Speed over accuracy as a testing strategy can help in better containment of Covid-19: Milind Tambe

Some of the problems we are facing are interdisciplinary challenges. What we find is we can help with modelling and simulation efforts, in understanding the impact of different measures such as opening lockdown, said Milind Tambe.

MUMBAI: Delays in diagnosing Covid-19 cases using the RT-PCR method can offset its effect, a shortcoming that a less-accurate but faster test like the rapid antigen tests can help with, a modelling study done by researchers from Harvard Chan School of Public Health and Harvard John A Paulson School of Engineering and Applied Sciences has found. In a yet to be peer reviewed paper the researchers write that given the pattern of viral load kinetics, they modelled surveillance effectiveness considering test sensitivities, frequency, and sample-to-answer reporting time. These results demonstrate that effective surveillance, including time to first detection and outbreak control, depends largely on frequency of testing and the speed of reporting, and is only marginally improved by high test sensitivity. The researchers conclude that surveillance should prioritize accessibility, frequency, and sample-to-answer time; analytical limits of detection should be secondary.

Milind Tambe, director of the Center for Research on Computation & Society at the Harvard Paulson School of Engineering and Applied Sciences, one of the authors of the study in an interview with Divya Rajagopal, talks about the details of the study, and the role of computer scientists and the use of artificial intelligence in dealing with the pandemic. Edited excerpts:

There is an intense debate on what are the appropriate tests that should be deployed to fight the pandemic. Could you tell us about the modelling study that you and your colleagues did in Harvard about this?
Our joint work was with Dr Michael Mina (epidemiologist and immunologist) of Harvard Chan School of Public Health and other collaborators, where we tried to look at what is the right testing strategy when you on one hand have accurate tests for low viral load but are expensive, like the PCR tests costs $50 and takes much longer to get results, and on the other hand there are rapid antigen tests that are less sensitive. So as universities open up, if you want to test students, what tests do you use? Should you go to accuracy or speed? So, our modelling shows that just a one-day delay in getting back results of the more accurate tests defeats its effects.

So, speed over accuracy is this what the modelling shows?
Yes, so if you do tests even with low sensitivity tests, the less-accurate tests so to speak, and if you get results fast, that is a much better way to go forward. It was interesting because this modelling shows that you don’t have to go for gold standard tests all the time. So, for universities being able to test all the students at high frequency at high cost will be a tremendous expense. In such setups, even though rapid tests can be less sensitive, because they are faster it helps in containment.

In India, too, we have deployed rapid antigen tests and there is a debate on whether we are missing out cases…
Well, so the other argument that epidemiologists, and it is also the point of view of Dr Michael Mina’s, is that when you catch people in sensitive range or low viral load, they might not be infectious, so you might be catching people at the tail end of their infection or when they are on their way to recovery as compared to when they are infected in their early stage. So, a sensitive test might not be valuable if the result comes back after a few days. Here in the US, getting the tests done, and getting back results takes three to five days. But key arguments in favour of rapid tests or point of care like antigen tests are cost and speed. So, I am hopeful that is the way.

How can AI experts help epidemiologists and public health officials in dealing with the pandemic?
Some of the problems we are facing are interdisciplinary challenges. What we find is we can help with modelling and simulation efforts, in understanding the impact of different measures such as opening lockdown. As computer scientists and AI researchers, we can build detailed models on the impact of testing strategies, which can be feasible in various settings. We can also look at vaccine deployment strategies when they come in.

Dr Tambe you also lead the AI for Social Good initiative at Google Research India, could you tell us about the work you do?
There is so much AI talent in India, a couple of years ago I was in IIT Patna in a room full of AI researchers and the main question is can we use this talent towards social good like public health, and other challenges. I see that the opportunity of using AI is exponential in India. So how do we bring these researchers who are hungry to bring their knowledge towards addressing societal needs. So, in 2019 we got computer scientists and not for profit organizations together to address some key societal challenges, we tried to do it on a smaller scale in Google Research India. So, the barriers we see is finding the right kind of match, the other barriers are that data is not readily available, so that is another research opportunity to look at. So, we expect this to do in a larger scale. So there are a lot of exciting opportunities in applying AI for social good and they are not just applicable in certain parts of the world.




More from our Partners

Loading next story
Text Size:AAA
This article has been saved