Tracking the Spatiotemporal Spread of the Ohio Overdose Epidemic with Topological Data Analysis

Nicholas Bermingham The Ohio State University, David White Denison University, Nathan Willey The Ohio State University · 2025

Read Tracking the Spatiotemporal Spread of the Ohio Overdose Epidemic with Topological Data Analysis by Nicholas Bermingham The Ohio State University, David White Denison University, Nathan Willey The Ohio State University (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-22705.txt.

Introduction

Nicholas Bermingham, David White, and Nathan Willey track Ohio's overdose epidemic with a topological lens. Topological data analysis captures spatial structure in complex data, but its use on data that moves through time is still rare. The team applies the Mapper algorithm to overdose data from 2007 to 2024. Mapper builds a graph that summarizes a data set's shape through overlapping cover sets.

Their new cover method respects geography and highlights time-dependent variables. A Mapper view of regional demographics shows how those factors relate to overdose deaths. The approach reveals temporal trends, hotspots, and lagged patterns tied to both place and community traits. It extends earlier topological studies of disease spread.

Catalog record

CREATORNicholas Bermingham The Ohio State University, David White Denison University, Nathan Willey The Ohio State University
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE54 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-22705

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Tracking the Spatiotemporal Spread of the Ohio Overdose Epidemic with Topological Data Analysis Nicholas Bermingham The Ohio State University, David White Denison University, Nathan Willey The Ohio State University Abstract In recent years, techniques from Topological Data Analysis (TDA) have proven effective at capturing spatial features of multidimensional data. However, applying TDA to spatiotemporal data remains relatively underexplored. In this work, we extend previous studies of disease spread by using the Mapper algorithm to analyze the Ohio drug overdose epidemic from 2007 to 2024. We introduce a novel method for constructing covers in Mapper graphs of spatiotemporal data that respects geographic structure and highlights the time-dependent variables. Finally, we generate a Mapper visualization of regional demographics to examine how these factors relate to overdose deaths. Our approach effectively reveals temporal trends, overdose hotspots, and time‑lagged patterns in relation to both geography and community demographics. \onlineid 0 \vgtccategory Research \vgtcinsertpkg \teaser Mapper graph created from county-level statistics of cumulative drug overdose deaths next to a map of Ohio with counties that form flares in the Mapper graph highlighted. The geometry of the Mapper graph corresponds to the geography of Ohio and flares in the graph highlight death counts that are substantially different from those of the surrounding counties. Introduction keywords Mapper, Topological Data Analysis, time-series data, spatiotemporal data, demographic data, drug overdose In this paper we investigate the effectiveness of applying the topological data analysis (TDA) tool known as Mapper to data relating to the spatiotemporal aspects of the Ohio overdose epidemic from January 2007 to September 2024 . We examine what insights into the Ohio overdose epidemic can be drawn from the topological and geometric features of Mapper graphs, such as connected components, flares and holes. Drug overdose remains a leading cause of preventable death in the United States. Between 1999 and 2022, annual overdose deaths rose more than sixfold, and in 2022 alone, over 107,000 Americans died from drug overdoses. The impact is not uniform: overdose mortality varies widely across time, geography, and demographic groups. Some of the hardest-hit areas, including counties in Ohio, have experienced multiple waves of the epidemic, each marked by distinct spatial and temporal patterns. As synthetic opioids make the drug supply increasingly volatile, it is important to understand where and when risk is increasing, and how it’s spreading from one county to another. Modeling the spatiotemporal spread of overdose deaths can help public health agencies identify emerging hotspots early, allocate limited resources more effectively, and implement tailored interventions before crises peak. In a policy environment of shrinking federal support, tools that can capture the evolving geography of the epidemic are critical to preventing future deaths. The spread of overdose deaths does not follow simple trajectories: it is shaped by changes in drug supply, local economic distress, demographic shifts, and public health infrastructure. Traditional approaches to modeling these patterns—including spatial regression and generalized linear mixed models—require strong assumptions about linearity, locality, and parametric form. High-dimensional models with large numbers of county-specific coefficients are often difficult to interpret and may obscure broader structural features. This landscape calls for flexible, non-parametric tools—like those from TDA—that can reveal large-scale structure in complex, high-dimensional data without relying on restrictive modeling assumptions. The Mapper algorithm is a tool, introduced in [ 30 ] and widely used in TDA, to construct graphs that encapsulate the broad topological features of a given dataset. Mapper constructs a simplified representation of a high-dimensional dataset by clustering nearby observations and connecting them into a graph that reflects both global and local structure. This allows one to more easily visualize data and draw both qualitative and quantitative insights. We apply the Mapper algorithm to county-level geographic, demographic, and drug overdose death counts in Ohio across a seventeen-year period with the goal of understanding the spatiotemporal spread of overdose deaths. Our methodology is informed by [ 6 ] , a similar analysis for Covid-19 spread using Mapper. To the best of our knowledge, drug overdose data have not yet been analyzed with TDA and so one of our goals is to determine its utility for both this specific application and more general spatiotemporal analyses. Additionally, we extend the experiments of [ 6 ] in a few ways. First, we explore whether there is more insight to gain from utilizing population-normalized death counts to better represent data from rural communities. Second, we use a novel method for selecting Mapper hyperparameters to best capture the evolution of spatiotemporal data. Third, going beyond spatiotemporal analysis, we investigate whether similar insights can be gained from building Mapper graphs from county-level demographic data and overdose data to create a “demographic landscape” of Ohio. We use visual properties of the resulting graphs in each setting to identify patterns of spread, population-normalized burden, and demographic correlates. Our findings confirm and extend known patterns in Ohio’s overdose epidemic. We identify the emergence of rural hotspots near the Ohio River and the increasing burden in counties that experienced high unemployment. We also observe a time-dependent trajectory in which overdose mortality spreads outward from urban centers into surrounding regions. These results align with prior work using conventional modeling frameworks (see \hyperref [subsec:related-works]Section 1.2). Notably, we find evidence of a lagged association between economic stressors and subsequent overdose mortality, as well as visual confirmation of unique regional trajectories (e.g., in Scioto County) consistent with known historical events and public health reports. By producing intuitive and interactive visualizations that highlight where and how the overdose crisis has unfolded, our approach has implications for future public health planning. Spatiotemporal analytics in Rhode Island helped motivate the state’s implementation of geographically targeted interventions, with overdose prevention resources deployed at the neighborhood level [ 22 ] . In that context, visual evidence of spatial spread prompted decision-makers to shift from reactive policies to proactive containment. We hope the ideas in this work can serve a similar function in Ohio and other states, both by providing a new lens on the overdose crisis and by demonstrating the potential of TDA methods in public health research. 0.1 Outline of the Paper In \hyperref [section:methods]Section 2, we first outline the Mapper algorithm and the features that allow us to customize it for specific analysis of spatiotemporal data. Next we discuss how we have sourced and organized our data on spatiotemporal spread of overdose deaths and county-level demographics in Ohio. We then detail the specific way we have implemented the Mapper algorithm to analyze these datasets, discuss how we have implemented a novel choice of hyperparameters to create more intuitive and interpretable Mapper graphs and note how our implementation differs from the work done in [ 6 ] . In \hyperref [section:results]Section 3, we display our Mapper graphs, analyze the factors contributing to the emergence of topological and geometric features in the graphs and identify the counties whose data contributes to the emergence of these features. Finally, in \hyperref [section:conclusions]Section 4 we draw conclusions regarding what our investigations tell us about the Ohio overdose epidemic, how our different methodologies could be used to study similar temporal datasets, and future directions. 0.2 Related Works This paper represents the first use of topological data analysis to visualize drug overdose data at the time of writing. Previous works have

Read the complete text