Justifying a shutdown: Covid-19 and the data story of the century

It seems likely that along with other firsts accompanying Covid19, it may be regarded as one of the biggest data stories of this century.

Whilst there have been policy missteps, it’s inspirational that many countries locked down their citizens with zero or handfuls of initial cases in hospitals, in what can be seen as a triumph of math and science over natural social gregariousness and economic interest.

We’re not healthcare workers but as data analysts we spend a lot of time helping firms persuade people by telling stories with data – and getting people to change behaviour based on data trends is a challenge – as is evident from photos and videos of social distancing fails.

Making it more of an uphill battle to sell the benefits of Covid19 policies like lockdowns to the community, health departments have detailed data models but aren’t necessarily sharing them, or even figures for the hospital resources they are working with, and likely have good reasons for doing that.

So to help with community support we built an online calculator that spells out Covid19 risks specifically for individuals and their ability to get medical care if they need it, as we think some policies like lockdowns may be hard to sustain support for outside the short term.

Three quick summary observations from our model

1. A relatively low spread of Covid19 amongst the general population of 3% exhausts hospital bed availability and suggests the need to build more hospital infrastructure to bolster it in many countries ASAP. OECD hospital bed occupancy averages normally run at 75% so we’ve assumed as a starting point 50% of theoretical beds are not available in our model.
2. It would take years to establish herd immunity at 60% population infection rates whilst managing the rate of infection to still be able to care for people in the healthcare system (with yo-yoing lockdowns). The implication is that either Covid19 eradication in individual countries (with strict post-infection border quarantine controls) or, a rapid vaccine, may be required to avoid huge numbers of deaths and even larger economic costs, hard as either option might be to achieve.
3. Difficult triage decisions for doctors exist around patients with different ages / pre-existing conditions. For Australia in our model, 89% of expected ICU beds are required for the over 60s (assuming a 3% population infection rate). How would you decide who doesn’t get a bed?

Covid19’s data backstory

Flatten the curve and exponential growth become memes

The graph for ‘Flatten the curve’ has become globally recognisable as has the exponential curve of cases of itself, so much so that there are now black humoured jokes about it:

Hundreds of respected institutions worldwide have put out detailed models showing information such as how social distancing compliance impacts the timeline of the virus in each country and businesses worldwide have built their own simpler models relying on number-of-days to double cases estimates to try and gauge how bad things would get by a certain date.

And whilst less glossy the 16th March paper from Imperial College’s Covid19 Response team in the UK has been extremely influential in setting health policy worldwide (and has also been used by us in our personal risk model).

However flatten the curve, whilst appealing in its simplicity, implies there’s a timeline where you can achieve herd immunity (60% infection population-wide) whilst not massively overloading the health system. Looking at the years required (see point 2 in our summary above) it’s possible that flatten the curve is a more theoretical than practical long term option.

Covid19 data agendas

Like all data, Covid19 data has been put to use to promote widely different agendas.

Some politicians, King Canute-like, have announced Covid19 would go away by certain dates based on concern about upcoming elections and the deep state’s influence amongst press and health advisers

Other politicians debated ‘stepped’ lockdown approaches versus ‘immediate’ lockdown approaches and will struggle to maintain community support and compliance for lengthy lockdowns.

60% herd immunity proponents wanted to lock down only vulnerable groups who we’ll protect (assuming limited viral mutation).

Businesses and opinion columnists argue that the economic costs of Covid19 far outweigh the health costs or that health treatments are too ineffective to make flattening the curve worthwhile.

We all have some friends and acquaintances who pay little heed to social distancing because ‘the odds are low that it’ll be bad and my friend Phil had it and said it was nothing’ (and at the other end of the scale armchair epidemiologists such as ourselves).

And finally healthcare spokespeople naturally emphasise the near term case avalanche about to hit their hospital.

The problems with national numbers of Covid19 cases data

Many of these individuals promoting the particular agendas above rely on growth in Covid19 case numbers.

But case numbers are problematic because:

• headline numbers focus on the number of cases in a country and not the per capita cases (Italy for example has 20% of the population of the United States)
• countries adopted widely varying testing policies with for example the UK mainly testing hospital admissions (understating the number of cases) versus other countries testing in the community far more widely
• contact- tracing amenable cases results in higher number of positive tests whereas cases with ‘unknown community transmission’ there’s just less people to test
• exponential growth in cases changes (with a lag) as you introduce distancing measures
• 40-50% of Covid19 cases appear to be asymptomatic so the person may not bother getting tested at all or even meet the testing criteria set by health departments (where tests are being rationed). We have assumed 50% in our model but it can be changed.

Case numbers are so problematic that some people have even built models that calculate implied case numbers from covid19 deaths.

Fatality data has issues as well

If it’s hard to accurately measure number of cases, it’s harder to measure fatality rates for Covid19 as you don’t know how many people had it in the first place. So studies like the Imperial College paper we’ve used in our model focused on hospitalisation/ICU rates for symptomatic cases.

The fatality rate is also different depending on where countries are in the infection curve (people requiring hospitalisation seem to need it about 10 days after they’re infected) and what age and condition people are in. For example, if the population is older or more obese they’re more at risk (the US ranks 16th worldwide by body mass index where Canada ranks 43rd). It’s obviously impractical to include these co-morbidities in our risk model.

And countries also determine cause of death differently: with Italy reportedly attributing deaths to Covid19 where patients tested positive but had pre-existing conditions that more likely killed them, and China reportedly not attributing Covid19 as cause of death if the person died with Covid19 symptoms before their test results were returned.

The Financial Times after consultation with epidemiologists estimated ‘real’ fatalities by simply comparing countries’ fatalities during the pandemic vs 5 year death average for the same period, and concluded that the reported number of Covid19 fatalities might be understated by 60%.

Finally the fatality rate is also likely impacted by how strained the health system is in handling the volume of Covid19 cases. As in Italy, doctors might have to stop giving beds to some groups of patients as hospitals hit their resourcing limits for delivering WHO-recommended Acute Respiratory Distress Syndrome treatments, thereby increasing the number of fatalities. Again, it’s not been practical to include the impact of an overloaded health system on the quality of healthcare in our model, but the last graph shows the numbers of beds being taken up by different age groups.

Animated 'Flatten the curve' graphic by Siouxsie Wiles and Toby Morris - Spinoff.co.nz, CC BY 4.0