|2021/10/19 Nathaniel Osgood
Service Delivery from Models: Production-grade real-time COVID-19 epidemiology and acute care demand monitoring and nowcasting via Particle-Filter & Particle MCMC-leveraged Transmission Models
- Speaker: Nathaniel Osgood, University of Saskatchewan
- Date and Time: Tuesday, October 19, 2021 - 1:00pm to 2:00pm
- Abstract: While COVID-19 transmission models have conferred great value in informing public health understanding, planning and response, the ability of public health decision-makers to rely purely upon traditional transmission models with pre-set assumptions - no matter how favourably evidenced when built - is challenged by numerous factors, including rapid changes of public commitment to social distancing and mask use, uncertainty regarding the impact of relaxation of measures, variant emergence and horizontal genetic transfer of mutations, unexpected occurrences of mass gatherings, and outbreaks in vulnerable communities. The ongoing replanning associated with rolling back and re-instituting measures, initiating surge planning, targeting vaccine rollout and conducting timely messaging can strongly benefit from approaches that continuously integrate unfolding time series into rigorous transmission models, so as to provide a consistent, integrated depiction of the evolving underlying epidemiology and acute care demand. Such replanning further benefits from the ability to project such depiction forward in a fashion suitable for triggering acute care surge planning and enhanced public health messaging and surveillance. The use of the Sequential Monte Carlo algorithm of Particle Filtering (PF) supports estimation via sampling of dynamic model system state on a daily basis on the basis of daily data on test volumes and positivity, endogenous and travel-related cases, hospital census and admissions flows, mortality and (most recently) wastewater concentrations of the SARS-Cov-2 virus. Particle Markov Chain Monte Carlo (PMCMC) approaches further support sampling from the posterior distribution of (static) parameter values jointly with the system state over time.
- We describe here the design, implementation of a purpose-built computational infrastructure and PF & PMCMC supported compartmental models that supports day-to-day use for public health and clinical support for COVID-19 decision making in each of 17 diverse jurisdictions.
Following a brief description of model design, we describe how PF is used to continually reground estimates of dynamic model state, to support probabilistic model projection and to probabilistically evaluate tradeoffs between potential intervention scenarios. For the first time, we also describe the role of PMCMC within this framework, and key resource and accuracy tradeoffs associated with practical PMCMC use. Important model outputs from both these approaches include estimates (via sampling) of the effective reproductive number, the count of undiagnosed infectives, the count of individuals at different stages of the natural history of infection, including via paucisymptomatic pathways. We further note aspects of model use and support in practice, including a sketch of the purpose-built distributed computing framework and associated processing pipeline handling multiple modeling frameworks. This modular infrastructure permits semi-automated model deployment, automatic post-scenario scripting and reporting for a wide variety of clients with differing reporting needs. Finally, responsive to Project 9 research foci, we mention work underway to inform such approaches using metrics drawn from online search and social media use, and characterize open research challenges whose resolution could aid future generations of such reporting frameworks.
- Nathaniel Osgood serves as Professor of Computer Science, Director of the Computational Epidemiology and Public Health Informatics Laboratory at the University of Saskatchewan & served as technical director of provincial COVID-19 modelling from March 2020-March 2021. His work combines tools from Systems Science, Data Science, Computational Science and Applied Mathematics to inform decision making in health & health care. Leveraging diverse data sources, transmission modelling and machine learning/computational statistical methods, CEPHIL delivers COVID-19 reporting for SK, for all provinces for PHAC and for Canada’s First Nations (via FNIHB), as well as other smaller SK jurisdictions. In addition to diverse application studies across the communicable, zoonotic, environmental, mental health and chronic disease areas, he has contributed a broad set of relevant methodological innovations to improve dynamic modelling quality and efficiency, introduced novel techniques hybridizing multiple simulation approaches and simulation models with decision analysis, computational statistics, and frameworks leveraging principles applied category theory, and leveraged models using data gathered from mobile sensors, social media, and online search volumes. Dr. Osgood is further the co-creator of two novel wireless sensor-based epidemiological monitoring systems, most recently the Ethica Data system for smartphones and wearables.