Branching Processes
For Infectious Disease Modeling

Dimitri Lopez
Mariah Boudreau
Michael Johansson

Guillaume St-Onge

Roadmap

Intro to
Branching
Processes
→
Disease Risk
+
Detection
→
Operational
Forecasting

Branching Processes Introduction

Instead of

extinction probability of a pathogen:

we ask

"Where is most vulnerable to disease introduction?"

“Where is most vulnerable to disease introduction?”

Importation Risk

Establishment Risk

Infectious Disease
Dynamics
Importation Risk
(Airline Data)
Establishment /
Outbreak Risk

"Nowcasting the Spread of Chikungunya Virus in the Americas" (Johansson 2014)

Instead of:

"Where is most vulnerable to disease introduction?"

we ask:

"How early can we detect a pandemic?"

“How early can we detect a pandemic?”

Detection Times

Global Surveillance Network

Seed Outbreaks
Around the World
Importation Risk
(Airline Data)
Pathogen Detection
(Waste Water
Surveillance)
Optimal Placement
of Surveillance?

"Pandemic monitoring with global aircraft-based wastewater surveillance networks" (St-Onge 2025)

Branching processes excel in early dynamics,
but they have a key limitation.

They assume the susceptible population is infinite.

Computational speed of branching processes for operational forecasting?

Explicitly track cumulative infections

Rt Inference Engine

Hospitalizations released weekly

Forecasts required in < 10 hours

FluSight Results

Beat the ensemble in the early season
Struggled in the later season

Upcoming Work

Improve outside of early outbreak dynamics
Computationally Efficient
Mechanistic Modeling Flexibility

What if you could generate mechanistic fitted forecasts for the entire US in an afternoon on a laptop?

Effectiveness of Branching Processes

Stochastic
Mechanistic Modeling Flexibility
Computationally Efficient

Q&A

Thank you!

Questions?