How AI can help eradicate COVID-19 and save the world’s economy

Original article can be found here (source): Artificial Intelligence on Medium

How AI can help eradicate COVID-19 and save the world’s economy

The global pandemic of coronavirus has led to lockdown of cities and nations around the world. Some form of lockdown and social distancing was enforced in cities and nations from Wuhan to London, New York to Los Angeles and Italy to India. As in 24th of March, one quarter of the world lived in some form of lockdown [1].

Lockdown has proven to be an effective solution to stop the exponential growth in the number of patients with coronavirus as reported in Wuhan and China [2]. However enforcing lockdown and social distancing has led to significant economic impact. One day after the prime minister told our citizens to practice social distancing, the United Kingdom announced to pour £350 billions (15% of our GDP) to shore up the economy [3]. Germany unveiled €750bn rescue package to save their shrinking economy [4]. On the other side of the Atlantic, the United States passed a $2 trillion stimulus bill, the largest emergency aid deal in history or 10% of their GDP, to rescue their economy [5, 6, 7]. In Asia, the world’s second biggest economy, Japan, is also working on a plan to lockdown Tokyo and to hand out $500bn to support households that are impacted by coronavirus outbreak [8, 9].

As a consequence, the world’s biggest economies are predicted to go to recession or deep recession, and this time it could be even worse than the financial crisis in 2018 [10, 11, 12]. The results of an economic recession are not only surge in unemployment that leads to people losing home but also degraded healthcare that indirectly causes more deaths. According to a Lancet study, the financial crisis in 2008 caused 500,000 extra cancer deaths [13]. Many other studies show strong correlation between unemployment or GDP per capita with life expectancy [26, 27, 29]. There are also unanswered questions about how sustainable lockdown and social distancing are? How long we can stay under lockdown and keep social distancing when vaccine likely takes at least 12 months to be available [24]? Will virus bounce back after lockdown is released? If so, what is our exit strategy?

The United Kingdom’s original plan was to apply different measures to “flatten the curve” and let the virus rip through the less vulnerable population to build up the immunity. Once an approximation of 60% of the population are immune, they then will achieve a state called herd immunity (Figure 1) [15]. At this stage enough population are immune and are not infectious. The virus is contained and finally eradicated [16]. The strategy is based on collected data showing that the virus is significantly more dangerous to people above 60 and people with underlying conditions [17, 18] and the assumption that recovered patients are immune from the virus (no strong evidence that this is not the case yet). However the strategy encountered massive criticism from critics that it will put our national health service (NHS) under stress that results in more deaths [19]. A research from Imperial College also shows that without social distancing, there will not be enough Intensive Critical Units (ICUs) at peak time [20]. The report has driven the UK government to immediately impose stricter measures to enforce social distancing to further flatten the curve and ensure the NHS can cope with the surge of patients who will need beds with ICUs [21].

Figure 1: How a virus spreads with different percentages of population are immune [16].

That arises a question that can we avoid lockdown and social distancing that vastly damage our economy by better selecting whom to build up the immunity (assuming that recovered patients are immune for a prolonged period of time)? By better selecting ones who unlikely need hospitalisation if infected with coronavirus, we’ll avoid putting stress on the NHS meanwhile build up the herd and eradicate the virus faster. The UK government’s original strategy is to enforce social distancing for people older than 70 but this still results in a massive need of ICUs at peak [20]. However we have also learned that people with underlying conditions such as hypertension, cardiovascular diseases, cerebrovascular diseases, diabetes, hepatitis B infections, chronic obstructive pulmonary disease, chronic kidney diseases, malignancy and immunodeficiency are significantly at risk [18]. Additionally patients reported having two or more co-morbidities are at greater risk than those who had a single comorbidity, and even more so as compared with those without [18].

Figure 2: Italy’s coronavirus fatalities (%) grouped by previous medical conditions [28]

Several studies show that men are more likely to die because of coronavirus than women [21], but that might be because men also might have higher chance of having mentioned underlying conditions. One study on blood group suggests that people with blood group A have a significantly higher risk for acquiring COVID-19 compared with non-A blood groups, whereas blood group O has a significantly lower risk for the infection compared with non-O blood groups [22]. A group from China further look at chest CT images in order to train Deep Neural Networks to predict which mild patients will later deteriorate into severe stage with high accuracy [23]. There might be many more studies looking at different patient’s data to predict who have higher risk of having severe symptoms caused by coronavirus. Some I can think of that might be worth looking at are our DNA, smoking habit, local air quality or whether their family members with coronavirus had severe symptoms.

Figure 3: Representative cases without malignant progression (A, B) and with malignant progression(C, D). A 32-year-old male with symptoms of fever, cough, and dyspnoea. CT1 (A) shows ground glass opacity lesion (yellow stars) in the left upper lobe and some fibrosis (red arrows) in the right upper lobe. The CT image of 3 days follow-up (B) shows the prior lesions shrunk in size and decrease in density. A 64-year-old male with symptoms of fever, dyspnoea. CT1 © shows ground glass opacity lesion (yellow stars) in bilateral upper lobes. The CT image of 3 days follow-up (D) shows the prior lesions progress to consolidation (red stars) [23].

Once all these data are available, it is possible that we can train an AI model that can give much better predictions of who will need hospitalisation with coronavirus than solely relying on age and co-morbidities. The governments can use the trained AI model to score their citizens with a fit score. Ones with high fit scores are free to go to work and socialise as usual while ones with low fit scores can practice social distancing. By gradually releasing groups with highest fit scores, nations can reach herd immunity stage quickly without sacrificing a big chunk of their economy. The fit score can be kept privately to each individual to avoid privacy breach.

Meanwhile building AI models can be quick and easy for many experts, I had difficulty in collecting data that are useful enough to build AI models. The lack of quality data has limited worldwide enthusiasts in quickly building potentially helpful solutions to assist governments to tackle the pandemic effectively. I hereby call out the administrations in charge to foster sharing their collected data publicly as soon as possible while keep the patients’ personal identifiable information and other sensitive information safe. By doing so, we could quickly come up with solutions that potentially save us trillions of dollars and, hence, save thousands or even millions of lives.