Approach: Modeling of global aircraft-based genomic surveillance networks (nasal swabs & wastewater) for global pathogen monitoring.
Advanced Modeling: Use of an epidemic and mobility model combined with probability generating function analytics to estimate the performance and optimization of global wastewater surveillance networks
Assessment and Optimization: Identification of the placement of sentinel sites for timely situational awareness and early warning, with optimization strategies to improve effectiveness and minimize resource use.
As seen in recent pandemics, airports are critical hubs for the global spreading of infectious diseases. At the same time, they could be used as primary sentinel points for novel pathogen detection. Implementing effective biosurveillance at airports involves navigating complex logistical challenges and employing modeling & analytical approaches able to optimize the system and estimate its performances.
The detection of specific pathogens among international travelers through nasal swabs upon arrival or via airplane wastewater surveillance at points of entry is a promising strategy for a global surveillance network against emerging infectious disease threats. This approach enhances our ability to monitor and respond to potential outbreaks effectively, leveraging both direct sampling from individuals and sampling from aircraft wastewater to provide an efficient early warning system (for example: wwwnc.cdc.gov/travel/page/travel-genomic-surveillance).
We developed a computational platform using the GLEAM model[1] and advanced analytics based on probability generating functions [2]. This model integrates the global air-travel network to simulate the transmission and spreading of pathogens and their detection at a given set of airports, referred to as the sentinel network[3]. Biosurveillance mechanisms include nasal samples from passengers and wastewater sampling from individual airplanes or airport triturators. The detection rate, the fraction of disease carriers leading to detection, varies from below 1% to over 16% depending on the methodology and airport settings. Given any initial conditions for an outbreak, the model generates stochastic realizations of the global epidemic spread. Simulated data includes international and domestic infection importations, incidence of infections, and individual level detection at sentinel sites with a daily resolution. These metrics provide a general framework for assessing the sentinel network effectiveness in real-time.
Planning mode: the model identifies optimal sentinel networks based on a given sampling strategy and surveillance network design and measures their efficiency by estimating epidemiological quantities of interest such as the time to first detection of a specific pathogen.
Response mode: the model provides estimates for the pathogen dispersal, outbreak source, reproduction number, and outbreak timing, given specific detections and a configured US surveillance network.
A critical metric for evaluating a sentinel network's performance is the time to first detection. For example, modeling the Alpha variant (B.1.1.7) of SARS-CoV-2[4], our simulations show that a traveler-based surveillance program consisting of sentinels in 8 airports could have detected importations 1.5 to 2 months before the first confirmed US case in December 2020, even with a 4% detection rate. The meantime to first detection varies globally, with significant spatial heterogeneity depending on the source of the outbreak. If an outbreak emerges in particular regions in Europe, the first case could be detected within 25 days, however, in certain parts of Central Africa it could take over 100 days, indicating blindspots in the sentinel system.
The time (in days) to detect a novel pathogen after its emergence (assuming a seeding cluster of 10 individuals) from anywhere in the world by performing traveler based genomic screening in the sentinel network identified by the airports indicated by the “symbol” on the map (JFK, EWR, BOS, IAD, MIA, LAX, SEA, and SFO). We use a SARS-CoV-2 like virus with a reproductive number R0 = 2.0. We consider each of the 3200+ subpopulations as the potential origin for an epidemic and evaluate the mean time to obtain a single detection by the surveillance system. We consider a detection rate of 4% for all sentinel sites only on International inbound flights.
Counterfactual detection of the Alpha variant by a US-based traveler surveillance network. We show plausible distributions for the time to first detection for different detection rates. The sentinels are the same network as above. We highlight December 29, 2020, the day a first confirmed case was identified in the US.
The model can systematically simulate the detection of pathogens through a travel-based genomic surveillance network at airports. The model adapts to various surveillance methods, including nasal-swab testing and wastewater from aircraft monitoring, and provides a robust framework for optimizing the placement and efficiency of sentinel sites.
Given the presence of blind spots and the varying performance of sentinel systems, optimizing the number and location of sentinels is crucial. The model shows that adding sentinels initially yields significant improvements, but marginal returns diminish as more than 10 to 20 sentinel sites are added to the surveillance network[3]. This balance between cost and efficiency informs the scaling and optimization of a travel-based surveillance network at airports, enhancing its overall performance for early pathogen detection. By strategically modeling and optimizing traveler-based surveillance, we can improve our preparedness and response to future infectious disease threats.
Our proposed application will develop innovative methodologies to integrate advanced statistical and analytical frameworks and machine intelligence with mechanistic modeling techniques, identifying new approaches that improve local, state, and regional forecasting and modeling capabilities and analytics tools. The proposed activities involve integrating novel data sources—including high-resolution mobility, airline travel, genomic and wastewater surveillance data—with agent-based, statistical, and deep learning forecasting models to increase the accuracy of outbreak analytic products. Importantly, our approach will consider population heterogeneities/disparities and will deliver outbreak analytic tools for rural/underserved populations and for diseases/locations with low prevalence. Finally, we will identify best practices for how to transfer and maintain the needed methodology, technical expertise, and data sources, and establish a comprehensive training program for public health workforce and emergency response decision makers (which includes embedding co-op students into public health/healthcare delivery agencies and developing collabathons).
1. Davis, Jessica T., et al. "Cryptic transmission of SARS-CoV-2 and the first COVID-19 wave." Nature 600.7887 (2021):127-132.
2. Miller, Joel C. "A primer on the use of probability generating functions in infectious disease modeling." Infectious Disease Modelling 3 (2018): 192-248.
3. St-Onge, Guillaume, et al. "Optimization and performance analytics of global aircraft-based wastewater surveillance networks." medRxiv (2024) 2024.08.02.24311418
4. Chinazzi, Matteo, et al. "A multiscale modeling framework for Scenario Modeling: Characterizing the heterogeneity ofthe COVID-19 epidemic in the US." Epidemics 47 (2024):100757.
.png)
