AI shows risk of third Covid wave is low
• But algorithm does not rule one out, Wits expert warns
The risk of a third wave of coronavirus infections in SA in the coming fortnight is low, according to an artificial intelligence (AI) algorithm designed by a consortium led by Wits University. “But these algorithms are not crystal balls,” said Wits Institute for Collider Particle Physics director Bruce Mellado.
The risk of a third wave of coronavirus infections in SA in the coming fortnight is low, according to an artificial intelligence (AI) algorithm designed by a consortium led by Wits University.
“The data is telling us the probability of a third wave is low, and is unlikely within the next two weeks, but these algorithms are not crystal balls,” said Wits Institute for Collider Particle Physics director Bruce Mellado, who is a member of Gauteng premier David Makhura’s advisory council on Covid-19.
“SA has done extremely well in controlling and bringing [infections] down to relatively low levels twice in a row. This is really commendable, but this does not mean that a third wave will not happen; absolutely not,” Mellado said.
The government is closely watching for signs of an increase in new infections, anticipating a third wave that could be more severe than the last, as has been the case in Brazil and several European countries. In March, finance minister Tito Mboweni said a third wave could disrupt the government’s attempts to stabilise its finances, and health minister Zweli Mkhize has consistently said the government was bracing for an increase in cases towards winter.
All the key Covid-19 indicators tracked by the government, such as new infections, hospital admissions and deaths have steadily fallen since their peak in early January, and are holding steady at relatively low levels. For example, the rolling sevenday average of new cases stood at 928 on April 11, compared to the peak rolling seven-day average of new cases of 19,042 on January 11. But officials are watching for signs of an increase after travel and religious gatherings grew at the Easter weekend in early April.
The AI algorithm designed by Wits, iThemba Labs, the Gauteng provincial government and York University in Canada predicts future daily confirmed cases, based on past infection data, mobility data from Google, Apple and Facebook, and epidemiological parameters such as the period during which the virus is transmissible. The researchers publish their AI-based analysis online, and update it daily.
Early detection algorithms such as the one designed by Mellado and his team can never be 100% accurate, but they can provide policymakers with early warning of an impending surge, he said, creating valuable time to prepare the health system and raise public awareness about the importance of nonmedical interventions such as wearing masks, physical distancing and hand hygiene.
“If we can postpone the third wave as long as possible, the pace of vaccination will pick up. We will be in a very different situation if we can vaccinate vulnerable people before the third wave. The longer we can postpone it, the more lives we can save. But we need to know when it is coming so we don’t waste resources,” he said.
The AI-based algorithm will be published in a preprint server later this week.
Alex Welte, a research professor with the SA Centre for Epidemiological Modelling and Analysis, said it is too soon to tell whether the increased social interaction and religious gatherings at the Easter weekend have seeded a surge in coronavirus transmission, or how profound an effect that might have on the trajectory of the pandemic in SA.
“It takes several weeks to get a substantial uptick,” he said.