Pith. sign in

REVIEW 3 major objections 4 minor 66 references

Vineyards can track and visualize evolving overdose hotspots across Ohio over time

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review

2026-07-08 20:00 UTC pith:B3IKYPFQ

load-bearing objection Solid applied-TDA use case on Ohio overdose deaths; the whole claim hangs on statistical tests we cannot audit from the abstract alone. the 3 major comments →

arxiv 2607.05710 v1 pith:B3IKYPFQ submitted 2026-07-07 math.AT

Visualizing Local Maxima of the Ohio overdose epidemic with Vineyards

Nicholas Bermingham , David White , Nathan Willey This is my paper
classification math.AT MSC 55N3162H1162P10
keywords vineyardstopological data analysisspatiotemporal dataoverdose epidemicOhiopersistence diagramshotspot evolutionpublic health
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper argues that vineyards—a tool from topological data analysis that tracks how connected features of a data set appear, merge, and disappear as a parameter changes—can reveal the spatiotemporal structure of drug-overdose deaths in Ohio. The authors first introduce statistical tests to decide whether a given spatiotemporal data set is suitable for vineyard analysis, then apply those tests to Ohio overdose counts and find the data pass. With that green light they construct vineyards that turn the yearly death maps into visual “vines” showing which local hotspots persist, intensify, or fade from year to year. Finally they explore further tests that flag which vines are statistically significant rather than noise. A sympathetic reader cares because the method supplies a concrete, visual language for the rise and migration of local overdose epidemics that ordinary heat-maps or county-level rates leave hidden.

Core claim

Vineyards constructed from a filtration of Ohio overdose-death counts over successive years produce diagrams whose vines track the birth, persistence, and death of local spatial hotspots, and the authors’ proposed statistical tests both confirm that the data are suitable for this analysis and can mark which vineyard features rise above chance.

What carries the argument

The vineyard: a continuous family of persistence diagrams obtained by computing topological features (connected components or holes) of the overdose-count point cloud under a filtration that varies both with a density threshold and with time; each feature’s birth and death times are joined across years into a “vine” that visualizes hotspot evolution.

Load-bearing premise

The authors’ proposed null models, filtrations, and significance criteria correctly decide when vineyards are suitable for count-based public-health data and correctly flag which vines are real rather than noise.

What would settle it

Apply the same vineyard pipeline and statistical tests to a synthetic data set that has the same marginal death counts per county but is deliberately free of spatiotemporal clustering; if the tests still declare the data “suitable” and flag significant vines, the method is not distinguishing signal from chance.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • Local overdose hotspots that persist across multiple years become visible as long vines, giving public-health officials a ranked list of enduring problem areas.
  • Hotspots that appear or disappear between consecutive years appear as short vines, marking the moments when an epidemic expands into or recedes from a new county.
  • The same statistical-suitability tests can be reapplied to other states’ overdose data or to other spatiotemporal public-health series to decide whether vineyard analysis is warranted.
  • Significant vines can be overlaid on geographic maps to produce an interactive, time-indexed visualization of the epidemic’s spatial trajectory.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Because the method is parameter-light once the filtration is fixed, it could serve as a near-real-time dashboard layer for state health departments that already publish yearly death counts.
  • The same vineyard construction should transfer, with only a change of input measure, to other count-based epidemics such as opioid-related emergency-department visits or naloxone administrations.
  • If the significance tests prove stable under modest changes of filtration, the vines themselves become candidate covariates for predictive models that forecast next-year hotspot locations.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The manuscript applies vineyards (time-parameterized persistence diagrams from topological data analysis) to spatiotemporal drug-overdose death counts in Ohio. It proposes statistical tests to assess whether vineyards are a suitable technique for a given spatiotemporal dataset, applies those tests to Ohio overdose data and concludes the data are suitable, then uses vineyards to visualize the evolution of local hotspots over time. It further explores statistical tests intended to verify the significance of features appearing in the resulting vineyard diagrams.

Significance. If the proposed suitability and significance tests are correctly specified and the visualizations are reproducible, the paper would offer a concrete applied demonstration that vineyards can track evolving local structure in public-health count data, together with a reusable testing pipeline for deciding when such methods are appropriate. That combination—methodological checks plus an applied case study on a high-impact epidemic—would be of interest to both the TDA-applications community and spatial epidemiology. The work is primarily empirical and methodological rather than a new theoretical construction; its value hinges on the soundness and transparency of the proposed tests and on the interpretability of the vineyard features for overdose hotspots.

major comments (3)
  1. The abstract and framing make the proposed statistical tests for vineyard suitability and for feature significance load-bearing for the central claim that vineyards are a “reasonable technique” for Ohio overdose data and that the visualized features are meaningful. From the available abstract alone those tests are only described as “proposing” and “exploring”; null models, filtrations, handling of discrete counts, spatial autocorrelation, population heterogeneity, and significance criteria are not specified. Without a clear statement of the nulls and of how they respect the structure of public-health death counts, the suitability conclusion and any declared significant vines cannot be audited. The manuscript must state the tests, null models, and decision rules explicitly (with section/equation references) so that the central claim can be checked.
  2. Relatedly, the suitability conclusion (“finding the data suitable”) is presented as a gate that licenses the subsequent vineyard analysis. If that gate rests on misspecified nulls (e.g., i.i.d. noise, pure temporal permutation, or homogeneous Poisson that ignore known spatial structure of counts), both the “data suitable” claim and the interpretation of hotspot vines can be spurious. The paper needs to justify why the chosen nulls are appropriate for overdose death counts and, ideally, report sensitivity to alternative nulls or to filtration choices.
  3. Visualization of local maxima / hotspots is the applied payoff, but the abstract does not indicate error bars, data-exclusion rules, population normalization, or how “local maxima” are identified from vineyard features. For the public-health claim to hold, the manuscript should document preprocessing (e.g., rates vs. raw counts, county vs. finer geography), the filtration used to build the vineyards, and how vines are mapped back to geographic hotspots so that the figures can be independently interpreted.
minor comments (4)
  1. The abstract is the only text provided in the query materials; the full manuscript should ensure that every claim in the abstract (suitability tests, significance tests, “local hotspots”) is backed by a numbered section, equation, or figure so that referees can cite them precisely.
  2. Clarify notation for vineyards vs. ordinary persistence diagrams early, and define any nonstandard terms (e.g., “local maxima of the epidemic”) in the introduction so that TDA and epidemiology readers share a common vocabulary.
  3. If code or data-processing scripts are available, a short reproducibility statement (repository, software versions, random seeds for any Monte Carlo nulls) would strengthen the empirical contribution.
  4. Ensure that any vineyard diagrams are labeled with time axes, birth–death scales, and geographic correspondence so that “evolution of local hotspots” is readable without the main text.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for a careful, constructive report. The three major concerns—explicit, auditable specification of the suitability and feature-significance tests (nulls, filtrations, decision rules); justification that those nulls respect the structure of public-health death counts, with sensitivity where feasible; and transparent documentation of preprocessing, filtration, and the map from vines to geographic hotspots—are well taken. These elements are load-bearing for the claim that vineyards are a reasonable technique for the Ohio overdose data and that the visualized features are meaningful. We will revise so that each is stated with clear section/equation references and so that the public-health interpretation of the figures is independently checkable. Point-by-point responses follow.

read point-by-point responses

  1. Referee: The abstract and framing make the proposed statistical tests for vineyard suitability and for feature significance load-bearing for the central claim. From the available abstract alone those tests are only described as “proposing” and “exploring”; null models, filtrations, handling of discrete counts, spatial autocorrelation, population heterogeneity, and significance criteria are not specified. The manuscript must state the tests, null models, and decision rules explicitly (with section/equation references) so that the central claim can be checked.

    Authors: We agree. Suitability and significance procedures are load-bearing, and the manuscript must make null models, filtrations, treatment of discrete counts, and decision rules fully auditable—not only at the level of the abstract’s high-level wording. In revision we will: (i) state the suitability and feature-significance procedures in dedicated methods subsections with numbered equations for the test statistics and Monte Carlo/permutation decision rules; (ii) specify the null models explicitly, including how discrete death counts enter the pipeline; (iii) name the filtration used to build the time-parameterized persistence diagrams; (iv) state significance criteria and any multiple-testing or vine-tracking thresholds; and (v) cross-reference these statements from the abstract, introduction, and results so the “reasonable technique” and “significant vines” claims can be checked without ambiguity. We do not claim new asymptotic theory; the procedures are simulation-based checks tailored to the vineyard pipeline, and we will make that design choice explicit rather than leave it implicit in “proposing/exploring” language. revision: yes

  2. Referee: Relatedly, the suitability conclusion (“finding the data suitable”) is presented as a gate that licenses the subsequent vineyard analysis. If that gate rests on misspecified nulls (e.g., i.i.d. noise, pure temporal permutation, or homogeneous Poisson that ignore known spatial structure of counts), both the “data suitable” claim and the interpretation of hotspot vines can be spurious. The paper needs to justify why the chosen nulls are appropriate for overdose death counts and, ideally, report sensitivity to alternative nulls or to filtration choices.

    Authors: We agree that a misspecified null would undermine both the suitability gate and the interpretation of hotspot vines. The revision will state the design principle of our nulls explicitly: they are intended to preserve marginal spatial intensity and overall temporal death volume while destroying the particular spatiotemporal dependence that vineyards are meant to track—not i.i.d. white noise or an unstructured homogeneous Poisson field. We will justify why that principle is appropriate for the geographic units and count nature of the overdose data, and we will discuss limitations relative to richer models that fully encode known spatial autocorrelation and population heterogeneity. We will also add at least one sensitivity check under an alternative null (e.g., a population-weighted or spatially structured permutation) and, where practical, a brief remark on sensitivity to filtration choices, so readers can assess robustness of the “data suitable” conclusion rather than take the gate as automatic. revision: yes

  3. Referee: Visualization of local maxima / hotspots is the applied payoff, but the abstract does not indicate error bars, data-exclusion rules, population normalization, or how “local maxima” are identified from vineyard features. For the public-health claim to hold, the manuscript should document preprocessing (e.g., rates vs. raw counts, county vs. finer geography), the filtration used to build the vineyards, and how vines are mapped back to geographic hotspots so that the figures can be independently interpreted.

    Authors: We agree that the applied claim requires self-contained documentation of preprocessing and of the correspondence between vineyard features and geography. The revision will document: (i) geographic unit of analysis and whether we use raw counts or population-normalized rates (and why); (ii) any data-exclusion, suppression, or incomplete-period rules; (iii) the precise filtration (and any density or smoothing step) used to construct the time-parameterized diagrams; (iv) the operational rule by which long-lived or significant vines are identified with local maxima/hotspots and mapped back to the map; and (v) what can and cannot be said about uncertainty (stability of vines under the stated nulls; where formal error bars are not available, we will say so rather than imply them). Figure captions and a dedicated methods subsection will make these choices independently interpretable for spatial epidemiology readers as well as the TDA-applications audience. revision: yes

Circularity Check

0 steps flagged

No significant circularity: empirical TDA application with proposed tests applied to external public-health data; no derivation reduces to its inputs by construction.

full rationale

This paper is an applied topological data analysis study of Ohio overdose death counts. It proposes statistical tests for whether vineyards are suitable for a spatiotemporal dataset, applies those tests to external public-health data, visualizes local hotspots via vineyards, and explores tests for feature significance. There is no claimed first-principles derivation of a numerical law, no fitted parameter renamed as a prediction of a closely related quantity, and no uniqueness theorem or ansatz imported from the authors’ prior work as a load-bearing external fact. Self-citations, if any, are ordinary methodological background rather than the sole support for a uniqueness or forced-choice claim. Suitability and significance tests are proposed and then applied to the same Ohio series—an empirical workflow, not a circular reduction of Eq. X to Eq. Y by definition. Residual risk that null models could be misspecified is a correctness concern, not circularity under the stated criteria. Score 0 is the honest finding for this genre.

Axiom & Free-Parameter Ledger

0 free parameters · 3 axioms · 0 invented entities

Abstract-only review; free parameters (e.g., filtration scales, bandwidths, significance thresholds, time-window choices) and any ad-hoc modeling choices cannot be enumerated from the available text. Standard TDA background (persistence modules, vineyards) and the assumption that overdose death counts form a suitable spatial filtration are domain assumptions. No new physical particles, forces, or dimensions are invented. Ledger is therefore sparse by necessity of incomplete source text.

axioms (3)
  • domain assumption Vineyards (time-parameterized persistence diagrams) are a valid topological summary of evolving spatial density of overdose deaths.
    Invoked throughout the abstract as the core analysis tool; standard in TDA but its appropriateness for count-valued public-health maps is what the paper’s own tests aim to check.
  • domain assumption Ohio drug-overdose death records, after unspecified preprocessing, form a spatiotemporal dataset to which the proposed vineyard pipeline and statistical tests apply.
    Abstract states the tests find the data suitable and then proceeds to vineyard analysis; exact data model and exclusions are not given in the abstract.
  • ad hoc to paper The proposed (unspecified) statistical tests correctly decide suitability of vineyards and significance of vineyard features.
    These tests are the paper’s methodological contribution and load-bearing for the claim that the data are suitable and that visualized features are meaningful; their validity is not independently established in the abstract.

reviewed 2026-07-08 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Visualizing Local Maxima of the Ohio overdose epidemic with Vineyards." pith.science (2026). https://pith.science/paper/B3IKYPFQ

@misc{pith2026260705710,
  author       = {Pith},
  title        = {Pith review of: Visualizing Local Maxima of the Ohio overdose epidemic with Vineyards},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/B3IKYPFQ}},
  note         = {Machine review of arXiv:2607.05710}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Understanding how spatial patterns evolve over time is a complex task that often arises in the analysis of public health data. In this work, we investigate the use of vineyards from topological data analysis (TDA) in this setting by applying them to time series data related to the overdose epidemic in the state of Ohio. We begin by proposing statistical tests that can be used in order to evaluate whether vineyards are a reasonable technique to study a spatiotemporal dataset. We then apply these tests to the data of drug overdose deaths in Ohio and, finding the data suitable, perform a subsequent analysis using vineyards to visualize the evolution of local hotspots in the Ohio overdose epidemic over time. We conclude by exploring statistical tests that can be used to verify the significance of features of our vineyard diagrams.

Figures

Figures reproduced from arXiv: 2607.05710 by David White, Nathan Willey, Nicholas Bermingham.

Figure 1
Figure 1. Figure 1: (Left) Vineyard of the six most persistent vines created using cumulative overdose deaths in the counties of Ohio from [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: (Top) A sequence of three different integer functions on [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The vineyard of H1 Persistence diagrams produced by lin￾early interpolating between the snapshot functions of [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: (Left) A top-down view of a vineyard constructed using [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: (Left) A top-down view of a vineyard constructed us [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: (Above) Vineyard Diagram of the two most persistent vines [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: 95% confidence band (shown in red) for the cumulative [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
Figure 9
Figure 9. Figure 9: (Left) Top-down view and (Right) angled view of the six [PITH_FULL_IMAGE:figures/full_fig_p008_9.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

66 extracted references · 66 canonical work pages

  1. [1]

    Abdallah, A

    H. Abdallah, A. Regalski, M. B. Kang, M. Berishaj, N. Nnadi, A. Chowdury, V . A. Diwadkar, and A. Salch. Statistical inference for persistent homology applied to simulated fmri time series data.Foun- dations of Data Science, 5(1):1–25, 2023. doi: 10.3934/fods.2022014 4

    work page doi:10.3934/fods.2022014 2023

  2. [2]

    Agresti.Categorical Data Analysis

    A. Agresti.Categorical Data Analysis. Wiley, 3 ed., 2013. 5

    work page 2013

  3. [3]

    Anselin.An Introduction to Spatial Data Science with GeoDa

    L. Anselin.An Introduction to Spatial Data Science with GeoDa

    work page

  4. [4]

    C. Arnade. ‘The pill mill of America’: where drugs mean there are no good choices, only less awful ones.The Guardian, 2017. 6

    work page 2017

  5. [5]

    S. V . Ault and J. Lu. Comparison of the spread of novel coronavirus: Topological data analysis of 13 countries.JIS, 6(2), 2022. 3

    work page 2022

  6. [6]

    Bahid, O

    Y . Bahid, O. Kutsenko, N. Rodr ´ıguez, and D. White. The statistical and dynamic modeling of the first part of the 2013-2014 Euromaidan protests in Ukraine: The revolution of dignity and preceding times. Plos one, 19(5):e0301639, 2024. 3

    work page 2013

  7. [7]

    Benjamini and Y

    Y . Benjamini and Y . Hochberg. Controlling the false discovery rate: A practical and powerful approach to multiple testing.Journal of the Royal Statistical Society: Series B (Methodological), 57(1):289–300,

    work page

  8. [8]

    doi: 10.1111/j.2517-6161.1995.tb02031.x 5

    work page doi:10.1111/j.2517-6161.1995.tb02031.x 1995

  9. [9]

    Bermingham, V

    N. Bermingham, V . Robins, and K. Turner. Planar symmetry detection and quantification using the extended persistent homology transform. In2023 Topological Data Analysis and Visualization (TopoInVis), pp. 1–9, 2023. doi: 10.1109/TopoInVis60193.2023.00007 3

    work page doi:10.1109/topoinvis60193.2023.00007 2023

  10. [10]

    Bermingham, D

    N. Bermingham, D. White, and N. Willey. Tracking the Spatiotem- poral Spread of the Ohio Overdose Epidemic with Topological Data Analysis . In2025 IEEE Workshop on Topological Data Analysis and Visualization (TopoInVis), pp. 22–31. IEEE Computer Society, Los Alamitos, CA, USA, Nov. 2025. doi: 10.1109/TopoInVis68599.2025 .00007 1, 3, 6

    work page doi:10.1109/topoinvis68599.2025 2025

  11. [11]

    P. Bubenik. Statistical topological data analysis using persistence landscapes.J. Mach. Learn. Res., 16(1):77–102, Jan. 2015. 3, 4, 8

    work page 2015

  12. [12]

    J. M. Buchanich, L. C. Balmert, K. E. Williams, and D. S. Burke. The effect of incomplete death certificates on estimates of unintentional opioid-related overdose deaths in the united states, 1999-2015.Public Health Reports, 133(4):423–431, 2018. 4

    work page 1999

  13. [13]

    Byers, N

    S. Byers, N. Pritchard, J. Turner, and T. Weighill. Detecting spatial dependence with persistent homology.Nonlinear Theory and Its Ap- plications, IEICE, 14(2):106–125, 2023. doi: 10.1587/nolta.14.106 3, 4, 5

    work page doi:10.1587/nolta.14.106 2023

  14. [14]

    Carlsson and V

    G. Carlsson and V . de Silva. Zigzag persistence.Foundations of Com- putational Mathematics, 10(4):367–405, Aug 2010. doi: 10.1007/ s10208-010-9066-0 3

    work page 2010

  15. [15]

    Chazal, B

    F. Chazal, B. T. Fasy, F. Lecci, A. Rinaldo, A. Singh, and L. Wasser- man. On the bootstrap for persistence diagrams and landscapes.Mod- eling and Analysis of Information Systems, 20(6):111–120, 2013. doi: 10.18255/1818-1015-2013-6-111-120 3

    work page doi:10.18255/1818-1015-2013-6-111-120 2013

  16. [16]

    Chazal, B

    F. Chazal, B. T. Fasy, F. Lecci, A. Rinaldo, and L. Wasserman. Stochastic convergence of persistence landscapes and silhouettes. Journal of Computational Geometry, 6(2):140–161, 2015. doi: 10. 20382/jocg.v6i2a8 3, 4

    work page 2015

  17. [17]

    Frontiers in Artificial Intelligence , volume=

    F. Chazal and B. Michel. An introduction to topological data analysis: Fundamental and practical aspects for data scientists.Frontiers in Ar- tificial Intelligence, 4:667963, 2021. doi: 10.3389/frai.2021.667963 4

    work page doi:10.3389/frai.2021.667963 2021

  18. [18]

    Chen and I

    Y . Chen and I. V oli´c. Topological data analysis model for the spread of the coronavirus.Plos one, 16(8):e0255584, 2021. 3

    work page 2021

  19. [19]

    J. I. Choi, J. Lee, A. B. Yeh, Q. Lan, and H. Kang. Spatial clustering of heroin-related overdose incidents: a case study in Cincinnati, Ohio. BMC public health, 22(1):1253, 2022. 2

    work page 2022

  20. [20]

    Choudhuri, K

    S. Choudhuri, K. Basu, K. Thomas, and A. Sen. Predicting future opioid incidences today, 2019. 2

    work page 2019

  21. [21]

    Ciocanel, R

    M.-V . Ciocanel, R. Juenemann, A. T. Dawes, and S. A. McKinley. Topological data analysis approaches to uncovering the timing of ring structure onset in filamentous networks.Bulletin of Mathematical Bi- ology, 83(3):21, Jan 2021. doi: 10.1007/s11538-020-00847-3 3, 4

    work page doi:10.1007/s11538-020-00847-3 2021

  22. [22]

    Cohen-Steiner, H

    D. Cohen-Steiner, H. Edelsbrunner, and J. Harer. Stability of persis- tence diagrams.Discrete & Computational Geometry, 37(1):103–120,

    work page

  23. [23]

    doi: 10.1007/s00454-006-1276-5 3

    work page doi:10.1007/s00454-006-1276-5

  24. [24]

    Cohen-Steiner, H

    D. Cohen-Steiner, H. Edelsbrunner, and J. Harer. Extending persis- tence using Poincar ´e and Lefschetz duality.Foundations of Compu- tational Mathematics, 9(1):79–103, Feb 2009. doi: 10.1007/s10208 -008-9027-z 3

    work page doi:10.1007/s10208 2009

  25. [25]

    Cohen-Steiner, H

    D. Cohen-Steiner, H. Edelsbrunner, and D. Morozov. Vines and vine- yards by updating persistence in linear time. InProceedings of the Twenty-Second Annual Symposium on Computational Geometry, SCG ’06, p. 119–126. Association for Computing Machinery, New York, NY , USA, 2006. doi: 10.1145/1137856.1137877 3, 6

    work page doi:10.1145/1137856.1137877 2006

  26. [26]

    J. P. Costa, P. ˇSkraba, D. Paolotti, and R. Mexia. A topological data analysis approach to influence-like illness.KDD Healthday Epidamik,

    work page

  27. [27]

    Curtis, J

    A. Curtis, J. Curtis, J. Ajayakumar, and E. Jefferis. Using spatial mixed methods to reveal the geographic nuances of opioid overdose patterns in small and rural towns. InNew Research in Crime Modeling and Mapping Using Geospatial Technologies, pp. 211–230. Springer,

    work page

  28. [28]

    A. Eck, A. Muradi, and M. Simoya. Utilizing supervised machine learning models for opioid hotspot prediction.preprint, 2024. 2

    work page 2024

  29. [29]

    A. M. Ertugrul, Y .-R. Lin, and T. Taskaya-Temizel. CASTNet: Community-attentive spatio-temporal networks for opioid overdose forecasting. InMachine Learning and Knowledge Discovery in Databases, pp. 432–448. Springer, 2020. doi: 10.1007/978-3-030 -46133-1 26 2

    work page doi:10.1007/978-3-030 2020

  30. [30]

    B. T. Fasy, F. Lecci, A. Rinaldo, L. Wasserman, S. Balakrishnan, and A. Singh. Confidence sets for persistence diagrams.The Annals of Statistics, 42(6):2301 – 2339, 2014. doi: 10.1214/14-AOS1252 3, 4, 7

    work page doi:10.1214/14-aos1252 2014

  31. [31]

    Gidea and Y

    M. Gidea and Y . Katz. Topological data analysis of financial time series: Landscapes of crashes.Physica A: Statistical Mechanics and its Applications, 491:820–834, 2018. 2

    work page 2018

  32. [32]

    Hepler, E

    S. Hepler, E. McKnight, A. Bonny, and D. Kline. A latent spatial fac- tor approach for synthesizing opioid-associated deaths and treatment admissions in Ohio counties.Epidemiology, 30(3):365–370, 2019. 2

    work page 2019

  33. [33]

    Hickok, D

    A. Hickok, D. Needell, and M. A. Porter. Analysis of spatial and spatiotemporal anomalies using persistent homology: Case studies with COVID-19 data.SIAM Journal on Mathematics of Data Science, 4(3):1116–1144, 2022. doi: 10.1137/21M1435033 3, 6

    work page doi:10.1137/21m1435033 2022

  34. [34]

    J. M. Hilbe.Negative Binomial Regression. Cambridge University Press, 2 ed., 2011. doi: 10.1017/CBO9780511973420 3, 7

    work page doi:10.1017/cbo9780511973420 2011

  35. [35]

    R. J. Hyndman and G. Athanasopoulos.Forecasting: Principles and Practice. 3 ed., 2021. Online textbook. 8

    work page 2021

  36. [36]

    Ji.A Joint Spatio-Temporal Model of Opioid Associated Deaths and Treatment Admissions in Ohio

    Y . Ji.A Joint Spatio-Temporal Model of Opioid Associated Deaths and Treatment Admissions in Ohio. Wake Forest University, 2019. 2

    work page 2019

  37. [37]

    Kline and S

    D. Kline and S. A. Hepler. Estimating the burden of the opioid epidemic for adults and adolescents in Ohio counties.Biometrics, 77(2):765–775, 2021. 2

    work page 2021

  38. [38]

    Kline, Y

    D. Kline, Y . Ji, and S. Hepler. A multivariate spatio-temporal model of the opioid epidemic in Ohio: a factor model approach.Health Services and Outcomes Research Methodology, 21(1):42–53, 2021. 2

    work page 2021

  39. [39]

    Z. R. Li, E. Xie, F. W. Crawford, J. L. Warren, K. McConnell, J. T. Copple, T. Johnson, and G. S. Gonsalves. Suspected heroin-related overdoses incidents in Cincinnati, Ohio: A spatiotemporal analysis. PLoS medicine, 16(11):e1002956, 2019. 2

    work page 2019

  40. [40]

    X. Liu, J. A. Carter, B. Ray, and G. Mohler. Point process modeling of drug overdoses with heterogeneous and missing data.The Annals of Applied Statistics, 15(1), 2021. doi: 10.1214/20-AOAS1384 2

    work page doi:10.1214/20-aoas1384 2021

  41. [41]

    Lo and B

    D. Lo and B. Park. Modeling the spread of the Zika virus using topo- logical data analysis.PloS one, 13(2):e0192120, 2018. 3

    work page 2018

  42. [42]

    L. Ma, L. Tran, and D. White. A statistical analysis of drug seizures and opioid overdose deaths in Ohio from 2014 to 2018.Journal of Student Research, 10(1), 2021. 2, 5

    work page 2014

  43. [43]

    B. D. Marshall, N. Alexander-Scott, J. L. Yedinak, B. D. Hallowell, W. C. Goedel, B. Allen, R. C. Schell, Y . Li, M. S. Krieger, C. Pratty, et al. Preventing overdose using information and data from the en- vironment (PROVIDENT): protocol for a randomized, population- based, community intervention trial.Addiction, 117(4):1152–1162,

    work page

  44. [44]

    McMaster, L

    R. McMaster, L. Masarweh-Zawahri, K. Coen Flynn, V . S. Deo, and D. J. Flannery. Drug overdose death among residents of urban cen- sus tracts: How granular geographical analyses uncover socioenvi- ronmental correlates in Cuyahoga County, Ohio.Journal of Urban Health, 102:445–458, 2025. doi: 10.1007/s11524-024-00939-8 2

    work page doi:10.1007/s11524-024-00939-8 2025

  45. [45]

    Moon and N

    C. Moon and N. A. Lazar. Hypothesis testing for shapes using vector- ized persistence diagrams.Journal of the Royal Statistical Society Se- ries C: Applied Statistics, 72(3):628–648, 2023. doi: 10.1093/jrsssc/ qlad024 4

    work page doi:10.1093/jrsssc/ 2023

  46. [46]

    A. D. Mullen, P. Rock, D. Harris, S. Slavova, J. Talbert, and V . K. C. Bumgardner. Forecasting opioid overdose incidents for rapid action- able data for opioid response in Kentucky (RADOR-KY).Drug and Alcohol Dependence, 267:111705, 2025. doi: 10.1016/j.drugalcdep. 2024.111705 2

    work page doi:10.1016/j.drugalcdep 2025

  47. [47]

    D. B. Neill and W. Herlands. Machine learning for drug overdose surveillance.Journal of Technology in Human Services, 36(1):8–14,

    work page

  48. [48]

    doi: 10.1080/15228835.2017.1416511 2

    work page doi:10.1080/15228835.2017.1416511 2017

  49. [49]

    Noriega, M

    I. Noriega, M. K. Bhullar, T. P. Gilson, D. J. Flannery, V . Deo, and S. Fulton. A case study for local data surveillance in opioid overdose fatalities in Cuyahoga County, OH 2016–2020.Drug and Alcohol De- pendence Reports, 8:100187, 2023. doi: 10.1016/j.dadr.2023.100187 2

    work page doi:10.1016/j.dadr.2023.100187 2016

  50. [50]

    Mortality data portal.https://data

    Ohio Department of Health. Mortality data portal.https://data. ohio.gov/wps/portal/gov/data/view/mortality, 2024. 4

    work page 2024

  51. [51]

    A. P. Robinson and K. Turner. Hypothesis testing for topological data analysis.Journal of Applied and Computational Topology, 1:241–261,

    work page

  52. [52]

    Rodr ´ıguez and D

    N. Rodr ´ıguez and D. White. An analysis of protesting activity and trauma through mathematical and statistical models.Crime Science, 12(1):17, 2023. 3

    work page 2023

  53. [53]

    Rosenblum, J

    D. Rosenblum, J. Unick, and D. Ciccarone. The rapidly changing US illicit drug market and the potential for an improved early warn- ing system: evidence from Ohio drug crime labs.Drug and alcohol dependence, 208:107779, 2020. 2

    work page 2020

  54. [54]

    Rudkin, D

    S. Rudkin, D. J. Webber, and P. Dlotko. Spatial disparities in infection rates at the dawn of a pandemic: Wealthy young workers mattered. Available at SSRN 4356837, 2023. 3

    work page 2023

  55. [55]

    R. H. Shumway and D. S. Stoffer.Time Series Analysis and Its Appli- cations: With R Examples. Springer, 4 ed., 2017. 8

    work page 2017

  56. [56]

    Soliman, V

    M. Soliman, V . Lyubchich, and Y . R. Gel. Ensemble forecasting of the Zika space-time spread with topological data analysis.Environ- metrics, 31(7):e2629, 2020. 2, 3

    work page 2020

  57. [57]

    Stewart, Y

    K. Stewart, Y . Cao, M. H. Hsu, E. Artigiani, and E. Wish. Geospatial analysis of drug poisoning deaths involving heroin in the USA, 2000– 2014.Journal of Urban Health, 94:572–586, 2017. 2

    work page 2000

  58. [58]

    Sumetsky, C

    N. Sumetsky, C. Mair, K. Wheeler-Martin, M. Cerd ´a, L. A. Waller, W. R. Ponicki, and P. J. Gruenewald. Predicting the future course of opioid overdose mortality: An example from two US states.Epidemi- ology, 32(1):61–69, 2021. doi: 10.1097/EDE.0000000000001264 2

    work page doi:10.1097/ede.0000000000001264 2021

  59. [59]

    Taylor, F

    D. Taylor, F. Klimm, H. A. Harrington, M. Kram ´ar, K. Mischaikow, M. A. Porter, and P. J. Mucha. Topological data analysis of conta- gion maps for examining spreading processes on networks.Nature communications, 6(1):7723, 2015. 3

    work page 2015

  60. [60]

    K. Turner. Representing vineyard modules, 2023. 3

    work page 2023

  61. [61]

    Turner, S

    K. Turner, S. Mukherjee, and D. M. Boyer. Persistent homology trans- form for modeling shapes and surfaces.Information and Inference: A Journal of the IMA, 3(4):310–344, 12 2014. doi: 10.1093/imaiai/ iau011 3

    work page doi:10.1093/imaiai/ 2014

  62. [62]

    Turner, V

    K. Turner, V . Robins, and J. Morgan. The extended persistent homol- ogy transform of manifolds with boundary.Journal of Applied and Computational Topology, 8(7):2111–2154, Nov 2024. doi: 10.1007/ s41468-024-00175-8 3

    work page 2024

  63. [63]

    Census Bureau

    U.S. Census Bureau. County population totals: 2010–

    work page 2010

  64. [64]

    population and housing unit estimates.https: //www.census.gov/data/datasets/time-series/demo/ popest/2010s-counties-total.html, 2020. 4

    work page 2020

  65. [65]

    White, L

    D. White, L. Ma, and L. Tran. Unintentional opioid overdose deaths in Ohio: Insights from SUDORS data.preprint, 2025. 3

    work page 2025

  66. [66]

    J. Yang, H. Fang, J. Dhesi, I. H. R. Yoon, J. A. Bull, H. M. Byrne, H. A. Harrington, and G. Grindstaff. Topological classification of tumour-immune interactions and dynamics.Journal of Mathematical Biology, 91(3):25, 2025. doi: 10.1007/s00285-025-02253-6 2

    work page doi:10.1007/s00285-025-02253-6 2025

This paper was first reviewed by grok-4.5 on July 8, 2026.