Thread by @PartTimePM

COVID-19, layered defense, and our border.

Many people seem to be criticising National's proposal about requiring a mandatory negative test before boarding a plane to NZ, often on the basis that positive cases will slip through. Here's why that doesn't matter.

#nzpol #covid19

1/40 When defending against a risk with potential for extremely adverse consequences (e.g. nationwide lockdowns, deaths, long-term health problems, major economic damage), the first thing to realise is that *no* single defensive measure is infallible.

2/ Given that even the best possible defensive measures do, in practice, have at least a small probability of failure, what matters most is decreasing the *overall* probability of adverse consequences. Ultimately, the risk profile of the whole system is what matters.

3/ For the purposes of this thread, let's assume that the goal is, with extremely high probability, to completely prevent uncontrolled circulation of COVID-19 within the NZ community, and to do so with as little impact on the normal day-to-day lives of Kiwis as possible.

4/ The first stage of this defense is to analyse the risk properly, and clearly define where the areas of vulnerability are. There are many ways that C19 might make its way from overseas into our community - we need to know them all, in as much detail as possible.

5/ Because you don't want to risk missing important risk surfaces, it's helpful to have many eyeballs on this step. The more people you have looking for problems, the less likely you are to miss them - what one person might overlook, ten probably won't.

6/ Diversity of thought is also useful at this stage - if you pick people from many different vocations and with different modes of thought, you'll catch more than if you have a group composed entirely of similar people.

7/ Assign rough threat levels to each identified area of vulnerability. The higher the probability of incursion via any one surface, the more important it is to protect (e.g. infected people arriving by plane are more dangerous than a bag of frozen peas on a ship).

8/ The total probability of uncontrolled community circulation is the sum of everything on the list of risks that might lead to such spread. As long as that total is higher than whatever probability is deemed acceptable (which in this case needs to be *extremely* low).

9/ Now we can design measures to mitigate the risks. Each individual measure should be *expected* to fail some of the time, and as such, the design of each measure is essentially a mini version of this whole process, with its own list of vulnerabilies & countermeasures.

10/ Because measures are expected to fail, it's important to have every point of risk covered by multiple measures. There should be processes in place to mitigate the consequences of failure, and if ideally notice failure when it occurs.

11/ Start with cheaper measures (in terms of cost per amount of overall risk reduction). These measures often have quite a wide scope, and will cover a large portion of the risk surface, but may have a high failure rate.

12/ Keep adding additional layers of defensive measures until the overall probability of uncontrolled community spread is low enough to be acceptable. There should be no identified risk vector with only one defensive measure protecting it.

13/ Layers should be diverse where possible. If two different layers can be simultaneously defeated by the same failure, they aren't a particularly useful combination. Failures in one layer should be catchable by other layers.

14/ Defensive measures with a low probability of being called into action are allowed to be more expensive. For example, contact tracing of cases that make it all the way through other layers, or regional lockdowns.

15/ The line between failure and what's acceptable isn't necessarily clear. The cost of some measures (e.g. regional lockdowns) is so high that, even though it reduces the risk of uncontrolled transmission, it's still best avoided, and should be a last resort.

16/ Ongoing audit and reporting on defensive measures is important. Implementing them often results in new risk surfaces, which must be detected. Defensive measures that are not properly implemented may lead to a false sense of security, and make things worse.

17/ National's proposal to require a negative pre-flight test is just one layer of defense. It doesn't need to be perfect, because it's merely one measure among many, and helps decrease the overall risk that the virus makes it through *every* layer into the community.

18/ Now let's look at the practical example of C19 arriving via a person travelling from overseas to NZ. With completely unrestricted travel, the likelihood of any one person being infected is low, but the likelihood of an infected person arriving is 100%.

19/ Defensive measures:
1. We have no viable way to sufficiently reduce the risk of an infected person arriving to an acceptable level under normal, unrestricted circumstances - so we must reduce the rate of arrivals to a low enough level that we can apply additional measures.

20/
2. Allowing only citizens & residents brings arrivals to a manageable rate. The probability of a positive case arriving soon (within days) is still 100%.
3. Isolating all arrivals from the rest of the community reduces the risk of community incursion.

21/
4. Arrivals obviously can't all be isolated away from the community indefinitely, so the probability of infected individuals being released needs to be lowered to an acceptable (very small) level.

22/
5. Isolated arrivals must be housed, fed etc, and some of them will be infected. New risks: transmission from isolated individuals to community via provided services in isolation, in transit to isolation facilities, and containment failure.

23/
6. Releasing arrivals after 14 days allows one full transmission cycle. Symptoms are likely to be noticeable within this period. New risks: unnoticed asymptomatic cases, isolated individuals infecting each other.

24/ At this point, you can see that we've introduced two layers of defense, and one of those layers has also brought several new risks along for the ride, which we also need to address. The overall risk is still too high, so let's add more layers.

25/
7. Don't release symptomatic individuals. Addresses: positive cases which do not recover within isolation period.
8. Test isolated individuals at days 3 & 12. Addresses: unnoticed asymptomatic cases, day 3 false negative. New risk: test refusal.

26/
9. Require negative test before release. Addresses: unrecovered positive cases. New risk: test refusal.
10. Isolate test-refusing individuals for further 28 days. Addresses: test refusal.
11. Add security guards. Addresses: containment failure.

27/
12. Add fencing. Addresses: containment failure.
13. Select premises to minimise egress points. Addresses: containment failure.
14. Add police with detainment powers. Addresses: containment failure.

28/
15. Require PPE for guests. Addresses: transmission between guests.
16. Require PPE for staff. Addresses: infection of staff.
17. Frequent health checks for staff. Addresses: transmission to community.
18. Regular testing of staff. Addresses: transmission to community.

29/
19. Criminal penalties for escapees. Addresses: containment failure.
20. Social distancing of guests. Addresses: transmission between guests.
21. Reduce services. Addresses: infection of staff.

30/
22. Contact tracing. Addresses: community spread. New risks: slow tracing reduces effectiveness, missed contacts, unreachable contacts.
23. Community testing. Addresses: community spread. New risks: false negative.
24. Genomic sequencing. Addresses: community spread.

31/
25. Increase tracing staff. Addresses: slow tracing.
26. NZ Covid Tracer app. Addresses: slow tracing, unreachable contacts, missed contacts.
27. Upstream tracing. Addresses: community spread.

32/ While far from a complete list, this is all stuff that NZ has done (albeit some of it quite poorly at times). You can see that risks are addressed by multiple measures, and that measures can introduce new risks of their own, which must be further addressed.

33/ It's also clear that there's lots of room to improve, and that the measures we have taken to date are insufficient. Our full-stack failure rate is about 1 in 40k arrivals, which - while it looks like a low number - is actually something we'll hit fairly frequently.

34/ Another big issue is that, while we've done a lot to reduce the risk of transmission to the community, we haven't done much to mitigate the extreme impact of failure in some of our defensive measures, as shown in the case of Auckland L3 lockdown.

35/ Some examples of things we could do to further decrease risk of overall failure, and reduce the impact of defensive measures:
28. Move MIQ facilities outside of urban areas. Addresses: impact of containment failure.

36/
29. Confine guests to room. Addresses: transmission between guests.
30. Land flights directly at MIQ. Addresses: transmission in transit.
31. Purpose-built MIQ facilities. Addresses: transmission to staff (because fewer staff needed).

37/ This is where requiring a pre-flight negative test comes in. It doesn't need to catch all cases before they board the plane, but it's a useful way to decrease the number of positive arrivals in NZ, which lowers the risk a bit for many of the downstream layers.

38/ Most of the objections to this policy seem to relate to it being imperfect, which it doesn't need to be, and its impracticality for many points of origin. Thankfully, because it doen't need to be perfect, it simply doesn't matter that it's impractical / unfair.

39/ Practicality doesn't matter, because you can simply exempt impractical origins or circumstances from the requirement. The same goes for excessive unfairness. Doing this reduces the effectiveness of the measure obviously, but most of the benefit still remains.

40/ For clarity, please note that the example above is a *very* partial one. There's an awful lot missing, both in terms of what's actually happening, and in terms of what could be done to improve it. It serves merely to illustrate the concept.

@threadreaderapp Please unroll this thread.