Detecting early‐warning signals of influenza outbreak based on dynamic network marker

Pei Chen, Ely Chen, Luonan Chen, Xianghong Jasmine Zhou, Rui LiuJournal of Cellular and Molecular Medicine 2018 · DOI

TL;DR

This paper introduces a novel "dynamic network marker" (DNM) method to predict seasonal influenza outbreaks by analyzing hospitalization data. The DNM method successfully identifies early-warning signals, providing a lead time of approximately 4 weeks before an outbreak, enabling proactive public health interventions.

What Problem Does This Paper Address?

Seasonal influenza outbreaks pose a significant global health and economic burden, causing widespread illness, death, and substantial economic losses due to healthcare costs and lost productivity. While timely interventions like vaccination and isolation can mitigate these impacts, accurately predicting the onset of these outbreaks is challenging. This difficulty stems from the complex interplay between biological and social systems that drive influenza transmission. Existing surveillance methods often rely on costly infrastructure or may not provide sufficient early warning. The paper aims to address this gap by developing a computational method that can predict outbreaks using readily available hospitalization data, simplifying information gathering and enabling proactive responses.

Methods

The core of this research is the Dynamic Network Marker (DNM) method. This approach treats influenza outbreaks as a "tipping point" or critical transition in a complex system. The DNM method is designed to identify this pre-transition state, characterized by low resilience, before the catastrophic shift into an outbreak.

The DNM method relies on three key statistical conditions observed in the data as a system approaches a critical point:

  1. Increased Correlation within a Group: The correlation between members of a specific group (potential DNM members) rapidly increases.
  2. Decreased Correlation with External Groups: The correlation between members of this group and external members (non-DNM) rapidly decreases.
  3. Increased Variability: The standard deviation or coefficient of variation for members within the DNM group drastically increases.

To apply this to influenza, the researchers developed a "landscape DNM score." This involves:

  • Network Construction: Mapping clinic hospitalization data to a network where each node represents a geographical district (e.g., wards in Tokyo).
  • Local Network Analysis: Analyzing subnetworks centered around each node, considering its immediate neighbors.
  • Calculating Local DNM Score (It): This score quantifies the criticality of a node's local network based on changes in standard deviation within the local network, correlation changes within the local network, and correlation changes between the local network and the rest of the network.
  • Landscape Visualization: Visualizing the It scores across all nodes over time to observe the emergence of critical states.

The method is "model-free," meaning it doesn't require pre-training or feature selection, relying solely on the observed statistical properties of the data.

Datasets

The study utilized historical clinic hospitalization data for influenza infections from:

  • Tokyo, Japan: Data from 23 wards, spanning January 1, 2009, to December 31, 2016. This dataset involved approximately 503 clinics and 18 million people.
  • Hokkaido, Japan: Data from 30 districts, also covering the period from 2009 to 2016.

This data was sourced from public records provided by the Tokyo Metropolitan Infectious Disease Surveillance Center and the Hokkaido Infectious Disease Surveillance Center.

Key Insights and Ideas

  • Predictive Power of Network Dynamics: The study demonstrates that analyzing the dynamic interactions within a network of geographical locations (represented by clinics/districts) can reveal critical transitions preceding large-scale events like influenza outbreaks.
  • Pre-Outbreak State as a Tipping Point: The DNM method effectively identifies a distinct "pre-outbreak state" that acts as a tipping point, offering a window for intervention before the full impact of an outbreak is felt.
  • Early Warning Lead Time: The DNM method consistently provided early-warning signals approximately 3-9 weeks ahead of the peak hospitalization counts, with an average lead time of 4 weeks. This is a crucial advantage for public health preparedness.
  • Model-Free Approach Advantage: The DNM method's independence from training data and feature selection makes it robust and less prone to overfitting, especially when dealing with limited or evolving epidemic data.
  • Visualizing Transmission Dynamics: The "landscape DNM" visualization provides insights into how influenza transmission dynamics evolve across a network of districts, offering a spatial and temporal understanding of the epidemic's spread.

Key Results

  • The DNM method successfully predicted seasonal influenza outbreaks in Tokyo between 2009 and 2016, with an average lead time of 4 weeks.
  • The method also showed promise in predicting outbreaks in the Hokkaido region, successfully identifying early-warning signals for six out of seven seasonal flu pandemics analyzed.
  • A comparison with a logistic regression-based surveillance system showed that the DNM method performed better, achieving a higher Area Under the Curve (AUC) of 89.80% compared to 83.85% for logistic regression, using only hospitalization records.
  • The study highlighted the emergence of significant changes in local DNM scores and inter-ward correlations as the system approached the outbreak point, serving as clear indicators.

Limitations and Future Work

  • Data Dependency: While the DNM method is model-free, its performance is still dependent on the quality and availability of hospitalization data.
  • Generalizability: The study focused on influenza in Japan. Further validation across different geographical regions and for other infectious diseases would be beneficial.
  • Integration with Other Data: The authors suggest that incorporating more clinic information and new discriminant input patterns could further enhance the sensitivity and accuracy of the DNM-based monitoring system.
  • Real-time Application: While the method provides early warnings, the practical implementation of real-time monitoring and intervention strategies based on these signals would require further development and integration into public health infrastructure. The authors mention a website for data gathering and processing, indicating a step towards practical application.

Summary updated

Abstract

The seasonal outbreaks of influenza infection cause globally respiratory illness, or even death in all age groups. Given early-warning signals preceding the influenza outbreak, timely intervention such as vaccination and isolation management effectively decrease the morbidity. However, it is usually a difficult task to achieve the real-time prediction of influenza outbreak due to its complexity intertwining both biological systems and social systems. By exploring rich dynamical and high-dimensional information, our dynamic network marker/biomarker (DNM/DNB) method opens a new way to identify the tipping point prior to the catastrophic transition into an influenza pandemics. In order to detect the early-warning signals before the influenza outbreak by applying DNM method, the historical information of clinic hospitalization caused by influenza infection between years 2009 and 2016 were extracted and assembled from public records of Tokyo and Hokkaido, Japan. The early-warning signal, with an average of 4-week window lead prior to each seasonal outbreak of influenza, was provided by DNM-based on the hospitalization records, providing an opportunity to apply proactive strategies to prevent or delay the onset of influenza outbreak. Moreover, the study on the dynamical changes of hospitalization in local district networks unveils the influenza transmission dynamics or landscape in network level.
BibTeX
@article{chen2018detecting,
  author = {Pei Chen and Ely Chen and Luonan Chen and Xianghong Jasmine Zhou and Rui Liu},
  title = {{Detecting early‐warning signals of influenza outbreak based on dynamic network marker}},
  journal = {Journal of Cellular and Molecular Medicine},
  year = {2018},
  doi = {10.1111/jcmm.13943},
}
Figure