Tech weapons to fight the coronavirus

OVER the past few months we experienced a series of Covid-19 outbreaks replicated across countries in almost identical ways: an initial phase with few infections and limited response, followed by a takeoff of the famous epidemic "curve" accompanied with a country-wide lockdown to "flatten the curve".

But once the curve peaks, governments ask what US President Donald Trump called "the biggest decision of my life" - when and how to manage de-confinement.

As a famous quote, often attributed to Einstein, says, "Insanity is doing the same thing over and over again and expecting different results".

The repetition of Covid-19 patterns indicates information weaknesses in our global health system that we may wish to fix. While attention is given to sharing critical information across countries - in particular from China - little has been said about how Covid-19 could have been better managed leveraging advanced data technologies that have transformed businesses over the past 20 years.

A "secret sauce" behind this transformation is extreme personalisation. Data-driven firms, from "Big Tech" to financial services, insurance, retail, media, and even healthcare, make personalised recommendations across purchases, pricing, risk and credit using predictions from models trained on customer data these firms amassed.

A technique behind many such predictions is known as classification, which splits individuals into several groups: for example those who will likely do something and those who will not.

The same could work for pandemics. Train a model to classify all individuals into either a high or a low clinical risk group (for example, would an ICU bed be required should they get infected?). Unlike current policies, use the predictions and information from the model to confine those with high risk, while impose lighter limitations to, and deconfine, those with lowest risk even if they may be infected - to also run the economy.

For Covid-19, for which data indicate probably about 90 per cent of the infected people do not experience severe (or any) symptoms, an important characteristic of this virus, a perfect severity-if-infected prediction test would allow the release of all these individuals from confinement - even if they "risk" infecting each other, as long as the remaining high risk individuals are still confined. Finally, do this in stages by varying the risk classification threshold - also gradually building herd immunity while limiting mortality.

This approach is different from, and complementary to, the current Covid-19 best practices based on diagnostic, antibodies, or other medical tests - also enabling a form of personalised policies, using such tests.

For example, diagnostic tests are used to support test-track-isolate type policies, used successfully in countries like Singapore, Taiwan and South Korea, where infected people and their traced contacts are quarantined, independent of the severity of their - and anyone they contact - symptoms or predicted symptoms-to-be.

Personalised clinical risks approach

Instead, with a personalised clinical risks approach, people with predicted mild symptoms (and their contacts) are not necessarily confined, depending on their clinical risk score and availability of healthcare resources to manage possible test errors, following standard medical liability and risk management best practices - hence likely enabling more deconfinement as the majority of Covid-19 cases (estimated to possibly be more than 90 per cent) have indeed mild symptoms.

But it could work even when such medical tests are unavailable at large scale or not accurate enough. This approach is, like for businesses, not about managing based on knowing who is, or has been actually infected; it is about managing based on the predicted consequences if someone and their contacts get infected, considering medical liabilities. Unlike medical tests which are scarce, expensive and slow to deploy, this approach is digital, fast and easy to scale.

Models for personalisation require data which may not be easily available. One approach to facilitate the requisite data is to ramp-up nations' own health data capabilities - implement electronic medical records, detailed electronic census.

This approach may be limited as it would take time before having any such data linked to a virus's symptoms. A conceptually better approach is to realise that, as key underlying data (biological and physiological) are nearly identical across countries, the goal should be to enable models development across countries - an extension from only sharing a few model parameters, such as the famous R0, today.

Consider how this could have played out for Covid-19. When Covid-19 emerged in Wuhan, data was initially non-existent; the lockdown approach therefore made sense, equivalently, model-based personalisation was infeasible due to lack of data: shut down the cities, implement one-size-fits-all social distancing, and monitor closely, making no major exceptions.

However, - and critically - collect all available data to train risk models to enable personalisation for both the pandemic epicentre and for the later-hit countries.

Such an ideal world can be achieved with the right technologies and policies. Today's machine learning and AI capabilities can already provide components needed to derive insights from data sources across countries or the ability to transfer learnt information from one region to another - some we have developed and used ourselves.

The existing pre-Covid-19, policies covering data privacy and cybersecurity largely prohibit leveraging these technologies for pandemics requiring data localisation or the prohibition of data sharing.

The policies largely do not differentiate between data used or derived at the different stages of modelling. Some policies retain provisions for data connectivity and exchange but may require legal or absolute certainty in the inability to deanonymise individuals, resulting in a de-facto prohibition to share models (and data). Intentionally or not, the last few years the world moved towards data isolation.

Tables may need to turn. Covid-19 seeded an environment for governments to seek a framework to facilitate mutualisation of learning and sharing of data and models, possibly only enabled at exceptional times of a pandemic "war".

During such times, many people will likely also be willing to exceptionally and temporarily provide their data, through appropriate and secure channels, for training models that can guide policy decisions - and their own de-confinement or allocation of medical resources.

This would allow the fast, scaleable and digital approach for managing pandemics, one based on the main technological innovations of recent decades: machine learning and AI. These innovations created trillions of dollars of business value. Now is the time to use them to save lives and support "the biggest decisions of our lives".