Without a crystal ball, Covid-19 modellers work with what they can
“I wish I had a crystal ball on my desk.”
These words by the head of the advisory committee on Covid-19, Prof Salim Abdool Karim, this week encapsulate what many are thinking amid the Covid-19 pandemic.
For modellers it has been a sleepless journey of using solid data to try to predict a feasible outcome.
As highlighted in a national health department briefing on Thursday, every country is grappling with an unknown enemy.
In diplomatic terms, minister Zweli Mkhize described how modellers are dealing with an equation that has more variables than any lay person could imagine, and that “no model can get it 100% right” in the face of such a sudden crisis.
Yet their models are attacked by armchair critics who jump up and say “but what about … ”, or by fellow modellers who are working with a different set of data.
In Abdool Karim’s words, “We are having to build the ship as we’re sailing it”.
Even so, the response to the crisis can only be based on what most of the models are saying, and that’s the bitter truth of a Covid-19 wave set to engulf the country, swallowing up human lives, hospital resources, and our economy in the process.
On the same day that Mkhize hosted the modelling webinar, SA recorded more than 1,100 cases in the previous 24 hours, bringing the country close to the 20,000 case mark.
Thirty South Africans died in the same period.
Mkhize had invited several high-end modellers to present their findings in the open webinar.
Models included predictions that:
- We are going to run out of ICU beds, and it could be as early as June;
- The situation would be a lot bleaker without lockdown: It has reduced the transmission of Covid-19 in SA by between 40 and 60%;
- Critical patients, on average, require six days in a hospital ward and another 10 in ICU if recovery happens;
- Should they die, it usually occurs after six days in ICU; and
- By July, 16.4 million N95 masks and 1.3 million testing swabs will be needed.
According to Stellenbosch University’s Prof Juliet Pulliam, speaking on behalf of the South African Covid-19 Modelling Consortium, lockdown has worked and yet we still face a shortage of ICU beds, and probably very soon.
According to the consortium, “the extension of the lockdown to five weeks bought us critical additional time to ramp up community testing and to prepare”.
Optimistically, lockdown reduced transmissibility by as much as 60%, or 40% at the lowest.
While there are many variables, there was one certainty: “Under almost all scenarios, hospital and ICU capacity will be exceeded.”
Pulliam says the modelling doesn’t capture super-spreading events like funerals (on the downside) or the positive effect, like behavioural change, which could reduce transmissions substantially.
Barry Childs, speaking on behalf of the Actuarial Society of South Africa, also emphasised the fact that with so many variables at play it was difficult to have a definitive answer.
The organisation had experienced wide-ranging and harsh feedback, but the overriding difficulty was “having to make such significant decisions amid such uncertainty”.
Despite these unknowns, it was crucial to “produce a model for use by the broader profession that is sound in its basis and provides guidance in its application”.
Models are not meant to predict the future perfectly - yet they’re still useful.
Deloitte has been focused on identifying the level of PPE, ventilators and hospital admissions that will be required as we move towards the peak.
Their modelling shows how the demand for such resources grows exponentially with each new week.
In June, we would need 10,197 ICU beds, but by August that figure would approximately double to 20,482.
Like the other modellers, Deloitte’s Ashleigh Theophanides says their models “are recalibrated every few weeks” and that their current estimates of deaths sit at about 40,000, but “the work the department of health has done to set up field hospitals could have a significant impact on the expected increase in fatalities”.
American biomedical mathematician Lester Caudhill, from the University of Richmond, makes the same point as Mkhize.
“Models are not meant to predict the future perfectly, yet they’re still useful,” he told The Conversation.
Modelling the spread of Covid-19 was so challenging because “modellers typically use data from prior outbreaks of the same infection to create their model”.
This worked well for infections like influenza “because scientists have decades of data that help them understand how flu outbreaks progress”.
In an ironic twist of fate, the last thing we have this year with Covid-19 is 20-20 vision.