Here is our first COVID-19 publication, submitted to @nature on February 18, and thereafter carefully reviewed. It’s out today. It’s got all kinds of goodness. Let’s talk about the role of “imported” cases in “community transmission” in an epidemic. nature.com/articles/s4158… 1/
Let’s also talk about travel bans, the use of #bigdata to forecast and monitor epidemics (something my lab #HNL is working on), and the accuracy of Chinese COVID19 case data (which this paper inter-alia also addresses). nature.com/articles/s4158… 2/
This paper reflects a longstanding collaboration between my lab #HNL and colleagues at @HKUniversity and @CUHKofficial, with whom we have been using phone data to study diverse phenomena, including the impact of earthquakes or of high-speed rail on human social interactions. 3/
In a bit of misfortune for our species, COVID19 pandemic started just before annual chunyun migration, in which 3 billion trips are taken in China as part of Lunar New Year (en.wikipedia.org/wiki/Chunyun). It’s like USA Thanksgiving. But SO MUCH bigger. Wuhan is a transport hub. 4/
Working with a Chinese teleco, we were able to count 11,478,484 cases of people egressing or transiting through the city of Wuhan from January 1 to 24, 2020. At that time, COVID19 was spreading throughout Wuhan. The first case had been detected on December 1, 2019. 5/
Chinese scientists published first report of first 41 COVID19 cases in @TheLancet on January 24, and had announced the likely existence of human-to-human transmission of COVID19 by then. thelancet.com/journals/lance… (Incidentally, this paper already has 3,307 citations!) 6/
Therefore, on January 25, China imposed a cordon sanitaire – first on Wuhan and then on the surrounding Hubei province (home to 58,000,000 people). Using phone data, we are able to observe the instantaneous cessation of human movement. 7/ 
Next, we are able to show that, as people had flowed from Wuhan in January 2020, they had already carried the virus with them to 296 other Chinese prefectures in China. These imported cases would seed "new" epidemics in the destinations. 8/
Since the virus had been spreading cryptically in Wuhan since November, 2019, there was no reason to believe that people going to one destination were more or less likely to be infected than those going to another. 9/
And the more people from Wuhan that went to a destination (providing “imported cases” in the destination prefectures), the bigger the consequent outbreak in that destination (the “community transmission” cases), across 296 prefectures in China. This is how epidemics spread. 10/
In fact, if you look at cumulative, verified cases of COVID19 reported to the Chinese CDC as of February 19 (N=74,279), the data on cumulative transits out of Wuhan from January 1 to 24 predicts them nearly perfectly, at the N=296 prefecture level. 11/
Incidentally, this result sheds light on accuracy of Chinese COVID19 reporting, because a totally different source of info (telco mobility) obtained from different source predicts case counts so well, in keeping with epidemiological expectations (at least until February 19). 12/
However, there are two things related to the accuracy of Chinese reporting our data could *not* evaluate. 13/
It’s possible (and IMHO likely) that the so-far reported case counts in Wuhan in January 2020 are an underestimate – mostly due to chaos of early days of epidemic, but also possibly due to some initial concealment. We cannot estimate rate of such an undercount with our data. 14/
Also, if indeed the central Chinese CDC had somehow divided the cases reported by each one of the 296 prefectures by the same number, we could not tell that either. But, given how data are collected in China, I think that this is exceedingly unlikely. 15/
However, those comments about the accuracy of Chinese case counts are an aside. This was not the main point of our paper! 16/
In this paper, we propose a novel spatio-temporal “risk source” model that operationalizes risk emanating from epidemic epicenters and allows predictions of the timing, location, and intensity of epidemics in destination locations based on population flows alone. 17/
This risk source model, which can be implemented with any archival or (even better) real-time data regarding population flows (whether from telecos or some other source, such as traffic tolls) also allows one to evaluate local success in suppressing community transmission. 18/
When people move, they take contagious diseases with them. Their movements are thus a harbinger of the future status of an epidemic in their destinations. This offers the prospect of using data-analytic techniques to forecast & control an epidemic before it strikes too hard. 19/
However, use of internal or external travel bans to entirely stop an epidemic of this nature is nearly impossible. The reason is that, by the time anyone is aware of such a pandemic – and, say, closes borders 30 days after becoming aware – it is usually too late to stop it. 20/
Even China, which closed its country to internal travel & imposed movement restrictions on 930M people as of January 25, 2020 () could not contain the virus in Wuhan. It still spread everywhere (and will come back to China, too, when restrictions ease). 21/
For rest of the world, it’s even harder. Models in 2006 @nature paper by @neil_ferguson et al show that, even if 99.9% of international flights are stopped on day 30, it only delays epidemic by 42 days. That’s something, but it’s not full insulation. nature.com/articles/natur… 22/
Such a virus as SARS-CoV-2 crosses political borders. It’s what it does. But we can get ahead of it by tracking the virus itself (with testing) or the people who (unavoidably) carry it (while respecting their privacy, as we did here!). 23/
There is a tradition of using telecommunication data to study public health and social processes. We have contributed some to this literature (eg, journals.plos.org/plosone/articl… & ncbi.nlm.nih.gov/pmc/articles/P…) but pioneers include @barabasi & @alexvespi. Many terrific labs do such work. 24/
The GLEAM model spearheaded by @alexvespi allows one to combine real word data on human mobility (e.g., via airline flights) with models of disease transmission to forecast the burden of disease, for instance: gleamviz.org [thread continues….] 25/
And here is a fine paper by in @PNASNews by @vcolizza et al on the role of the airline transportation network in the prediction and predictability of global epidemics: pnas.org/content/103/7/… 26/
Here again is a link to our latest @nature paper on human movement in China and COVID-19 cases, across 296 prefectures, in the early days of the pandemic, through February 19, 2020: nature.com/articles/s4158… 27/
Our prior work has evaluated other strategies for forecasting timing and intensity of epidemics, such as by exploiting what we have called “network sensors,” whereby people in particular positions in a network can be seen as harbingers, via @PLOSONE. journals.plos.org/plosone/articl… 28/
Working on this new paper out today, plus all the other not-yet-announced efforts in our lab #HNL, has made me less able to keep up my former pace of twitter threads on the COVID19 pandemic. I’m sorry about that. But more work (and threads) will be emerging soon. 30/
