Covid Accounting: International Arrivals
There are some spectacular parallels between getting science into decisions to manage the coronavirus crisis, and the much slower issue of growth-beyond-limits in greenhouse gas emissions and freshwater nutrient accounting. In general, I think one of the most critical needs science can help with is fairly simple, unglamorous tables of accounts. Let's hide that in the back for a moment, and consider the interesting and somewhat surprising result.
Social media furore suggests a need for this as we try to understand if New Zealand's level 4 lockdown is working. The problem I see is that a lot people continue to assume growth means cases are spreading, and possibly that there are many undetected cases.
Half New Zealand's cases are still from overseas travel, and this was the vast majority of cases until the last 10 days. So with arrival stats now available, that can be paired with infection rates in countries of origin (I used Johns Hopkins CSSE time series), what does a baseline accounting look like?
It's fairly easy to do this, and I haven't seen any version publicly available. So I had a beer and did it last night. It's accounting, so despite dreams of doing something fancy, I used a spreadsheet – which you can see or download. (Also note, no one has reviewed it.) Overall, it is good news - even with gross assumptions the peak should be passing.
I coupled locations (province/state or country) with airports feeding New Zealand to expected sources of travellers and infection rates (reported numbers per capita). That's an exercise because there are 25 main source localities ranging from many with little infection (Polynesia, modelled using Fijian data), to relatively obvious links (Australian states), to challenging (Europeans arrive via Los Angeles, Doha, Singapore, etc). I used the best paper on incubation periods I'm aware of and assumed an additional 2 days before infections are reported.
The first thing we learn is that it doesn't look exactly like reality, since "Case cause by date" peaked for international travel on 26 March. A good reason for that is a simple fudge factor (a constant) used to make the accounting work: arriving travellers are 50 times more likely to be infected than the population in the country they arrive from.
This makes sense, because travellers usually do see lots of people, eat out, attend events, and runs lots of errands. They also may be more likely to be in areas with large numbers of travellers and higher rates of infection. And until testing ramps up in some countries, infection rates are severely underestimated. Ultimately, travellers also pack into airports and planes with other people who have been doing the same thing, enabling further spread. In technical terms, they have a much higher number of close contacts (and therefore a higher effective Ro). It is also very likely that travellers and the countries they are in have become much more careful, reducing this factor considerably in big steps through the second half of March. So, there are many obvious but unknown causes for this factor to exist.
I modelled the factor describing how much more likely travellers are to be infected than the population as a constant, because it makes it easy to understand what I did. Further modelling can examine nuances that describe how it may have have changed over time.
What we really learn here is just how much the rocketing exponential of rates of infection overseas has potentially overwhelmed the strict restrictions and reductions in arrivals.
From my perspective, more transparency around this baseline transmission of international cases to New Zealand looks really helpful to enhance public understanding of how well the response has worked, and what we learn for managing what comes next. It's probably also useful as a baseline for real models (who may already be doing this) – understanding needs some representation of the driver of half the cases.
The most obvious point: the factor is 50. Travellers really matter.
This work is licensed under a Creative Commons Attribution 4.0 International License.
Social media furore suggests a need for this as we try to understand if New Zealand's level 4 lockdown is working. The problem I see is that a lot people continue to assume growth means cases are spreading, and possibly that there are many undetected cases.
Half New Zealand's cases are still from overseas travel, and this was the vast majority of cases until the last 10 days. So with arrival stats now available, that can be paired with infection rates in countries of origin (I used Johns Hopkins CSSE time series), what does a baseline accounting look like?
It's fairly easy to do this, and I haven't seen any version publicly available. So I had a beer and did it last night. It's accounting, so despite dreams of doing something fancy, I used a spreadsheet – which you can see or download. (Also note, no one has reviewed it.) Overall, it is good news - even with gross assumptions the peak should be passing.
I coupled locations (province/state or country) with airports feeding New Zealand to expected sources of travellers and infection rates (reported numbers per capita). That's an exercise because there are 25 main source localities ranging from many with little infection (Polynesia, modelled using Fijian data), to relatively obvious links (Australian states), to challenging (Europeans arrive via Los Angeles, Doha, Singapore, etc). I used the best paper on incubation periods I'm aware of and assumed an additional 2 days before infections are reported.
The first thing we learn is that it doesn't look exactly like reality, since "Case cause by date" peaked for international travel on 26 March. A good reason for that is a simple fudge factor (a constant) used to make the accounting work: arriving travellers are 50 times more likely to be infected than the population in the country they arrive from.
This makes sense, because travellers usually do see lots of people, eat out, attend events, and runs lots of errands. They also may be more likely to be in areas with large numbers of travellers and higher rates of infection. And until testing ramps up in some countries, infection rates are severely underestimated. Ultimately, travellers also pack into airports and planes with other people who have been doing the same thing, enabling further spread. In technical terms, they have a much higher number of close contacts (and therefore a higher effective Ro). It is also very likely that travellers and the countries they are in have become much more careful, reducing this factor considerably in big steps through the second half of March. So, there are many obvious but unknown causes for this factor to exist.
I modelled the factor describing how much more likely travellers are to be infected than the population as a constant, because it makes it easy to understand what I did. Further modelling can examine nuances that describe how it may have have changed over time.
What we really learn here is just how much the rocketing exponential of rates of infection overseas has potentially overwhelmed the strict restrictions and reductions in arrivals.
From my perspective, more transparency around this baseline transmission of international cases to New Zealand looks really helpful to enhance public understanding of how well the response has worked, and what we learn for managing what comes next. It's probably also useful as a baseline for real models (who may already be doing this) – understanding needs some representation of the driver of half the cases.
The most obvious point: the factor is 50. Travellers really matter.
This work is licensed under a Creative Commons Attribution 4.0 International License.
Comments
Post a Comment