Towards Inferring Network Properties from Epidemic Data.

Area of Science:

• Epidemiology
• Network Science
• Statistical Modeling

Background:

• Traditional epidemic models struggle with high-dimensional network data.
• Mean-field models like the pairwise model (PWM) simplify analysis but have limited use in statistical inference.
• Inferring both disease and network parameters from epidemic data is challenging.

Purpose of the Study:

• To assess the effectiveness of the pairwise model (PWM) coupled with susceptible-infected-recovered (SIR) dynamics for inferring disease and network parameters.
• To compare statistical inference methods using population-level versus individual-level epidemic data.
• To evaluate the robustness of inference methods against model mismatch in real-world scenarios.

Main Methods:

• Utilized the pairwise model (PWM) with susceptible-infected-recovered (SIR) epidemic dynamics.
• Employed maximum likelihood estimation (MLE) for population-level data (e.g., daily new cases).
• Applied dynamical survival analysis (DSA) for individual-level data (e.g., recovery times).

Main Results:

• Both MLE and DSA performed well with simulated data (no model mismatch).
• DSA demonstrated robustness to model mismatch with real-world data (e.g., Foot-and-Mouth, H1N1, COVID-19), yielding plausible epidemiological parameters.
• MLE struggled with real-world data, showing issues with parameter unidentifiability and sensitivity to population size and under-reporting.

Conclusions:

• Dynamical survival analysis (DSA) is a more robust method for inferring parameters from individual-level epidemic data on networks, especially with real-world data.
• Network-based mean-field models can be adapted for approximate likelihoods, enabling inference of both disease dynamics and network structure.
• Future research should focus on efficient inference schemes for network-based epidemic models.