learning model using local interpretation method: An example of SHAP and XGBoost. Computers, Environment and Urban Systems, 96: 101845

    2022

  18. [26]

    Liu, T.; Liu, Y.; Su, Y.; Hao, J.; and Liu, S. 2024. Air pollution and upper respiratory diseases: an examination among medically insured populations in Wuhan, China. International Journal of Biometeorology, 68

    2024

  19. [27]

    M.; and Lee, S.-I

    Lundberg, S. M.; and Lee, S.-I. 2017. A unified approach to interpreting model predictions. Advances in neural information processing systems, 30

    2017

  20. [28]

    G.; and Lieberman, M

    Menaghan, E. G.; and Lieberman, M. A. 1986. Changes in Depression following Divorce: A Panel Study. Journal of Marriage and the Family, 48

    1986

  21. [29]

    P.; and Rahwan, I

    Obradovich, N.; Migliorini, R.; Paulus, M. P.; and Rahwan, I. 2018. Empirical evidence of mental health risks posed by climate change. Proceedings of the National Academy of Sciences, 115(43): 10953--10958

    2018

  22. [30]

    M.; Li, Z.; Kang, W.; Wolf, L

    Oshan, T. M.; Li, Z.; Kang, W.; Wolf, L. J.; and Fotheringham, A. S. 2019. MGWR: A python implementation of multiscale geographically weighted regression for investigating process spatial heterogeneity and scale. ISPRS International Journal of Geo-Information, 8

    2019

  23. [31]

    A.; Meltzer, G

    Potter, N. A.; Meltzer, G. Y.; Avenbuan, O. N.; Raja, A.; and Zelikoff, J. T. 2021. Particulate Matter and Associated Metals: A Link with Neurotoxicity and Mental Health. Atmosphere, 12(4): 425

    2021

  24. [32]

    Qasrawi, R.; Polo, S. P. V.; Al-Halawa, D. A.; Hallaq, S.; and Abdeen, Z. 2022. Assessment and Prediction of Depression and Anxiety Risk Factors in Schoolchildren: Machine Learning Techniques Performance Analysis. JMIR Formative Research, 6

    2022

  25. [33]

    Qiu, T.; Jiang, Z.; Chen, X.; Dai, Y.; and Zhao, H. 2023. Comorbidity of Anxiety and Hypertension: Common Risk Factors and Potential Mechanisms

    2023

  26. [34]

    S.; Katoch, V.; Bhardwaj, S.; Kaur-Sidhu, M.; Gupta, M.; and Mor, S

    Ravindra, K.; Bahadur, S. S.; Katoch, V.; Bhardwaj, S.; Kaur-Sidhu, M.; Gupta, M.; and Mor, S. 2023. Application of machine learning approaches to predict the impact of ambient air pollution on outpatient visits for acute respiratory infections. Science of the Total Environment, 858

    2023

  27. [35]

    V.; Sadeghi-Niaraki, A.; and Choi, S

    Razavi-Termeh, S. V.; Sadeghi-Niaraki, A.; and Choi, S. M. 2021. Effects of air pollution in Spatio-temporal modeling of asthma-prone areas using a machine learning model. Environmental Research, 200

    2021

  28. [36]

    T.; Große, T.; Sonnenwald, D.; Fuchs, M.; and Walker, B

    Scarpone, C.; Brinkmann, S. T.; Große, T.; Sonnenwald, D.; Fuchs, M.; and Walker, B. B. 2020. A multimethod approach for county-scale geospatial analysis of emerging infectious diseases: A cross-sectional case study of COVID-19 incidence in Germany. International Journal of He...

    2020

  29. [37]

    Scepanovic, S.; Obadic, I.; Joglekar, S.; Giustarini, L.; Nattero, C.; Quercia, D.; and Zhu, X. 2023. MedSat: A Public Health Dataset for England Featuring Medical Prescriptions and Satellite Imagery. NeurIPS

    2023

  30. [38]

    Semenova, L.; Rudin, C.; and Parr, R. 2022. On the Existence of Simpler Machine Learning Models. In ACM Conference on Fairness, Accountability, and Transparency ( ACM FAccT )

    2022

  31. [39]

    Shao, Q.; Xu, Y.; and Wu, H. 2021. Spatial Prediction of COVID-19 in China Based on Machine Learning Algorithms and Geographically Weighted Regression. Computational and Mathematical Methods in Medicine, 2021

    2021

  32. [40]

    M.; Tsai, R

    Shockey, T. M.; Tsai, R. J.; and Cho, P. 2021. Prevalence of Diagnosed Diabetes Among Employed US Adults by Demographic Characteristics and Occupation, 36 States, 2014 to 2018. Journal of Occupational and Environmental Medicine, 63

    2021

  33. [41]

    Sun, H.; Chen, S.; Li, X.; Cheng, L.; Luo, Y.; and Xie, L. 2023. Prediction and Early Warning Model of Mixed Exposure of Air Pollution and Meteorological Factors on Death of Respiratory Diseases Based on Machine Learning. Environmental Science and Pollution Research

    2023

  34. [42]

    S.; Lu, P.; Choi, D.; Kobayashi, L

    Varghese, J. S.; Lu, P.; Choi, D.; Kobayashi, L. C.; Ali, M. K.; Patel, S. A.; and Li, C. 2023. Spousal Concordance of Hypertension Among Middle-Aged and Older Heterosexual Couples Around the World: Evidence From Studies of Aging in the United States, England, China, and India...

    2023

  35. [43]

    Wang, J.; Wei, Z.; Yao, N.; Li, C.; and Sun, L. 2023. Association Between Sunlight Exposure and Mental Health: Evidence from a Special Population Without Sunlight in Work. Risk Management and Healthcare Policy, 16

    2023

  36. [44]

    Zhang, L.; Zhou, S.; Qi, L.; and Deng, Y. 2022. Nonlinear Effects of the Neighborhood Environments on Residents’ Mental Health. International Journal of Environmental Research and Public Health, 19

    2022

  37. [45]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all...

  38. [46]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.