Here is the link to the paper:

doi.org/10.1073/pnas...

Takeaway: Pandemic respiratory viruses spread fast and stochastically, often before we can clearly detect them.
Preparedness needs to plan for uncertainty, and surveillance must be broad, not just focused on a few major hubs.

Grateful to an amazing group of collaborators across institutions!

We also ran simulations for future pandemics. Results suggest that wastewater surveillance limited to a few major hubs isn’t enough - broader coverage is needed to meaningfully slow early geographic spread.

Main finding: both pandemics spread to most US metro areas within weeks, leaving a very narrow window for early detection and containment.

The two viruses followed different transmission routes, but shared key spread hubs.

Key question: How fast did the last two pandemics spread in the US? Did they follow the same spatial transmission routes?

Using high-resolution disease data and human mobility, we built an ensemble inference framework that explicitly accounts for stochasticity and superspreading in early outbreaks.

This approach helps address common issues like filter divergence and underestimation of uncertainty in data assimilation. More importantly, we can reconstruct epidemic curves with time-varying Rt using forward simulations, which are essential for running counterfactual analyses.

Our results show that ensemble filter/smoother methods with adaptive inflation give more accurate and robust Rt estimates, especially around sudden changes in transmission dynamics.

Accurately estimating Rt and its uncertainty is central to understanding infectious disease dynamics and informing public health decisions. We systematically evaluated multiple data assimilation methods for estimating Rt using both synthetic epidemic simulations and real COVID-19 case data.

Safe travels! What a heavy snow ❄️

Read a story from Mailman on this study: www.publichealth.columbia.edu/news/scienti...

end/

Huge thanks to my collaborators and co-authors for their incredible work on this project!

Code and examples are available here 👉 github.com/SenPei-CU/AM...

6/

We found that even limited sequence data can meaningfully improve carrier inference when integrated with other data. By linking patient mobility, genomics, EHR, and culture data, we move closer to spotting the silent spreaders of AMROs in hospitals and intervening more strategically. 5/

The inference framework was validated using both simulated outbreaks and real-world data on carbapenem-resistant Klebsiella pneumoniae in a large hospital. Inference with multiple data streams can better identify carriers and inform more effective target interventions. 4/

We then built an inference framework that combines patient movement, clinical cultures, whole-genome sequencing, and risk factors in electronic health records to estimate who is likely colonized. 3/