The Coronavirus outbreak has called into question the ability of governments in predicting, containing and even preventing epidemics. AI and big data can be the answer
In 2019, the Global Preparedness Monitoring Board said that the world was not prepared for the next big pandemic, which could potentially spread between continents in 36 hours and kill up to 80 million people. It pointed out that the climate crisis, global migration and humanitarian conflicts were all breeding grounds for newer outbreaks. The Coronavirus took us by surprise, simultaneously pointing out that systems of the past decade might not be effective in tackling challenges that are taking novel shapes and forms. With one of the largest populations and widening socio-economic divides, India is vulnerable to contagions. Experts claim that this susceptibility hasn’t changed much in the last 100 years, when India was one of the countries worst hit by the Spanish Influenza in 1918. We have been struggling to contain influenza and encephalitis outbreaks for the last four decades. There is a pressing need to overhaul how we respond to outbreaks, both globally and in India, keeping technology at the centre. Coronavirus’ example itself shows that it was two Artificial Intelligence (AI)-enabled enterprises, BlueDot and Metabiota, that provided the key and life-saving insights into the containment of this viral eruption. By using Natural Language Processing (NLP), BlueDot sifts through and analyses over 10,000 articles in 65 languages, including foreign news reports, plants and animal disease networks and official proclamations, to issue warnings. It was able to catch and warn about the impending outbreak days before the official announcement was made. Similarly, Metabiota applied AI to the used travel itineraries and flight patterns to determine the likely spread of the disease. It was accurate insofar as it predicted that South Korea, Japan, Taiwan and Thailand had the highest risk of the viral outbreak.
AI for comprehensive disease surveillance: As epidemics spread in phases from introduction to amplification and finally contained transmission, the response and sequence of interventions flow accordingly. Even before diseases erupt, a public health system attempts to anticipate new and re-emerging disease through early detection. Thereafter, the response strategy changes to containment, control and mitigation, followed by re-focus on eradication. Creating a technology-based intervention into pandemic management also needs to consider the response strategies commensurate with the stage of epidemical evolution.
However, technologies have a critical and more successful role to play in some phases of the response cycle as opposed to others. One such is prediction and early warnings, for which integrated disease surveillance programmes exist both in the public and private sectors. These surveillance strategies include Event-Based Surveillance (EBS) and Risk Modelling. EBS systems use unstructured data from multiple sources like internet, official reports, social media and so on, to detect and trace the evidence of an emerging threat and overlay it with traditional surveillance systems to issue public health warnings and formulate mitigation strategies. The Global Public Health Intelligence Network (GPHIN), and HealthMap, an automated electronic information system, are EBS tools that use NLP, text processing algorithms and Machine Learning (ML). These can increase the speed of identifying signs, filtering information, enhance capacity for consuming information and increase accuracy manifold.
The Program for Monitoring Emerging Diseases (ProMED), Medical Information System (MedISys) and Pattern-based Understanding and Learning System (PULS) are similar EBS tools. These systems are proving absolutely essential in disseminating information at breakneck speed, characterising transmissibility patterns, contagiousness, illnesses and deaths caused by the pathogen, aiding quicker emergency response. Risk modelling uses statistical tools to characterise and identify factors in populations or individuals that enhance their vulnerabilities to contracting a particular disease. Overlaying this data with open source internet data and climate data accounting for presence, distribution and movement of pathogens can help identify correlations that were invisible to us before. For example, in China, the cases of hand, foot and mouth disease in children were best predicted by AI models that utilised data on weekly temperature and precipitation as well as data on disease-related queries from the Chinese Baidu search engine.
Compared to these, India’s Integrated Disease Surveillance Programme (IDSP) uses manual surveillance, where data on some 24 epidemic-prone diseases is collected on a weekly basis at the primary health centre level and reported upwards. Whenever there is a rising trend of illnesses in any area, it is investigated by the Rapid Response Teams (RRT) to diagnose and control the outbreak. This system is slow and doesn’t use any risk modelling for predictions. A study by the Indian Institute of Public Health, Hyderabad in 2016 pointed out that this procedure suffered a time lag from anything between three to 64 days.
The Media Scanning and Verification Cell of the IDSP was established in 2008 for early warning signals through media reports. However, this activity is limited to manually scanning newspapers/electronic media. Some technologically-adept States use Google alerts to automate the process. Needless to say, State media verification cells are unable to exhaustively scan all media sources with the swiftness that’s needed to respond to outbreaks. The Health Ministry is piloting an Integrated Health Information Platform (IHIP) that can enable near real-time data reporting and hopes to apply modelling and GIS tools to enhance the IDSP. There exists immense scope for integrating emerging technologies within the IHIP platform. But at present these discussions are nascent.
Focussing on holistic pandemic management: While most technology applications in pandemic response are limited to surveillance, there is a dire need to identify intervention areas in overall pandemic management. AI can not only help us predict where the disease might be travelling to, but it can also offer insights into how people take up health services during emergencies. Accounting for health-seeking behaviours in designing response strategies can substantially boost effectiveness and success of responses.
Learning from behavioural data, ML models can identify less obvious patterns in human behaviour and disease transmission, which could enable a targetted response. This is called infodemiology, where you can integrate internet data into public health informatics to examine individual health-seeking patterns during emergencies. Google Trends Data as well as more-specific Google Dengue Trends have also been used by researchers to develop a holistic understanding of behavioural aspects of citizen response to pandemics. Such insights can be used to determine where health services are imperative to be delivered in times of shortages that mandate trade-offs. In a more mature system, one would also be able to determine if an individual or a group of people is likely to change its location, go out and seek formal assistance, or if they will adhere to treatment routines — all insights that can help authorities make better decisions.
AI-assisted genomics research is slowly emerging to be a game changer in the rapid development of treatments and vaccines for contemporary infections. Baidu has developed an algorithm that can significantly speed up RNA structure prediction and subsequently unlock the key to the virus. Developing models that can match patterns embedded in the viral genomes to their animal host and vectors that carry the virus can be a breakthrough in narrowing the search for diseases. This can lead to early interventions in controlling disease upsurge or preventing their emergence altogether.
Building blocks for AI in pandemic response: Building a comprehensive and accurate surveillance system requires massive amounts of quality data from different sources. Meteorological data, for instance, needs to be overlaid with vector-movements and population mobility data to accurately identify hotspots for outbreaks. This requires Government departments coming together in mission-mode and pooling their data into one consolidated programme. It also requires AI-enabled technologies to be integrated into existing workflows for pandemic management and creation of new protocols with adequate capacity-building, so its benefits can flow to all levels.
As pandemics are not constrained by national boundaries, a global surveillance system is imperative, something where multilaterals like the WHO and international foundations can play a catalytic role. Nations need to be propelled to make disease information public and if possible move towards interoperable surveillance networks that can communicate with their global counterparts.
Can AI fight the next Coronavirus is a question up for debate. But it can bolster our capacity to respond substantially, especially in disease pre-emption, design of accurate interventions and in some cases with more research investment, even prevention of outbreaks.
Therefore, in a fight against outbreaks — one that the global community is not winning — channelising the power of data through AI can be the perfect weapon. Coronavirus and the swiftness that AI-enabled solutions showed here were necessary proof of the concept to incentivise greater resources and collaborations in AI-assisted outbreak management. Now it is upon governments and multilaterals to lead the way in mobilising AI against the next virus.
(Verma is a lawyer and public policy consultant and Sharma is a software engineer. Both work with the International Innovation Corps, University of Chicago.)
