Risk Thinking for Covid’s Radical Uncertainty

If nothing else, Covid-19 has taught us a hard lesson in radical uncertainty. While one scientific paper back in 2007 warned ominously of a coronavirus “time bomb” from China, no one predicted the devastating health crisis and unprecedented macroeconomic shock that blew through 2020 like a hurricane.

More than six months on from the outbreak, cases are still on the rise in several developed countries. And without evidence-based guidance, policymakers around the world struggle to weigh up the cost of intervention policies. Too strict and they tank the economy. Too mild and mortality skyrockets as ICUs flood.

Policymakers are paralysed by the radical uncertainty of the pandemic and the unknown consequences of their decisions. What percentage of the population will obey a strict lockdown? How many will wear masks if asked? When will a vaccine be available, and will people take it? How many people will lose their jobs under a new lockdown?

Overcoming Radical Uncertainty

To approach this problem, policymakers must learn how to risk think. They must learn how to adaptively strategize in the face of a constantly changing threat.

https://covidwisdom.riskthinking.ai

This is where COVIDWISDOM comes in. It’s a comprehensive and science-driven decision support tool that weighs up both economic and epidemiological risk factors to assess the impact of potential intervention policies.

Combining information from epidemiological and economic models, it estimates the effects of seven different intervention levels on public health metrics (like cumulative fatalities and peak demand for ICU beds) and economic metrics (like employment and per-capita GDP).

Importantly, it’s highly granular. Currently focused on Canada, it goes right down to the individual health units in each province. By providing daily updated region-specific epidemiological information and correlating this with economic data at that same level, it presents highly localized cost-benefit analysis for individual public health officials tasked with assigning an intervention policy.

This allows for decision makers to chart a strategy that is appropriate for their specific circumstances, removing the reliance on the dichotomized, blanket interventions of either total nationwide lockdown or complete non-interference.

Using the Right Model

When it comes to epidemiological modelling, existing tools are focused on too great a scale. The widely regarded DELPHI model developed at MIT forecasts infections, hospitalizations, and deaths. But it does so on a national and global scale. So things get a bit weird when you try to apply it on a micro level.

For COVIDWISDOM, Riskthinking.ai has made significant extensions to the DELPHI model in order to get more accurate predictions at smaller scales. Riskthinking’s algorithm not only takes more data into account, but it is trained using hyper-parametrization to account for Canada-specific variations in the data. Where traditional models are trained on a subset of data and create a correlation matrix that stays static, the Riskthinking.ai model constantly reads new data and optimizes itself — in other words, it’s highly adaptive.

Importantly, this specialization doesn’t prevent COVIDWISDOM from being transferrable. You can plug its architecture into just about any bank of health data with a bit of formatting, and it will start churning out highly accurate results in minutes. It was recently trialled in Connecticut and up and running so quickly that Governor Lamont was able to present it in a state assembly to discuss intervention policy.

Correlating Health Data with Economic Data

One of COVIDWISDOM’s biggest leaps forward in terms of decision making is that it can correlate epidemiological data with economic impact. Select an intervention on the app and you can see both the change in cases and the impact on employment. But how, exactly?

Largely, this is about finding the right kind of data. And knowing where to look is just as important as knowing where to avoid looking. Historical data is almost inexistent, and where it does exist it’s practically useless. When was the last time a pandemic like this happened? And how much applicable data do we have from it?

When strategizing under radical uncertainty, we must turn to forward-looking data. And we find that on the ground. To understand people’s behaviour and how they would react to interventions, Riskthinking.ai did a survey of 300,000 Canadians back in October that asked all manner of questions. Do you wear a mask when you go out? Do you see others wearing a mask? Are you socially distanced? Are you retail shopping? And the responses help build up correlations and uncover the uncertainty in the different intervention scenarios.

Of course, you have to take this data with a pinch of salt. While there’s a lot of theory involved in asking the right kind of questions and knowing how to interpret the answers, inaccuracies inevitably emerge. But that’s part of risk thinking. In a situation as rapidly evolving and as wildly unknowable as a global pandemic, there is no perfect strategy — no perfect methodology for predicting the future. You have to do the best you can with the tools available.

What COVIDWISDOM manages to do so successfully is not guarantee an outcome, but help policymakers chart a course through the unknown with the best data available. And that’s the essence of risk thinking.