Network memory in the movement of hospital patients carrying antimicrobial-resistant bacteria

Ashleigh C. Myall, Robert L. Peach, Andrea Y. Weiße, Siddharth Mookerjee, Frances Davies, Alison H. Holmes, Mauricio BarahonaAppl. Netw. Sci 2021 · DOI · OpenReview

Introduction

Standard epidemiological models often treat patient movement within hospitals as a "memory-less" process. In these first-order Markov models, a patient's next destination is assumed to depend only on their current location. However, clinical reality is governed by structured pathways—from admission to surgery to recovery—meaning a patient's history significantly dictates their future movement. This study demonstrates that ignoring this "network memory" leads to inaccurate predictions of pathogen spread, particularly for Carbapenemase-producing Enterobacteriaceae (CPE).

Modeling Patient Movement with Memory

To capture the nuances of hospital journeys, the researchers utilized Higher-Order Memory Networks. Unlike traditional physical networks where nodes represent single wards, memory networks use "state nodes" to represent sequences of locations (e.g., Ward A $\rightarrow$ Ward B). This structure allows the model to "remember" where a patient has been.

Illustration of transitive dependence encoded into memory network. (a) Patient pathways, (b) First-order representation, (c) Memory network representation.

The researchers introduced several key methodological innovations:

  • Entropy Rate Analysis: By measuring the entropy rate of random walks, they quantified how constrained patient movement is. A lower entropy rate in higher-order models indicates that memory significantly restricts possible next steps compared to a random, memory-less walk.
  • Lumped State Networks: To manage the "curse of dimensionality" inherent in higher-order models, a novel "lumping" technique was used to merge redundant state nodes with similar connectivity, preserving accuracy while reducing computational complexity.
  • Dynamical Tools: Markov Stability (MS) and Multiscale Centrality (MSC) were applied to identify functional communities of wards and rank influential hubs across various timescales.

Key Insights and Results

The study analyzed anonymized records from a 1000-bed London teaching hospital Trust, focusing on 526 patients who tested positive for CPE. The results highlight the stark differences between traditional and memory-aware models:

The "Small World" Illusion

First-order models suggest a "small world" effect where a patient could reach nearly any ward (71 out of 79) in just three steps. In contrast, the memory-aware model ($k=2$) reveals a much more restricted reality, with only 9 wards reachable in the same timeframe. This suggests that pathogens do not spread as "leakily" as traditional models predict.

Identifying Hidden Centrality

In memory-less models, acute admission units appear as the primary hubs simply because of high throughput. However, memory networks reveal that Renal wards are the true persistent hotspots for CPE. These patients frequently cycle back to specialized care, creating a "memory" in the network that traditional models overlook.

Predictive Power and Constraint

Moving to a second-order model ($k=2$) significantly improved predictive power, increasing the Kendall-Tau rank correlation from 0.49 to 0.60. Furthermore, the transition caused a massive drop in entropy (from 2.70 to 0.57), proving that patient trajectories are highly non-random and governed by strong transitive dependencies.

Conclusion and Future Directions

Accounting for network memory provides a more realistic map of how infections like CPE move through a hospital. While the study was limited to $k=2$ or $k=3$ due to data sparsity, it clearly demonstrates that memory-aware models can better inform Infection Prevention and Control (IPC) resource allocation. Future work may incorporate "temporal memory"—the duration of stay—to further refine these transmission models.

Reference: Myall, A.C., Peach, R.L., Weiße, A.Y. et al. Network memory in the movement of hospital patients carrying antimicrobial-resistant bacteria. Appl Netw Sci 6, 34 (2021).

Summary updated

Abstract

Hospitals constitute highly interconnected systems that bring into contact an abundance of infectious pathogens and susceptible individuals, thus making infection outbreaks both common and challenging. In recent years, there has been a sharp incidence of antimicrobial-resistance amongst healthcare-associated infections, a situation now considered endemic in many countries. Here we present network-based analyses of a data set capturing the movement of patients harbouring antibiotic-resistant bacteria across three large London hospitals. We show that there are substantial memory effects in the movement of hospital patients colonised with antibiotic-resistant bacteria. Such memory effects break first-order Markovian transitive assumptions and substantially alter the conclusions from the analysis, specifically on node rankings and the evolution of diffusive processes. We capture variable length memory effects by constructing a lumped-state memory network, which we then use to identify individually import wards and overlapping communities of wards. We find these wards align closely to known hotspots of transmission and commonly followed pathways patients. Our framework provides a means to focus infection control efforts and cohort outbreaks of healthcare-associated infections.
BibTeX
@article{myall2021network,
  author = {Ashleigh C. Myall and Robert L. Peach and Andrea Y. Weiße and Siddharth Mookerjee and Frances Davies and Alison H. Holmes and Mauricio Barahona},
  title = {{Network memory in the movement of hospital patients carrying antimicrobial-resistant bacteria}},
  journal = {Appl. Netw. Sci.},
  year = {2021},
  doi = {10.1007/s41109-021-00376-5},
}
Figure