COVID-19 dynamics across the US: A deep learning study of human mobility and social behavior

Abstract

This paper presents a deep learning framework for epidemiology system identification from noisy and sparse observations with quantified uncertainty. The proposed approach employs an ensemble of deep neural networks to infer the time-dependent reproduction number of an infectious disease by formulating a tensor-based multi-step loss function that allows us to efficiently calibrate the model on multiple observed trajectories. The method is applied to a mobility and social behavior-based SEIR model of COVID-19 spread. The model is trained on Google and Unacast mobility data spanning a period of 66 days, and is able to yield accurate future forecasts of COVID-19 spread in 203 US counties within a time-window of 15 days. Strikingly, a sensitivity analysis that assesses the importance of different mobility and social behavior parameters reveals that attendance of close places, including workplaces, residential, and retail and recreational locations, has the largest impact on the basic reproduction number. The model enables us to rapidly probe and quantify the effects of government interventions, such as lock-down and re-opening strategies. Taken together, the proposed framework provides a robust workflow for data-driven epidemiology model discovery under uncertainty and produces probabilistic forecasts for the evolution of a pandemic that can judiciously inform policy and decision making. All codes and data accompanying this manuscript are available at https://github.com/PredictiveIntelligenceLab/DeepCOVID19.

Global and local mobility as a barometer for COVID-19 dynamics

“…In the outbreak dynamics, reduced local mobility induces a reduction of the effective reproduction number R t and with it convergence to an enforced equilibrium state, a converged state under given constraints, long before herd immunity is achieved in the entire population. The speed and magnitude by which the reproduction number drops are a measure of the effectiveness of public health interventions (Bhouri et al 2020;. For example, in Austria, a country that is known for its strict response to the pandemic R t dropped from 4.0 to 1.1 in only 16 days; in Sweden, a country that implemented relatively lose public restrictions, it dropped from 2.0 to 1.1 in as much as 33 days.…”

“…On the behavioral side, there has been a dramatic change in everyday habits with widespread adoption of new rules to prevent both close contact between individuals and the exchange of contaminated bodily fluids. The remaining factors, frequency and type of contact, depend on human activity (Bhouri et al 2020). Work, school, and leisure inevitably increase the number of contacts, and hence the risk of transmission that occurs in everyday life (Kraemer et al 2020).…”

Multitask learning and nonlinear optimal control of the COVID-19 outbreak: A geometric programming approach

“…Several recent epidemic prediction methods opt to set some (or all) of the parameters in their models to estimations from the medical and virology literature (e.g. ( Achterberg et al, 2020 , Aleta et al, 2020 , Bhouri et al, 2020 , Birge et al, 2020 , Chang et al, 2020 )). However, these parameters often have wide confidence intervals and are commonly inferred from statistical models that do not take into account the effects of social distancing and hospital capacity ( Ali et al, 2020 ).…”

“…However, these parameters often have wide confidence intervals and are commonly inferred from statistical models that do not take into account the effects of social distancing and hospital capacity ( Ali et al, 2020 ). Other recent works learn these parameters from data ( Bhouri et al, 2020 , Van den Broeck et al, 2011 , Chang et al, 2020 , Hota et al, 2020 ), but do not explicitly model the infectivity of the epidemic using mobility patterns, making the design of related control strategies difficult. In contrast to these approaches, our model is entirely data-driven in that all parameters used (including initial conditions) are learned directly from data, and we learn an explicit mapping between mobility patterns and the spread of the epidemic.…”

“…Several epidemic models have been proposed in the recent literature to simulate the effects of social distancing on the evolution of the pandemic; see, e.g., ( Achterberg et al, 2020 , Bhouri et al, 2020 , Chang et al, 2020 ). Although the majority of epidemic models in recent years are variations of the seminal mathematical models on theoretical epidemiology (see Nowzari, Preciado, and Pappas (2016) and references therein), the availability of rich datasets describing human mobility and behavior is rapidly changing the field of mathematical modelling of epidemics.…”