Pandemics, Data, and Analytics (PANDA) coordinators said there has been a breathtaking failure by COVID-19 modellers whose coronavirus death predictions for South Africa amounted to scaremongering.
Speaking to Biznews, PANDA co-ordinator Nick Hudson said the official modelling of South Africa’s projected mortalities from COVID-19 started at 375,000.
Hudson said this extremely high mortality figure “scared the living daylights” out of President Cyril Ramaphosa.
“When he saw forecasts which entailed a two-and-a-half million shortage of hospital beds – it was that forecast which made us enter into a state of disaster and eventually the lockdown,” he said.
He said that, over time, these forecasts have been significantly reduced to 40,000 mortalities, which they still believe is a drastic over-estimate.
Hudson referred to recent modelling which predicts that South Africa could see about 40,000 COVID-19 deaths by November.
The models also show there will be between 1.2 million and 1.6 million symptomatic COVID-19 cases during peaks in July or August.
“South Africa is likely to see a peak demand for hospital and intensive care unit (ICU) beds between August and September,” the research said.
Despite the reduction in predicted symptomatic COVID-19 cases and deaths, Hudson and PANDA remain sceptical.
“The actual format and structure of these models are wrong and have been proven to be wrong every day. This model class should not be used at all,” said Hudson.
He said PANDA was involved in Health Minister Zweli Mkhize’s recent modelling symposium and made their case as to why the models which are currently used are wrong.
Despite highlighting very specific problems with the COVID-19 forecasting models, they have had no engagement from any of the parties involved.
“In our mind, it is an admission of guilt. They know their models are wrong and not appropriate,” said Hudson.
Two major flaws with current COVID-19 models
Hudson explained there are two major flaws with the COVID-19 models which are used to forecast the impact of the virus in South Africa:
- The forecast of continuous exponential growth which was not seen anywhere in the world.
- An incorrect assumption about the number of people who are susceptible to the disease and dangerous infection.
Hudson explained the way the COVID-19 models work is that you build a complicated exponentiating state model – something that treats people as random variables.
- Exponentiation is a mathematical operation involving two numbers – the base number and the exponent (power). It explains the expected exponential growth.
- The states of the COVID-19 models are typically susceptibility, being exposed to the virus, being infected, and then either recovering or dying.
These models, which are based on exponential growth, have to be calibrated. This is done by making assumptions.
“What an exponentiating growth involves is that each day you get a constant extra percentage more cases or deaths. That results in the rapid escalation of the number of cases and deaths,” said Hudson.
“This has not been observed anywhere in the world. Normally the exponentiating phase, if there really is one at all, is very short and then the growth rate of cases starts dropping off,” he said.
The second problem, Hudson said, is that COVID-19 modellers seem to make an assumption that all people are susceptible to the disease and dangerous infection.
“In the real world, however, the actual story is much lighter. You have 10%, maybe 20%, maybe as high as 25%. That makes a big difference,” said Hudson.
“If you have 30%, 50% or 80% of people in your population who deal with the virus very easily, it makes a big difference.”
He said every population in the world has many people who are simply resistant to COVID-19. “That is a robust finding. You will not find a single data point which refutes it.”
Predicted death rate
While the number of projected mortalities were lowered from 375,000 to 40,000, Hudson said this is still much higher than their models.
“Our observation is that estimates of South African deaths in the 40,000 plus range are outlandish,” Hudson said in an open letter.
He added that the model used to predict the number of infections and mortalities in South Africa has not been made public, which is deeply problematic.
“It is, in our view, scandalous that the models the South African government has been presented with could ever have produced a number of 351,000 fatalities,” Hudson said.
He added that the adjustment of the best-case scenario – i.e. with a hard lockdown – from 89,000 fatalities to 40,000 without any explanation is equally scandalous.
Hudson said the lower-bound estimates profoundly overestimate the magnitude of the COVID-19 epidemic in South Africa – by a factor of at least four – and the timing of the peak by several months.
He said their modelling shows South Africa will have around 18,000 deaths based on the age distribution in the country.
Hudson also dismissed concerns about the rapid rise in confirmed COVID-19 cases in recent weeks.
He argued that the government is not reporting the number of COVID-19 infections in South Africa, but rather the number of detections.
He said there are can be over a million COVID-19 infections in South Africa – mostly undetected – where most of these people will have mild or no symptoms and simply get on with their lives.
Considering the inflated COVID-19 forecasts, PANDA urged Ramaphosa to “not to risk further harm to the economy by opting for a gradual reduction to Level 2, but instead to reduce the lockdown to Level 1 as the next step”.
Nick Hudson Biznews Interview
PANDA’s Open Letter to President Cyril Ramaphosa
Dear Mr. President,
As South Africa enters Level 3 of the Coronavirus disease 2019 (COVID-19) lockdown, we write to you with a deep sense of concern and dire urgency.
You may recall from our previous communications that Pandemic ~ Data and Analytics’ (PANDA) technical team brings to bear knowledge from the fields of actuarial mathematics, economics, data science and medicine, amongst others.
PANDA aims to assist stakeholders in their efforts to make the best-possible informed decisions. The central premise of our approach is that the impact of lockdown must be understood and quantified in order to make informed decisions. We need to understand not just what the impact of lockdown will be on COVID-19, but what the impact will be on livelihoods and the lives that are inextricably linked thereto.
Our estimates of the long-term economic damage and loss of life that South Africa could still endure—appended to this letter and previously shared with Your Excellency’s office, colleagues on the National Coronavirus Command Council (NCCC) as well as other spheres of government—have made it increasingly clear that a more rapid de-escalation of the COVID-19 lockdown is required to save our economy, the livelihoods of millions of South Africans and ultimately lives. We agree with Your Excellency’s remark that lockdown has become a “blunt tool”, and would add that the blunt edge of that tool strikes the poor hardest.
We note that during the virtual forum with the South African National Editors’ Forum (SANEF) held on Sunday, 31 May, Your Excellency mentioned that:
“(The medical scientists who have been advising the NCCC) also said once we went through Level 5 and Level 4, they also said the lockdown has served its purpose. In fact, what they were also advising was that you could quite easily go to Level 1.”
Our scientists concur with them.
We note that the World Health Organization (WHO), whose current estimate is 23,661 South African deaths from COVID-19, recommends a gradual removal of lockdown conditions that weighs against this. It is our strong view that this advice is misguided. Each of the preconditions suggested by the WHO are unachievable by almost all countries and is neither based on science, nor adequately takes into account the negative impacts of lockdown. Indeed, it has become abundantly clear in our country as in others that lifting lockdowns does not result in the feared resurgence of cases and deaths that the WHO and various modelling teams have predicted.
The WHO’s recommended guidance is a broad framework which fails to take into account the specific circumstances of any individual country and as such is entirely impractical in the South African context. The lockdown is increasingly unpopular in South Africa and with 230,000 of our citizens having already been charged for contravention of lockdown regulations, it is instilling a culture of lawlessness.
Another question weighing on Your Excellency’s mind is likely the projections of the various modelling teams. At Minister Mkhize’s COVID-19 Modellers’ Symposium of 21 May, we pointed out that all of the models presented were inconsistent with evidence that had emerged internationally, and we invited engagement with these teams to help them correct their models. Even as contradicting evidence has continued to mount, making the case against continued deployment of these models overwhelming, we disappointingly see no material effort on behalf of the modelling teams to apply the basic principles of the scientific method or to lift the veil of secrecy that shrouds their work.
It remains our opinion that even their lower-bound estimates profoundly overestimate the magnitude of the COVID-19 epidemic in South Africa (by a factor of at least four) and the timing of the peak (by several months). We advise caution in premising the allocation of scarce resources on these models, which appear to us to represent pseudoscience.
We remain deeply sensitive to the onerous position that policymakers occupy. The PANDA team remains willing to support decision-making and serve at Your Excellency’s pleasure.
We urge Your Excellency not to risk further harm to the economy by opting for a gradual reduction to Level 2, but instead to reduce the lockdown to Level 1 as the next step.
We trust our request will receive due consideration and would appreciate an acknowledgement of receipt together with a possible date at which we can present our critical findings.
Open letter to Prof. Juliet Puliam of SACEMA
Dear Professor Puliam
Following Thursday 21 May’s symposium with the Minister of Health, we set out below our comments.
At the outset, we note that, as at the date of this letter, your model has not been made public. This is, in our view, deeply problematic. By insulating your model from scrutiny, you forego the valuable contributions that could be made to refining the model or assessing your model against others. Given that the model serves a public interest imperative, and that decisions affecting all aspects of citizens’ lives are being taken on the basis of the predictions your model generates, the public should have insight into its workings. In this regard, the experience of other countries cannot be ignored. Imperial College’s SEIR model (which predicted over 500,000 deaths in the UK and over 2,000,000 deaths in the United States) was released publicly and experts set to work improving what was described as “totally unreliable” and “a buggy mess” in a manner that clearly benefited decision-making.
Before even getting into the details of model structures and parameters, our salient observation is that estimates of South African deaths in the 40,000 plus range are outlandish:
1. International population mortality, infection fatality and case fatality data all paint a picture of a disease whose severity is profoundly age-related. Current global death rates vary from 0% for children under 10 to 0.2% for people aged 10-39 years, up to 14.8% for people over 80 years of age. Failing to reflect the age distribution of South Africa’s population in forecasts is a profound problem and renders any model that does not do so inappropriate.
2. It has been clear for some time that serology studies in countries where fatalities have peaked and begun to subside return infection percentages that don’t exceed 30%.This could be either due to the fact that the majority of humans are not susceptible to the disease or enjoy significant t-cell immunity that prevents them even from generating antibodies when infected. In either case, models need to take into account the lower attack rate.
When we age-adjust relative to the very worst-hit nations in the world, we are challenged to understand how any model for South Africa could reasonably produce a death forecast of more than 10,000, particularly given that the observed age distribution of deaths in South Africa so far is almost precisely what would arise if the country did not embed any sui generis features (such as HIV or TB as comorbidities).
Our article of Thursday last also highlighted seven serious falsifications of the corollaries of standard SEIR models, any one of which would be material enough to void their legitimate application.
One of the consequences of models entailing these large fatality forecasts is that they encourage policymakers to sustain lockdown strategies that embed their own humanitarian crisis. PANDA may have been first to attempt to put a number to the mortality implications of the economic devastation, but our results have been picked up and echoed elsewhere, here and abroad. The assumption in modelling circles seems to be that low fatality estimates are intrinsically more dangerous than high estimates. This is patently and demonstrably incorrect because lockdown strategies that are justified by the high fatality numbers the models are generating also have health impacts. Excessive estimates of fatalities produced by your model will result in harm being suffered.
It is, in our view, scandalous that the models the South African government has been presented with could ever have produced a number of 351,000 fatalities and that they have been adjusted from a best case scenario (i.e. with hard lockdown) of 89,000 fatalities to 40,000 without any explanation.
We therefore regard it as being in the public interest that you should immediately describe the parameters used for your model at all points in the context of the above observations. In the absence of a sound hypothesis for why any of these observations should not be considered highly relevant for South Africa, we expect that you will revise your model with great haste and make public the results at the nearest opportunity.
We trust that you will respond positively to our request to release your model publicly. In case you choose not to do so, we attach to this letter a request for access under the Promotion of Access to Information Act.