Insight – the Control Cycle
As society gradually re-opens, the government has to make assumptions about the coronavirus. So we should describe what we expect up-front and then go about aggressively testing that expectation. We are effectively conduct a real-world experiment, along with everyone else in the world, about how the virus is transmitted. We should use that experiment to learn as much as we can, and then use what we learn to minimise the harm of the virus as much as we can.
Actuaries are used to this kind of process, as they are used to building models about the future – future claims, investment earnings, customer behaviour etc. We call this test and learn process the actuarial control cycle.
The most important part of the cycle is to describe what you expect to happen in ways it can be tested. It is always tempting, when something happens, to fit an explanation to the outcome. It’s much harder to do that if you have described your assumptions up front.
For example, as schools are reopened we could:
- describe what we expected to happen (eg very low to no transmission between children and from children to adults, children don’t get sick, even if they catch the virus, some transmission between adults)
- design a test to check whether it was happening (pick a school at random and do regular testing of children, teachers and parents)
- review the results compared with the expectations and adjust the approach.
That test might find that children don’t transmit the virus to each other, but can transmit it to adults. That might mean that schools could stay open, but take extra steps to keep children away from teachers. Or the test might find that children do transmit it to each other, which might mean that children would have be tested very regularly for schools to stay open without risk to the rest of society.
There are likely to be some natural experiments as different states adopt slightly different parts of the government’s three step framework of reopening. Ideally, we would look at those differences and form hypotheses about what the different rules are likely to lead to in outcomes – and then test them.
Do any of my readers have suggestions for natural experiments? What are the hypotheses you would be testing as we gradually reopen?
Link
Last week a glimpse of the genuine possibilities of big data – this paper from the UK analyses the pseudonymised health data of over 17.4 million UK adults to discover the key factors associated with death from COVID-19.
There were 5683 deaths attributed to COVID-19. In summary after full adjustment, death from COVID-19 was strongly associated with: being male (hazard ratio 1.99, 95%CI 1.88-2.10); older age and deprivation (both with a strong gradient); uncontrolled diabetes (HR 2.36 95% CI 2.18-2.56); severe asthma (HR 1.25 CI 1.08-1.44); and various other prior medical conditions.
And our own Actuaries Institute approached the question of risk factors from a different angle – comparing the underlying population data (what is known about underlying risk factors) with the deaths in New York. They found that not all of the co-morbidities, or risk factors, seemed to lead to higher death rates than the underlying population, particularly when adjusted for age. But people with diabetes in particular, had much higher fatality rates than their representation in the broader population would predict, especially at younger ages.
Two studies from completely different data and methodologies both lead to diabetes being a serious risk factor, which is something I hadn’t seen before in the broader literature.
Life Glimpses
The lockdown rules have slightly changed here in NSW, so that more visiting between people is allowed. But the foreshadowing of bigger changes to come meant that our local shopping streets were full of people this weekend. I found it quite offputting how much dodging I had to do. Citymapper shows quite clearly that people in Sydney, in particular, have started being more active again on the weekends.
On the plus side, for my weekend walk today, I visited a park that is quite hidden away, on the foreshores, but not well known, and it was much more relaxing. There were a few people there, but no more than a normal weekend. When I lived nearby I went there most weekends, and so it was lovely to be reminded of one of the many little bits of beauty around Sydney Harbour.
Bit of Beauty
Yet another glorious sunset the other evening. It is always hard to capture the light, but I think I’ve got a bit of it.
The authorities here have eased restrictions. Nothing is very clear. As the weather is perfect
lots of people go to the beach. Supposedly for sport activities. On TV, you see crowds very
Near each other, in spite of very clear instructions, distancing of 2 metres should be observed.
Public transport is not working in full capacity.
How do you manage to order such sunsets. Will you enlarge the photo
Love
Sensible suggestion but that would mean evidence based policy, which only rarely surfaces in Australia.
I really like that Citymapper tool. No idea what’s behind it, so how reliable it is. Working for a company headquartered in Chicago I was very interested to see how much more severe the lock-down was in Chicago and the rest of the US compared to here. It was interesting looking for the city most analogous to us — looks like Toronto (frankly closer even than Melbourne), with Vancouver not far behind (Canada — surprise surprise) or Copenhagen or… Sao Paulo 🙁
You can clearly see Sydney getting out and about quicker than Melbourne. We noticed the uptick in outdoor activity over the weekend too — I was hoping it was just a Mother’s Day phenomenon, but I guess if you relax the lockdown — which it effectively has been, well… that’s what it looks like. And it still is nowhere near as busy as a typical weekend. I imagine the ultimate question is, can they keep R0 below 1 with a relaxed level of lockdown? I mean I suppose, off of a base of about four new cases a day in NSW, that would be acceptable?
Personally we took the opportunity to introduce Isabel to driving in traffic on main roads, and I feel that Sydney is slowly throwing more advanced levels at her as she advances through the game.
I would like to hear a lot more about what the various levels of government learned from the first lockdown. What was effective, what made no difference. Early days, but I do suspect that people will get a bit loosey-goosey and there will be a second wave and I’d like to think that the government could implement a lockdown with a bit more finesse and better targeted towards Covid-19 than the first one implemented with, you know, forty-eight hours notice towards a disease whose characteristics were only loosely known?
I felt that a lot of the restrictions were too much of an all or nothing affair — you close the coastal walk, or open slather — nothing in between. And opening beaches only at certain times funnelling people through a single entry point seems like it will exacerbate problems not mitigate them. Surely there are better ways to ration access? Odds and evens like when petrol is rationed — that sort of thing?
Other restrictions, maybe there was no point in. Maybe closing playgrounds? Picnics? Sunbathing? Bush-walking?
Eating kebabs on park benches? Maybe that will be my Ph.D. thesis: “Eating kebabs on park benches — its role in the spread of infectious diseases in the Hunter region during March 2020.”
Or maybe these were very efficacious measures. But surely now the data is out, can be analysed and reimposed if necessary on the basis of data, not just what seems likely.
On the other side maybe there should be much greater emphasis on regulating access to old-age homes. And arresting Nazis.
On a final note. I see you have completely failed in any of your so-called “Reflections” to address the role of 5G in the spread of this microbial scourge. Are you in the pocket of big telco?
Hi Jennifer, the other interesting thing coming out of the UK study is that Asian and black groups are at higher risk of death, which is only partially explained by pre-existing conditions and socio-economic factors. I also read a small study about the UK children with Kawasaki/toxic shock – 6/8 were black and 2/8 were asian. Not sure if there is anything in this….Cheers, Karen
To start, we could look at the figures for the past two weeks as a base expectation for the next two. I’ve been using the daily NSW Health media releases as they have a consistent structure and the figures will be measured in a comparable manner. Over the past two weeks, NSW has gone from 3009 cases on April 27 to 3053 on May 10, so an increase of 44. We can also see that the ‘overseas’ figures increased by 15 and the local with confirmed contact by 32. There were 4 that were under investigation which blurs the figures a bit, but it is safe to assume a ratio of around one third overseas and two thirds local for new cases.
Unless there’s a massive increase, the next two weeks are unlikely to provide cases related to a school in a volume sufficient to draw conclusions. More likely will be cases in parents returning to workplaces that they could not attend while looking after children. In schools, I think the closest to a natural experiment will be where year 12 are attending five days per week while year 11 are only there one day. At five times the exposure, if there is a risk in schools, it should appear in that cohort at five times the rate.