Expert estimates of South African COVID-19 mortality continue to plummet, with the Actuarial Society of SA (Assa) this week slashing by almost in half an earlier estimate of 48-88,000 deaths by end August, writes MedicalBrief. Mortality modelling projections for SA have ranged between 40-175,000 by the end of the year.
Assa’s revised model now puts the probable range at 27-50,000. ASSA’s new model is fitted to death data from the National Institute for Communicable Diseases (NICD), but it states that the official death count is most likely understated. This week SA’s official COVID-19 death toll edged past 15,000 with the 7-day rolling average at 123 deaths daily.
SA government policy regarding the COVID-19 response has been based on projections compiled by a consortium of experts under the aegis of the NICD. In June, the NICD model predicted 40,000 deaths by end November, in terms of its “optimistic” scenario.
Pandemic Data Analysis (Panda), a private sector group of actuaries, statisticians and economists, dismissed NICD earlier estimates as “outlandish”, writing in a May open letter to the NICD demanding that the model be made available for public scrutiny: “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.”
In August Panda wrote a scathing letter to the SA COVID-19 Modelling Consortium suggesting that the official consortium reconsider its methods, as its official model “has now departed too far from reality to be useful”.
Business Day reports Assa as saying that the earlier version of its model had produced a limited range of scenarios, and many people had focused on the higher death estimate of 80,000. “It was intended to show the sensitivity of the model to key assumptions, but people took it as the worst-case scenario,” said Assa’s COVID-19 modelling working group co-ordinator Barry Childs, who is joint CEO of Insight Actuaries.
The newest Assa model has been updated to allow a user to be able to input a range of assumptions about key parameters, such as the extent to which the population is susceptible to the disease, and the extent to which they may be asymptomatic.
In an Assa media statement, Lusani Mulaudzi, healthcare actuary and president of Assa, says that the updated model takes into consideration emerging research as well as the effects that the lockdown had on the spread of the Coronavirus and mortality rates. Modelling the mechanics of Covid-19 remained complex as the science of the virus and disease is still evolving and data is incomplete and inconsistent. He adds that even “hard data” such as mortality rates is incomplete, whether comparing provinces or countries, due to differing reporting standards.
“Due to the high levels of uncertainty, it remains impossible to present a consensus view on expected Covid-19 related mortality numbers. The Assa COVID-19 working group will, however, continue to refine the model in an effort to provide actuaries with a useful tool to better understand the dynamics of the disease.”
Mulaudzi says the true number of deaths probably lies between the NICD figures and the South African Medical Research Council (SAMRC) reported excess deaths. He says future work on the model would look at using adjusted death rates.
One of the key changes in assumption from the first version of the model is around the levels of susceptibility to the Coronavirus. While the original version assumed that everyone was equally at risk of contracting the virus and developing Covid-19, the revised version of the model allows for different contact and infection rates (heterogeneity and susceptibility).
“The reality is that not everyone comes into contact with the same number of people, and not everyone is equally susceptible to becoming ill after coming into contact with the virus.”
Therefore, while the baseline scenario of the first version of the model assumed that 75% of the population would remain asymptomatic, the updated version introduces a range of percentages of the population contracting the Coronavirus without developing symptoms. “We had to accommodate widely differing views from actuaries involved in the modelling process,” explains Mulaudzi.
The excess deaths report published weekly by SAMRC provides statistics of all natural deaths per week of people older than one year. Mulaudzi explains that the differences between excess deaths and confirmed Covid-19 deaths by province together with variations in the Case Fatality Rates by province suggest that there is indeed significant under reporting of Covid-19 deaths, particularly outside of the Western Cape. He adds that the precise degree of under reporting is difficult to determine given the quality of data and reporting delays.
While some aspects regarding the Covid-19 disease have become clearer since the first version of the ASSA model was released in April, many other aspects remain unclear. Material uncertainties include the following:
- There is no certainty around the true infection fatality rate by age. Early indications are that in South Africa the mortality experience is heavier than the international baseline in the 40 to 70 age range.
- There have been major developments in effective treatment options since April, improving survival of Covid-19 patients treated in hospital. However, the extent of improvements and the timeframe remain unclear.
Mulaudzi says 30 different scenarios were modelled using an expanded range of assumptions provided for by the updated version of the ASSA Covid-19 Model to show to actuaries using the model the sensitivities to key assumptions. Projections have been made to the end of 2020, but the model can also be used to project to the end of 2021.
The expanded range of assumptions includes the following:
- Proportion of asymptomatic infections set at 35%, 50% or 75%.
- Five different levels of susceptibility to infection ranging from 25% to 100% of the population being susceptible to the virus.
The two scenarios represent two possible narratives about the progression of the virus in South Africa. Scenario 1 is consistent with a narrative of high asymptomatic proportion of the population and low levels of interactions between infected and uninfected people. Scenario 2 is consistent with a narrative of high proportion of non-susceptible people, moderate proportion of asymptomatic people and moderate levels of interactions between infected and uninfected people.
This scenario assumes low infection rates due to lockdown and other non-pharmaceutical interventions with 75% of those infected remaining asymptomatic. In this scenario the number of deaths by the end of the year would be around 27 000, with an infection fatality rate of 0.28%.
The mortality per million lives would be 481. Currently South Africa’s death rate per million people is 249 compared to 570 in the U.S. and 612 in the U.K. Belgium recorded a level of 855 deaths per million lives on 6 September 2020, which is among the highest for a country with a population of over a million people.
This scenario assumes that a large proportion (60%) of the population is not susceptible to the virus, but that only 35% of those who do get infected are asymptomatic.
With this scenario, the number of deaths by the end of the year will be approximately 50 000, with a comparatively high infection fatality rate of 0.49%. The mortality per million lives would be 876, which would currently be amongst the highest in the world.
Mulaudzi comments that the excess mortality figures by year end could come in close to this number. The variance in the number of deaths under the two scenarios above prove how difficult it is to come up with an accurate model given the high level of uncertainties.
The table below summarises the scenarios described above. It is important to note that the cumulative number of infections* in the table below cannot be compared to confirmed number of infections given the low number of tests being conducted countrywide due to testing limitations.
|By the end of 2020||Scenario 1
|Cumulative no. of infections*||15 million||10 million|
|Cumulative no. of deaths||27 000||50 000|
|No. of deaths per million lives||481||876|
Assa documentation Full Business Day report (Subscription Required)
See also:Previous ASSA and Panda estimates Panda open letter to Ramaphosa