Pith. sign in

REVIEW 4 major objections 5 minor 26 references

Statistical Learning of Pediatric Mental Health-Related Emergency Department Visits Across COVID-19 Pandemic Periods

T0 review · 4 major / 5 minor · reviewed 2026-08-01 · deepseek-v4-flash

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

Pith's one-line read The paper develops a stepwise statistical framework for analyzing pediatric mental health–related emergency department (MHED) visits as zero-truncated recurrent event data, where each child is observed only if they had at least one visit, a

desk verdict Competent applied extension of the authors' own recurrent-event methods to a timely public-health question; the main caveats are a potentially violated censoring assumption the authors themselves flag, and some interpretive overreach relative to the paper's own confidence intervals. read the letter →

arxiv 2607.26210 v1 pith:NWXQHY2Q submitted 2026-07-28 stat.AP

Anqi A. Chen , X. Joan Hu , Rhonda J. Rosychuk This is my paper · ORCID
classification stat.AP MSC 62N0162P10
keywords conditionalintensityfunctionhealthadministrativedatastratifiedregressionanalysiszero-truncatedrecurrenteventspatientmentalCOVID-19pandemictime-varyingcoefficients
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

The paper claims that a carefully staged modeling approach can recover how pediatric mental health–related emergency department visits changed across the COVID-19 pandemic periods, despite two data limitations: children with no visits are missing from the records, and exact birthdates are unavailable. Starting with nonparametric rate estimates to guide model choice, the analysis moves to Cox-type intensity models with time-varying coefficients, first stratified by pandemic period and then by whether the child has already had a visit. Applying this to 82,481 children with 161,026 visits, the framework yields period-specific visit rates and risk-factor effects, showing, for example, that annual visit counts rose during the pandemic while visits per person fell, and that the age at which female risk exceeds male risk dropped from about 11 to about 10 years. A sympathetic reader would care because the approach offers a practical template for extracting reliable conclusions from imperfect administrative health records.

What carries the argument

The key machinery is the adaptation of Cox-type recurrent-event regression to zero-truncated, coarsened data. Because exact birthdates are masked, the method generates plausible birthdates from the intervals implied by integer ages at visits, assuming a uniform distribution, and averages estimates over replicates. Because non-visitors are absent from the data, population census counts supply the at-risk denominators. The final model uses age as the time scale, stratifies by pandemic period and a history indicator (first visit vs. subsequent visits), and lets regression coefficients vary with age via local linear kernel smoothing.

What would settle it

A sensitivity analysis that adds birth-cohort terms or uses inverse probability of censoring weighting and checks whether the estimated period-specific rate curves and sex-effect crossover ages change materially; if they do, the independent-censoring assumption does not hold.

Watch

Extended reading notes

Core claim

The central claim is that a stepwise statistical learning framework—nonparametric marginal rate estimation followed by Cox-type regression models stratified by pandemic period and event history—can characterize how pediatric MHED visit patterns evolved across the pre-, during-, and post-COVID periods. The analysis finds that the pandemic period saw higher annual visit counts but fewer visits per person, a larger female share of visits, and a shift in the sex-effect crossover age from roughly 11 years before the pandemic to roughly 10 years during and after it. The history-stratified model further shows that after a first visit, the risk of subsequent visits is substantially higher and has di

Load-bearing premise

The assumption that a child's birthdate, which determines their observation window, is independent of their mental-health visit pattern; if birth cohort is associated with the visit process, the period-specific estimates are biased.

Editorial extensions

If this is right

  • Provides a reusable template for analyzing recurrent healthcare utilization data when exact event origins are masked and non-users are missing.
  • Demonstrates that event-history stratification matters: the risk factors for a first MHED visit differ from those for subsequent visits, and the cumulative risk after a first visit is much higher.
  • Quantifies COVID-era changes: per-year visit volume rose during the pandemic while per-person visit frequency fell, and the female share of visits increased.
  • Shows that the age at which girls overtake boys in MHED risk moved earlier during the pandemic, pointing to age-specific windows for intervention.
  • The framework can be transferred to other zero-truncated recurrent event settings with evolving temporal patterns, such as chronic disease care episodes.

Reading between the lines

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

  • If the independent-censoring assumption fails—because birth cohort is associated with visit patterns—the period-specific rate and covariate estimates could be biased; an inverse probability of censoring weighting extension would test and correct this.
  • The prespecified cutoffs (WHO pandemic declaration and school-mask lifting) treat the period boundaries as known; treating them as unknown change-points could alter which differences are attributed to the pandemic versus background trends.
  • The same framework could be applied to condition-specific MHED visits (e.g., self-harm, mood disorders) or to other regions with similar administrative data, potentially revealing whether the observed sex and deprivation patterns are generalizable.
  • The shift in the sex crossover age during the pandemic might reflect differential pandemic-related stressors on adolescent girls, an inference the paper does not make but that could motivate targeted mental health screening in schools or primary care.
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. This paper develops a stepwise statistical framework for pediatric mental health-related emergency department (MHED) visits in Alberta, 2010–2025. The data are treated as zero-truncated recurrent events on the age scale. The framework proceeds from a nonparametric marginal rate estimator (with multiple imputation for coarsened birthdates and census-based denominators) to Poisson and history-stratified Cox-type models with age-varying coefficients, with all models stratified by three COVID-19 pandemic periods. The application to 82,481 subjects and 161,026 visits yields estimated rate functions, coefficient curves, and cumulative baseline intensities. The paper claims that the framework provides insights into how visit frequencies and covariate effects evolved across the pre-, during-, and post-pandemic periods and how previous MHED visits influence subsequent visits.

Significance. If the results are valid, the framework offers a practical way to analyze recurrent healthcare utilization data that feature zero truncation and coarsened birthdates, a setting common in administrative databases. The manuscript is transparent about its data sources, and the use of census denominators to account for subjects with no MHED visits is a genuine strength. The multiple-imputation approach for missing birthdates is also sensible. However, the methodological novelty is limited because the core estimating equations are adapted from the authors' earlier work; the new contribution is mainly the workflow and its application. The central scientific claim about changes across pandemic periods rests on an assumption that the authors themselves describe as possibly violated, and it is not supported by formal tests or simulation evidence. Reproducibility is also limited by the empty Code Availability section.

major comments (4)
  1. [§2.2.2 and Discussion] The independent-censoring assumption (birthdate B_i independent of N_i) is load-bearing because the age time scale is defined as calendar time minus birthdate, so birth cohort and calendar period are collinear. The authors acknowledge in the Discussion that this assumption 'may be more restrictive' and cite Xiong et al. [18] for generational differences in MHED visit patterns, but no sensitivity analysis or simulation is provided; IPCW is only promised as future work. If birth cohort affects the event rate, the period-specific estimates in Eqs. (2)–(3) and the stratified comparisons in Models (4) and (7) could be qualitatively biased. I recommend adding a simulation or sensitivity analysis that introduces a cohort effect and quantifies the resulting bias in the period comparisons.
  2. [§3.3 and Abstract] The central claim that covariate effects 'evolved' across the three pandemic periods is supported only by visual inspection of pointwise 95% confidence intervals. There is no formal test for differences between periods, no joint confidence bands, and no adjustment for the many age-grid comparisons. Some statements are stronger than the displayed intervals warrant; for example, the text notes 'substantial overlap' between age-varying and age-constant confidence intervals yet concludes that the results 'support the use of age-varying regression coefficients,' and several 'higher/lower risk' claims are based on overlapping pointwise intervals. Formal sup-norm tests or simultaneous confidence bands are needed, or the claims should be explicitly reframed as descriptive.
  3. [§2.2.2–§2.2.4] No simulation study is reported for any of the proposed estimators. The estimating equations involve multiple imputation of birthdates, census-based denominators, and conditional probabilities for history-based strata, but the finite-sample bias, variance, and coverage of the resulting estimators are unexamined. Given that the methodology is adapted from prior work and this paper presents it as a practical framework, a simulation study under realistic zero-truncated, coarsened-birthdate settings is necessary to assess whether the proposed workflow reliably recovers the true rate functions and coefficients.
  4. [Equation (1), §2.2.2] The sentence following Eq. (1) states that 'the summation over O can be equivalently restricted to O1' because Y_i^{(c)}(·|B_i)dN_i(·)=0 for subjects outside the MHED cohort. This is true for the numerator, but the denominator in Eq. (1) contains no dN_i term and must include the full target population O. If the restriction were applied to the denominator, the zero-truncation correction would be lost. This needs to be clarified, especially because the denominator is the mechanism that uses census information.
minor comments (5)
  1. [§2.2.3–§2.2.4] The text refers to 'Approach A' in Section 2.2.3 and 'Procedure A' in Section 2.2.4; these names are not defined and appear inconsistent. Please harmonize.
  2. [Introduction] There are typographical spacing issues: 'TheAndersen-Gill model' and 'Prentice-Williams-Peterson models' should have spaces after 'The' and 'Prentice'.
  3. [Table 1] The table header reads 'T able 1'; the extra space should be removed.
  4. [§3.3] The sentence 'Set the pre-determined constant τ_L to 9 units...' is missing a subject; it should be 'We set...'. Also, the definition of age units (years vs. two-month units) should be stated more explicitly in the main text.
  5. [Code Availability] The manuscript states that analyses were conducted in R and C++ via Rcpp, but the Code Availability section is empty. Since the data are not public, providing the analysis code is important for reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: all estimates are directly data/census-driven; self-citations provide estimation machinery, not the conclusions.

full rationale

The paper's derivation chain does not recycle a fitted input as a prediction. Period cutoffs (March 11, 2020 and February 14, 2022) and census denominators are external, prespecified inputs. The nonparametric estimator (Eq. 3) is a direct ratio of observed MHED visits to census population counts; the Cox-type estimators (Eqs. 5-10) are standard local-linear/Breslow-type procedures adapted from the authors' earlier published papers. Those citations supply algorithms and identifying assumptions (e.g., uniform birthdate imputation), but they are not used to manufacture the scientific conclusions: the period-specific rate and coefficient estimates are computed from the Alberta NACRS and census data. The flowchart's choice between age-constant and age-varying coefficients is a data-driven model-selection step, not an equation-level reduction of the final estimates to the selection criterion. The Discussion explicitly acknowledges the independent-censoring assumption may be violated by birth-cohort effects and proposes IPCW as future work; that is a validity threat, not circularity. No specific reduction of a claimed result to its own inputs was found.

Assumptions & free parameters 6 free parameters · 7 assumptions · 0 invented entities

The central estimates depend on several structural assumptions inherited from the administrative data setting and the authors' previous methods: independence, independent censoring, complete NACRS recording, uniform birthdate imputation, census denominators, fixed covariates, and correct intensity model specification. The free parameters are tuning/grouping choices rather than fitted scientific constants, but they affect the reported curves and conclusions.

free parameters (6)
  • Kernel bandwidth h = 9 units (1.5 years)
    Epanechnikov kernel bandwidth chosen by the authors; controls smoothness of age-varying coefficient estimates and affects all reported curves.
  • Boundary trimming constants tau_L and tau_R = tau_L=9 units (1.5 y), tau_R=105 units (17.5 y)
    Chosen to avoid boundary problems; estimates are flattened outside this interval, influencing interpretation at young and old ages.
  • Age discretization unit = two months
    Ages discretized into two-month units; affects the local constant fitting and the handling of tied event ages.
  • Pandemic period cutoffs = 2020-03-11 and 2022-02-14
    Prespecified policy dates, but any alternative cutoffs could change period-specific estimates and conclusions. The paper acknowledges this in the Discussion.
  • Birthdate imputation count R = not specified
    R independent birthdate draws are averaged over, but the actual number used is not reported; the choice affects Monte Carlo error in the estimators.
  • Deprivation grouping = quintiles 1-3 plus unknown vs 4-5
    A hand-chosen binary collapse of the deprivation index; different groupings could change the deprivation effect estimates.
assumptions (7)
  • domain assumption Independence among subjects in the target population P
    Assumed in Section 2.2.2 and listed as assumption (i) in the Discussion; within-community correlation is dismissed because regions are large, but no test is provided.
  • domain assumption Independent censoring: birthdate and extraction window independent of the event process
    Stated in Section 2.2.2; the Discussion admits this is potentially restrictive and cites prior evidence of generational differences.
  • domain assumption NACRS completeness: absence of an MHED visit in administrative data means absence in practice
    Stated in Section 2.1; if records are missing or non-mandatory, the zero-truncated formulation is invalid.
  • ad hoc to paper Uniform birthdate distribution over the inferred interval
    Section 2.2.2: 'we assume that birthdate B_i follows a uniform distribution, Unif(I_i)' following previous work; no sensitivity analysis is reported.
  • domain assumption Census counts are valid denominators for the zero-truncated population at risk
    Equation (3) uses population census counts for age/covariate groups as denominators, implicitly assuming these counts correspond to the same at-risk population captured by NACRS numerators.
  • domain assumption Covariates are time-independent and fixed at first recorded visit
    Section 2.1 states characteristics are treated as time-independent because they were 'relatively stable over the study period'; no verification is presented.
  • domain assumption Correct specification of the Poisson and stratified intensity models
    Equations (4) and (7) assume a multiplicative age-varying Cox form for the intensity/rate; misspecification would bias the estimated covariate effects and period comparisons.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Statistical Learning of Pediatric Mental Health-Related Emergency Department Visits Across COVID-19 Pandemic Periods." pith.science (2026). https://pith.science/paper/NWXQHY2Q

@misc{pith2026260726210,
  author       = {Pith},
  title        = {Pith review of: Statistical Learning of Pediatric Mental Health-Related Emergency Department Visits Across COVID-19 Pandemic Periods},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NWXQHY2Q}},
  note         = {Machine review of arXiv:2607.26210}
}
read the original abstract

This article presents a statistical learning framework for studying the evolution of pediatric mental health-related emergency department (MHED) visit patterns across the pre-, during-, and post-COVID-19 pandemic periods using population-based administrative health records. The MHED records are formulated as zero-truncated recurrent event data, partitioned into three successive time periods. We develop the modeling framework in a stepwise manner, guided by model fit using a collection of MHED records. The resulting framework progresses from nonparametric marginal rate models to more structured Cox-type regression models for characterizing visit patterns. We ultimately apply stratified regression analysis to investigate changes in visit frequencies and covariate effects across pandemic periods, accounting for prespecified period cut-off points and coarsened individual follow-up information. The proposed framework is motivated by and illustrated using pediatric MHED data throughout the article, providing a practical approach for analyzing recurrent healthcare utilization data with evolving temporal patterns.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

26 extracted references · 4 canonical work pages

  1. [18]

    Journal of the Royal Statistical Society Series C (Applied Statistics), 001 (2026)

    Xiong, Y., Hu, X.J., Rosychuk, R.J.: Learning differences between two decades of mental health related emergency department visits by youth via recurrent events data analyses. Journal of the Royal Statistical Society Series C (Applied Statistics), 001 (2026)

    2026

  2. [1]

    Licence: CC BY-NC-SA 3.0 IGO

    World Health Organization: World mental health today: latest data. Licence: CC BY-NC-SA 3.0 IGO. https://iris.who.int/server/api/core/bitstreams/ 31714489-1345-4439-8b37-6cbdc52e15ca/content (2025)

    2025

  3. [2]

    https: //mentalhealthcommission.ca/catalyst/brave-new-world/ Accessed 2026-07-22

    Chevrier, N.: Brave New World: Integrated Service Hubs Are an Innovative Approach to Transforming Youth Mental Health Care in Canada (2023). https: //mentalhealthcommission.ca/catalyst/brave-new-world/ Accessed 2026-07-22

    2023

  4. [3]

    Canadian Medical Association Journal184(12), 665–674 (2012)

    Newton, A.S., Rosychuk, R.J., Dong, K., Curran, J., Slomp, M., McGrath, P.J.: Emergency health care use and follow-up among sociodemographic groups of children who visit emergency departments for mental health crises. Canadian Medical Association Journal184(12), 665–674 (2012)

    2012

  5. [4]

    Canadian Journal of Emergency Medicine21(1), 75–86 (2019) https://doi.org/10.1017/cem.2017.416

    Cappelli, M., Cloutier, P., Newton, A.S., Fitzpatrick, E., Ali, S., Dong, K.A., Gray, C., Kennedy, A., Lyons, J.S., Polihronis, C., Rosychuk, R.J.: Evaluating mental health service use during and after emergency department visits in a multisite cohort of canadian children and youth. Canadian Journal of Emergency Medicine21(1), 75–86 (2019) https://doi.org...

    work page doi:10.1017/cem.2017.416 2019

  6. [5]

    Academic Pediatrics19(4), 386–393 (2019) https://doi.org/10

    Hoffmann, J.A., Stack, A.M., Samnaliev, M., Monuteaux, M.C., Lee, L.K.: Trends in visits and costs for mental health emergencies in a pediatric emergency depart- ment, 2010–2016. Academic Pediatrics19(4), 386–393 (2019) https://doi.org/10. 1016/j.acap.2019.02.006

    2010

  7. [6]

    Academic pediatrics22(6), 889–891 (2022)

    Cutler, G.J., Bergmann, K.R., Doupnik, S.K., Hoffmann, J.A., Neuman, M.I., Rodean, J., Zagel, A.L., Zima, B.T.: Pediatric mental health emergency depart- ment visits and access to inpatient care: A crisis worsened by the COVID-19 pandemic. Academic pediatrics22(6), 889–891 (2022)

    2022

  8. [7]

    JAMA329(17), 1469–1477 (2023) https://doi.org/10.1001/jama.2023.4809

    Bommersbach, T.J., McKean, A.J., Olfson, M., Rhee, T.G.: National trends in mental health–related emergency department visits among youth, 2011-2020. JAMA329(17), 1469–1477 (2023) https://doi.org/10.1001/jama.2023.4809

    arXiv 2011

Show all 26 references
  1. [8]

    Academic emergency medicine31(8), 739–754 (2024)

    Hoffmann, J.A., Carter, C.P., Olsen, C.S., Ashby, D., Bouvay, K.L., Duffy, S.J., Chamberlain, J.M., Chaudhary, S.S., Glomb, N.W., Grupp-Phelan, J.,et al.: Pediatric mental health emergency department visits from 2017 to 2022: A multi- center study. Academic emergency medicine3...

    2017

  2. [9]

    Springer, New York, NY (2007)

    Cook, R.J., Lawless, J.F.: The Statistical Analysis of Recurrent Events, 1st edn. Springer, New York, NY (2007)

    2007

  3. [10]

    The Annals of Statistics10(4), 1100–1120 (1982) 17

    Andersen, P.K., Gill, R.D.: Cox’s regression model for counting processes: A large sample study. The Annals of Statistics10(4), 1100–1120 (1982) 17

    1982

  4. [11]

    Journal of the Royal Statistical Society, Series B (Statistical Methodology)34(2), 187–220 (1972)

    Cox, D.R.: Regression models and life-tables. Journal of the Royal Statistical Society, Series B (Statistical Methodology)34(2), 187–220 (1972)

    1972

  5. [12]

    Biometrika68(2), 373–379 (1981)

    Prentice, R.L., Williams, B.J., Peterson, A.V.: On the regression analysis of multivariate failure time data. Biometrika68(2), 373–379 (1981)

    1981

  6. [13]

    Cancer121(24), 4389–4397 (2015)

    Sutradhar, R., Agha, M., Pole, J.D., Greenberg, M., Guttmann, A., Hodgson, D., Nathan, P.C.: Specialized survivor clinic attendance is associated with decreased rates of emergency department visits in adult survivors of childhood cancer. Cancer121(24), 4389–4397 (2015)

    2015

  7. [14]

    BMC Geriatrics18(1), 157–11 (2018)

    Gruneir, A., Cigsar, C., Wang, X., Newman, A., Bronskill, S.E., Anderson, G.M., Rochon, P.A.: Repeat emergency department visits by nursing home residents: A cohort study using health administrative data. BMC Geriatrics18(1), 157–11 (2018)

    2018

  8. [15]

    Journal of the American Geriatrics Society (JAGS)71(12), 3731–3743 (2023)

    Jones, A., Watt, J.A., Maclagan, L.C., Swayze, S., Jaakkimainen, L., Schull, M.J., Bronskill, S.E.: Factors associated with recurrent emergency department visits among people living with dementia: A retrospective cohort study. Journal of the American Geriatrics Society (JAGS)7...

    2023

  9. [16]

    Accident Analysis & Prevention39(2), 290–299 (2007) https://doi.org/10.1016/j.aap.2006.07.009

    Lim, H.J., Liu, J., Melzer-Lange, M.: Comparison of methods for analyzing recurrent events data: Application to the emergency department visits of pedi- atric firearm victims. Accident Analysis & Prevention39(2), 290–299 (2007) https://doi.org/10.1016/j.aap.2006.07.009

    2007 doi

  10. [17]

    Biometrics72(4), 1113–1122 (2016)

    Hu, X.J., Rosychuk, R.J.: Marginal regression analysis of recurrent events with coarsened censoring times. Biometrics72(4), 1113–1122 (2016)

    2016

  11. [19]

    Journal of the American Statistical Association91(433), 300–310 (1996)

    Hu, X.J., Lawless, J.F.: Estimation of rate and mean functions from truncated recurrent event data. Journal of the American Statistical Association91(433), 300–310 (1996)

    1996

  12. [20]

    The Annals of Applied Statistics20(1), 434–451 (2026)

    Chen, A.A., Hu, X.J., Rosychuk, R.J.: Stratified regression analysis of zero- truncated recurrent event data. The Annals of Applied Statistics20(1), 434–451 (2026)

    2026

  13. [21]

    arXiv preprint arXiv:2507.23060 (2025)

    Chen, A.A., Hu, X.J., Rosychuk, R.J., Zeng, L.: Analyzing zero-truncated recur- rent events by stratified regression with time-varying coefficients. arXiv preprint arXiv:2507.23060 (2025)

    2025 arXiv

  14. [22]

    https://icd.who.int/browse10/Content/statichtml/ ICD10Volume2 en 2019.pdf (2019)

    World Health Organization: International statistical classification of diseases and 18 related health, 10th revision. https://icd.who.int/browse10/Content/statichtml/ ICD10Volume2 en 2019.pdf (2019)

    2019

  15. [23]

    https://www.cihi.ca/en/ national-ambulatory-care-reporting-system-nacrs-metadata

    Canadian Institute for Health Information: National Ambulatory Care Reporting System (NACRS) metadata. https://www.cihi.ca/en/ national-ambulatory-care-reporting-system-nacrs-metadata

  16. [24]

    Journal of the Royal Statistical Society, Series B (Statistical Methodology)34, 216 (1972)

    Breslow, N.E.: Discussion of Professor Cox’s paper. Journal of the Royal Statistical Society, Series B (Statistical Methodology)34, 216 (1972)

    1972

  17. [25]

    Eddelbuettel, D., Francois, R., Allaire, J., Ushey, K., Kou, Q., Russell, N., Ucar, I., Bates, D., Chambers, J.: Rcpp: Seamless R and C++ Integration. (2025). https://doi.org/10.32614/CRAN.package.Rcpp . R package version 1.1.0. https: //CRAN.R-project.org/package=Rcpp

    2025 doi

  18. [26]

    Eddelbuettel, D., Francois, R., Bates, D., Ni, B., Sanderson, C.: RcppArmadillo: ’Rcpp’ Integration for the ’Armadillo’ Templated Linear Algebra Library. (2025). https://doi.org/10.32614/CRAN.package.RcppArmadillo . R package version 14.6.0-1. https://CRAN.R-project.org/packag...

    2025 doi

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.