Jiaqi (Jackie) Chen

Jiaqi (Jackie) Chen

PhD student at the University of Hong Kong

Predicting the Time-Varying Population Immunity and Importation Risk and Their Impact on Co-Circulation of Influenza and COVID-19 in Hong Kong

Published in Nature Human Behaviour, 2026

📝 Under Submission   First Author   Target: Nature Human Behaviour

Authors

Jiaqi Chen, Hao Wang, Dong Wang, Yiu Chung Lau, M. Pear Hossain, Sukhyun Ryu, Peng Wu, Lulla Opatowski, Benjamin J. Cowling, Sheikh Taslim Ali

Status

Manuscript currently under submission. Full preprint and publication links will be added after peer review.

Three-Model Bayesian Benchmarking Framework

The central methodological contribution of this work is a side-by-side benchmarking of three Bayesian transmission model families, each representing a distinct philosophy of how influenza dynamics should be formalised:

① Fully Mechanistic
Compartmental SIR-type dynamics with explicit biological parameters (contact, waning, antigenic drift). Strong priors, full interpretability, but sensitive to model misspecification.
② Semi-Mechanistic
Hybrid framework retaining a mechanistic transmission core while letting flexible latent processes absorb unmodelled heterogeneity — consistently recovers subtype-resolved immunity distributions where the other two degrade.
③ Purely Statistical
Data-driven Bayesian regression with minimal structural assumptions. Flexible and low-bias under dense data, but weak extrapolation beyond the observed regime.

Benchmarking all three against identical syndromic, genomic, serological and air-travel data reveals that only the semi-mechanistic framework consistently recovers subtype-resolved immunity distributions (Kolmogorov–Smirnov distance 0.133–0.242), whereas the fully mechanistic and purely statistical alternatives each degrade two-fold for at least one subtype — establishing hybrid inference as methodologically essential for subtype-resolved influenza epidemiology.

Abstract

Whether the COVID-19 pandemic only transiently interrupted seasonal influenza, or durably reshaped its transmission ecology, remains an unresolved question rendered urgent by anomalously severe post-pandemic influenza seasons. We reconstruct subtype-specific population immunity against influenza in Hong Kong across fifteen years (2010–2025) by benchmarking three Bayesian transmission models — fully mechanistic, semi-mechanistic, and purely statistical — against syndromic surveillance, genomic, serological and international air-travel data from twelve countries.

Inferred immunity follows a subtype-specific sawtooth pattern in which each epidemic initiates only after immunity returns to a seasonally reproducible, lineage-stable floor, predicting epidemic magnitude, speed and peak timing — and explaining up to 72% of within-subtype variance. Joint modelling of influenza–COVID-19 co-circulation (2019–2025) reveals durable post-pandemic shifts, including a twenty-fold pandemic collapse of the COVID-19–influenza interaction parameter and a four-fold attenuation of seasonal travel–influenza coupling despite near-complete mobility recovery.

Key Contributions

  • Built a three-model Bayesian benchmarking framework (fully mechanistic + semi-mechanistic + purely statistical) to reconstruct subtype-resolved influenza immunity over 15 years
  • Established the semi-mechanistic model as methodologically essential — the only framework that consistently recovers the distributional profiles of immunity across all subtypes
  • Integrated multi-source surveillance: syndromic, genomic, serological, and international air-travel data from 12 countries
  • Uncovered a subtype-specific sawtooth immunity pattern that predicts epidemic magnitude, speed and peak timing (up to 72% within-subtype variance explained)
  • Identified a Simpson’s paradox in pooled influenza data, challenging the treatment of influenza as a single pathogen
  • Quantified subtype- and outcome-specific antigenic drift effects across A(H1N1)pdm09, A(H3N2) and influenza B
  • Demonstrated a post-pandemic regime change in influenza transmission: 20× collapse in the COVID-19–influenza interaction parameter, asymmetric restructuring of mainland-China importation dynamics, and 4× attenuation of travel–influenza seasonal coupling

Technologies

Python   Stan / PyMC (Bayesian Inference)   Fully Mechanistic ODE   Semi-Mechanistic Hybrid Inference   Statistical Bayesian Regression   Genomic Surveillance Integration   Multi-Source Data Fusion   Kolmogorov–Smirnov Diagnostics

Recommended citation: Chen, J., Wang, H., Wang, D., Lau, Y. C., Hossain, M. P., Ryu, S., Wu, P., Opatowski, L., Cowling, B. J., & Ali, S. T. (2026). "Predicting the time-varying population immunity and importation risk and their impact on co-circulation of influenza and COVID-19 in Hong Kong." Manuscript under submission.