Reducing size bias in epidemic network modelling.

Area of Science:

• Epidemiology
• Network Science
• Computational Biology

Background:

• Epidemiological models inform disease control policy.
• These models often use sampled contact networks.
• Common Random Walk (RW) sampling creates size-biased, over-representative samples of highly connected individuals, skewing disease spread estimates.

Purpose of the Study:

• Compare Metropolis-Hastings Random Walk (MHRW) and RW sampling algorithms.
• Evaluate their effectiveness in reducing size bias in network sampling.
• Assess their impact on disease spread simulation accuracy across different network structures.

Main Methods:

• Simulated disease spread using a stochastic Susceptible-Infected-Recovered (SIR) framework.
• Compared RW and MHRW sampling algorithms on Erdös-Rényi (ER), Small-world (SW), Negative-binomial (NB), and Scale-free (SF) networks.
• Analyzed real-world cattle movement and human contact network data.

Main Results:

• RW overestimated infections and secondary infections, and underestimated time-to-infection in NB networks.
• MHRW significantly reduced size bias across ER, SW, and NB networks.
• Both algorithms yielded non-representative and variable estimates on SF networks.
• MHRW provided disease spread estimates closer to the underlying network for real-world data.

Conclusions:

• MHRW sampling is more suitable than RW for slower, low-severity epidemics and heterogeneous networks (NB).
• RW is appropriate for fast-spreading epidemics in homogeneous networks (ER, SW).
• Algorithm choice depends on network structure and epidemic characteristics for reliable disease modeling and policymaking.