A geospatial approach to identify patterns of antibiotic susceptibility at a neighborhood level in Wisconsin, United States

Laurel Legenza, Kyle McNair, Song Gao, James P. Lacy, Brooke J. Olson, Thomas R. Fritsche, Lucas Schulz, Samantha LaMuro, Frances Spray-Larson, Tahmeena Siddiqui, Warren E. RoseScientific 2023 · DOI

Abstract

The global threat of antimicrobial resistance (AMR) varies regionally. This study explores whether geospatial analysis and data visualization methods detect both clinically and statistically significant variations in antibiotic susceptibility rates at a neighborhood level. This observational multicenter geospatial study collected 10 years of patient-level antibiotic susceptibility data and patient addresses from three regionally distinct Wisconsin health systems (UW Health, Fort HealthCare, Marshfield Clinic Health System [MCHS]). We included the initial Escherichia coli isolate per patient per year per sample source with a patient address in Wisconsin (N = 100,176). Isolates from U.S. Census Block Groups with less than 30 isolates were excluded (n = 13,709), resulting in 86,467 E. coli isolates. The primary study outcomes were the results of Moran's I spatial autocorrelation analyses to quantify antibiotic susceptibility as spatially dispersed, randomly distributed, or clustered by a range of - 1 to + 1, and the detection of statistically significant local hot (high susceptibility) and cold spots (low susceptibility) for variations in antibiotic susceptibility by U.S. Census Block Group. UW Health isolates collected represented greater isolate geographic density (n = 36,279 E. coli, 389 = blocks, 2009-2018), compared to Fort HealthCare (n = 5110 isolates, 48 = blocks, 2012-2018) and MCHS (45,078 isolates, 480 blocks, 2009-2018). Choropleth maps enabled a spatial AMR data visualization. A positive spatially-clustered pattern was identified from the UW Health data for ciprofloxacin (Moran's I = 0.096, p = 0.005) and trimethoprim/sulfamethoxazole susceptibility (Moran's I = 0.180, p < 0.001). Fort HealthCare and MCHS distributions were likely random. At the local level, we identified hot and cold spots at all three health systems (90%, 95%, and 99% CIs). AMR spatial clustering was observed in urban areas but not rural areas. Unique identification of AMR hot spots at the Block Group level provides a foundation for future analyses and hypotheses. Clinically meaningful differences in AMR could inform clinical decision support tools and warrants further investigation for informing therapy options.

Introduction

When clinicians encounter a patient with a suspected infection, they often initiate "empiric therapy"—prescribing antibiotics before laboratory results confirm the specific pathogen and its drug sensitivity. To guide these decisions, clinicians rely on antibiograms, which summarize local resistance rates. However, traditional antibiograms are typically aggregated at the level of an entire hospital or health system. This broad approach fails to account for the fact that antimicrobial resistance (AMR) is influenced by hyper-local factors, including socioeconomic conditions, land use, and community prescribing habits. This study addresses the need for granular, "spatially-enabled" surveillance data to better inform clinical practice.

Methods and Datasets

This multicenter observational study utilized a longitudinal dataset spanning 2009–2018 from three distinct Wisconsin health systems: UW Health (urban/suburban Madison), Marshfield Clinic Health System (MCHS; rural north-central Wisconsin), and Fort HealthCare (metropolitan Fort Atkinson).

Geographic distribution of the three participating health systems across Wisconsin.

Researchers mapped 86,467 Escherichia coli isolates to specific U.S. Census Block Groups—neighborhood units typically containing 600 to 3,000 people. To ensure statistical reliability, only Block Groups with at least 30 isolates were included in the final analysis. Two primary statistical tools were used:

  1. Global Moran’s I: To determine if antibiotic susceptibility was clustered, dispersed, or randomly distributed.
  2. Getis-Ord Gi*: To identify specific "hot spots" (areas of significantly higher susceptibility) and "cold spots" (areas of significantly lower susceptibility) relative to the regional average.

Key Results and Insights

The analysis revealed that antibiotic resistance is not a monolith; rather, it varies significantly between neighboring communities. In the UW Health system, susceptibility was significantly clustered (Moran’s I = 0.180 for TMP/SMX, p < 0.001), suggesting that a patient's neighborhood is a strong predictor of their resistance profile.

Hot spot and cold spot analysis for Ciprofloxacin and Sulfamethoxazole/Trimethoprim susceptibility.

Key findings include:

  • Urban vs. Rural Dynamics: Spatial clustering was predominantly observed in urban areas like Madison. In contrast, distributions in the rural MCHS and Fort HealthCare systems were largely random, indicating different environmental or social drivers of resistance in rural settings.
  • Susceptibility Variances: While mean susceptibility for Ciprofloxacin across the systems ranged from 84% to 90%, neighborhood-level data showed dramatic swings from 62% to 100%.
  • Statistical vs. Clinical Significance: The study distinguished between statistical "cold spots" (lower susceptibility than neighbors) and clinical thresholds. For example, a neighborhood might have dangerously low susceptibility (<80%) but not be a statistical cold spot if the entire surrounding region shares that low susceptibility.

Discussion and Future Directions

While this geospatial approach offers a powerful new lens for AMR surveillance, several challenges remain. Data gaps exist in regions where health systems did not participate, and the "30-isolate rule" often excludes rural areas with low population density, potentially masking resistance patterns in those communities. Furthermore, while the study identifies where resistance clusters occur, it does not yet explain why.

Future research will aim to link these maps to local antibiotic prescribing volumes, agricultural land use, and socioeconomic indices. Ultimately, the goal is to integrate these granular maps into Clinical Decision Support tools, providing clinicians with neighborhood-specific risk profiles to optimize antibiotic selection at the point of care.

Summary updated

Abstract

The global threat of antimicrobial resistance (AMR) varies regionally. This study explores whether geospatial analysis and data visualization methods detect both clinically and statistically significant variations in antibiotic susceptibility rates at a neighborhood level. This observational multicenter geospatial study collected 10 years of patient-level antibiotic susceptibility data and patient addresses from three regionally distinct Wisconsin health systems (UW Health, Fort HealthCare, Marshfield Clinic Health System [MCHS]). We included the initial Escherichia coli isolate per patient per year per sample source with a patient address in Wisconsin (N = 100,176). Isolates from U.S. Census Block Groups with less than 30 isolates were excluded (n = 13,709), resulting in 86,467 E. coli isolates. The primary study outcomes were the results of Moran's I spatial autocorrelation analyses to quantify antibiotic susceptibility as spatially dispersed, randomly distributed, or clustered by a range of - 1 to + 1, and the detection of statistically significant local hot (high susceptibility) and cold spots (low susceptibility) for variations in antibiotic susceptibility by U.S. Census Block Group. UW Health isolates collected represented greater isolate geographic density (n = 36,279 E. coli, 389 = blocks, 2009-2018), compared to Fort HealthCare (n = 5110 isolates, 48 = blocks, 2012-2018) and MCHS (45,078 isolates, 480 blocks, 2009-2018). Choropleth maps enabled a spatial AMR data visualization. A positive spatially-clustered pattern was identified from the UW Health data for ciprofloxacin (Moran's I = 0.096, p = 0.005) and trimethoprim/sulfamethoxazole susceptibility (Moran's I = 0.180, p < 0.001). Fort HealthCare and MCHS distributions were likely random. At the local level, we identified hot and cold spots at all three health systems (90%, 95%, and 99% CIs). AMR spatial clustering was observed in urban areas but not rural areas. Unique identification of AMR hot spots at the Block Group level provides a foundation for future analyses and hypotheses. Clinically meaningful differences in AMR could inform clinical decision support tools and warrants further investigation for informing therapy options.
BibTeX
@inproceedings{legenza2023geospatial,
  author = {Laurel Legenza and Kyle McNair and Song Gao and James P. Lacy and Brooke J. Olson and Thomas R. Fritsche and Lucas Schulz and Samantha LaMuro and Frances Spray-Larson and Tahmeena Siddiqui and Warren E. Rose},
  title = {{A geospatial approach to identify patterns of antibiotic susceptibility at a neighborhood level in Wisconsin, United States}},
  booktitle = {Scientific},
  year = {2023},
  doi = {10.1038/s41598-023-33895-5},
}
Figure