This blog displays part of an article by Matthew Edwards of Willis Towers Watson. You can read the full original through the link in the tweet below. Thanks To Matthew and WTW
“In an ideal world, actuaries would wait for near-certainty before changing any major assumptions. In the first half of 2020, such an approach is unfortunately not feasible.” https://t.co/REfDjIpjHA
— Henry Tapper (@henryhtapper) March 13, 2020
Why I read Stuart Macdonald’s tweets is because they allow me to plan ahead
Updated odds of catching COVID-19:
Assumes continue current trajectory, without radical interventions.
Hopefully this gives context to Govt comments about timing interventions right.
— stuart mcdonald (@ActuaryByDay) March 13, 2020
As the world grapples with the growing COVID-19 (coronavirus) threat, a useful reference point is the Spanish Flu of 1918. Given what we now know about COVID-19, are we looking at something of similar severity to the Spanish Flu, or materially less severe, or perhaps worse?
Up until now, the Spanish Flu has been generally regarded as a “worst case” reference point in pandemic modelling: This is as bad as it could get.
The thinking would be that healthcare is so much more advanced now, and we are not recovering from four years of strategic war as was the case then. These positive factors should more than compensate for the (then unimaginable) extent of international air travel and globalization of trade. But perhaps that rationale no longer holds, and we could be heading for something similar.
Plausible mortality impacts
An actuary’s immediate reaction to COVID-19 is, of course, to model it. But even now, with more than two months of data, there are considerable obstacles to any plausible modelling using the typical pandemic modelling methods (e.g. the ‘SIR’ approach of splitting into the states of susceptible, infected, recovering, with age/gender-specific transition rates) and this article does not attempt any such modelling.
Most pandemic models involve a “spread” assumption, R0, which represents how many individuals an infected person will transmit the virus to (in an otherwise perfectly susceptible population). An R0 parameter of one, for instance, equates to a stable state; a value greater than one (at the start of an outbreak) implies growth in infections, with the potential for exponential growth in the absence of interventions. Normal influenza spread implies a parameter of two to three whereas measles is one of the most contagious viruses with an R0 of 12 to 16.
At the moment, even this fundamental element is uncertain by a wide range. Various publications so far have it ranging from 1.4 to over six, while even the World Health Organization’s (WHO’s) suggested range allows almost 100% variation (from 1.4 to 2.5). In addition, as a further complication, the effective rate of reproduction will differ between countries and also vary over time as people shift their routines to avoid infection (hopefully decreasing as people and Governments become more aware of the problems).
Then we have the problem of mortality, the case fatality rate (CFR), which measures the mortality of infected cases. Overall estimates are around 2%, but here there is wide variation between the experience in Wuhan (above 5%) and outside that area (of the order of 0.7%). Reporting and data issues add further uncertainty; even determining the number of cases is challenging, especially given the asymptomatic cases whose number cannot accurately be measured.
These figures compare with the following reference points:
Experience CFR COVID-19 (Coronavirus) Circa 2%? Spanish Flu 2% Major flu pandemics (1957, 1968) 1% Typical flu season 0.1%
The overall CFR in any country is likely to be heavily influenced by the availability of appropriate healthcare resources. However, the age variability seen so far is likely to be similar across countries.
The Chinese experience showed mortality rates varying broadly and exponentially by age, from circa 0.2% for ages up to 40 increasing to 4% for the 60 to 69 age band, 8% for the next decade of age, and roughly doubling again to 15% for anyone 80 or older. (This concentration of deaths at higher ages is supported by observation of the ages of reported European fatalities.)
So what sort of impact is plausible? To paint a very broad picture, if we assume a CFR in the 0.5% to 1.0% range (in line with the ‘outside Hubei’ COVID-19 figures to date, and the 1957/68 pandemics), this could lead to, as a plausible order-of-magnitude conjecture, a doubling of mortality in a country such as the U.K. – that is, an overall doubling of population mortality, not a doubling of mortality for infected cases. This would be a bad outcome (and is not a ‘best estimate’ prediction), but it is plausible.
Moreover, the above broad-brush perspective deliberately overlooks various likely and material indirect impacts. The likely surge in patient volumes at healthcare facilities related to a sustained COVID-19 outbreak in any country could mean a corresponding squeeze on access to care (and hence, increased mortality) for patients already suffering from other serious conditions in secondary care. There will be an equivalent but less severe effect at the primary care level. Quarantine and self-isolation will also have a detrimental effect on those suffering from chronic conditions at home. The healthcare workforce will itself be depleted by absences. Furthermore, supplies of medicine taken for granted will potentially be adversely affected by production supply chain issues.
There is one material mitigating point from an insurance perspective, if not from that of the individuals affected: the coronavirus will disproportionately affect the less healthy (in particular, people already with chronic conditions), and hence the mortality impact does not consist entirely of ‘new’ deaths but will include a material proportion of ‘accelerated’ deaths (hence leading to correspondingly reduced mortality in later years).