Iterative data-driven forecasting of the transmission and management of SARS-CoV-2/COVID-19 using social interventions at the county-level

Abstract

The control of the initial outbreak and spread of SARS-CoV-2/COVID-19 via the application of population-wide non-pharmaceutical mitigation measures have led to remarkable successes in dampening the pandemic globally. However, with countries beginning to ease or lift these measures fully to restart activities, concern is growing regarding the impacts that such reopening of societies could have on the subsequent transmission of the virus. While mathematical models of COVID-19 transmission have played important roles in evaluating the impacts of these measures for curbing virus transmission, a key need is for models that are able to effectively capture the effects of the spatial and social heterogeneities that drive the epidemic dynamics observed at the local community level. Iterative forecasting that uses new incoming epidemiological and social behavioral data to sequentially update locally-applicable transmission models can overcome this gap, potentially resulting in better predictions and policy actions. Here, we present the development of one such data-driven iterative modelling tool based on publicly available data and an extended SEIR model for forecasting SARS-CoV-2 at the county level in the United States. Using data from the state of Florida, we demonstrate the utility of such a system for exploring the outcomes of the social measures proposed by policy makers for containing the course of the pandemic. We provide comprehensive results showing how the locally identified models could be employed for accessing the impacts and societal tradeoffs of using specific social protective strategies. We conclude that it could have been possible to lift the more disruptive social interventions related to movement restriction/social distancing measures earlier if these were accompanied by widespread testing and contact tracing. These intensified social interventions could have potentially also brought about the control of the epidemic in low- and some medium-incidence county settings first, supporting the development and deployment of a geographically-phased approach to reopening the economy of Florida. We have made our data-driven forecasting system publicly available for policymakers and health officials to use in their own locales, so that a more efficient coordinated strategy for controlling SARS-CoV-2 region-wide can be developed and successfully implemented.

Assessing the mechanism of citywide test-trace-isolate Zero-COVID policy and exit strategy of COVID-19 pandemic

“…We estimate the following parameters, including the probability of transmission per contact and the contact rate in each stage, by using a Monte-Carlo-based Bayesian modeling framework [ 37 , 38 ]. We sample 100,000 parameter vectors from the uniform prior distribution (Additional file 1 : Table S1), and we choose the 1000 best-fitting parameter vectors based on the modified normalized mean square error (NMSE) indicator by comparing data and model predictions of cumulative confirmed cases and cumulative isolated individuals.…”

Projections of the transmission of the Omicron variant for Toronto, Ontario, and Canada using surveillance data following recent changes in testing policies

“…The model is calibrated using publicly available surveillance data of daily new reported cases and deaths, daily hospitalized and ICU cases in Toronto, Ontario, and Canada from January 1 to February 9, 2022 ( City of Toronto, 2022a , 2022b ; Government of Canada, 2022 ; Public Health Ontario, 2022 ). The proportion of nHRS mild symptomatic infections conducting self-testing and PCR testing, the transmission risk, the proportion of hospitalized, ICU and deaths from ICU are estimated using a Monte-Carlo-based Bayesian melding framework ( Newcomb et al, 2022 ) and are reported in Table S6 . The estimations are based on a forwarding weekly time window to capture in a timely fashion the changes in transmission risk and severity.…”

Combining predictive models with future change scenarios can produce credible forecasts of COVID-19 futures

“…Thus, while some workers have highlighted the difficulty of anticipating and accommodating novel, unknown, or previously unsuspected changes in the drivers of future viral transmission to allow the making of reliable long-term projections by models [ 10 , 11 ], others have pointed to the value of these models as tools for being able to integrate information on the diverse structures and processes related to transmission dynamics in order to propagate forecasts that are more accurate than predictions afforded by common sense alone [ 12 , 23 ]. Such assessments have also, for example, pinpointed the need for continual model refinement and for the use of data for making predictions to counter the effects of changes in local risk factors [ 12 , 14 , 24 ]. These studies ultimately suggest that, as for other studies investigating socio-ecological futures, the better use of models for forecasting the plausible futures that could be followed by the pandemic is to combine simulations within a scenario framework in order to focus on explorations of possible trajectories in the evolution of the system as a result of changes predicted for key drivers rather than employing them to make precise predictions about the extent or duration of disease burdens [ 25 , 26 ].…”