Hi all, we have a new research paper titled “Intracounty modeling of COVID-19 infection with human mobility: Assessing spatial heterogeneity with business traffic, age, and race” published in the Proceedings of the National Academy of Sciences of the United States of America (PNAS). Using SafeGraph data, a human mobility flow-augmented stochastic SEIR epidemic modeling framework is developed, which combines with data assimilation and machine learning to reconstruct the historical growth trajectories of COVID-19 infection in the two largest counties in Wisconsin https://www.pnas.org/content/118/24/e2020524118
Congratulations @Song_Gao_UW-Madison! I’m very much looking forward to reading this when I get a chance.
@Song_Gao_UW-Madison Interesting study. The blue network graph is mesmerizing. When you say “clustered regions”, how’d you derive those? And a separate question: on the transition rate, did you try alternative assumptions outside of the Ornstein–Uhlenbeck?
@Thomas_Young_Econometric_Studios_Utah_Legislature Thanks! The clustered regions are derived based on the travel flow interaction between subregions using the graph-based community detection algorithm.
You can find more details about our other assumptions in the SI: https://www.pnas.org/highwire/filestream/986990/field_highwire_adjunct_files/0/pnas.2020524118.sapp.pdf
Hello @Song_Gao_UW-Madison – another great paper from your group, and thanks for ref # 60 
Your paper got me curious about a few ideas and questions. Please feel no urgency to respond, but I appreciate your thoughts if you have time.
-
You describe a non-supervised spatial clustering to re-organize census block groups and tracts into sub-county regions defined empirically by O-D traffic flows. I understand how this is superior to just using somewhat arbitrary government boundaries. That said, how important is this re-clustering for the results of the model described in this paper? E.g., would it be possible to generate similar results with your model simply using empirical O-D traffic flows between government-defined census tracts? Or is the re-clustering necessary for the results of the paper?
-
I would guess that government policies such as re-opening primarily impact effective reproduction (R) via O-D traffic flow (n). But that doesn’t have to be the case (e.g., maybe government policies don’t change O-D traffic flow, but affect close-contact social distancing and mask-wearing behavior). Can your model, which features the empirical O-D traffic flow data via SafeGraph, provide any causal analysis of the impact of government policies on effective transmission, specifically mediated by O-D traffic flow (as opposed to other measured or unmeasured mediators)?
-
Similar vein as 2: in the counterfactual scenarios you report that some govt policies, e.g. the phase 2 reopening in Dane County made significant impacts for Regions 3, 6 and 7, but less or negligible impacts on regions 1, 2, 4, and 5. The spatial heterogeneity of these results surprised me. What causes them? Are you able to tease apart whether the diff between Regions 3,6,7 vs 1,2,4,5 are driven primarily by differences in O-D traffic flows themselves vs other socioeconomic variables like age?
Again, thank you so much for sharing the great work!
Excellent questions! @Ryan_Fox_Squire_SafeGraph1 (1) Yes, it is possible to generate similar results with our model simply using empirical O-D traffic flows between government-defined census tracts; however, the key assumption is that the derived regions with more intra-county flow interactions have homogenous infection parameters. If we use default tract stats, there are a lot of zeros and any arbitrary aggregation doesn’t make sense.
(2) We model both transmissibility parameters and travel flows as different components in the ODE system. We actually did such evaluation in another state-specific work State-specific projection of COVID-19 infection in the United States and evaluation of three major control measures | Scientific Reports